Regional forestry economic evaluation based on neural network and fuzzy model

Abstract

In the process of people’s development, many resources in natural resources provide convenience and help for people’s survival and development, and forest is an important part of ecological balanced development. In the process of economic development, forest resources can also bring people a lot of economic benefits. People develop regional forestry economy by planting trees with economic value. Asset assessment of forests can better promote the development of forestry economy and help people better use and manage resources. During the process of forestry economic development, the technology of asset evaluation established on the basis of BP neural network enriches the theoretical content of forestry economic development to a certain extent and provides a more solid theoretical basis for development. At the same time, the qualitative and quantitative methods are combined to ensure that the data are more accurate and the key influencing factors can be better analyzed. Based on the analysis of the relevant research results, this paper takes some trees of the same growth year as the object of study. Using BP neural network to analyze the factors that will change the overall value of forest resources, a special BP neural network model belongs to forestry and economic development is established.

Keywords

Neural network fuzzy model regional forestry economy forest resources assets

1 Introduction

Whether in nature or in people’s daily life, the role of forests is indispensable. In the structure of ecosystem, forest is an important part of purifying air, maintaining soil moisture and promoting the development of ecological environment in a balanced and healthy direction. In people’s life, trees can provide people with wood to build houses, and they can also become greening resources to decorate urban streets. Trees can be seen everywhere in daily life. With the increasing protection of ecological environment, people also realize the important role of forest, forestry and other related industries have received more and more attention. The world-wide conferences held at the international level have created more opportunities for forestry development.

In the evaluation of forest resources, people through a long period of continuous exploration and the formulation of relevant laws and regulations, the specific content of the value measurement of forest resources has been stipulated. In order to better assess forest values, when measuring and assessing forest values, attention should be paid to forest resources, including ecological values and economic development values, both of which are central to forest value assessment and measurement. As early as 1992, the relevant agreements of the International Development Conference held at the United Nations had begun to incorporate the assessment and measurement of forest resources into the value of social development and economic development, which was also the content of evaluating the overall value of a country’s economic development. In the system of laws and regulations of our country, there are clear provisions on asset evaluation of forest resources, which provide a lot of support for related work.

Forests play a particularly important role in people’s daily life and social development, so the asset assessment of forest resources should also be paid attention to. Without advanced technical support in the past when people evaluate and measure, it may lead to a certain difference between the estimated value and the actual value. For better improving the effect of forest management and making more accurate judgment on the value of assets, people began to use the method of BP neural network to analyze. The use of this method can be carried out in combination with qualitative and quantitative analysis techniques, and it can greatly reduce the errors caused by some factors related to input. Using this method can set up a platform for forestry and economics development on the network, which can save a lot of time. The important role of this paper is to discuss the method of BP neural network, to find out the problems that will affect the value evaluation of forest resources, and to select the BP neural network model more suitable for forestry development, and to further advance the value and benefit of forestry development.

2 Related work

In order to strengthen the theoretical basis of this article, the author studies the relevant literature on regional forestry economic development according to the specific direction of the article. According to the literature [1] to evaluate the specific assets and values of forest resources, different methods should be developed for the growth years of different trees in the forest. If there are more trees in a forest, the time of cutting and the number of times of cutting should be considered when estimating its value. In trees with similar growth years, the valuation method for trees with shorter growth years is mainly the replacement cost method, and the income present value method is used to estimate the trees in the middle of the growth year. Trees growing to maturity are estimated at market prices. In the current market economy system, if the relevant information records of the year of tree growth are incomplete, then no matter what type of forest, the method of market comparison can be used to estimate the value. That is to say, they can be determined according to the specific conditions of market pricing, which is more flexible. With the deepening of people’s research on related contents, many people also begin to pay attention to the impact of the final purpose of the evaluation process on the process of value estimation and the selection of valuation methods. Including valuation after the better management or for sale and so on. [2] the literature, the new evaluation method, the average cost profit method, is explained, and the specific operation flow of the method in practical use and the method of replacement cost mentioned in the first literature are compared. Finally, the conclusion is that the new method is more feasible and can be applied better. According to the literature [3] the asset evaluation process of forest resources is compared with the ordinary type of asset evaluation process. It is considered that the two assessments are similar and have a similar operation process. The specific income of forest resources is evaluated and an ordinary resource will produce the same income in the future long-term investment. In the evaluation of trees, more attention is paid to the growth height of trees in infancy, and to the assessment of volume during growth for trees in the middle and longer years. Based on the analysis of the relevant factors [4] the impact of the growing period on trees in the early and mid-term or mature periods, including the problems arising in the management and management of forests, the impact of the development of the market economy system, the differences in the adjustment of the quality of stands, and so on, the suggestions that the price of trees and the rate of return on investment can be adjusted, the criteria for the final harvest evaluation should be adjusted appropriately, and when the trees in the shorter growing period are evaluated, Adjustments to some regulations and functional devaluations can reduce differences in the assessment process to some extent. [5] the literature, the replacement cost method used in forest management, which is specially used to provide wood, is analyzed. The characteristics and process of using this method are summarized. In the process of re-analysis, a compound interest formula must be used to calculate the return on investment. [6] literature does not particularly agree with the method of evaluating the value of forest resources by reference to market prices. This method can not count all the specific costs involved in forest management. There will be a great deviation between the total estimated value and the actual total value. A grey correlation analysis method [7] used in the literature to calculate the important proportion of the specific factors affecting forest value. It is concluded that the change of stand volume has the greatest and most obvious effect on forest value, and the order of influence of other factors on final value is: the grade of regional development conditions, the utilization efficiency of trees as wood, the average DBH size of trees in trees, the height of trees and the growth year of their own trees.

As for the establishment of the model, the literature [8] after analyzing the growth characteristics of young trees, middle trees and mature trees, the value of the forest is estimated by the analysis technology of multivariate linear regression equation, and the batch model of asset evaluation is established according to the characteristics of each growth stage of trees. Literature [9] mainly put forward that in the selection of evaluation methods and evaluation process, it is suggested that the evaluation method used in estimating the value of forest resources is not fixed. Different evaluation methods and processes can be selected according to the final evaluation purpose, which can greatly increase the flexibility of evaluation. The grey correlation degree method can be used to determine the proportion of influencing factors in the asset price decision of trees, and then a suitable reference case can be selected by using fuzzy clustering technology. After the analysis, the estimated values can be verified by hierarchical analysis, and the incorrect ones can be corrected in time. According to the literature [10] the cost and cost of managing and managing the trees with time limit and managing the trees without time limit are different. The corresponding calculation formula is given according to the relevant characteristics. In addition, it is suggested that the number and intensity of wood cutting will have a significant impact on the final estimated value. Literature [11] when evaluating the assets of artificially cultivated tree species, the method of multiple linear regression and the method of BP neural network are used to construct the model, and according to the age of tree growth, The values generated by these two methods in the process of analysis are compared to determine which method has high accuracy. The literature [12] applied the analysis method of multivariate linear regression and the analysis method of BP neural network to practice, evaluated the Chinese fir forest in Shunchang County, and constructed the corresponding mathematical model. For specific applications, the multivariate linear regression method can analyze the linear regression relationship between multiple independent variables and strain quantities in the established model. The model established BP neural network can analyze the nonlinear regression relationship between independent variables and strain quantities.

Through the analysis and understanding of many documents, we can see that China has made some achievements in the evaluation of forest resources, but it still lacks more scientific theoretical research, which is lacking at present. On the one hand, the evaluation practice is mostly in the man-managed forest, and the research and practice of the forest contains a variety of complex types; on the other hand, in the course of the research, the research is mostly in the qualitative aspect, and the research on the quantitative aspect is less, especially on the influence of economic and social development on the value, such as the choice of the management mode of the forest, the degree of accessibility and so on; The last aspect of the problem is that the evaluation of the majority at the same time can save a lot of time and improve the efficiency of the evaluation under certain conditions, but this method has not yet been applied to practice. And has not formed a very complete evaluation system, need to continue to study in future practice.

3 Model design based on neural network and fuzzy model

3.1 BP Neural network model establishment

3.1.1 Network operation steps

(1) Waiting for system initialization.

(2) Randomly select a set of targets into the network in the system.

(3) The output values between the units included in the middle layer are calculated by function: $\begin{matrix} s_{j} = \sum_{i = 1}^{n} w_{ij} a_{i} - θ_{j} \end{matrix} j = 1, 2, . . ., p$ (1) $b_{j} = f (s_{j}) \begin{matrix} j = 1, 2, . . ., p \end{matrix}$ (2)

(4) Transfer the calculated values to the next layer $L_{t} = \sum_{j = 1}^{p} v_{jt} b_{j} - γ_{t} \begin{matrix} t = 1, 2, . . ., q \end{matrix}$ (3) $C_{t} = f (L_{t}) \begin{matrix} t = 1, 2, . . ., q \end{matrix}$ (4)

(5) Calculation of output layer error $d_{t}^{k} = (y_{t}^{k} - c_{t}) (C_{t} (1 - C_{t}) \begin{matrix} t - 1, 2, . . ., q \end{matrix}$ (5)

(6) Calculation of errors in the middle layer $e_{j}^{k} = [\sum_{t = 1}^{q} d_{t} \cdot v_{jt}] b_{j} (1 - b_{j})$ (6)

(7) Error in validation: $\begin{matrix} v_{jt} (N + 1) = v_{jt} (N) + α \cdot d_{t}^{k} \cdot b_{j} \\ γ_{t} (N + 1) = γ_{t} (N) + α \cdot d_{t}^{k} \\ t = 1, 2, . . ., q \begin{matrix} j = 1, 2, . . ., p \begin{matrix} 0 < α < 1 \end{matrix} \end{matrix} \end{matrix}$ (7)

(8) Correction of errors: $\begin{matrix} w_{ij} (N + 1) = w_{ij} (N) + β e_{j}^{k} a_{j}^{k} \\ θ_{j} (N + 1) = θ_{j} (N) + β e_{j}^{k} \\ \begin{matrix} i = 1, 2, . . ., n j = 1, 2, . . ., p 0 < β < 1 \end{matrix} \end{matrix}$ (8)

After determining that there is no error, enter the next random target and cycle until there is no error.

3.1.2 Learning algorithm

Suppose the function of this algorithm is: $E (w) = \frac{1}{2} \sum_{i = 1} e_{i}^{2} (w) = \frac{1}{2} \sum_{i = 1} | Y_{i} - Y_{i}^{'} |^{2}$ (9) $Δ w = {[J^{T} (w) J (w) + μ I]}^{- 1} J^{- T} (w) e (w)$ (10)

The next step is to: $MSE = \sqrt{\frac{\sum ɛ^{2}}{n}}$ (11)

The corresponding function of this method is SSE. to Trainbr, the training performance $SSE = λ E_{W} + λ E_{D}$ (12)

3.2 Input layer determination

3.2.1 Alternative factor determination

When using BP neural networks to estimate forest assets, the specific meaning of the input layer refers to the factors that will affect the final evaluation value. When estimating the asset value of forest resources, it is affected by many factors, because the final estimated value is different from the actual value or large or small. Through the specific classification and study of the influencing factors, they can be roughly divided into three categories. The first type of influencing factors is the geological conditions of tree growth. The quality of trees can be greatly improved, and the living conditions of different tree species are different. This factor can affect the final harvest quality of trees. The second kind of influence factor is the influence of the final harvest quality of trees on the value of trees. In the process of cutting, if the trees are damaged, it may affect the final value of the trees to some extent. The value of the trees used for sale in the final cutting is also related to the type of trees, the growth years, the height of the trees, the size of the DBH, the accumulation size and so on. The third factor is the market economy environment, the pricing trend in the market and the quantity of wood in the market will also affect the final number of transaction prices to some extent. When considering market factors, we should also pay attention to the impact of the cost of managing and operating trees, the cost of selling trees and the pricing of sales. There is also a relationship between different factors, which play a decisive role in the value evaluation of forests.

(1) Young forest:

When assessing the number of young people in the growing period, the replacement cost calculation method is used. The relevant calculation formulas include the age of the trees, the number of trees in the trees, the height of the trees, the cost and interest rate of managing and operating the trees. This is the five main factors in the formula. In the process of calculation, the numerical value can be directly replaced into the formula for calculation. When constructing the model, it must enter the network from the input layer. Different tree species are planted during the growth of trees. The number of trees, the height of trees, and certain management costs in the trees have a great impact on the evaluation process. The alternative input part of the input layer is the type of tree, the age of growth, the number of plants, the height of tree growth, the management cost, the interest rate.

(2) Middle age forests:

In the assessment of trees growing to the medium term, the formula of the income present value method is used, the age of the trees in the formula, the accumulation, the timber efficiency of the trees in the specification, the timber efficiency of the trees in the specification, the selling price of the trees in the specification, the selling price of the trees in the specification, the cost of the trees in the process of sale, the cost of managing the trees, the year of cutting the trees, the interest rate, etc.

(3) Mature forests:

In the evaluation of trees growing to maturity, the calculation formula of market price inversion is used. The items included in the formula are volume, timber efficiency of specification trees, timber efficiency of non-specification trees, sales price of specification trees, sales price of non-specification trees, and related costs incurred during sales. In addition, the influence of different kinds of trees on the final production efficiency and sales price of trees, and the effect of tree growth age on the volume of storage should be considered. These two factors are added to the input layer of the model.

Analysis of variance of the alternative factors in the third age group showed that the selected factors were significantly correlated with the evaluation value (p < 0.05).

3.2.2 Sensitivity analysis and input layer factor screening

In the process of using neural networks to analyze data, the operation of black boxes is not observed by researchers. Therefore, professionals can not fundamentally know the internal relationship between independent variables and strain in the model, and there is no way to explain the biological or economic relationship between many data. In order to improve the running speed of neural network, a reasonable method should be adopted to ensure the relationship between the input factors, so as to improve the efficiency and accuracy of the model analysis data. $S_{n} = \frac{\sum_{j = 1}^{L} (w_{ij} v_{jk} / \sum_{r = 1}^{N} w_{rj})}{\sum_{i = 1}^{N} \sum_{j = 1}^{L} (w_{ij} v_{jk} g / \sum_{r = 1}^{N} w_{rj})}$ (13)

The value has the positive and negative difference, will have the certain influence in the calculation. Through the improved formula: $S_{n} = \frac{\sum_{j = 1}^{L} (| w_{ij} v_{jk} | / \sum_{r = 1}^{N} | w_{rj} |)}{\sum_{i = 1}^{N} \sum_{j = 1}^{L} (| w_{ij} v_{jk} | / \sum_{r = 1}^{N} | w_{rj} |)}$ (14)

3.3 Model evaluation indicators

The prediction effect and ability of the evaluation network can also refer to the following indicators.

3.3.1 Mean absolute error

The formula for this error is: $MAE = \frac{1}{N} \sum_{i = 1}^{N} \sum_{i = 1}^{N} | y_{i} - \overset{\land}{y_{i}} |$ (15)

3.3.2 Average relative error

The formula is: $MAPE = \frac{1}{N} \sum_{i = 1}^{N} | \frac{y_{i} - \overset{\land}{y_{i}}}{y_{i}} | \times 100$ (16)

3.3.3 Determining factor

The formula is: $R^{2} = 1 - \sum_{i = 1}^{N} (y_{i} - \overset{\land}{y_{i}})^{2} / \sum_{i = 1}^{N} (y_{i} - \overset{\land}{y_{i}})^{2}$ (17) The smaller the calculated value, the higher the accuracy.

3.4 Construction of a comparative judgment matrix

After the model is constructed, it is very important to calculate the related factors again.

In order to further highlight the role of quantitative analysis in the decision-making process, the scale used in this paper is shown in Fig. 2.

Fig. 1

Effect factor sensitivity coefficient of young forest.

Fig. 2

Comparison of the accuracy of different hidden layer number models in young forest.

3.4.1 Computing the comparison judgment matrix between adjacent layers

The nine scales are compared to each other, as follows: $\begin{matrix} a_{ij} = \frac{1}{a_{ij}} \\ A = (a_{ij}) \\ n \times n = [\begin{matrix} a_{11} & a_{12} & \dots & a_{1 n} \\ a_{21} & a_{22} & \dots & a_{2 n} \\ \dots & \dots & \dots & \dots \\ a_{n 1} & a_{n 2} & \dots & a_{nn} \end{matrix}] \end{matrix}$ (18)

3.4.2 Calculation weights

The characteristics of the matrix can be calculated by computer, the specific formula is as follows: $M_{i} = Π_{j = 1}^{a} b_{ij}$ (19) Step 2: $W_{i} = \sqrt[n]{M_{i}}$ (20) Formalize the calculated data by the following formula: $W = \frac{W_{i}}{\sum_{i = 1}^{n} W_{j}}$ (21) Step 3: Calculate the largest characteristic root $λ_{max} = \frac{1}{n} \sum_{i = 1}^{n} \frac{(BW)_{i}}{W_{i}}$ (22)

3.4.3 Consistency detection

Step 1: Compare the consistency of the matrix structure $λ_{max} = \frac{1}{n} \sum_{i = 1}^{n} \frac{(BW)_{i}}{w_{i}}$ (23) Step 2: Calculate the indicators to measure consistency $CI = \frac{λ_{max} - n}{n - 1}$ (24) Step 3: Randomly select the consistency ratio $CI = \frac{CI}{RI}$ (25)

3.4.4 Composite index evaluation method

Calculating the indices at all levels, $Y_{i - 1} = \sum_{i = 1}^{n} D_{i} X_{i}$ (26) To distinguish between new factors, we need to add new weights, $Y_{i - j}^{[n]} = \sum_{i = 0, j = 1, [n^{(i = 1)}] = 1}^{k, l, s} W_{(n + 1)_{j}}^{n^{(i + 1)}} D_{(I + 1)_{J}}^{n^{(i + 1)}} X_{(n + 1)_{j}}^{n^{(i + 1)}}$ (27)

3.5 Modeling language programming

By establishing a neural network model, researchers can determine the number of relevant training samples and prediction samples themselves, but predicting the same training samples and prediction samples must be selected, and there is no way to control them. The establishment of the network through language programming can realize the editing of the custom content of the network during initialization, or the authorization of the network when selecting the training sample and the prediction sample, and the user will be prompted if the network runs incorrectly. Therefore, this method to establish the network is faster.

4 Modeling results and analysis

4.1 Batch evaluation model of forest and wood resources in young age

4.1.1 Model structure optimization

The details of the sensitivity coefficients calculated when the number is analyzed in infancy are shown in Table 3, as shown in Fig. 1.

Table 1
The model of matrix

A B1 B2 B3 B4

B1 B11 B12 B13 B14

B2 B21 B22 B23 B24

B3 B31 B32 B33 B34

B4 B41 B42 B43 B44

A	B1	B2	B3	B4
B1	B11	B12	B13	B14
B2	B21	B22	B23	B24
B3	B31	B32	B33	B34
B4	B41	B42	B43	B44

Table 2

Level scale

Serial number	Level of importance Results (Bij)	Comparison Assignment
1	i,j two elements are equally important	1
3	i element is more important than J element	3
5	i element is more important than J element	5
7	i element is more important than J element	7
9	i element is more important than J element	9
2,4,6,8	represents the intermediate value of the above judgment	/

Table 3

Effect factor sensitivity coefficient of young forest

Factor sequence	Factor name	Sensitivity coefficient
1	dominant species	0.0461
2	Age	0.1547
3	Tree height	0.2182
4	Number of plants	0.1832
5	Interest rate	0.2336
6	Forestry costs	0.1642

4.1.2 Determine number of hidden layer nodes

Run each model 20 times, as shown in Fig. 2. The results of the calculation are shown in Table 4.

Table 4
Comparison of the accuracy of node number model of different hidden layer in young forest

Number of hidden layer nodes SSE MAE (hm yuan/²) MAPE (%)

3 0.2988 198.60 6.35

4 0.1640 158.25 5.64

5 0.0345 110.70 4.04

6 0.0207 94.50 3.14

7 0.0145 81.00 279

8 0.0754 104.40 4.02

9 0.0137 89.40 3.03

10 0.0124 97.35 3.38

11 0.0118 87.75 3.65

12 0.0164 93.00 3.82

Number of hidden layer nodes	SSE	MAE (hm yuan/²)	MAPE (%)
3	0.2988	198.60	6.35
4	0.1640	158.25	5.64
5	0.0345	110.70	4.04
6	0.0207	94.50	3.14
7	0.0145	81.00	279
8	0.0754	104.40	4.02
9	0.0137	89.40	3.03
10	0.0124	97.35	3.38
11	0.0118	87.75	3.65
12	0.0164	93.00	3.82

4.1.3 Model precision verification

Through calculation, we can clearly see the relevant numerical value, which can facilitate the analysis of tree value in infancy.

Calculate the specific values of 30 samples in the model, and the results are shown in Table 5. It can be seen that there is a significant linear relationship between the evaluated value and the predicted value, and there is a significant linear correlation between the evaluation value and the predicted value. The smaller the value calculated by this model, the higher the accuracy. It can be used to evaluate the value of young trees.

Table 5
Prediction accuracy of young forest model

Sample serial number number Absolute error (Meta/hm) Total error²) (%) relative error error

1 5.55 0.10

2 –43.35 0.37

3 25. 05 0.52

4 –1.95 0.04

5 69.30 0.85

6 –101.40 1.00

7 –6.15 0.36

8 95.70 1 42

9 85.80 2. 66

10 –4.95 0.10

11 –73.50 0.80

12 2.55 0.05

13 54.45 1.18

14 101.40 1.66

15 32 3.25 12.11

16 –9.45 0.05

17 –4.35 0.06

18 –9.60 0.23

19 –1.80 0.04

20 398. 5 5 9.85

21 –63.45 0.95

22 42.75 1.38

23 –1 73.40 0.91

24 –81.60 2.70

25 113.85 1.52

26 –5.40 0.12

27 7.05 0.19

28 58.05 0.93

29 –26.85 0.57

30 –25.35 0.68

Absolute average 67.20 1.45

Sample serial number number	Absolute error (Meta/hm) Total error²)	(%) relative error error
1	5.55	0.10
2	–43.35	0.37
3	25. 05	0.52
4	–1.95	0.04
5	69.30	0.85
6	–101.40	1.00
7	–6.15	0.36
8	95.70	1 42
9	85.80	2. 66
10	–4.95	0.10
11	–73.50	0.80
12	2.55	0.05
13	54.45	1.18
14	101.40	1.66
15	32 3.25	12.11
16	–9.45	0.05
17	–4.35	0.06
18	–9.60	0.23
19	–1.80	0.04
20	398. 5 5	9.85
21	–63.45	0.95
22	42.75	1.38
23	–1 73.40	0.91
24	–81.60	2.70
25	113.85	1.52
26	–5.40	0.12
27	7.05	0.19
28	58.05	0.93
29	–26.85	0.57
30	–25.35	0.68
Absolute average	67.20	1.45

4.1.4 The relationship between absolute error and relative error is shown in Fig. 3

Fig. 3

Relationship between the absolute error and the relative error of the model of different hidden layer number in young forest.

4.2 Medium-aged forest and wood resources asset batch evaluation model

4.2.1 Model structure optimization

When using Bayesian calculation method to calculate, when the value of hidden node is 12, the minimum error of network prediction is minimized. The specific results are shown in Table 6 and shown in Fig. 4.

Table 6
Sensitivity factors of middle age forest

Factor sequence Cause Child Name of name Sensitivity coefficient

1 Age 0.2652

2 Interest rate 0.1638

3 Accumulation 0.1167

4 dominant species 0.0988

5 Main cutting age 0.0549

6 Sales price of specifications 0.0522

7 Specification material yield 0.0512

8 Forestry costs 0.0510

9 Non-specification material yield 0.0418

10 Sales costs 0.0533

11 Non-specification price 0.0511

Factor sequence	Cause Child Name of name	Sensitivity coefficient
1	Age	0.2652
2	Interest rate	0.1638
3	Accumulation	0.1167
4	dominant species	0.0988
5	Main cutting age	0.0549
6	Sales price of specifications	0.0522
7	Specification material yield	0.0512
8	Forestry costs	0.0510
9	Non-specification material yield	0.0418
10	Sales costs	0.0533
11	Non-specification price	0.0511

Fig. 4

Sensitivity coefficient of influencing factors in middle age forest.

4.2.2 Determine number of hidden layer nodes

It can be seen from Figs. 5 and 7 that the number of hidden nodes in the model has a certain effect on the final accurate value of the model, but there is no absolute relationship between the two. As a result, the optimal structure of forest resource asset assessment is BR9-10-1.

Fig. 5

Comparison of the accuracy of the model of different hidden layer nodes in the aged forest.

Fig. 6

Relationship between the absolute error and relative error of the model of the number of nodes in different hidden layers of middle age forest.

Fig. 7

Influence factor sensitivity coefficient of mature forest.

4.2.3 Model precision verification

According to the above model structure optimization, BR9-10-1, is the optimal structure of the batch evaluation model for middle-aged forest resource assets The learning algorithm is Bayesian regularization; The input layer factors are age, interest rate, accumulation, tree species, main cutting age, selling cost, selling price of specification material, producing rate of specification material, selling price of irregular material, A total of 9; The number of hidden layer nodes is 10; The output layer is the evaluation value, The number of nodes is 1. After 25 steps of convergence, Fitting performance function SSE 0.000941, Below target error (goal = 0.001), The number of effective parameters in the network is 38.3, The sum of the square of network weights is 12.4.

Validate the model with the reserved 30 prediction samples, the results are shown in Table 8. A significant linear correlation was found between the predicted value and the evaluation value, and R coefficient of determination² = 0.9997. The trained BP model BR9-10-1 high prediction accuracy and strong generalization ability, which can meet the requirements of medium-aged forest batch evaluation.

Table 7
Comparison of accuracy of node number model of different implicit layer in age forest

Number of hidden layer nodes SSE MAE (hm yuan/²) MAPE (%)

4 0.0169 75.21 4.88

5 0.0180 68.59 3.68

6 0.0194 101.74 5.14

7 0.0023 50.78 2.50

8 0.0019 38. 95 2.25

9 0.0019 37.66 1.87

10 0.0009 35.55 1.29

11 0.0025 42 : 76 1.78

12 0.0019 46.92 1.97

13 0.0074 52.76 4.50

Number of hidden layer nodes	SSE	MAE (hm yuan/²)	MAPE (%)
4	0.0169	75.21	4.88
5	0.0180	68.59	3.68
6	0.0194	101.74	5.14
7	0.0023	50.78	2.50
8	0.0019	38. 95	2.25
9	0.0019	37.66	1.87
10	0.0009	35.55	1.29
11	0.0025	42 : 76	1.78
12	0.0019	46.92	1.97
13	0.0074	52.76	4.50

Table 8

Prediction accuracy of middle age forest model

Sample serial number	Absolute error (Meta/hm) Total error²)	(%) relative error
1	–21.50	1.70
2	0.03	0.49
3	44.16	1.03
4	65.84	0.38
5	16.30	0.37
6	29.74	0.04
7	1.35	0.79
8	–22.37	0.53
9	48.07	0.42
10	18. 06	0.11
11	7.63	0.42
12	–6.66	0.07
13	–19.55	0.14
14	5.42	0.35
15	11.92	1.70
16	–13.10	0.49
17	41.66	0.89
18	–40.83	0.54
19	142.35	2. 09
20	4.12	0.17
21	–76.36	0.55
22	–15.59	0.21
23	32.11	0.38
24	18.31	0.89
25	–62.76	17.10
26	–1.61	0.03
27	53.09	0.11
28	149. 93	4.05
29	3.32	2.31
30	32.46	0.14
Absolute average	41.66	1.28

The relationship between absolute error and relative error is shown in Fig. 6.

4.3 Batch evaluation model of forest and wood resources in mature forest

4.3.1 Model structure optimization

The selection factors of the input layer of mature forest modeling are dominant tree species, age, accumulation, specification material yield, non-specification material yield, specification material sales price, non-specification material sales price, sales cost, total 8. The sensitivity coefficient of each factor is calculated by formula (3–15). The results are shown in Table 9 and shown in Fig. 7.

Table 9
Influence factor sensitivity coefficient of mature forest

Factor sequence CauseChildName of name Sensitivity coefficient

1 dominant species 0.1309

2 Age 0.0294

3 Accumulation 0.3392

4 Sales costs 0.1217

5 Specification material yield 0.1157

6 Non-specification material yield 0.0891

7 Sales price of specifications 0.1099

8 Sales price of non-specification materials 0.0641

Factor sequence	CauseChildName of name	Sensitivity coefficient
1	dominant species	0.1309
2	Age	0.0294
3	Accumulation	0.3392
4	Sales costs	0.1217
5	Specification material yield	0.1157
6	Non-specification material yield	0.0891
7	Sales price of specifications	0.1099
8	Sales price of non-specification materials	0.0641

The factor sensitivity coefficient is arranged in order from big to small: accumulation, dominant tree species, sales cost, specification material output rate, specification material sales price, non-specification material output rate, non-specification material sales price, age. Accumulation is the most direct determinant of the asset value of mature forest, so the sensitivity coefficient is the highest and the contribution to the evaluation value is the largest, which is 0.3392. Tree species factors affect the yield and sales price, and the sales cost (tax and fee) is different because of the difference of tree species, so the contribution of tree species to the evaluation value is the second, which is 0.1309. When the stand grows to mature forest, the accumulation increases with age, so the contribution of age to the evaluation value is the smallest, less than 0.03.

4.3.2 Determine number of hidden layer nodes

According to the above input layer factor screening results, Mature forest batch evaluation model input layer node number N = 7. Comparing the accuracy of different hidden layer node number models, The number of nodes affects the fitting performance and prediction ability of the model. The results are shown in Fig. 8 and Table 10. When the number of hidden layer nodes is 7,9, Model MAPE to a minimum, It is 0.54. Then comparing the fitting ability of the two models to the training samples, SSE(9)<SSE(7), Therefore, the optimal number of nodes in the hidden layer of the BP model L = 9, BR7-9-1. is the model

Fig. 8

Comparison of Accuracy of Node Number Model of Different Implicit Layer in Mature Forest.

Table 10

Comparison of the number of nodes in different hidden layers of mature forest

Number of hidden layer nodes	SSE	MAE (hm yuan/²)	MAPE (%)
3	0.000882	88.55	0.79
4	0.000868	80.85	0.69
5	0.000844	.85.20	0.56
6	0.000894	75.60	0.59
7	0.000846	99.75	0. 54
8	0.000837	81.90	0.56
9	0.000799	75.75	0.54
10	0.000853	85.95	0.62
11	0.001266	72.30	0.57
12	0.000789	73.20	0.58

Table 11

Prediction accuracy of mature forest model

Sample serial number	Absolute error (Meta/ hm) Total error²)	(%) relative error
1	–24.87	0.26
2	–103.62	0. 34
3	–4.45	0.01
4	132.83	1.51
5	27.03	0.07
6	52.41	1.50
7	9.97	0.03
8	65.70	0.34
9	–69.66	0.46
10	–117.89	0.52
11	–80.57	0.34
12	104.98	0.77
13	–11.33	0.04
14	46.24	1.13
15	–1.70	0.01
16	264. 94	2.93
17	–118.23	0.54
18	20.32	0.04
19	–58. 37	0.23
20	–23. 81	0.09
21	–154 99	0.98
22	–52.46	0.19
23	–71.82	0. 35
24	–2192	0.08
25	–25.70	0.10
26	163.41	2.41
27	237.94	2.46
28	208.42	0.41
29	–36.74	0.33
30	–9.07	0.03
Absolute average	77.40	0.62

4.3.3 Modeling results

After optimizing the constructed model, the best evaluation model can be obtained for the number of different growth stages. The algorithm used in the analysis of data in the model is Bayesian regularization method. When the trees grow to maturity, the number of nodes in the hidden layer between the input layer factors is nine. After the model converges quickly, the error will reach the minimum value in nine steps, and the value will not change again in the later calculation.

The relationship between absolute error and relative error is shown in Fig. 9.

Fig. 9

Relationship between the absolute error and relative error of the model of the number of nodes in different hidden layers of mature forest.

4.4 Qualitative test of the optimal model

After the concrete model is constructed, the evaluation and detection of the model should include qualitative test in addition to the quantitative detection of the accuracy of the model. The content of qualitative test is to compare the predicted results with the possibility of certain changes in the future and the known biological research results or the law of economic development to see if they are consistent.

5 Measures to improve economic development in forest areas

5.1 Optimizing economic development policies in forest areas

In the process of forestry economic development, it is necessary to further achieve sustainable and safe and effective development. In maintaining and managing the environment of tree-growing areas, the funds used can not only rely on state allocations. Forestry development departments in various regions should also attach importance to the development of forestry economy. The better the preferential policy is, the more relevant enterprises and some very professional technical personnel can be attracted to manage and manage the forest area. The introduction of advanced technology and talents can improve the quality of forestry economic development to a certain extent and create a lot of wealth.

5.2 mproving the development technology and management level of forestry economy

With the continuous development of economy and society, the socialist market mechanism of our country is constantly improving. In the process of developing forestry economy, if we want to have a certain development prospect and occupy a certain position in the market, We should constantly observe the changing trend of the market, adjust our development direction and improve our competitive strength.

5.3 Sustaining sustainable development strategies

In the process of developing our country’s economy, we should not only pursue the speed of economic development, but also maintain the balance of ecological development. Therefore, in the process of management, forestry workers should be encouraged to combine development with ecological protection. Actively learn related forestry development and environmental protection technology, constantly improve the quality of management.

6 Conclusion

The batch evaluation of forest resources assets is to use the same data to establish an automatic evaluation model in a certain period of time, and to put all forest resources in the same evil market. Or in a market with the same characteristics as the market to evaluate the data and assets. In the traditional evaluation process, only one influencing factor can be evaluated, and the new method can fundamentally improve the efficiency and quality of the evaluation. By connecting with the construction process of BP neural network and fuzzy model, this paper helps to evaluate the assets of forest resources, and analyzes the influence of different influence factors on the size of the final evaluation value. Finally, the optimal evaluation model of trees in different growth years is obtained.

Footnotes

Acknowledgments

This paper was supported by (1) The project of research and planning fund for Humanities and social sciences of the Ministry of Education (17yja630094) “Research on the transformation mode and path of forestry industry in Northeast State-owned Forest Region Based on industrial ecosystem”; (2) Fund project of State Forestry Administration (jyc2016-47) “Research on Problems and Countermeasures of state-owned forest reform”.

References

Pathak

, Pakray

and Das

, Context guided retrieval of math formulae from scientific documents, Journal of Information and Optimization Sciences 40(8) (2019), 1559–1574.

Pathak

, Pakray

and Gelbukh

, Binary vector transformation of math formula for mathematical information retrieval, &Fuzzy Systems 36(5) (2019), 4685–4695.

Guidi

and Coen

C.S.

, A survey on retrieval of mathematical knowledge, Mathematics in Computer Science 10(4) (2016), 409–427.

Sahi

and Gupta

, A novel technique for detecting plagiarism in documents exploiting information sources, Cognitive Computation 9(6) (2017), 852–867.

Baba

, Nakatoh

and Minami

, Plagiarism detection using document similarity based on distributed representation, Procedia Computer Science 111 (2017), 382–387.

Kenter

and De Rijke

, Short text similarity with word embeddings, Proceedings of the 24th ACM International on Conference on Information and Knowledge Management. ACM 72(12) (2015), 1411–1420.

, Zhou

, Ye

, et al., Math Expression Image Retrieval via Attention-Based Framework, Proceedings of the 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI). IEEE 2019, 259–264.

Davila

, Joshi

, Setlur

, et al., Tangent-V: Math Formula Image Search Using Line-of-Sight Graphs, Proceedings of the European Conference on Information Retrieval. Springer, Cham, 2019, 681–695.

Alvaro

, Sánchez

J.A.

and Benedí

J.M.

, An integrated grammar-based approach for mathematical expression recognition, Pattern Recognition 51 (2016), 135–147.

10.

Youssef

A.S.

, Methods of relevance ranking and hit-content generation inmath search, Towards Mechanized Mathematical Assistants. Springer, Berlin, Heidelberg, 2007, 393–406.

11.

Shatnawi

and Youssef

A.S.

, Equivalence detection using parse-tree normalization for math search, Proceedings of the 2007 2nd International Conference on Digital Information Management. IEEE 2 (2007), 643–648.

12.

Miner

and Munavalli

, An approach to mathematical search through query formulation and data normalization, Towards Mechanized Mathematical Assistants. Springer, Berlin, Heidelberg, 2007, 342–355.

13.

Egomath [EB/OL]. http://egomath.projekty.ms.mff.cuni.cz, 2019. 12. 4.

14.

Mišutka

and Galamboš

, Mathematical extension of full text search engine indexer, Proceedings of the 2008 3rd International Conference on Information and Communication Technologies: From Theory to Applications. IEEE, 2008, 1–6.

15.

Mišutka

and Galamboš

, Extending full text search engine for mathematical content, Towards Digital Mathematics Library. Birmingham, United Kingdom, July 27th, 2008 (2008), 55–67.

16.

Mišutka

and Galamboš

, Systemdescription: Egomath2 as a tool for mathematical searching on wikipedia. org, Proceedings of the International Conference on Intelligent Computer Mathematics. Springer, Berlin, Heidelberg, 43(2) (2011), 307–309.

17.

Culpepper

J.S.

and Moffat

, Efficient set intersection for inverted indexing, ACM Transactions on Information Systems (TOIS) 29(1) (2010), 1–25.

18.

Stalnaker

and Zanibbi

, Math expression retrieval using an inverted index over symbol pairs, Proceedings of the Document Recognition and Retrieval XXII, International Society for Optics and Photonics 9402 (2015), 940207.

19.

Tian

, Yang

, Li

, et al., An indexing method of mathematical expression retrieval, Proceedings of 2013 3rd International Conference on Computer Science and Network Technology. IEEE 54(6) (2013), 574–578.

20.

Yang

and Tian

, A maintenance algorithm of FDS based mathematical expression index, Proceedings of the 2014 International Conference on Machine Learning and Cybernetics. IEEE 2 (2014), 888–892.

21.

Prodescu

C.C.

and Kohlhase

, Math Web Search 0.5 An Open Formula Search Engine, Proceedings of the LWA, 2011, 257–264.

22.

Sojka

and Líška

, Indexing and searching mathematics in digital libraries, Proceedings of the International Conference on Intelligent Computer Mathematics. Springer, Berlin, Heidelberg, 2011, 228–243.

23.

Asebriy

, Raghay

and Bencharef

, An Assistive Technology for Braille Users to Support Mathematical Learning: A Semantic Retrieval System, Symmetry 10(11) (2018), 547.

24.

Schubotz

, Meuschke

, Hepp

, et al., VMEXT: a visualization tool for mathematical expression trees, Proceedings of the International Conference on Intelligent Computer Mathematics. Springer, Cham, 2017, 340–355.

25.

, Gao

, Lin

, et al., Wikimirs: a mathematical information retrieval system for wikipedia, Proceedings of the 13th ACM/IEEE-CS Joint Conference on Digital Libraries. ACM, 2013, 11–20.

26.

Wang

, Gao

, Wang

, et al., Wiki Mirs 3.0: a hybrid MIR system based on the context, structure and importance of formulae in a document, Proceedings of the 15th ACM/IEEE-CS Joint Conference on Digital Libraries. ACM, 2015, 173–182.

27.

Lin

, Gao

, Hu

, et al., Amathematics retrieval system for formulae in layout presentations, Proceedings of the 37th International ACM SIGIR Conference on Research & Development in Information Retrieval. ACM, 2014, 697–706.

28.

Shi

, Wang

, Lin

C.Y.

, et al., Automatically solving number word problems by semantic parsing and reasoning, Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing 12(3) (2015), 1132–1142.

29.

, Su

, Cheng

, et al., N-gram Index Structure Study for Semantic Based Mathematical Formula, Proceedings of the 2014 Tenth International Conference on Computational Intelligence and Security. IEEE 55(13) (2014), 293–298.

A	B1	B2	B3	B4
B1	B11	B12	B13	B14
B2	B21	B22	B23	B24
B3	B31	B32	B33	B34
B4	B41	B42	B43	B44

A	B1	B2	B3	B4
B1	B11	B12	B13	B14
B2	B21	B22	B23	B24
B3	B31	B32	B33	B34
B4	B41	B42	B43	B44

Regional forestry economic evaluation based on neural network and fuzzy model

Abstract

Keywords

1 Introduction

2 Related work

3 Model design based on neural network and fuzzy model

3.1 BP Neural network model establishment

3.1.1 Network operation steps

3.2.1 Alternative factor determination

3.2.2 Sensitivity analysis and input layer factor screening

3.3.1 Mean absolute error

4 Modeling results and analysis

4.1 Batch evaluation model of forest and wood resources in young age

4.1.1 Model structure optimization

Table 1 The model of matrix A B1 B2 B3 B4 B1 B11 B12 B13 B14 B2 B21 B22 B23 B24 B3 B31 B32 B33 B34 B4 B41 B42 B43 B44

4.2.1 Model structure optimization

4.3.1 Model structure optimization

5 Measures to improve economic development in forest areas

5.1 Optimizing economic development policies in forest areas

5.2 mproving the development technology and management level of forestry economy

5.3 Sustaining sustainable development strategies

6 Conclusion

Footnotes

Acknowledgments

References

Table 1
The model of matrix

A B1 B2 B3 B4

B1 B11 B12 B13 B14

B2 B21 B22 B23 B24

B3 B31 B32 B33 B34

B4 B41 B42 B43 B44

A	B1	B2	B3	B4
B1	B11	B12	B13	B14
B2	B21	B22	B23	B24
B3	B31	B32	B33	B34
B4	B41	B42	B43	B44