Intelligent decision support system to increase the operational reliability of the hydrocarbon pipeline transport system of a Mexican oil industry

Abstract

This research is carried out in the Mexican oil and gas industry. An Intelligent Decision Support System (IDSS) is proposed, through support modules for the human operator (fuzzy expert system and artificial neural network) that simulate, forecast and standardize operational decision criteria of a sequential pipeline pumping system, with problems of vandalism, mechanical deterioration in the face of a complex topographic profile, in order to minimize operational subjectivity and prevent contingencies. The research provides new control and monitoring alternatives that guarantee the operational reliability of a pumping station, minimizing the effects of risk by managing the knowledge of the experts involved in the problem, data mining and association of results, which allow to unify criteria decision. The originality of the work focuses on the ability to model, identify and adapt variables to current international parameters, considering previous works through a comprehensive perspective.

Keywords

Hydrocarbon pipeline transport system (PTS)fuzzy expert system (FES)artificial neural network (ANN)intelligent decision support system (IDSS)

Nomenclature A

Diameter

Artificial Intelligence

ANN

Artificial neural network

APSC

Automated pumping station control with SCADA or similar.

ASPS

Auxiliary systems and components of the pumping station.

API

American Petroleum Institute

ARH

Availability of human resources in E-5A

ATB

Availability of the E-5A turbopumps.

AATV

Availability of alarms, tanks, recirculation and safety valves in E-5A.

AFF&PG

Availability of firefighting equipment and auxiliary power generator in E-5A

AMF

Availability of measurement and filtering equipment for E-5A

ACV

Availability of Check valve in E-5A.

ALSV

Availability of line section valves in E-5A.

ADSRT

Availability of Devil Shipping or Receipt Traps in E-5A.

AMIFI

Availability and management of interconnection equipment for flow improver in E-5A.

AOC-SCADA

Availability of open communication and SCADA equipment in PTS.

Bias

COVID-19

Coronavirus disease

COA

Area Centroid method

Contingencies

CTP

Characteristics of the transported product.

Cognitive knowledge.

CPS-TB

Total ideal suction pressure capacity of the turbo pumps in the E-5A.

CPD-TB

Total ideal discharge pressure capacity of the turbo pumps in the E-5A.

CPS-E6

Total suction pressure capacity for the station to receive the product.

Nomenclature b

CPD-E5

Total discharge pressure capacity of the station that sends the product.

instrumentation and characteristics of dynamic equipment (turbo pumps).

DSS

Decision support System

DBT

Daily Barrels Transported

EPP

Exploration, Extraction and Production.

EOP

Established Operational Program

HMD

E-5A Historical Maintenance Data.

FES

Fuzzy expert System

Fuzzy logic

Flow

Flow rate of a continuous fluid.

Finance and Human Resources

Gas and basic petrochemicals

Horse Power

Historical data.

Kilometer

Information Technology

IPS

Instrumentation and characteristics of a pipe segment.

IDSS

Intelligent Decision Support System

IUDS

instrumentation and characteristics of upstream or downstream stations.

IW_ji

Coefficients for the neuron weights j

Logistics

LMT

Linear Modeling Techniques or Mathematical formulation.

Max

Maximum

MPSC

Manual pumping station control (Suction and Discharge).

MATLAB

MATrix LABoratory

MAOP

Maximum Allowable Operating Pressure

MDC

Mendoza distribution center

MIS

Mechanical integrity study

MSE

Mean square error

Mechanical integrity

Min

Minimum

MLP

Multilayer Perceptron

NMLT

Nonlinear or Dynamic modeling techniques.

OPC

Ole for process control

OPEC

Organization of the Petroleum Exporting Countries

P safe

Maximum safe pressure for a pipeline with an indication of metal loss

Psi

Pounds-force per square inch

PTS

Pipeline Transport System

POL

Permissible Operative Limits

PLC

Controllable logic programmers

Refining

RSG

Response Surface Graphs

R²

Numerical Correlation

SAP/R3

Supports Business Management

S-1

Station 1 (KM 0) - Nuevo Teapa, Veracruz, Mexico.

S-2

Station 2 (KM 166) - Loma Bonita, Oaxaca, Mexico.

S-3

Station 3 (KM 277) - Arroyo Moreno, Veracruz, Mexico.

S-4

Station 4 (KM 317) - Zapoapita, Veracruz, Mexico.

S-5A

Station 5A (KM 351) - Ciudad Mendoza, Veracruz, Mexico.

S-5

Station 5 (KM 362) - Maltrata, Veracruz, Mexico.

S-6

Station 6 (KM 488) - San Mart $\overset{´}{ı}$ n Texmelucan, Puebla, Mexico.

S-7

Station 7 (KM 572) - Venta de Carpio, State of Mexico, Mexico.

SCADA

Supervisory Control And Data Acquisition

SLR

Systematic literary review

U.S.

United States

TBD

Thousand barrels per day

UVs

Uncertain variables

X_j

Iinput Variables

Output Values

1 Introduction & literature review

Spot crude oil prices rebounded in November 2020 after a rally in crude oil futures contracts after positive news about COVID-19 vaccines sparked optimism about the recovery in oil demand in the coming months. Nonetheless, weak refining margins, an increase in crude oil supply, also in the USA, have constrained prices [1]. The oil and gas industry in Mexico accounts for about 70% of the energy in the country’s primary sector [2]. However, barrel price constraints, technological demands, and constant vandalism on Pipeline Transportation System (PTS) hamper the company’s ability to respond to new global supply and demand. Hydrocarbon theft, mechanical deterioration, cavitation, and the lack of technological updates all add to the operational instability problems that the PTS currently experiences as the conditions in its dynamic equipment (Turbopumps) have to be constantly readjusted. The pumping stations impact communities and the surrounding ecosystem on various occasions [3 –5].

The technological progress in the world represents the visible face of the systems and forms the means of communication between the actors (humans) of an interactive process (computer) via interfaces. In particular, intelligent interfaces for monitoring and controlling industrial processes have distinct features, many of which are critical and which strongly influence the communication and decisions that the operator has to make in the equipment [6]. Artificial intelligence (AI) techniques, show their capacity for dynamic adaptability in complex issues before known methodologies, demonstrating their efficiency in a particular way or as a whole.

Recent research in the oil and gas industry shows applications of AI to model control parameters in dynamic and auxiliary equipment in sequential pipeline pumping stations [7 –15]. For example, the major advantage of an FL-based control system is its ease of implementation and the lack of dependence on complex or extensive mathematical equations. These models can easily approximate the control behavior of people who function well in such poorly defined environments [16]. On the other hand, the ANNs offer a learning time to achieve the best results at the output [17]. The equivalent relationship between FL and ANN was demonstrated in this study, as well as the importance of using the advantages of both techniques and their comparison [18, 19]. AI collaborative decision support systems generate Intelligent Decision Support System (IDSS), efficient and practical utility in the decision-making process [20], IDSSs are implemented in several research studies to efficiently predict the price of oil, manage hydrocarbon spills, and determine classes in generated thickness maps. on the pollution of oil slicks in the well-known case of Deepwater Horizon [21 –23]. In the Mexican oil and gas industry, IDSS uses intelligent agents to optimize the control and logistics planning of contingent liabilities in the maritime zone of the Gulf of Mexico [24]. In other countries, IDSSs are reliable tools for determining trends in equipment deterioration, failures in specific segments, analyzing their effects, and determining the likelihood of occurrence, as well as planning of human resources in oil companies [25 –27]. Table 1 shows some relevant applications of IDSS in various areas in the oil and gas industry, however the authors highlight the application of control and monitoring in pipelines as it is the most used, efficient and economical means of transport today. Through a Systematic Literary Review (SLR), which deepens the current state of the art with 245 works published between 1994 and 2018, the graph [A] represents the growing trend of the subject of study [28].

Table 1
IDSS in the oil and gas industry. Source: Based on [28]

IDSS in the oil and gas industry Areas analyzed

EEP R GP L C FR

[29] •

[27] •

[25] •

[22] • •

[24] • •

[26] •

[30] •

[31]. •

[23] • •

[32] •

[33] •

[34] •

[21] • •

[35] •

[36] •

[37] • •

[38] •

[39] •

IDSS in the oil and gas industry	Areas analyzed
[29]						•
[27]						•
[25]	•
[22]	•				•
[24]	•				•
[26]				•
[30]	•
[31].	•
[23]	•				•
[32]	•
[33]	•
[34]				•
[21]		•				•
[35]				•
[36]	•
[37]			•	•
[38]				•
[39]				•

Abbreviations: Exploration, Extraction and Production (EEP), Refining (R), Gas and basic petrochemicals (GP), Logistics “Pipelines, ships and tank cars” (L), Contingencies (C), Finance and Human Resources (FR).

IDSS can be designed using AI-based methods such as Fuzzy Logic (FL) and ANN [20]. Both the fuzzy expert systems (FES) with applications for volume prediction and the diagnosis of circulation losses in oil wells show excellent results [30, 33]. The FES as an inference engine of adaptive fault-tolerant control in a gas turbine reduces the operational variability [11]. ANNs provide inductive means of collecting, PTS, and using cognitive knowledge [40]. Integrating both methods for the interpretation, processing, and regulation of sources of cognitive and historical information for the standardization of decision criteria is not an isolated case in the world in which comparison and evaluation are often consistent [29, 41]. The authors establish the following questions :How does subjectivity impact the human operator who establishes decision criteria for a sequential pipeline pumping system, with problems of vandalism, mechanical deterioration in the face of a complex topographic profile? Why is it important to standardize operational decision criteria for the PTS? Do the operational monitoring and control technologies used clearly justify their operation and adaptive work dynamics in the event of any failure? Below the mention the Purpose, Design/methodology / approach, Findings, Practical implications, Originality / Value, which generated the research.

Purpose - Develop an IDSS, through support modules for the human operator that simulate, forecast and standardize operational decision criteria of a sequential pipeline pumping system, with problems of vandalism, mechanical deterioration in the face of a complex topographic profile, in order to minimize operational subjectivity and prevent contingencies.

Design/methodology/approach - The dyna-mic adaptability of Artificial Intelligence (AI), allowed the design of modeling strategies, which can be replicated around the world, unifying knowledge of specialists in the operation, monitoring and control of oil pipelines, capable of identifying the behavior in the suction capacities and discharge from a sequential pumping system. Through a fuzzy expert system and an artificial neural network, the cognitive knowledge of the personnel and the operational behavior of the system are modeled using historical data from 30”Ø pipes, through the dynamic simulation of the equipment and operational resources of a station to obtain reliable parameters. The proposed IDSS allows the reconfiguration and management of new operating practices in oil pipelines, under a comprehensive approach that maximizes the operational reliability of the system and minimizes risk management. It should be noted that the intelligent modules that make up the IDSS generate independent and comparable results with current knowledge, due to the comprehensive perspective shared by their information sources, unified in a man-machine interface that facilitates their interpretation.

Findings - The proposed system starts from the unification of cognitive knowledge and historical data, supported by field tests, control and monitoring, which allows predicting and simulating the operational behavior of the pipeline transport system in the pumping station. To increase the effectiveness and reduce the errors of the FES, it is recommended to frequently review the behavior of the input variables of the system and the inference rules with the experts so that the decision-making of the FES is similar to that of the specialists. With regard to the ANN, it is recommended to update the system database constantly to ensure the representation and consistency of the data and the ANN correctly identifies the behavior pattern of the system and can predict assertively. The authors do not establish a calculation of the net benefit of the research to determine if the failure in the models is acceptable according to the current state of knowledge, but to interpret the presented results with flexibility and variability, because some non-quantifiable criteria may be important. to assess overall risk, such as the cost of protecting life and ecosystems.

Practical implications - This system is built from a case study in the oil and gas industry in Mexico, designed through data mining and the cognitive knowledge of 58 specialists in the areas of operation, mechanical integrity, maintenance, planning, flow improver., safety and instrumentation. The FES is supported by 348 surveys that describe the dynamic behavior of the station in its various areas as well as specific data generated by the expertise of the personnel, considering aspects of the latest internal inspection severity analysis using ultrasonic beam technology inspection equipment. That establish the operational capacity for pipes with metal loss. Where 730 tests determined by a factorial design of experiments, served to evaluate and monitor their level of confidence. The ANN, constitutes historical information of the main dynamic equipment of the station (gas turbo pumps) modeling 4 years of suction and discharge records of the daily operating flow as well as the injection of Flow improver in the system, maintenance and a wide base of control data and administration of operational times in SAP/R3. Where 56 networks were trained using 45,260 operational data, for training, 0.9629 for tests 0.9882 and 0.985 validation. The IDSS integrates the human resources experience and historical data of an oil company. This study will be useful to compare the results between ANN and FES and to unify decision criteria in a complex, deteriorated and vandalized system. The results of the IDSS prioritize maximizing the operational reliability of the PTS, considering a comprehensive approach to operational risk, under various perspectives never previously addressed jointly in the current literature, where in addition to academic interest, the work is expected to have positive implications for the oil and gas industry, encouraging future research.

Originality/Value - The article provides an IDSS, with the aim of generating new control and monitoring alternatives that guarantee the operational reliability of a pumping station, minimizing the risk effects generated by the operational subjectivity of the personnel. The IDSS is capable of managing the knowledge of the experts involved in the problem, identifying the behavior pattern of the variables involved through data mining and associating results, which allow unifying decision criteria. The originality of the work focuses on the ability to model, identify and adapt the variables to current international parameters, considering previous works through a comprehensive perspective. The FES interprets and predicts the behavior of a sequential hydrocarbon pumping station before a complex topographic profile through goodness of fit tests, design of experiments and inferential statistics using a Mandani model, capable of interpreting the dynamics of the behavior of the station equipment and systems, efficiently in the face of systems in force in the literature. The ANN proposes an alternative method that assumes that the measurement error is inherent to the process (noise) and then processes the results to draw valid conclusions from the available data and determine the operational capabilities of the turbopumps, considering previous studies with known variables that determine the characteristics of the equipment. The IDSS is developed considering proven methodologies in the oil sector, which is not a contribution comparable in its entirety with the works presented by the dynamic innovation in its behavior. By demonstrating the originality of smart modules in a global context, IDSS is a genuine contribution to current knowledge.

2 Methodology and results

The methodology underpins its success by considering the eight key issues for the discipline of decision support systems shown in Fig. 1 [20].

Fig. 1

Methodology. Based on: [20, 28].

1- The relevance of DSS research, by connecting the practical part [7 –15] with the theory [28]. 2-DSS research methods and paradigms, with case study research and design sciences [21–27 , 29–39]. 3-The theoretical foundations of DSS research, considering explicit bases in the decision-making judgment through theoretical and practical foundations applied in a case study in Mexico. 4- The role of the information technology (IT) artifact in DSS research, fundamental in all the processes to analyze, interpret and connect the information through a human-machine interface that is easy to understand. 5- Funding for DSS research, no implicit support from any organization. Compliance with keys 6,7 and 8 is generated by considering the largest number of interpretive case studies and the academic rigor to support the research designs [7 –15].

The IDSS comprises two AI techniques with which robust databases and cognitive knowledge can be modeled via a human-machine interface, which guarantees the operational reliability of the PTS. A four-stage structure is proposed for the IDSS:

Stage 1. Operational analysis of the PTS. The operational performance of the PTS is analyzed; the sources of information, the variables to be considered, are determined with the help of experts to model the system correctly.

Stage 2. Capitalization of knowledge with an FES. Modeling the knowledge and operational structure of the dynamic equipment of a hydrocarbon pumping station.

Stage 3. Interpretation of historical data using an ANN. The neural system predicts the operating trend of turbopumps via a database which minimizes the control factors for the personnel.

Stage 4. Interconnection of IDSS modules. Evaluation and integration of smart modules through standardization of communication with OPC & Simulink. The IDSS was developed, considering the applicable international operating procedures. The Matlab toolbox was used for the design and development of the FES and the ANN as well as non-invasive equipment to standardize communication protocols and not to change manufacturer parameters.

2.1 Stage 1: Operational analysis of the PTS

Mexico’s oil and gas industry has developed a strategy to commercialize oil and petrochemical products under international supply and demand. The case study is being developed at the PTS distribution center (DMC) with 62 years of service and based in Mendoza City, Veracruz, Mexico, with an operating program of 140 to 220,000 barrels per day (TBD) [42]. Hydrocarbons are transported through pipes with a diameter of 24 and 30 “Ø, and a length of 572 km. The process is carried out with a sequential hydrocarbon pumping system that allows the flow of the product and is powered by 6 stations with 96 lines isolation valves, 17 check valves, 28 devil traps, and pieces of equipment for remote instrumentation assisted by telemetry and monitoring, data acquisition and control system (SCADA), the functional strategy of a control station pumping the company’s products, comprising measuring, filtering and recovery equipment is shown in Fig. 1. [A] The operating capacity of the DMC is shown in Table 2.

Table 2
POL established by Industry Mexican of gas and oil. Source: Based on [43]

Parameters (Kg/cm²) E-2 E-3 E-4 E-5A E-5 E-6

Maximum operating pressure 53 66 79 55 74 63

High pressure discharge alarm 54 67 80 56 75 64

High pressure discharge trip 54.5 67.5 80.5 63 75.5 64.5

Discharge relief valve trip 55 68 81 63.5 76 65

Low suction pressure alarm 6 6 6 14 5.5 6

Low suction pressure trigger 4 4 4 12 4 4

Number of dynamic teams 6 6 7 4 7 7

Installed power (HP) 30,000 30,000 35,000 20,000 35,000 35,000

Parameters (Kg/cm²)	E-2	E-3	E-4	E-5A	E-5	E-6
Maximum operating pressure	53	66	79	55	74	63
High pressure discharge alarm	54	67	80	56	75	64
High pressure discharge trip	54.5	67.5	80.5	63	75.5	64.5
Discharge relief valve trip	55	68	81	63.5	76	65
Low suction pressure alarm	6	6	6	14	5.5	6
Low suction pressure trigger	4	4	4	12	4	4
Number of dynamic teams	6	6	7	4	7	7
Installed power (HP)	30,000	30,000	35,000	20,000	35,000	35,000

The process starts with the receipt of the hydrocarbon from the pumping station (E-1) in Nuevo Teapa, Veracruz (KM 0), receives and sends sequentially to (E-2) in Loma Bonita, Oaxaca (km 166), to (E- 3) in Arroyo Moreno, Veracruz (km 277), to (E-4) in Zapoapita, Veracruz (km 317), to (E-5A) in Mendoza City, Veracruz (km 351) as a power station due to the topographical profile of the area, continue to (E-5) in Maltrata, Veracruz (km 362), to (E-6) in San Martín Texmelucan, Puebla (km 488) and end the pumping process of the distribution center with the (E-7) in Venta de Carpio, State of Mexico (km 572). At this point, the hydrocarbon is transferred to transformation processes, sold and marketed nationally and internationally. A topographical and hydraulic profile of the DMC for a pumping rate of 150 TBD in a 24 “Ø pipeline, the MAYA crude oil with a constant viscosity of 21 ° API and a specific gravity of 0.88 Km/m³ transported considering the operational limits, critical areas and the location of the main facilities at sea level are shown in Fig. 2 [B]. The DMC requires the stabilization of the operational reliability of the PTS, derived from the variable abnormal operating conditions of flow, suction, and discharge that the PTS has over a length of 90 km in the Veracruz and Puebla areas from the E-5 to E-6 which is a critical point due to the topographical profile and fluence of hydrocarbon theft. On average, there are 1403 clandestine intakes every 5 hours and 13 minutes as well as 799 every 8 hours and 13 minutes [5]. The company analyzes the dynamic adaptability of the operating personnel of the E-5A Ruston 5000 turbopumps Fig. 1[C], due to the complexity of the topographical profile of the area as a power station and, given the current problem, crucial for the creation of an intelligent decision support system (IDSS) by AI considering equipment specifications, current regulations and an approach to operational risk that minimizes operational uncertainties and is reflected in the flow differences between pumping and receiving, valve throttling, hydrocarbon spills, ram effects and cavitation in a mechanically deteriorated and constantly vandalized system which generates results given the pressure variability.

Fig. 2

PTS operation [A], Topographic and hydraulic profile [B] and Ruston TB5000 gas turbine [C]. Source: Based on: [5, 44].

The mathematical modeling of all the systems that make up a sequential pumping station by pipelines in said research, consider the contribution in Table 3 [9], where the centrifugal forces of the turbo pumps are characterized by the dependence of the differential head produced by the pump on the fluid flow ΔH = F(Q), in suction condition (P_in) and discharge (P_ex) in a certain section of the line (P_g) represented in Equation (1), segment [A]. Pumps connected in series or parallel provide the basis of oil-pumping stations intended to produce driving pressure [B]. The (Q – ΔH) characteristics of pumps connected in series aresummarized, the fluid flow rates of the pumps are identical Q₁ = Q₂ = Q and the differential heads are given by ΔH =ΔH₁ + ΔH₂. If ΔH₁ = a₁ – b₁ * Q² is the characteristic of the first pump and ΔH₂ = a₂ - b₂ * Q² the characteristic of the second pump, the characteristic of a system of two pumps connected in series is equal to (2).

Table 3

Operation oil-pumping station. Source: Based on: [9]

[A] Operating principle of a centrifugal pump.

Δ H = \frac{ρ_{ex -} ρ_{in}}{ρ g} = F (Q) r \frac{Po}{Pg} = \frac{Pu}{Pg} + F (v) (1)

[B] Series connection of pumps &Parallel connection of pumps.

Δ H = (a_{1} + a_{2}) - (b_{1} + b_{2}) * Q^{2} (2) \sqrt{\frac{(a_{1} - Δ H)}{b_{1}}} + \sqrt{\frac{(a_{2} - Δ H)}{b_{2}}} = Q (3)

[C] Combined Operation of Linear Pipeline Section and Pumping Station.

{\begin{matrix} (\frac{P_{o}}{P_{g}} + Z_{o}) - (\frac{P_{L}}{P_{g}} + Z_{o}) = λ (Re, ɛ) * \frac{L}{d} * \frac{v^{2}}{2 g} \\ \frac{P_{o}}{P_{g}} = \frac{P_{u}}{P_{g}} + (α - 1.296 * 10^{7} S^{2} b * v^{2} \end{matrix} (4)

\frac{P_{u}}{P_{g}} - \frac{P_{L}}{P_{g}} + (Z_{o} - Z_{L}) + α - 1.296 * 10^{7} S^{2} b * v^{2} = λ (Re, ɛ) * \frac{L}{d} * \frac{v^{2}}{2 g} (5)

\frac{P_{u}}{P_{g}} - \frac{P_{L}}{P_{g}} + (Z_{o} - Z_{L}) + α = (λ * \frac{L}{d} * \frac{1}{2 g} + 1.296 * 10^{7} S^{2} b) * v^{2} (6)

[D] Calculations on the Operation of a Pipeline with Intermediate Oil-Pumping Stations

{\begin{matrix} [Z_{1} + h_{n 1} + F_{1} (Q)] - [Z_{2} + h_{n 2}] = h_{1 - 2} (Q), \\ [Z_{2} + h_{n 2} + F_{2} (Q)] - [Z_{3} + h_{n 3}] = h_{2 - 3} (Q), \\ \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots . . \\ [Z_{n} + h_{n, n} + F_{n} (Q)] - [Z_{L} + h_{L}] = h_{n - (n - 1)} (Q), \end{matrix} (7)

In parallel connection of pumps their (Q - ΔH) characteristics are different. Fluid discharges in pumps are given by Q = Q₁ + Q₂ but the heads produced by each pump are identical ΔH =ΔH₁=ΔH₂. If ΔH = a₁ – b_1 * Q² is the characteristic of the first centrifugal pump and ΔH = a₂ – b_2 * Q² that of the second one, the characteristic of the system of two pumps connected in parallel is equal to (3). To calculate the combined operation of a linear pipeline section and the pumping station located at the beginning of the pipeline section [C] the Bernoulli Equation (4) is used, in which the pressure P_o = _p (0) at the initial cross section of the pipeline section is excluded with the help of boundary condition. Where a and b are the approximation factors of the pumping station (Q - ΔH) characteristic, the velocity v is measured in (m s^- 1). After eliminating P_o from these equations we obtain (5). This equation is called the head balance equation. At given values of the head before the pumping station p _u and pressure at the pipeline section end p_L Equation (6) serves to determine the unknown velocity v of the fluid flow in the pipeline. Consider a pipeline consisting of n successive sections separated by oilpumping stations (D).

The transportation of fluid is performed in the socalled pump-to-pump regime. When intermediate fluid dumping and pumping are absent we can write the Bernoulli equation for each section (7) where ΔH = F ₁ (Q), ΔH = F ₂(Q),..., ΔH = F_n (Q) are the hydraulic (Q - ΔH) characteristics of oil-pumping stations; h_{j - (j - 1)}(Q) the head losses in the sections between the oil-pumping stations dependent on the pumping flow rate Q; Z_1, Z_2, ... ,Z_n the elevations of the oil-pumping stations; h_u,1, h_u,2,_… . .h_u,n the heads before the oil-pumping stations equal to h_u,i = p_u,i/(P_g);Z_L, h_L = p_L/(P_g);the elevation and piezometric head at the pipeline end (x = L), respectively. In determining the head losses in the pipeline sections it is necessary to account for the possibility of existing transfer points and segments of gravity flow in thesesections.

2.2 Stage 2: Capitalization of knowledge with an FES

The architecture of the FES is made up of a system of fuzzy inference and the process of capitalization of the knowledge of the operators of the Oil and Gas Industry, using operational data only from 30 “Ø pipes, for its validation and subsequent application to 24 “Ø pipes [45]. A fuzzy set is a general form of a sharp set. A fuzzy number belongs to the closed intervals 0 and 1, for which one refers to full membership and zero expresses non-membership [46].

The FES comprises deterministic variables (inputs) and uncertain variables (outputs) that model a station’s equipment and operational resources to obtain reliable suction and discharge parameters, as shown in Table 4.

Table 4
Description of the input and output variables

Input variables

CPS E-6 Total suction pressure capacity for the station to receive the product.

CPD E-5 Total discharge pressure capacity of the station that sends the product.

HMD E-5A Historical Maintenance Data.

ARH Availability of human resources in E-5A.

ATB Availability of the E-5A turbopumps.

AATV Availability of alarms, tanks, recirculation and safety valves in E-5A.

AFF&PG Availability of fire fighting equipment and auxiliary power generator in E-5A.

AMF Availability of measurement and filtering equipment for E-5A

ACV Availability of Check valve in E-5A.

ALSV Availability of line section valves in E-5A.

ADSRT Availability of Devil Shipping or Receipt Traps in E-5A.

AMIFI Availability and management of interconnection equipment for flow improver in E-5A.

AOC-SCADA Availability of open communication and SCADA equipment in PTS.

Output variables

CPS-TB Total ideal suction pressure capacity of the turbo pumps in the E-5A.

CPD-TB Total ideal discharge pressure capacity of the turbo pumps in the E-5A.

Input variables
CPS E-6	Total suction pressure capacity for the station to receive the product.
CPD E-5	Total discharge pressure capacity of the station that sends the product.
HMD	E-5A Historical Maintenance Data.
ARH	Availability of human resources in E-5A.
ATB	Availability of the E-5A turbopumps.
AATV	Availability of alarms, tanks, recirculation and safety valves in E-5A.
AFF&PG	Availability of fire fighting equipment and auxiliary power generator in E-5A.
AMF	Availability of measurement and filtering equipment for E-5A
ACV	Availability of Check valve in E-5A.
ALSV	Availability of line section valves in E-5A.
ADSRT	Availability of Devil Shipping or Receipt Traps in E-5A.
AMIFI	Availability and management of interconnection equipment for flow improver in E-5A.
AOC-SCADA	Availability of open communication and SCADA equipment in PTS.
Output variables
CPS-TB	Total ideal suction pressure capacity of the turbo pumps in the E-5A.
CPD-TB	Total ideal discharge pressure capacity of the turbo pumps in the E-5A.

The input variables are determined with the support of the company to establish and structure the ideal operational requirements of the E-5A, which are approved by the Energy Regulatory Commission [47]. The two output variables use the uncertainty value that the turbopumps operator must check promptly to deduce risk factors. This demonstrates the structural and operational reliability of the PTS and guarantees compliance with the company’s daily hydrocarbon transportation program [43].The architecture is based on the retrieval of operating parameters from the PTS (Fig. 3) as well as on the linguistic and mathematical classification of the Mamdani model (Table 5A & B). Membership functions are established from goodness and fit tests, as well as non-parametric tests, to determine if the data observed in a random sample fit with a minimum level of significance (variance) to a given probability distribution. Table 5B, presents in section [A], a representative sample of the behavior of 600 population data in the fit tests between the observed data “y(t)” and the estimated “ỹ(t)” with ranges of slack “σ” between the observed and estimated data: σ= y(t) - ỹ(t) ranging from 2 to 8%. Section [B], presents a plot of normalized outliers, identifies the characteristics of the components that are numerically different from the rest of the data. The graph indicates that there are several outliers above the reference line (1.00), however it shows us how well a large agglomeration of the observed (o) and estimated (Δ) data overlap each other in a correlation sample.

Fig. 3

The architecture of the FES.

Table 5A

FES parameters

Deterministic variables (Inputs)	Linguistic label	Membership function	Interval
CPS-E6 (Kg/cm²)	Out of operation	1-Gama	(0, 0.01, 3,99, 4.00)
	Low	Trapezoidal	(3.99, 4.01, 18.49, 18.50)
	Average	Triangular	(18.49, 37.50, 55.50)
	High	Gama	(55.49, 55.51, 75.50, 85.50)
CPD-E5 (Kg/cm²)	Out of operation	1-Gama	(0, 0.01, 3.99, 4.00)
	Low	Trapezoidal	(3.99, 4.01, 20.10, 20.11)
	Average	Triangular	(20.10, 40.25, 60.38)
	High	Gama	(60.37, 60.39, 80.50, 90.50)
HMD (%)	Unavailable	1-Gama	(0, 0.01, 10.97, 10.98)
	Bad	Trapezoidal	(10.97, 11.00, 44.99, 45.00)
	Regular	Triangular	(44.99, 65.00 80.00)
	Good	Gama	(79.99, 80.01, 99.99, 100.00)
ARH (1:1)	Unavailable	1-Gama	(0,0.01,0.99,1.00)
	Low	Trapezoidal	(0.99, 1.01,10.99,11.00)
	Average	Triangular	(10.99, 22.00, 33.00)
	High	Gama	(32.99, 33.01, 44.99,45.00)
ATB (1:1)	Unavailable	1-Gama	(0,0.01, 0.99, 1.00)
	Low	Trapezoidal	(0.99, 1.01, 1.99, 2.00)
	Average	Triangular	(1.99, 3.00, 4.00)
	High	Gama	(3.99, 4.00, 4.98, 4.99)
AATV (%)	Unavailable	1-Gama	(0, 0.01, 10.99, 11.00)
	Low	Trapezoidal	(10.99, 11.01, 44.99, 45.00)
	Average	Triangular	(44.99, 65.00, 80.00)
	High	Gama	(79.99, 80.01, 99.99, 100.00)
AFF&PG (%)	Unavailable	1-Gama	(0, 0.01, 14.99, 15.00)
	Low	Trapezoidal	(14.99, 15.01, 44.99, 45.00)
	Average	Triangular	(44.99, 65.00, 80.00)
	High	Gama	(79.99, 80.01, 99.99, 100.00)
AMF (%)	Unavailable	1-Gama	(0, 0.01, 19.99, 20.00)
	Low	Trapezoidal	(19.99, 20.01, 44.99, 45.00)
	Average	Triangular	(44.99, 65.00, 80.00)
	High	Gama	(79.99, 80.01, 99.99, 100.00)
ACV (1:1)	Unavailable	1-Gama	(0,0.01, 0.99, 1.00)
	Low	Trapezoidal	(0.99, 1.01, 1.99, 2.00)
	Average	Triangular	(1.99, 3.00, 5.00)
	High	Gama	(4.99, 5.01, 5.99, 6.00)

Table 5B

FES parameters

Deterministic variables (Inputs)	Linguistic label	Membership function	Interval
ALSV (1:1)	Unavailable	1-Gama	(0,0.01, 3.99, 4.00)
	Low	Trapezoidal	(3.99, 4.01, 7.99, 8.00)
	Average	Triangular	(7.99, 12.00, 16.00)
	High	Gama	(15.99, 16.01, 18.99, 19.00)
ADSRT (1:1)	Unavailable	1-Gama	(0,0.01, 0.99, 1.00)
	Low	Trapezoidal	(0.99, 1.01, 1.99, 2.00)
	Average	Triangular	(1.99, 3.00, 4.00)
	High	Gama	(3.99, 4.00, 4.98, 4.99)
AMIFI (Y/N)	Not	1-Gama	(0, 0.01, 0.98, 0.99)
	Yes	Gama	(1.00, 1.01, 1.98, 1.99)
AOC-SCADA (Y/N)	Not	1-Gama	(0, 0.01, 0.98, 0.99)
	Yes	Gama	(1.00, 1.01, 1.98, 1.99)
Uncertain variables (Outputs) CPS-TB (Kg/cm²)	Out of operation	1-Gama	(0, 0.1, 11.99, 12.00)
	Low	Trapezoidal	(11.99, 12.01, 14.11, 14.12)
	Average	Triangular	(14.11, 28.25, 42.36)
	High	Gama	(42.35, 42.36, 56.50, 66.50)
CPD-TB (Kg/cm²)	Out of operation	1-Gama	(0, 0.01, 11.99, 12.00)
	Low	Trapezoidal	(11.99, 12.01, 15.86, 15.87)
	Average	Triangular	(15.86, 31.75, 47.61)
	High	Gama	(47.60, 47.62, 63.50, 73.50)
[A]

[B]

Fuzzification comprises the process of transforming crisp values into degrees of membership for linguistic terms of fuzzy sets. The membership function is used to assign a degree to each linguistic term [34]. The knowledge base comprises the formulation of the fuzzy rules with the support and experience of specialists. The Mamdani method uses the concepts of fuzzy sets and fuzzy logic to translate a completely unstructured set of linguistic heuristics into an algorithm [48]. The general if-then rule form of the Mamdani algorithm [34] is given below:

$\begin{matrix} If x_{1} is A_{i 1} and x_{2} is A_{i 2} and \dots x_{r} is A_{ir} \\ then y is B_{i} (for i = 1, 2, \dots k) \end{matrix}$ (8)

$Y = ξ_{{iW}_{i} B_{i}}$ (9)

Where (X_i) is the input variable, (A_ir) and (B_i) are linguistic terms, (Y) is the output variable, and (k)is the number of rules. Fuzzy rules work in the precedent consequential way. These rules refer to prepositions that contain linguistic variables. A common fuzzy rule relates (m) precedent variables X₁, ... X_m to n consequential variables Y₁, ... Y_n. Finally, the defuzzification process is used to transfer fuzzy sets to a precise value, where the output (Y) is expressed by the sum between the normalized weights (w_i) of the variables and the linguistic terms (B_i), to equation (9). The defuzzification process that uses the Matlab software calculates the center of gravity of images (centroid method) that are generated at the moment of decomposing each of the linguistic input variables [50].

2.2.1 Tests, validation, and results of the FES.

730 tests were conducted to evaluate and monitor the confidence level of FES the results showed a high degree of certainty, indicating that the system meets realistic expectations. In Fig. 4, the correlation rates 0.9501 and 0.9449 for the output variables CPS-TB [A] and CPD-TB [B] are observed at the EOP, respectively. Response surface graphs (RSG) allow one to determine the impact that the input variables can have on the output variable [51], and the objective is to maximize the response [52].

Fig. 4

Correlation coefficients and response surface plots of the FES outputs.

If the combination of these variables activates an optimal zone (light colors), the FES works well. However, if the combinations activate a sub-optimal area (dark colors) it means that the FES is underperforming. Two examples of RSGs developed for the FES are shown in Fig. 4. The output variable CPS-TB is used in the graph [C]. After influencing two input variables of the model (HMD and CPD-TB), a gradual increase is achieved for the variable (CPS-TB). For case [D], the same input variables are used as in case [C], which are affected by the second output variable CPD-TB. As a result, the FES is trained to perform functions that facilitate decision-making for operations personnel, as well as conduct operational response and control assessments to improve personnel skills.

2.3 Stage 3: Interpretation of historical data using an ANN.

ANNs were used in this research because of their many advantages such as non-linearity, addictiveness, and a high degree of robustness [53]. The model estimates the CPS-TB and CPD-TB values of the E-5A using databases from the PTS operating registry. The aim is to work with FES and maximize operational reliability, by modeling historical operational information from the PTS, thereby reducing operational human resource subjectivity when interacting with turbopumps.

The ANN model is a feed-forward with a multilayer perceptron type (MLP) that uses numerical variables from the operating history of the E-5A turbopumps in its inference mechanism, as opposed to the Fes, which use the experience of the personnel. The input variables use the historical 4-year suction and discharge capacity of the E-5A (HCPS & HCPD E-5A), the hydrocarbon flow (Flow), and the injection of Flow Improver into the system (Flow Improver) and analyze detailed operating times and availability, Unavailability, maintenance and accumulation of the station’s 4 turbopumps, which can be used to model various operating scenarios and predict the ideal operating conditions for station suction and discharge (CPS-TB & CPD-TB). The IDSS integrates the experience of human resources and the historical data of an oil company. This study will be useful to compare the results between ANN and FES and to unify decision criteria in a complex, deteriorated, and vandalized system.

2.3.1 Architecture, Tests, validation and results of the ANN

The architecture of the ANN comprises thirty-one neurons in its input layer, sixteen neurons in the hidden layer, and two neurons for the output layer, as shown in Fig. 5 [A]. Index X_j stands for the input variables; IW_ji shows the coefficients for the neuron weights j. The bias is represented by b, which is considered as a skew value, which allows the neuron weights to be adjusted to get a minimal output error, and finally the index Ys is used for the output values. This Y value is obtained by a Sigmoid-type function since the initial value is normalized using IW values [54]. To determine relative contribution (RC) of the 31 input variables of the ANN, Garson’s Algorithm is used, defined in Equation (10) for the training process (Trainlm) applied to a Perceptron supervised learning architecture, and the differences between the known and estimated outputs are analyzed using mean square error (MSE). RC determine in how much each input variable affect the behavior of the discharge and suction capacity outputs of the E-5A. Where (p) is the input which is calculated its relative contribution (CPS-TB or CPD-TB), (n) is the number of hidden neurons, j the (j-nth) hidden neuron, (I) is the number of ANN inputs, (W_jp) is the synaptic weight of input (p) to the input j, (V_j) is the synaptic weight of neuron (j) to the output, and (W_jk) is the synaptic weight of output (k) to theneuron (j).

Fig. 5

Architecture and evaluation of the ANN.

${RC}_{p} = \frac{\frac{ξ_{j = 1}^{n} | w_{jp} | | v_{j} |}{ξ_{k = 1}^{I} | w_{jk} |}}{ξ_{i = 1}^{I} ξ_{j = 1}^{n} \frac{| w_{jp} | | v_{j} |}{ξ_{k = 1}^{I} | w_{jk} |}}$ (10)

Figure 5 [B] shows that the selected supervised learning architecture performed best, given the 18 available from MATLAB software using normalized data, various transfer functions, weight-related connections, and back-propagation training algorithms categorized into heuristics and numerical optimization has the best performance (0.94%) before those evaluated.

56 networks were trained, for training, 0.9629 for tests 0.9886, and 0.9857 for validation (Fig. 6). The coefficients are close to 1, which indicates that the ANN is very efficient and can be used with great certainty for IDSS predictions. During training, 45,260 data received normalization and desnormalization treatment, and the expected value of POL was obtained, and the subjectivity of operation was minimized through the company’s historical data.

Fig. 6

Tests, validation and results of the ANN.

2.4 Stage 4: Interconnection of IDSS modules

The smart module developed in this research works according to POL, ensuring the operational reliability of E-5A, the versatility of FES (for the cognitive knowledge of modeling specialists), the operational requirements of the station, and the ANN for modeling the conditions. The historical operation of a turbopumps is being supplemented to standardize the criteria in IDSS. The results of the modules are compared with 300 data from the POL of the CPS-TB and 300 data from the CPD-TB of the E-5A. Table 6 shows a fraction of 20 tests in which developing both modules produces very similar results when assessing each of the uncertain parameters that affect the output variables and their direct impact on the expected result.

Table 6
POL tests against ANN outputs and FES outputs (Kg/cm²)

Test EOP FES CPS-TB ANN CPS-TB Test EOP FES CPD-TB ANN CPD-TB

1 18.36 20.13 19.20 11 0.73 3.45 0.04

2 39.83 42.77 39.01 12 42.79 43.45 40.83

3 22.73 24.56 21.85 13 44.90 48.12 42.12

4 62.84 63.77 61.90 14 51.83 54.22 53.94

5 24.02 27.55 23.10 15 15.26 16.78 12.05

6 13.80 15.33 13.50 16 0.43 2.37 3.32

7 59.15 62.77 61.30 17 21.53 24.33 25.91

8 51.20 55.22 50.08 18 34.11 36.55 33.80

9 11.42 13.55 10.41 19 50.48 54.66 52.47

10 52.02 54.55 53.96 20 5.98 7.89 5.87

Test	EOP	FES CPS-TB	ANN CPS-TB	Test	EOP	FES CPD-TB	ANN CPD-TB
1	18.36	20.13	19.20	11	0.73	3.45	0.04
2	39.83	42.77	39.01	12	42.79	43.45	40.83
3	22.73	24.56	21.85	13	44.90	48.12	42.12
4	62.84	63.77	61.90	14	51.83	54.22	53.94
5	24.02	27.55	23.10	15	15.26	16.78	12.05
6	13.80	15.33	13.50	16	0.43	2.37	3.32
7	59.15	62.77	61.30	17	21.53	24.33	25.91
8	51.20	55.22	50.08	18	34.11	36.55	33.80
9	11.42	13.55	10.41	19	50.48	54.66	52.47
10	52.02	54.55	53.96	20	5.98	7.89	5.87

The resulting numerical correlation (R2) of 40 data for each of the outputs (FES VS ANN) is shown in Fig. 7. An acceptance level of 0.9692 is obtained when comparing the uncertain variable CPD-TB and 0.9767 for CPS-TB. After performing this process in various tests, it is determined that the values generated by the modules are reliable, allowing their operational use within the IDSS.

Fig. 7

A fraction of the results of the tests of the outputs of the ANN vs. FES (Kg/cm²).

The IDSS man-machine interface is developed in Simulink, Matlab, through remote information from the suction and discharge meters of the turbopumps in SCADA connected to the control room by Siemens 57–1200 controllable logic programmers (PLC) for monitoring and control. Information are taken in real-time using data Access 3.0 with an Ole for process control (OPC), in parallel connection sequences to standardize the interconnectivity between servers, SAP / R3 historical information bases, and unchanged parameters of the cognitive knowledge factory. The interconnection model between FES and ANN shows the results of uncertain variables and deterministic variables that operators must attend to through interfaces. For this purpose, operational manuals and training campaigns have been developed. The diagram in Fig. 8 shows the interconnection sequence of IDSS.

Fig. 8

Interconnection of IDSS modules.

3 Discussion

The research specifically contributes to the current state of the art by the ability to model, identify and adapt its variables to current international parameters, considering previous works through a comprehensive perspective, capable of standardizing decision criteria and generating new control and monitoring alternatives in a sequential pipeline pumping station. Table 7 presents the parameters considered in the research for the modeling and improved control of pumping station operations, based on the capabilities offered by dynamic modeling through fuzzy logic to model the suction and discharge capabilities of the pumps. turbo pumps [8] considering its equipment and instrumentation [7, 44] in manual or assisted operating procedures [13, 15] through the use of cognitive knowledge [14, 55] and modeling the capabilities of all auxiliary systems to maximize station reliability [9, 12]. Determining operating conditions through mechanical integrity studies of the pipe [8, 56] product characteristics and capacities of upstream and downstream stations [13, 14]. However, the authors agree that the generation of a system capable of interpreting the dynamic behavior of a team backed by historical data, will maximize the reliability of the IDSS, highlighting the learning capabilities that an ANN offers.

Table 7
FES parameters

Models for the control of oil, derivatives and gas pumping stations. Parameters considered

LMT NLMT HD CK MPSC APSC DE ASPS IPS IUDS CTP Other features

[7] • • • • The influence of the type (thermodynamic cycle) of the gas turbine to improve its performance.

[8] • • • • • • • Modeling the friction factor in pipes using an artificial neural network.

[9] • • • • • • • • Mathematical modeling processes for the transportation of oil, gas and derivatives in pipelines.

[55] • • • • • • Nonlinear Model Predictive Control formulation for optimizing the operational costs of gas pipeline networks using a steady-state real-time optimizer.

[11] • • • • • • • • Fault tolerant control of a Based on industrial gas turbines in a Fuzzy Adaptive hybrid Unscented Kalman filter to maximize reliability.

[12] • • • • • • • • Modeling and simulation of a pipeline transportation process using Reynolds-AveragedNavier-Stokes (RANS) equations and tram function.

[13] • • • • • • • • • This Predictive Control Model involves four pumping stations separated by three pipe sections, based on physical and thermodynamic laws.

[14] • • • • • • • • Multidimensional logic control of a pumping station using fuzzy logic.

[15] • • • • • The classical control system presents the discrete Laplace transform for the function considered and an autoregressive linear model to establish a proportional, integral and differential control method of an oil and gas pumping system under conditions of uncertainty and lack of information.

Proposed IDSS • • • • • • • • • • Comprehensive perspective, capable of standardizing decision criteria.

Models for the control of oil, derivatives and gas pumping stations.	Parameters considered
[7]	•		•		•		•					The influence of the type (thermodynamic cycle) of the gas turbine to improve its performance.
[8]		•	•		•		•	•	•		•	Modeling the friction factor in pipes using an artificial neural network.
[9]	•		•		•		•	•	•	•	•	Mathematical modeling processes for the transportation of oil, gas and derivatives in pipelines.
[55]		•	•	•	•	•					•	Nonlinear Model Predictive Control formulation for optimizing the operational costs of gas pipeline networks using a steady-state real-time optimizer.
[11]	•	•	•	•	•	•	•	•				Fault tolerant control of a Based on industrial gas turbines in a Fuzzy Adaptive hybrid Unscented Kalman filter to maximize reliability.
[12]	•	•	•	•	•		•	•			•	Modeling and simulation of a pipeline transportation process using Reynolds-AveragedNavier-Stokes (RANS) equations and tram function.
[13]		•	•	•	•	•	•	•		•	•	This Predictive Control Model involves four pumping stations separated by three pipe sections, based on physical and thermodynamic laws.
[14]		•	•	•	•	•	•	•		•		Multidimensional logic control of a pumping station using fuzzy logic.
[15]	•		•		•		•				•	The classical control system presents the discrete Laplace transform for the function considered and an autoregressive linear model to establish a proportional, integral and differential control method of an oil and gas pumping system under conditions of uncertainty and lack of information.
Proposed IDSS		•	•	•	•	•	•	•	•	•	•	Comprehensive perspective, capable of standardizing decision criteria.

Abreviaturas: LMT - Linear Modeling Techniques or Mathematical formulation. NLMT - Nonlinear or Dynamic modeling techniques. HD - Historical data. CK - Cognitive knowledge. MPSC - Manual pumping station control (Suction and Discharge). APSC - Automated pumping station control with SCADA or similar. DE - Instrumentation and characteristics of dynamic equipment (turbo pumps). ASPS - Auxiliary systems and components of the pumping station. IPS - instrumentation and characteristics of a pipe segment. IUDS - instrumentation and characteristics of upstream or downstream stations. CTP - Characteristics of the transported product.

A quantitative comparison of the proposed IDSS, given the papers presented would be wrong, because there are papers that do not declare success rates and model conventional operating pressures without considering auxiliary systems [8, 57]. The FES shows rates of 0.9501 for the suction capacity and 0.9442 for the discharge capacity of the turbopumps considering auxiliary systems. The ANN models the performance of the system using percentages of 0.9629 for training, 0.9886 tests and 0.9857 for validation, through data mining. However, an adaptive tolerant control of the neurodiffuse type shows its versatility when modeling the operating capacities of a turbopump with indicators from 67.3 to 82.9% [11], which does not generate a direct or indirect comparison, with the present work, on the contrary, it represents the flexibility of current methods to solve a problem. The IDSS was developed to increase operational reliability between the suction and discharge capacities of pumping stations and to minimize operational subjectivity, given the vandalism of the region [5]. However, it does not rule out the application of linear systems directly to optimize their function and consider the possibility of implementing other methodologies such as fault-tolerant control to improve the control of the proposed system under fault conditions [9 , 15]. The IDSS can be designed using techniques such as FL and ANN [30 , 35], However, there are success cases that have been developed using genetic algorithms, Bayesian networks, and Monte Carlo simulations to make their theories reliable to maturity [58 –60]. Integrating AI in dynamic equipment such as gas turbines represents intelligent automation processes that could one day bring the machine into full autonomy [11 , 62].

An IDSS does not replace the work of human resources but generates support that minimizes subjectivity in operational processes in contingency scenarios. The company currently uses offline databases and uses IDSS as a strategy to train and develop new skills with its operators. The representativeness and coherence of the data make it impossible to compare the IDSS directly with another current one [7 –15], However, the oil sector is an exceptional field for testing new technologies.

4 Conclusions

With the implementation of IDSS, improved indicators used to measure the hydrocarbon distribution process have been obtained, and the decision-making criteria for controlling and monitoring dynamic equipment in the station have been standardized, giving personnel new operating skills, thereby reducing operative subjectivity by 82% according to 39 evaluations of personnel trained with the IDSS in 6 months of use. More research is needed to expand the database and modeling capabilities to optimize IDSS. Due to space limitations, some graphs and tables of the study were not given. In this case, no calculation method of the net income of the study was established to determine whether the failure of the model is acceptable (based on current knowledge), but to interpret with variable flexibilities and results, because certain non-quantifiable criteria may be important for assessing overall risks (such as the cost of protecting lives and the naturalenvironment).

Footnotes

Acknowledgments

The research was developed with the support of the National Technological Institute of Mexico through the project 14222.22P “Development of a monitoring system tolerant to behavior patterns that generate corrosion in a pipeline transport system” and by the scholarship for doctoral studies Conacyt (CVU-660805).

Disclosure statement

No potential conflict of interest was reported by the authors.

References

OPEC, 2020, OPEC MONTHLY OIL MAR KET REPORT 14 December 2020, Vienna, Austria.

Pemex, 2016, Anuario Estadístico 2016, Cd. de México, México.

Infobae, 2018, Un Derrame Petrolero En México Provocó LaMuerte de Especies y Graves Daños Ambientales, 8/10/2018, pp. 2.

El sol de Puebla, 2019, In Environmental Emergency, 54 Sites in Puebla Due to Oil Spills, June 30, 2019.

Petroleos Méxicanos, 2020, Robo de Hidrocarburos En México Enero - Septiembre Del 2019.

Kaklauskas

, Intelligent Decision Support Systems, Intell Syst Ref Libr 2015.

Urban

L.A.

, Parameter Selection for Multiple Fault Diagnostics of Gas Turbine Engines, J Eng Gas Turbines Power 1975.

Neuroth

, MacConnell

, Stronach

and Vamplew

, Improved Modelling and Control of Oil and Gas Transport Facility Operations Using Artificial Intelligence, Knowledge-Based Syst 13(2) (2000), pp. 81–92.

Lurie

M.V.

, Modeling of Oil Product and Gas Pipeline Transportation, Wiley-VCH, 2009.

10.

Gopalakrishnan

and Biegler

L.T.

, Economic Nonlinear Model Predictive Control for Periodic Optimal Operation of Gas Pipeline Networks, Comput Chem Eng 52 (2013), 90–99.

11.

Mirzaee

and Salahshoor

, Fault Tolerant Control of an Industrial Gas Turbine Based on a Hybrid Fuzzy Adaptive Unscented Kalman Filter, J Eng Gas Turbines Power (12) (2013), 135.

12.

Cabas

J.J.

and Ariza

J.D.R.

, Modeling and Simulation of a Pipeline Transportation Process, J Eng Appl Sci 13(9) (2018), 2614–2621.

13.

Cardenas-Cabrera

, Diaz-Charris

, Torres-Carvajal

, Castro-Charris

, Romero-Fandiño

, Ruiz Ariza

J.D.

and Jiménez-Cabas

, Model Predictive Control Strategies Performance Evaluation over a Pipeline Transportation System, J Control Sci Eng, 2019(4538632) (2019), 1–11.

14.

Sagdatullin

, Development of an Intelligent Control System Based on a Fuzzy Logic Controller for Multidimensional Control of a Pumping Station, Advances in Intelligent Systems and Computing 1127 AISC (2020), 76–85.

15.

Sagdatullin

, Analysis of a Discrete Object Control of Oil and Gas Pumping System under Uncertainty and Unfull Information Conditions, Proceedings - 2020 International Russian Automation Conference, RusAutoCon 2020, Institute of Electrical and Electronics Engineers Inc., (2020), pp. 405–409.

16.

Kim

, An Implementation of Fuzzy Logic Controller on the Reconfigurable FPGA System, IEEE Trans Ind Electron 47(3) (2000), 703–715.

17.

Zhang

, Patuwo

and Hu

M.Y.

, Forecasting with Artificial Neural Networks, Int J Forecast 14 (1998), 35–62.

18.

Takagi

, Cooperative Systems of Neural Networks and Fuzzy Logic and Its Applications, IEEE (510) (1992), 771–775.

19.

Schipor

, Geman

, Chiuchisan

and Covasa

, From Fuzzy Expert System to Artificial Neural Network: Application to Assisted Speech Therapy, Artif Neural Networks - Model Appl 1(8) (2016), 165–193.

20.

Arnott

and Pervan

, Eight Key Issues for the Decision Support Systems Discipline, Decis Support Syst 44(3) (2008), 657–672.

21.

Chiroma

, Zavareh

A.A.

, Baba

M.S.

, Abubakar

A.I.

, Gital

A.Y.

and Zambuk

F.U.

, Intelligent Decision Support Systems for Oil Price Forecasting, Int J Inf Sci Manag 13 (2015), 47–59.

22.

Keramitsoglou

, Cartalis

and Kassomenos

, Decision Support System for Managing Oil Spill Events, Environ Manage 32(2) (2003), 290–298.

23.

Leifer

, Lehr

W.J.

, Simecek-Beatty

, Bradley

, Clark

, Dennison

, Hu

, Matheson

, Jones

C.E.

, Holt

, Reif

, Roberts

D.A.

, Svejkovsky

, Swayze

and Wozencraft

, State of the Art Satellite and Airborne Marine Oil Spill Remote Sensing: Application to the BP Deepwater Horizon Oil Spill, Remote Sens Environ 124 (2012), 185–209.

24.

Sheremetov

L.B.

, Contreras

and Valencia

, Intelligent Multi-Agent Support for the Contingency Management System, Expert Systems with Applications 26(1) (2004), 57–71.

25.

Yam

R.C.M.

, Tse

P.W.

, Li

and Tu

, Intelligent Predictive Decision Support System for Condition-Based Maintenance, Int J Adv Manuf Technol 17(5) (2001), 383–391.

26.

Dey

P.K.

, Decision Support System for Inspection and Maintenance: A Case Study of Oil Pipelines, IEEE Trans Eng Manag 51(1) (2004), 47–56.

27.

Mohanty

R.P.

and Deshmukh

S.G.

, Evolution of a Decision Support System for Human Resource Planning in a Petroleum Company, Int J Prod Econ 51(3) (1997), 251–261.

28.

Cid-Galiot

J.J.

, Aguilar-Lasserre

A.A.

, Rodriguez-Jarquin

J.P.

, Villanueva-Vasquez

, Posada-Gomez

and Juarez-Martinez

, Fault-Tolerance Techniques and Global Positioning Systems Based on Artificial Intelligence for the Analysis of Corrosion Impact on the Pipelines of Oil and Gas Industry, J Press Vessel Technol Trans ASME 143(2) (2021), 1–15.

29.

and Love

P.E.D.

, Combining Rule-Based Expert Systems and Artificial Neural Networks for Mark-up Estimation, Constr Manag Econ 17(2) (1997), 169–176.

30.

Sheremetov

, Batyrshin

, Filatov

, Martinez

and Rodriguez

, Fuzzy Expert System for Solving Lost Circulation Problem, Appl Soft Comput J 8(1) (2008), 14–29.

31.

Almeida

L.F.

, Vellasco

M.M.B.R.

and Pacheco

M.A.C.

, Optimization System for Valve Control in Intelligent Wells under Uncertainties, J Pet Sci Eng 2010.

32.

Mirzaei-Paiaman

and Salavati

, The Application of Artificial Neural Networks for the Prediction of Oil Production Flow Rate, Energy Sources, Part A Recover, Util Environ Eff 34(19) (2012), 1834–1843.

33.

Ghallab

S.A.

, Badr

, Hashem

, Salem

A.B.

and Tolba

M.F.

, A Fuzzy Expert System for Petroleum Prediction, Recent Adv Comput Sci Appl (2013), pp. 70–75.

34.

Jamshidi

, Yazdani-Chamzini

, Yakhchali

S.H.

and Khaleghi

, Developing a New Fuzzy Inference System for Pipeline Risk Assessment, J Loss Prev Process Ind 26(1) (2013), 197–208.

35.

Bal Beşikçi

, Arslan

, Turan

and Ölçer

A.I.

, An Artificial Neural Network Based Decision Support System for Energy Efficient Ship Operations, Comput Oper Res 66 (2016), pp. 393–401.

36.

Seletkov

, Application of matrix approach of fuzzy logic for decision support in oil mining equipment service, Appl Math Control Sci 2020.

37.

Raeihagh

, Behbahaninia

and Macki Aleagha

, Risk Assessment of Sour Gas Inter-Phase Onshore Pipeline Using ANN and Fuzzy Inference System – Case Study: The South Pars Gas Field, J Loss Prev Process Ind 68 (2020), 104238.

38.

Zemenkova

M.Y.

, Gladenko

A.A.

and Zemenkov

Y.D.

, InnovativeIntelligent Technologies for Predictive Reliability and RiskManagement in Oil and Gas Transport and Storage Systems, AIPConference Proceedings. 2020.

39.

Zemenkova

M.Y.

, Zemenkov

Y.D.

, Golik

V.V.

and Belsky

, Technology for Developing Expert Modules and DSS Comprehensive Platform for Reliability and Risks Monitoring in the Oil and Gas Engeneering, IOP Conference Series: Materials Science and Engineering. 2020.

40.

Schocken

and Ariav

, Neural Networks for Decision Support:. Problems and Opportunities, Decis Support Syst 11(5) (1994), 393–414.

41.

Tasdemir

, Saritas

, Ciniviz

and Allahverdi

, Artificial Neural Network and Fuzzy Expert System Comparison for Prediction of Performance and Emission Parameters on a Gasoline Engine, Expert Syst Appl 38(11) (2011), 13912–13923.

42.

Pemex Refinación, 2011, FILOSOFíA OPERATIVA DELOLEODUCTO NUEVO TEAPA - VENTA DE CARPIO - TULA - SALAMANCA 24 30, México City.

43.

Pemex, Procedimiento Para La Operación de La Estación de Rebombeo No. 5 ‘a’ Mendoza Oleoducto 24 – 30 Ø Nuevo Teapa – Venta de Carpio – Tula - Salamanca 583-51540-PEO-04.Pdf, pp. 51.

44.

Turbines

R.G.

, 1961, Manual Tb Ruston 5000.Pdf, pp. 203.

45.

Siler

, Buckley

J.J.

, Fuzzy Expert Systems and Fuzzy Reasoning, John Wiley & Sons, Inc., Hoboken, New Jersey. 2005.

46.

Yazdani-Chamzini

and Yakhchali

S.H.

, Tunnel Boring Machine (TBM) Selection Using Fuzzy Multicriteria Decision Making Methods, Tunn Undergr Sp Technol 30 (2012), pp. 194–204.

47.

Pemex, 2017, Normas de Referencia Vigentes [Online]. Available: http://www.pemex.com/procura/procedimientos-de-contratacion/normas-referencia/Paginas/normas-ref-vigentes.aspx. [Accessed: 03-Dec-2017].

48.

Mamdani

E.H.

and Assilian

, An Experiment in Linguistic Synthesis with a Fuzzy Logic Controller, Int J Man Mach Stud 7(1) (1975), 1–13.

49.

Zadeh

L.A.

, Soft Computing and Fuzzy Logic, IEEE Softw 37(3) (1994), 77–84.

50.

Domínguez Hernández

K.R.

, Aguilar Lasserre

A.A.

, Posada Gómez

, Palet Guzmán

J.A.

and González Sánchez

B.E.

, Development of an Expert System as a Diagnostic Support of Cervical Cancer in Atypical Glandular Cells, Based on Fuzzy Logics and Image Interpretation, Comput Math Methods Med 2013 (2013), 17.

51.

Meza-Palacios

, Aguilar-Lasserre

A.A.

, Ureña-Bogarín

E.L.

, Vázquez-Rodríguez

C.F.

, Posada-Gómez

and Trujillo-Mata

, Development of a Fuzzy ExpertSystem for the Nephropathy Control Assessment in Patients with Type2 Diabetes Mellitus, Expert Syst Appl 72(1) (2017), 335–343.

52.

Montgomery

D.C.

, 2012, Design and Analysis of Experiments.

53.

Pavlekovic

, Zekic-Susac

and Djurdjevic

, Integration of Expert Systems and Neural Networks in Recognizing Mathematically Gifted Children, Proc Int Conf Inf Technol Interfaces, ITI (2008), pp. 557–562.

54.

Rico-Contreras

J.O.

, Aguilar-Lasserre

A.A.

, Méndez-Contreras

J.M.

, López-Andrés

J.J.

and Cid-Chama

,Moisture Content Prediction in Poultry Litter Using Artificial Intelligence Techniques and Monte Carlo Simulation to Determine the Economic Yield from Energy Use, J Environ Manage 202(2017), 254–267.

55.

Gopalakrishnan

and Biegler

L.T.

, Economic Nonlinear Model Predictive Control for Periodic Optimal Operation of Gas Pipeline Networks, Comput Chem Eng 52 (2013), 90–99.

56.

Weatherford, 2011, Reporte Final Sección Cd. Mendoza – Maltrata Oleoducto de 30 Pulgadas..

57.

Liu

, Application Research on Intelligent Management Technology Based on Oil and Gas Pipeline, Journal of Physics: Conference Series. 2020.

58.

Agwu

O.E.

, Akpabio

J.U.

, Alabi

S.B.

and Dosunmu

, Artificial Intelligence Techniques and Their Applications in Drilling Fluid Engineering: A Review, J Pet Sci Eng 167 (2018), 300–315.

59.

Mohaghegh

, Virtual-Intelligence Applications in Petroleum Engineering: Part 1—Artificial Neural Networks, J Pet Technol 52(9) (2000), 64–73.

60.

Mohaghegh

S.D.

, Al-Fattah

S.M.

, Popa

A.S.

, Artificial Intelligence & Data Mining Applications in the E&P Industry, Society of Petroleum Engineers, Texas, United States. 2011.

61.

Bartolini

C.M.

, Caresana

, Comodi

, Pelagalli

, Renzi

and Vagni

, Application of Artificial Neural Networks to Micro Gas Turbines, Energy Convers Manag 52(1) (2011), 781–788.

62.

Bettocchi

, Pinelli

, Spina

P.R.

and Venturini

, Artificial Intelligence for the Diagnostics of Gas Turbines - Part II: Neuro-Fuzzy Approach, J Eng Gas Turbines Power 129(3) (2007), 711–719.

Models for the control of oil, derivatives and gas pumping stations.	Parameters considered
	LMT	NLMT	HD	CK	MPSC	APSC	DE	ASPS	IPS	IUDS	CTP	Other features
[7]	•		•		•		•					The influence of the type (thermodynamic cycle) of the gas turbine to improve its performance.
[8]		•	•		•		•	•	•		•	Modeling the friction factor in pipes using an artificial neural network.
[9]	•		•		•		•	•	•	•	•	Mathematical modeling processes for the transportation of oil, gas and derivatives in pipelines.
[55]		•	•	•	•	•					•	Nonlinear Model Predictive Control formulation for optimizing the operational costs of gas pipeline networks using a steady-state real-time optimizer.
[11]	•	•	•	•	•	•	•	•				Fault tolerant control of a Based on industrial gas turbines in a Fuzzy Adaptive hybrid Unscented Kalman filter to maximize reliability.
[12]	•	•	•	•	•		•	•			•	Modeling and simulation of a pipeline transportation process using Reynolds-AveragedNavier-Stokes (RANS) equations and tram function.
[13]		•	•	•	•	•	•	•		•	•	This Predictive Control Model involves four pumping stations separated by three pipe sections, based on physical and thermodynamic laws.
[14]		•	•	•	•	•	•	•		•		Multidimensional logic control of a pumping station using fuzzy logic.
[15]	•		•		•		•				•	The classical control system presents the discrete Laplace transform for the function considered and an autoregressive linear model to establish a proportional, integral and differential control method of an oil and gas pumping system under conditions of uncertainty and lack of information.
Proposed IDSS		•	•	•	•	•	•	•	•	•	•	Comprehensive perspective, capable of standardizing decision criteria.