Using hybrid-CBR for intelligence monitoring and decision-making systems on SMART grid

Abstract

Smart grid systems are being actively developed and implemented all over the world. However, along with developed systems for monitoring and data analysis, decision support functions are not fully implemented. Wherein decision support is necessary due to the complexity of possible emergencies. In this work, we offer the concept of an intelligent decision support system (IDSS) for the SMART grid, which is based on the hybrid Case-Based Reasoning (CBR) method. This method combines models of knowledge-based systems and models of neural networks and machine learning, which simplifies realization on complex changing objects of the SMART grid. In the first part of the research, we describe the concept of the proposed hybrid-CBR method, the principle of formalizing the situation at the objects of the SMART grid systems and present the involved neural network architecture Comparator-Adder. The second parts of the research reveal the concept of applied IDSS and also show the results of an experiment of retrieving precedent from a knowledge base with using a neural network. Experimental results show that our architecture successfully copes with the task of selecting the most similar situation. We believe that the MAPE error in this incident does not play a key role; the efficiency of the neural network is confirmed primarily by the coherence with the results of the expert choice and the absence of collisions.

Keywords

Case-based reasoning decision making decision support system neural network neural network architecture SMART grid perceptron comparator adder control of complex systems

1. Introduction

Smart Grid, by definition, is real-time data monitoring and active control of microgrids through introduction of technological decision in the power supply infrastructure. Integration exists between microgrids and inside the electricity company, renewable energy generating system, consumer load devices and third-party consumers, suppliers or regulatory organization. Smart Grid includes an intellectual monitoring system that monitors over electricity flow and includes the use of cables or lines to manage power fluctuations, losses and the integration of cogeneration from solar energy, fuel element and wind [1].

SMART grids solve a key problem in the service and exploitation of the energy industry:

1.
Emergency situation emerged because of breakdowns. The developed monitoring system allows the transmission of signals about malfunctions to the control center, where it will be shown which part is damaged on the facility.
2.
The complication of the control of energy complex. Most of the energy facilities are located in hard-to-reach places. Getting to them to conduct technical monitoring is difficult, and sometimes impossible.
3.
Losses due to irrational consumption of fuel and electricity. Due to the fact that energy facilities are difficult to control by employees, more resources are spent than necessary. Power station generates energy with excess and loses money for this reason.

Deployment of the SMART grid system includes the integration of SCADA systems with data transmission systems [1], the introduction of a digital counterpart [2], which will reflect the full cycle of the system and show timing of equipment deterioration depending on the operating conditions. The basis for the functioning of the SMART grid is a series of tools for monitoring and analyzing collected data. Among the main:

•
remote monitoring with using drones
•
remote monitoring through IoT systems
•
artificial intelligence based on predictive analytics with using machine learning.

The main trend of modern SMART grid systems is the collection and analysis of primary data. Their completeness and accuracy allow making a conclusion about condition of the monitoring object, about actual demand for electricity. Along with development of digitalization, monitoring system, collecting and analyzing data, decision support is insufficiently developed.

There are works concerning to methodological decision support for planning service of system elements in Electric Distribution Networks [3, 4], suggestions about use methods of intellectual analysis data and (or) neural networks for forecasting demand [5, 6], forecasting malfunctions of individual elements [7]. Wherein laboriousness of identifying and formalizing knowledge restricts use of these methods to relatively simple objects and situations.

At the same time, dangerous emergency situations at SMART grid objects are usually unique and rare incidents that are not limited to separate elements. They can be fatal consequences for a person, nature and infrastructure. The application of the above methods with a high probability will be ineffective in connection with absence of sufficient experience and training data.

Adoption of integrated operational decision, especially in emergency situations is a laborious and responsible task, which complicated by the deficit of time and a lot of participants. There are high risks of a wrong decision or no decision, because complex support is required

The diversity of environment, the dynamism of condition leads to the fact that a significant capacity of data must be analyzed, systematized and further used to support decision-making. As a result, the tasks of creating knowledge bases and tools for applied intelligent decision support systems (IDSS) have a high laboriousness and science intensity, which slows down their creation and introduction in real conditions.

We suggest that creating an applied IDSS for Smart grid systems requires a hybrid approach combining knowledge-based systems models and neural network and machine learning models. We consider the Case-based reasoning method (CBR) as a basis for this combination. The proposed hybrid CBR allows eliminating the limitations restrictions of separate using of knowledge-based systems (require laborious work to identify and formalize knowledge) and machine learning methods (require a large capacity of teaching examples).

The purpose of this work is to develop the IDSS concept for SMART-grid systems based on the hybrid-CBR method.

The IDSS concept for SMART grid systems based on our previous works:

•
situation define on a complex object through aggregate conditions of its elements and connections between them [8, 9];
•
research of neural network architecture to evaluate situations similarity on a complex technological object in urban infrastructure [10].

The article is organized as follows. Section 2 reveals the essence of the proposed hybrid-CBR method and provides a brief overview of the research on hybrid CBR models. Further, the concept of presentation and formalization of the situation on the SMART-grid object is introduced. Then the architecture of the Comparator-Adder neural network is described, which is used to select a use case from the knowledge base. Section 3 describes the concept of the proposed IDSS for SMART-grid systems and provides an operating algorithm. The following are the results of a model experiment of retrieving the most similar precedent from the knowledge base using a neural network, and some conclusions are given. Section 4 discusses the results obtained and formulates conclusions.
2. Methods and models

2.1 The hybrid-CBR method

The case-based reasoning method is widely used in different subject areas. One of the prospective directions is decision-making when managing complex technological objects [11, 12]. CBR foresees accessing a database and retrieving a precedent meaning decision of a previously occurred situation which purpose to use for the current problem. At the same time, the fixed solving problem way may be adapted for the current situation.

The procedure for evaluating the similarity of situations with another has high importance in CBR for retrieving the most similar situation in heterogeneous complex object cases. In these cases arises need to take into account the states of heterogeneous elements and the connections between them which describe by quantitative and categorical parameters.

Methods for learning similarity measures have been a topic of research in the CBR community for many years [13, 14], including the possibility of using neural networks to assess the similarity of situations [15, 16, 17].

An attempt to compare situations in the multidimensional attributive space of complex objects faces some problems [10] associated with time-consuming tasks that require expert intervention. At the same time, the more complex the object, the higher the labor intensity, and the higher the probability of error and collisions.

Modern research in the field of CBR in applied problems is considering the possibilities of improving known models and precedent inference algorithms by creating hybrid models using ontological engineering methods [18, 19], reinforcement machine learning [20], and neural networks [21].

Our work develops research of hybrid CBR models. For the implementation of the IDSS applied system, we propose a hybrid CBR model that uses neural networks, machine learning, and ontological modeling methods Artificial intelligence will speed up the decision-making process, which is especially important in critical situations, eliminate the human factor, and reduce the labor intensity of processes. Ontological modeling methods allow unifying applied IDSS for objects of SMART-grid systems.

2.2 Representation and formalization of the situation at the SMART-grid object

Representation and formalization of the situation are necessary directly for the implementation of the CBR method. Uniform formalization makes it possible to compare situations with each other for a further selection of a precedent.

Energy supply infrastructure objects related to SMART-grid systems are complex, heterogeneous, and dynamically changing monitoring objects with a variety of connections and states of the internal and external environment. According to the methods of situational management, such a monitoring object belongs to complex technological objects.

Earlier in [10] we introduced the Definition: The situation at a complex technological object is a set of those states in which the elements are at a given time.

Wherein elements of a complex object are formed into groups

•
technological;
•
providing;
•
personnel;
•
environment.

The elements of the “personnel” and “environment” groups are not directly related to the object but are considered part of it since they can influence it. For example, snowfall can make it difficult for personnel to access the facility and affect the composition of the solution to a problem situation.

In order to take into account the peculiarities of a complex technological object, we [9] introduce a formal representation of a complex object O through its elements and relations between them:

$\displaystyle O=\{O_{i}|i=1,2,\ldots,N\},$ (1)

where $O_{i}$ with $i\in$ I ${}_{1}$ denotes complex object elements, and $O_{i}$ with $i\in I_{2}$ denotes relations between complex object elements; $I_{1}\cap I_{2}=\emptyset$ denotes sets of indices of elements and relationships between them; N denotes the number of considered elements and connections in a complex object.

Let each of $O_{i}\in$ O corresponds to its set of possible states $S_{i}=\{S_{ij}|j=1,2,\ldots,M_{j}\}$ and at any time moment, the elements $O_{i}$ can be in one of these states.

The situation can be formally represented through the matrix of states, where one in the column corresponds to the state of the element. Thus, it is possible to formally show the situation on the SMART-grid object using the matrix presented in Table 1. This matrix displays the normal situation on the SMART grid object – step-down substation.

Table 1
State matrix that reflects the current situation at the step-down substation

Sit ${}_{\textit{act}}$ Correctly Broken In work Stopped Not available Available Influences Doesn’t affect

Reserve power line 1

Power transformer 1

Voltage and amperage transformer 1

Transformer own needs 1

Relay protection and automatics devises 1

Communication server 1

Telemechanics devises 1

Sectional switcher 1

Person on duty 1

Firefighting system 1

Buildings 1

Natural objects 1

Natural phenomena 1

Rooms 1

Figure 1.
Neural network architecture Comparator-Adder.

A set of matrix pairs needs to convert to the embeddings in order to bring the required form for the functioning of neural networks. The situation presented in the matrix of states (Table 1) will have the vector form formalized, in which the length is 112 positions:

$\displaystyle|1,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,1,0,$ $\displaystyle 0,0,0,0\ldots 0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,$ $\displaystyle 1,0,0,0,0,0,0,0,1|$
2.3 Neural network architecture

The main application of neural network for the proposed IDSS concept is the retrieving of the most similar situation from the knowledge base. There are using neural network architecture Comparator-Adder, which a detailed study with source code snippet is presented in the previous work [10].

The neural network architecture Comparator-Adder is based on the idea of Siamese neural networks [16, 22, 23, 24], which are used to compare images or other signals. It may be considered an example of such neural networks development with regard to tabular data. In the networks are organized two channels of neural network calculations to encode input signals with their subsequent comparison at the final decisive element, which determines the class “similar” or “dissimilar”. The proposed architecture Comparator-Adder is shown in Fig. 1.

Our architecture compares the individual parts of the input vectors of two situations. Each part corresponds to its own element of a complex technological object. Thus, N comparison channels are organized in the form of N trained neural network comparators, the outputs of which are then fed to the Adder neural network. The adder at its output calculates the value of the similarity function, by which one can judge the degree of similarity of the two situations in their complex representation.

The experiments showed that under conditions of a small training sample (in the study [10], we used a training date set of 150 examples of situations), the trained neural network gives a sufficiently high accuracy (MAPE less than 10%) when assessing the similarity of situations in the validation file.

Figure 2.

IDSS work concept for SMART grid

3. Results

3.1 General concept of IDSS

The proposed concept of the application IDSS is to use the hybrid-CBR method and imply access to the knowledge base for retrieving solutions. It is envisaged several functional modules, the main ones of which are:

•
Module for identifying the elements states;
•
Module for identifying the object state;
•
Module for retrieving a precedent;
•
Module for adaptation/issuance of a solution.

At the same time, to identify the current situation on the SMART-grid object and its subsequent formalization, the states of the elements will be classified based on the input data:

•
For technological elements: parameters of the device (temperature, rotation frequency, input voltage, etc.); parameters of work substances, necessary for assessing the state of operability of the element (temperature, pressure, stream speed, etc.);
•
For providing elements (for example, "electricity connection point"): information about the presence/absence of the flow of the providing resource (electricity, oil, etc.);
•
For elements of the “personnel” category: information on the availability/unavailability of personnel;
•
For elements of the “environment” category: information about the influence of the elements on the object (a complication of personnel access, physical impact on other elements, etc.).

Information processing to identify the state of each element is performed in a “classifier” module using neural networks or a logical inference system based on expert rules in relation to technological elements from which information comes in the form of parameter signals.

At the facilities of SMART-grid systems, the implementation of data transmission channels is simplified by the existing network using microprocessor technology. The information arriving at the dispatching point makes it possible to determine, for example, which connects at the facility is damaged.

The output response of IDSS is a reasoned decision contained in the precedent or generated based on its transformation. Argumentation is based on analogy when the user is explained: where, when, by whom, under what conditions a similar solution was applied (recommended).

The solution is an instruction on how to transfer an object from a critical situation to a target serviceable one. The solution can be addressed to end-users: the dispatcher on duty, the operational field team, and others.

Figure 3.
Calculation results of sim.

The general structure of the solution is represented by the following components:

•
D1 – instructions for actions in a situation (flow-chart, process-mapping);
•
D2 – contact list of responsible persons and required organizations;
•
D3 – required reporting documents (templates, forms, acts, etc.);
•
D4 – additional background information (link to similar situations, expert advice, etc.).

Thus, the IDSS concept can be described by the following algorithm:

1.
Collecting data I from elements E of object O
2.
Identifying the elements E states
3.
Identifying the object O state
4.
Accessing the knowledge base (KB) and retrieving the most similar case
5.
Decision adaptation (if necessary)
6.
Providing decision D to end-users (participants) P
7.
Retention new precedent to KB

The diagram in Fig. 2 shows how IDSS works. The input from the elements E receives information I, the output information flow - decisions D intended for participants P.
3.2 Experiment: Retrieving a situation

An experiment was carried out on the basis of the SMART grid object “substation” – the most similar situation retrieving from the knowledge base. The composition of the object’s elements is presented in Table 1. For the experiment, a base of 27 situations was taken, formalized according to Section 2.2, and a situation simulating a malfunction was modeled. The purpose of the experiment is to evaluate the effectiveness of a neural network for selecting a situation from a database in the event of an emergency at the facility.

The similarity estimation Sim results obtained using the neural network were coared with the indicators of additive convolution and expert judgment. The similarity values obtained using expert judgment are taken as a template. Expert judgment implies the input of dynamic coefficients: the weights of each element and state to evaluate the similarity of the situation under consideration. The similarity of situations was determined expertly using the method previously described in our study [9], with which use the following Eq. (2):

$\displaystyle\text{Sim}(\text{Sit}_{Z},\text{Sit}_{\text{Act}})=\sum\beta_{i}% \ast\left({1-d_{i}}\right)$ (2)

where $\beta_{i}$ is the normalized element importance factor; $d_{i}$ is the distance between the states in which the $i$ -th element of a complex object is in the compared situations.

The distance between the states of an element is determined by the following Eq. (3):

$\displaystyle d_{i}=||S_{i,\textit{act}}-S_{i,z}||,$ (3)

where $S_{i,\textit{act}}$ , S ${}_{i,z}$ is the state of the $i$ -th element in the current situation and the $z$ -th situation, respectively.

The results of comparing the generated situation with the situations from the base are presented in the graph (Fig. 3). The ordinate shows the similarity indicators, the abscissa shows the conditional number of the situation from the knowledge base. The blue line is values obtained with the participation of an expert, the red line is values obtained using additive convolution, the green line is values obtained using a neural network based on architecture Comparator – Adder.

During the experiment, the MAPE (mean absolute percentage error) was 15%, which is higher than the value in the previous research work [10] obtained on the validation data set. However, we believe that MAPE value is not decisive in this case, the neural network choice of the most similar situation is much more important.

The graph shows that the result of our neural network is consistent with the choice of the expert. This means that a neural network, in contrast to additive convolution, allows avoids collisions. Collisions are situations with the same Sim, but in practice with a different solution [9].

The highlighted situations of zone A contain the most similar situation No. 3. In the same zone, collisions are clearly observed: the additive convolution shows the same similarity value in situations 3, 5, and 6, while the expert singled out namely the third situation as the most similar. In this case, the neural network repeats the choice of the expert and does not output controversial results.

Zone B also demonstrates the advantages of a neural network over additive convolution in terms of collision avoidance. Selection trends are consistent with expert selection.

4. Discussion

In this work, we examined a decision support system for SMART-grid systems. The diversity of the environment, the dynamism of the states of elements lead to the need to develop decision support systems with the participation of artificial intelligence.

We propose a new concept of an intelligence decision support system based on the hybrid-CBR method. Combining artificial intelligence and machine learning will speed up the decision-making process, which is especially important in critical situations, eliminate the human factor, and reduce the labor intensity of operation processes.

The possibility of implementing the proposed system correlates with the existing monitoring and data analysis systems implemented at the SMART-grid facility.

From a scientific point of view, our approach offers a new opportunity to create hybrid case-based reasoning models [25, 26, 27, 28, 29] and contributes to solving the urgent problem of integrating two concepts of artificial intelligence – knowledge-based systems and machine learning [30, 31].

The experimental results confirm the prospects of using neural networks to determine the similarity of situations [15, 16, 17]. The experiment showed the effectiveness of the proposed neural network architecture for retrieving the most similar situation from the knowledge base. The neural network repeated the choice of the expert, while, unlike the additive convolution, it did not output collisions and ranked the situations by similarity in accordance expert choices.

However, the proposed neural network has some limitations, which associated with the uncertainty of the elements state. The implementation of one-hot encoding is impossible in such a situation. The architecture modernizing for uncertainty cases is a plan for further research.

This work does not investigate the issues of generating solutions in the absence of similar precedents in the knowledge base. We believe that to solve these problems, will be using an approach that implies collection elements of action programs from different precedents for the formation of a new program. The implementation of this functionality will also allow machine learning methods and methods of knowledge-based systems. The further research plan includes detailing these methods for different conditions and cases.

Footnotes

Acknowledgments

The research was funded by RFBR and Tyumen Region, project number 20-47-720004.

References

Sayed

Gabbar

. SCADA and smart energy grid control automation. Smart Energy Grid Engineering. 2017; 481-514. doi: 10.1016/B978-0-12-805343-0.00018-8.

Jiang

Guo

. A novel application architecture of digital twin in smart grid. Journal of Ambient Intelligence and Humanized Computing. 2021. doi: 10.1007/s12652-021-03329-z.

Barriquello

Garcia

Schmitz

Bernardon

Fonini

. A decision support system for planning and operation of maintenance and customer services in electric power distribution systems. System Reliability. 2017. doi: 10.5772/intechopen.69721.

Chelaru

Grigoras

. Decision support system to determine the replacement ranking of the aged transformers in electric distribution networks. 2020 12th International Conference on Electronics, Computers and Artificial Intelligence (ECAI). 2020. doi: 10.1109/ECAI50035.2020.9223122.

Dogdu

Murat Ozbayoglu

Benli

Akinc

Erol

Atasoy

et al. Ontology-centric data modelling and decision support in smart grid applications a distribution service operator perspective. 2014 IEEE International Conference on Intelligent Energy and Power Systems (IEPS). 2014. doi: 10.1109/IEPS.2014.6874179.

Kulkarni

Shingare

. Decision support system for smart grid using demand forecasting models. In: Kulkarni

Shingare

, ed. by. Smart Network Inspired Paradigm and Approaches in IoT Applications. Singapore: Springer; 2021.

Nina

da Fonseca Neto

Ferreira

Miranda dos Santos

. Hybrid support system for decision making based on MLP-ANN, IED and SCADA for disturbances analysis of electrical power distribution transformers. 2013 UKSim 15th International Conference on Computer Modelling and Simulation. 2013. doi: 10.1109/UKSim.2013.147.

Glukhikh

. Case based reasoning for managing urban infrastructure complex technological objects. CEUR Workshop Proceedings. 2021; 2843: 038.

Glukhikh

. Situations representation and retrieve in the case-based reasoning system for managing a complex technological object. CEUR Workshop Proceedings. 2021; 2922: 017.

10.

Glukhikh

. Case-based reasoning with an artificial neural network for decision support in situations at complex technological objects of urban infrastructure. Applied System Innovation. 2021; 4(4): 73. doi: 10.3390/asi4040073.

11.

Eremeev

Varshavskiy

Alekhin

. Case-based reasoning module for intelligent decision support systems. Proceedings of the First International Scientific Conference “Intelligent Information Technologies for Industry” (IITI’16). 2016; 207-216. doi: 10.1007/978-3-319-33609-1_18.

12.

Varshavskii

Eremeev

. Modeling of case-based reasoning in intelligent decision support systems. Scientific and Technical Information Processing. 2010; 37(5): 336-345. doi: 10.3103/S0147688210050096.

13.

Aha

. Case-based learning algorithms. In Proceedings of the 1991 DARPA Case-Based Reasoning Workshop, Washington, USA, 1991 31 May 1, 147-158.

14.

Chen

Burrell

. Case-based reasoning system networks: A review. Neural. Comput. Applic. 2001; 10: 264-276.

15.

Gabel

Godehardt

. Top-down induction of similarity measures using similarity clouds. Proceedings of the International Conference on Case-Based Reasoning. 2015; 149-164. doi: 10.1007/978-3-319-24586-711.

16.

Mathise

Bach

Aamodt

. Using extended siamese networks to provide decision support in aquaculture operations. Applied Intelligence; 2021. doi: 10.1007/s10489-021-02251-3.

17.

Kenny

Keane

. Twin-systems to explain artificial neural networks using case-based reasoning: Comparative tests of feature-weighting methods in ANN-CBR twins for XAI. Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI-19). 2019; 2708-2715. doi: 10.24963/ijcai.2019/376.

18.

Bouhana

Zidi

Fekih

Chabchoub

Abed

. An ontology-based CBR approach for personalized itinerary search systems for sustainable urban freight transport. Expert Systems with Applications. 2015; 42(7): 3724-3741. doi: 10.1016/j.eswa.2014.12.012.

19.

Iqbal

Murad

Mustapha

Sharef

. An analysis of ontology engineering methodologies: a literature review. Research Journal of Applied Sciences, Engineering and Technology. 2013; 6(16): 2993-3000. doi: 10.19026/rjaset.6.3684.

20.

Huang

Nie

Luo

. Scenario-based marine oil spill emergency response using hybrid deep reinforcement learning and case-based reasoning. Applied Sciences. 2020; 10(15): 5269. doi: 10.3390/app10155269.

21.

Leak

Crandall

. The supporting case-based reasoning with neural networks: An illustration for case adaptation. ceur workshop proceedings. 2021; 2846: 1.

22.

Ntalampiras

. One-shot learning for acoustic diagnosis of industrial machines. Expert Systems with Applications. 2021; 178: 114984. doi: 10.1016/j.eswa.2021.114984.

23.

Deshpand

Minai

. Kumar

. One-shot recognition of manufacturing defects in steel surfaces. procedia manu-facturing. 2020; 48: 1064-1071. doi: 10.1016/j.promfg.2020.05.146.

24.

Hsiao

S-C

Kao

D-Y

Liu

Z-Y

Tso

. Malware Image Classification Using One-Shot Learning with Siamese Net-works. Procedia Computer Science. 2019; 159: 1863-1871. doi: 10.1016/j.procs.2019.09.358.

25.

Abdelwahed

Mohamed

Saleh

. Solving the motion planning problem using learning experience through case-based reasoning and machine learning algorithms. Ain Shams Engineering Journal. 2020; 11(1): 133-142. doi: 10.1016/j.asej.2019.10.007.

26.

Guo

Zhang

Sun

Jiang

. Machine learning based feature selection and knowledge reasoning for CBR system under big data. Pattern Recognition. 2021; 112(6): 107805. doi: 10.1016/j.patcog.2020.107805.

27.

Zhai

Martínez

Díaz

. Applying case-based reasoning and a learning-based adaptation strategy to irrigation scheduling in grape farming. Computers and Electronics in Agriculture. 2020; 178: 105741. doi: 10.1016/j.compag.2020.105741.

28.

Guo

Chen

Zhu

Y-X

Guo

Y-Q

. Research on the integrated system of case-based reasoning and Bayesian network. ISA Transactions. 2019; 90: 213-225. doi: 10.1016/j.isatra..

29.

Leake

Crandall

. Supporting case-based reasoning with neural networks: an illustration for case adaptation. CEUR Workshop Proceedings. 2021; 2846: 1.

30.

Bhatia

Pinto

. Supporting automated construction of knowledge-bases for safety critical applications: Challenges and opportunities. CEUR Workshop Proceedings. 2021; 2846: 37.

31.

Saker

Zhou

Eberhart

Hitzler

. Neuro-symbolic artificial intelligence: Current trends; 2021. arXiv2105.05330: [csAI].

Sit ${}_{\textit{act}}$	Correctly	In work	Available	Doesn’t affect
Reserve power line	1
Power transformer		1
Voltage and amperage transformer		1
Transformer own needs		1
Relay protection and automatics devises		1
Communication server		1
Telemechanics devises		1
Sectional switcher			1
Person on duty			1
Firefighting system			1
Buildings				1
Natural objects				1
Natural phenomena				1
Rooms				1