Urban green travel level index system construction and evaluation method

Abstract

The invention provides an urban green travel level index system construction and evaluation method, and the method comprises the steps: S1, extracting green travel indexes of each city, and building a green travel index system; selecting a plurality of indexes from comprehensive evaluation, infrastructure, transportation equipment, operation service and capability construction, and determining a green travel index system; s2, carrying out correlation analysis on each index, and carrying out primary screening on the indexes; s3, calculating the relative weight of each index, and identifying a key index; s4, performing normalization processing, and calculating expected output of the sample data; s5, constructing a green travel evaluation model, and evaluating a green travel effect; and optimizing the weight and the threshold value of the BP neural network by using a white shark optimization algorithm, constructing a green travel evaluation model, and evaluating the green travel effect. The urban green travel evaluation system provided by the invention quantitatively evaluates the development level and effect of urban green travel.

Description

Technical Field

[0001] This application relates to the technical field of green travel, and particularly to a method for constructing an index system and evaluating the level of urban green travel. Background Art

[0002] Urban transportation, as the artery of urban development, has a serious impact on the environment due to the carbon emissions it generates. Green travel is a civilized behavior that saves resources and protects the environment, and it is the consensus of social development. Clearly defining the development stage of green travel, judging the development level of green travel, and making scientific responses are the fundamental ways to reduce carbon emissions from urban transportation.

[0003] The research on green travel evaluation methods provides guidance for scientifically formulating urban green travel development goals and planning schemes. There are few existing research results on green travel evaluation, and they mainly focus on green transportation research. The index system is large and broad, the index screening lacks correlation analysis, the index weights are mostly determined qualitatively, with strong subjectivity. At the same time, the evaluation methods mostly use single models, the evaluation effect is unstable, and the guidance for urban green travel development is not strong.

[0004] Therefore, proposing a method for constructing an index system and evaluating the level of urban green travel, clarifying the relationship between indicators, scientifically establishing an evaluation index system, and scientifically judging the level of urban green travel are urgent problems to be solved at present. Summary of the Invention

[0005] This application provides a method for constructing an index system and evaluating the level of urban green travel, which can be used to solve the technical problem of the lack of a scientific evaluation method for green travel at present.

[0006] This application provides a method for constructing an index system and evaluating the level of urban green travel, and the method includes:

[0007] Step S1: Extract green travel indicators for each city and build a green travel index system; select multiple indicators from comprehensive evaluation, infrastructure, transportation equipment, operation services, and capacity building to determine the green travel index system;

[0008] Step S2: Conduct a correlation analysis on each indicator and conduct a preliminary screening of the indicators;

[0009] For the constructed index system, use the SHAP model to analyze the correlation between indicators;

[0010] Step S3: Calculate the relative weights of each indicator and identify key indicators;

[0011] The weights of the indicators are confirmed by two methods: the Analytic Hierarchy Process and the Entropy Weight Method. To ensure the comprehensiveness of the weight assignment method, the combined weight method is used. Drawing on the idea of game theory, the balance point between subjective and objective weights is found, and the weights of the indicators are comprehensively determined to identify the key indicators.

[0012] Step S4: Normalization processing, calculating the expected output of sample data;

[0013] The original data is normalized using the min-max normalization function, and the expected output values of each indicator are determined in combination with the weights in step S3.

[0014] Step S5: Construct a green travel evaluation model to evaluate the effect of green travel;

[0015] Utilize the characteristics of the white shark optimization algorithm, such as strong optimization ability and fast convergence speed, to optimize the weights and thresholds of the BP neural network, construct a green travel evaluation model, and evaluate the effect of green travel.

[0016] Furthermore, the method for constructing the green travel indicator system in S1 is as follows:

[0017] S11: Construct an indicator library:

[0018] Sort out the existing indicators for green travel. The indicator types include infrastructure, transportation equipment, operation services, capacity building, and comprehensive evaluation, initially forming a three-level indicator system; among them, infrastructure and operation services account for more than 81%, and public transportation facilities account for 30% in the infrastructure monitoring indicators.

[0019] In the field of operation service monitoring, there are lacks of monitoring indicators for slow travel, rail, etc., and there are also lacks of monitoring indicators for safe operation services.

[0020] S12: Construction of the indicator system framework:

[0021] The indicator system is divided into a target layer and three-level indicators. Study the composition and influencing factors of the green travel structure, and construct the indicator system framework.

[0022] S13: Selection of statistical indicators:

[0023] After completing the construction of the alternative indicator library, the logical analysis method is used to preliminarily associate the specific monitoring indicators in the alternative indicator library with the secondary indicators of the green travel indicator system, construct a green travel evaluation logic tree, and select multiple important indicators from comprehensive evaluation, infrastructure, transportation equipment, operation services, and capacity building to construct the green travel indicator system:

[0024] The comprehensive evaluation indicators include the proportion of green travel, the sharing rate of urban public transportation motorized trips, the satisfaction rate of green travel services, and the decline rate of urban public transportation carbon emission intensity;

[0025] Infrastructure indicators include the coverage rate of urban rail transit stations and bus and trolleybus stations, the mileage of bus-only lanes per 100 standard buses, the connection rate of transportation transfer facilities around urban rail transit stations and bus and trolleybus transfer stations, the opening rate of adjacent buses, and the proportion of sidewalks;

[0026] Transport equipment indicators include the proportion of new energy and clean energy buses, the proportion of newly added and updated new energy and clean energy buses, and the proportion of newly added and updated low-floor and low-entry urban buses;

[0027] Operation service indicators include the average operation speed of buses and trolleybuses during the morning and evening rush hours, the bus network density, the proportion of feeder and micro lines in regular buses, the proportion of customized lines, the on-time rate of public transportation, the real-time prediction rate of arrival information of buses and trolleybuses, the application of the green travel information platform, and the death rate of public transportation operation liability accidents;

[0028] Capacity building indicators include establishing a cross-departmental and cross-field green travel coordination mechanism and organizing theme publicity activities such as green travel and bus travel every year.

[0029] Furthermore, the method for analyzing the correlation between indicators by the SHAP method in step S2 includes:

[0030] The SHAP method uses an additive feature attribution method to generate an interpretable model. Assuming that the input variables of the model are X = (x 1 , x 2 ,..., x L ), then the interpretation model g(x') of the original model f(x) is expressed as:

[0031]

[0032] In the formula, φ 0 represents the constant value when there is no input, φ i represents the contribution value of feature i, x' represents the simplified input variable, N represents the number of input features, L{i} is the set of all input features, H is the set of all possible combinations, and |H| represents the number of non-zero terms in H.

[0033] Furthermore, in step S3, the weights are extracted by the analytic hierarchy process, entropy weight method, and comprehensive weighting method, including:

[0034] Step S31: Calculate the weights of the green travel indicators initially screened by the SHAP method in step S2 using the analytic hierarchy process;

[0035] Step S32: Calculate the weights of the green travel indicators initially screened by the SHAP method in step S2 using the entropy weight method;

[0036] Step S33: Calculate the weights of the green travel indicators initially screened in Step S2 using the SHAP method by means of the comprehensive weighting method;

[0037] Step S34: Calculate the mean of the weights of the green travel indicators calculated by the three methods in Step S31, Step S32, and Step S33 to obtain the comprehensive indicator weight.

[0038] Furthermore, Step S31: Calculate the weights of the green travel indicators initially screened in Step S2 using the SHAP method by means of the analytic hierarchy process, including:

[0039] Suppose to compare the influence magnitudes of n factors X = {x 1 , …, x n} on a certain factor Z. To provide more credible data, the method of establishing a pairwise comparison matrix for the factors is adopted; that is, each time two different factors x i and x j are taken, and a ij represents the ratio of the influence degrees of x i and x j on the target. Among them, all comparison results are represented by the matrix C = (a ij ) n*n . C is the judgment matrix between Z - X; if the ratio of the response degrees of x i and x j to Z is a ij , then the ratio of the influences of x i and x j on Z should be a ji = 1 / a ij . The value of a ij is scaled by referring to the numbers 1 to 9 and their reciprocals;

[0040] Table 1 Assignment criteria for elements in the judgment matrix

[0041] <![CDATA[Judgment scale (a ij )]]> Definition 1 <![CDATA[x i and x j are equally important]]> 3 <![CDATA[x i is more important than x j slightly 5 <![CDATA[x i is more important than x j significantly]]> 7 <![CDATA[x i more important than x j strongly important]]> 9 <![CDATA[x i than x j is extremely important]]> 2、4、6、8 At the intermediate value between the above two adjacent comparison scales

[0042] Conduct n(n - 1) / 2 pairwise judgments; conducting n(n - 1) / 2 judgments can obtain more information at various levels. In this way, through the repeated comparison of each index, an accurate ranking can be derived.

[0043] For the judgment matrix of the single sorting of the hierarchical indicators C corresponding to its largest eigenvalue λ max the obtained eigenquantity K, after normalization, the ranking weight value of the corresponding factors at the same level relative to a certain factor at the upper level is obtained, which is the single sorting of the hierarchy.

[0044] The steps for the consistency test of the judgment matrix are as follows:

[0045] Since the judgment matrix is the basis for finally calculating the weights, it is required that the matrix generally has consistency to avoid the unreasonable judgment of "A is significantly more important than B, B is significantly more important than C, and C is significantly more important than A", which violates the principle of authenticity of evaluation. Therefore, it is necessary to analyze the error and compatibility of the judgment matrix;

[0046] Let the consistency index be C.I (Consistency Index), that is:

[0047]

[0048] And find out the corresponding random consistency index R.I (Random Index), as shown in Table 2:

[0049] Table 2 R.I values corresponding to different matrix orders n

[0050] n 1 2 3 4 5 6 7 8 9 R.I 0 0 0.58 0.9 1.12 1.24 1.32 1.41 1.45

[0051] The consistency ratio is determined by the following method:

[0052]

[0053] Normally, if C.R (Consistency Ratio) < 0.1, the judgment matrix is a consistent matrix and the consistency value meets the expectation, otherwise the judgment matrix needs to be modified again;

[0054] The method for the total sorting of hierarchical indicators is as follows: Suppose the first upper layer A contains A 1 , …, A m , a total of m factors, and the weight values of the final hierarchical total sorting are a 1 , …, a m ; Also assume that the layer B behind the first layer A contains n elements B 1 , …, B n , and their single sorting hierarchical weight values regarding A j are b 1j , …, b nj respectively. When B i is not relevant to A j , b ij = 0; It is required to find the weights of each element in layer B for the total target layer A, that is, to find the weight values b 1 , …, b n of the hierarchical total sorting of each element in layer B. The calculation method is as follows:

[0055]

[0056] Conduct a consistency test on the overall hierarchical ranking. The test method is still the same as the previous single hierarchical ranking, and it is carried out step by step from the highest level to the lowest level for all levels; assume that among the elements in layer B of the criterion layer that are related to A j The judgment matrices of each relative comparison of the relevant elements have passed the consistency test in the final single ranking test, and the consistency hierarchical indicators C.I(j) of the single ranking are obtained, j = 1,..., m. The corresponding randomly selected consistency indicators are R.I(j) and C.I(j) respectively; R.I(j) has also been obtained during the single hierarchical ranking. Then the random consistency ratio of the final overall ranking of layer B is:

[0057]

[0058] When C.R < 0.1, it is considered that the result of the overall hierarchical ranking meets the expectation, and the weight value under the analytic hierarchy process is calculated.

[0059] Furthermore, step S32: Calculate the weights of the green travel indicators initially screened by the SHAP method in step S2 using the entropy weight method; including:

[0060] In different systems, the probability of each indicator appearing in each state is P i , i = 1, 2,..., m =, then the entropy of the system is:

[0061]

[0062] P i = 1 / m, i = 1, 2,..., m, that is, when the effects produced in multiple states are exactly the same, take the highest value; that is

[0063] e = -lnm

[0064] If there are m evaluation objects and n evaluation indicators now, and the initial decision matrix R = (r ij ) m×n , for a certain indicator r j :

[0065]

[0066] In the formula:

[0067]

[0068] Research shows that the magnitude of the entropy value is closely related to the change of the indicator. The smaller the entropy value of a specific indicator, the greater the variability of the indicator. If the entropy value of a specific indicator is larger, the change of the indicator is smaller; the more insufficient the basic data information is. After calculating the entropy value, calculate the entropy weight for evaluation.

[0069] The entropy weight method includes the following steps:

[0070] Suppose there are n evaluation objects and m evaluation indicators, and the generated initial decision matrix R = (r ij ) n×mm ;

[0071]

[0072] r ij represents the evaluation value of the j-th indicator of the i-th evaluation object;

[0073] Step S321, standardize the matrix;

[0074] Since the meanings represented by each indicator are different, the indicator evaluation criteria are different, and the indicator data is different, it is necessary to standardize the indicators. For safety evaluation, the larger the indicator value, the more representative it is, so the indicators with larger values do not need to be changed.

[0075] Step S322, the proportion P of the indicator eigenvalue of the i-th evaluation object under the j-th indicator ij :

[0076]

[0077] Step S323, the entropy value e under the j-th indicator j :

[0078]

[0079] where: k = 1 / ln n;

[0080] Step S324, calculate the entropy weight value W of each indicator j :

[0081]

[0082] The more uniform the distribution of the indicators or data, the larger the entropy value, and vice versa, that is, the greater the distribution difference between the indicators. The difference coefficient of the indicators is calculated as: d j = 1 - e j ;

[0083] Step S325, calculate the comprehensive evaluation value of each evaluation object:

[0084]

[0085] Obtain the green travel weight value based on the entropy weight method.

[0086] Furthermore, step S33: Calculate the weights of the green travel indicators initially screened in step S2 using the SHAP method by the comprehensive weighting method, including:

[0087] First, based on the subjective weight B of AHP i and the objective weight W of the entropy weight method i , introduce two linear coefficients δ 1 and δ 2 to construct the possible comprehensive weight V i , that is:

[0088] V i = δ 1 B i + δ 2 W i

[0089] Secondly, construct the objective function and constraint conditions, find the equilibrium point, and optimize the linear coefficients δ 1 and δ 2 to minimize the deviation between the combined comprehensive weight V i and B i and W i , that is

[0090] min(||V i - B i || 2 + ||V i - W i || 2 ) = min(||δ 1 B i + δ 2 W i - B i || 2 + ||δ 1 B i + δ 2 W i - W i || 2 )

[0091]

[0092] According to the properties of matrix differentiation, the condition for the optimal first derivative of the objective function to obtain the minimum value is

[0093]

[0094] Finally, normalize the optimized linear coefficients δ 1 and δ 2 to determine the optimal comprehensive weight of the green traffic evaluation index of the central city, that is:

[0095]

[0096] Among them, is the normalized subjective weight coefficient; is the normalized objective weight coefficient; is the comprehensive weight of evaluation index i in the evaluation index system.

[0097] Furthermore, in the normalization process of step S4, the method for calculating the expected output of sample data is as follows:

[0098] S41: Use the min-max normalization function to normalize the original data:

[0099]

[0100] x i represents the original value of the index, and y i represents the normalized index value. x max and x min respectively represent the maximum and minimum values obtained for the same index;

[0101] S42: After normalizing the sample data in S41, multiply the normalized sample data by the weight obtained in step S3 to calculate the expected output of the sample data.

[0102] Furthermore, the method for constructing the green travel evaluation model in step S5 is as follows:

[0103] S51: Data preprocessing, dividing the training samples and test samples;

[0104] S52: Determine the number of input layer neurons and output layer neurons of the evaluation model neural network according to the number of input features and the number of evaluation features;

[0105] S53: Set the initial weights and thresholds of the neural network;

[0106] S54: Use the training set as the input vector of the evaluation model, introduce the objective function, train the evaluation model, and introduce the white shark optimization algorithm to optimize the weights of the evaluation model;

[0107] The BP neural network propagates forward layer by layer, compares the final output obtained with the expected output, and adjusts the weights and thresholds through the gradient descent strategy to finally establish a neural network model with the output result within the allowable error range:

[0108]

[0109] Among them, w(m) represents the weight after the mth iteration, and η is the iteration step size, that is, the learning rate; the value of η largely determines the accuracy of the iteration result. If the value of η is too large or too small, the probability of finding the optimal solution will be reduced. To avoid errors caused by inappropriate selection of the η value, the white shark optimization algorithm is introduced to optimize the weights and thresholds of the evaluation model;

[0110] S541: Initialize the white shark population and randomly generate the individual positions of the white shark population that are uniformly distributed. That is, the weights of the evaluation model:

[0111]

[0112] In the formula is the position of the j-th dimension of the i-th white shark, i = 1, 2,..., n, where n is the size of the white shark population; u j , l j are the upper and lower limits of the j-th dimension search space respectively; R is a random number between 0 and 1;

[0113] S542: Velocity update. White sharks spend most of their time hunting and tracking prey. In the WSO algorithm, white sharks sense the position of the prey based on its movement and update their own velocity:

[0114]

[0115] In the formula: is the velocity of the i-th white shark at the (k + 1)-th iteration; is the velocity of the i-th white shark at the k-th iteration; is the global optimal position obtained by the white shark in the k-th iteration so far; is the position of the i-th white shark at the k-th iteration; The optimal position corresponding to the velocity of the i-th white shark at the k-th iteration; c 1 , c 2 are random numbers between 0 and 1; p 1 , p 2 are respectively control coefficients, described as k and K are the current iteration number and the maximum iteration number respectively; p max , p min are the maximum and minimum values of the control coefficients respectively; μ represents the contraction factor;

[0116] S543: Position update: 1) Update based on the position of the optimal prey. In the white shark optimization algorithm, white sharks update their own positions by moving towards the optimal prey to find the optimal or sub-optimal prey:

[0117]

[0118] In the formula: is the position of the i-th white shark at the (k + 1)-th iteration; is the negation operator; is the bit operator; a and b are one-dimensional binary vectors, described as sgn(-) is the sign return function; μ and l are the upper and lower limits of the search space; ω 0 is a logical vector, described as f is the wave motion frequency of the great white shark; m v is the power that increases with the number of iterations when the great white shark approaches its prey; the meanings of other parameters are the same as above;

[0119] 2) Update based on the position of the best great white shark. In the great white shark optimization algorithm, the great white shark moves towards the position of the optimal great white shark to approach the optimal position of the prey:

[0120]

[0121] In the formula: is the new position of the i-th great white shark relative to the prey at the (k + 1)-th iteration; r 1 and r 2 and r 3 are random numbers between 0 and 1; D k is the distance between the prey and the great white shark; S s is the olfactory and visual intensity parameter of the great white shark when approaching the best prey; the meanings of other parameters are the same as above.

[0122] S544: By simulating the feeding behavior of the great white shark group, the first two optimal solutions are retained, and the positions of other great white shark individuals are updated according to these optimal positions:

[0123]

[0124] S545: Through the iterative calculation of the great white shark optimization algorithm, the candidate solution corresponding to the optimal great white shark position obtained is the optimal weight of the evaluation model, which is substituted into the BP neural network to complete the training of the model;

[0125] S546: Input the green travel index to be evaluated into the trained model to obtain the corresponding evaluation value of the green travel index.

[0126] Quantitatively evaluate the development level and effectiveness of urban green travel through the urban green travel evaluation system, which can horizontally compare the gaps in the green travel development levels between different cities, thus contributing to the mutual learning and reference between cities; it can also vertically compare the green travel development status of a city at different times, timely adjust the current strategy, formulate targeted improvement measures, and help predict and master the future green travel development trend. Description of the Drawings

[0127] Figure 1 is the overall process schematic diagram of an index system construction and evaluation method for urban green travel level;

[0128] Figure 2It is a logical structure diagram for selecting indicators of an index system construction and evaluation method for the level of urban green travel;

[0129] Figure 3 It is an evaluation index system diagram of an index system construction and evaluation method for the level of urban green travel. Specific implementation manners

[0130] To make the purpose, technical solutions and advantages of this application clearer, the following will further describe the implementation manners of this application in detail with reference to the accompanying drawings.

[0131] First, the embodiments of this application will be introduced below with reference to the accompanying drawings.

[0132] To solve the above problems, the technical solutions adopted by this invention are as follows:

[0133] Step S1: Taking Jiangsu Province as an example, extract the green travel indicators of each city and build a green travel indicator system;

[0134] Select 22 indicators from aspects such as comprehensive evaluation, infrastructure, transportation equipment, operation service, and capacity building to determine the three-level indicator system of green travel in Jiangsu Province;

[0135] Step S2: Conduct a correlation analysis on each indicator and conduct a preliminary screening of the indicators;

[0136] For the constructed three-level indicator system, use the SHAP model to analyze the correlation between the indicators;

[0137] Step S3: Calculate the relative weights of each indicator and identify the key indicators;

[0138] Confirm the indicator weights through two methods: the Analytic Hierarchy Process (AHP) and the entropy weight method. To ensure the comprehensiveness of the weight assignment method, through the comprehensive weight assignment method, draw on the idea of game theory, find the balance point of subjective and objective weights, and comprehensively determine the indicator weights to identify the key indicators;

[0139] Step S4: Conduct normalization processing and calculate the expected output of the sample data;

[0140] Use the min-max normalization function to normalize the original data, and combine the weights in Step S3 to determine the expected output values of each indicator;

[0141] Step S5: Build a green travel evaluation model to evaluate the green travel effect;

[0142] Utilize the characteristics of the white shark optimization algorithm, such as strong optimization ability and fast convergence speed, to optimize the weights and thresholds of the BP neural network, build a green travel evaluation model, and evaluate the green travel effect.

[0143] The method for constructing the green travel index system in S1 is as follows:

[0144] S11: Construct an index library:

[0145] Sort out the existing green travel indicators. The indicator types include infrastructure, transportation equipment, operation services, capacity building, and comprehensive evaluation, initially forming an index library with a three-level index system and 82 specific monitoring indicators. Among them, the proportion of infrastructure and operation services exceeds 81%. Among the infrastructure monitoring indicators, the proportion of public transportation facilities is 30%. In the field of operation service monitoring, there are lacks of monitoring indicators for slow travel, rail transit, etc., and there are also lacks of monitoring indicators for safe operation services;

[0146] S12: Construction of the index system framework:

[0147] The index system is divided into a target layer and three-level indicators. By analyzing the connotation and basic theory of green travel, studying the composition and influencing factors of the green travel structure, construct the index system framework;

[0148] S13: Selection of statistical indicators:

[0149] After completing the construction of the alternative index library, use the logical analysis method to preliminarily associate the specific monitoring indicators in the alternative index library with the secondary indicators of the green travel index system, construct a green travel evaluation logic tree, and select multiple important indicators from comprehensive evaluation, infrastructure, transportation equipment, operation services, and capacity building to construct the green travel index system:

[0150] The comprehensive evaluation indicators include the proportion of green travel, the sharing rate of urban public transport motorized trips, the satisfaction rate of green travel services, and the decline rate of urban public transport carbon emission intensity;

[0151] The infrastructure indicators include the coverage rate of urban rail transit stations (800 meters) and bus and trolleybus stations (300 meters), the mileage of bus lanes per 100 standard buses, the connection rate of traffic transfer facilities around urban rail transit stations and bus and trolleybus transfer stations, the opening rate of adjacent buses, and the proportion of sidewalks;

[0152] The transportation equipment indicators include the proportion of new energy and clean energy buses, the proportion of newly added and updated new energy and clean energy buses, and the proportion of newly added and updated low-floor and low-entry urban buses;

[0153] The operation service indicators include the average operation speed of buses and trolleybuses during peak hours, the bus network density, the proportion of regular bus feeder and micro lines, the proportion of customized lines, the on-time rate of public transportation, the real-time forecast rate of bus arrival information, the application of the green travel information platform, and the death rate of public transportation operation liability accidents;

[0154] The capacity building indicators include establishing a cross-departmental and cross-field coordination mechanism for green travel and organizing theme publicity activities such as green travel and bus travel every year;

[0155] The green travel indicators emphasize the assessment of two aspects: the decline rate of carbon emissions intensity of urban public transport and the connection rate of traffic transfer facilities around urban rail transit stations and bus and trolleybus transfer stations. Among them, the decline rate of carbon emissions intensity of urban public transport is expressed as the reduction ratio of the carbon emissions intensity of the urban public transport system relative to the base period within a specific time period. Carbon emissions intensity is usually expressed as the carbon emissions generated per unit of transport turnover or per unit of operating mileage, such as kg CO₂ equivalent / person-km or kg CO₂ equivalent / vehicle-km, etc.; the connection rate of traffic transfer facilities around urban rail transit stations and bus and trolleybus transfer stations is expressed as the sum of the number of urban rail transit station entrances with bus and trolleybus stops within a transfer distance of 100 meters and the number of bus and trolleybus transfer stations with bicycle stops within a transfer distance of 50 meters divided by the total number of urban rail transit station entrances and bus and trolleybus transfer stations. Among them, bus and trolleybus transfer stations include bus stops, bus terminals, and bus hubs that converge 3 or more bus stops.

[0156] The method for analyzing the correlation between indicators by the SHAP method in step S2 is as follows:

[0157] The SHAP method uses an additive feature attribution method to generate an interpretable model. Assuming that the input variables of the model are X = (x 1 , x 2 ,..., x L ), then the interpretation model g(x') of the original model f(x) is expressed as:

[0158]

[0159] In the formula, φ 0 represents the constant value when there is no input, φ i represents the contribution value of feature i, x' represents the simplified input variable, N represents the number of input features, L{i} is the set of all input features, H is the set of all possible combinations, and |H| represents the number of non-zero terms in H.

[0160] In step S3, the method for extracting weights by the Analytic Hierarchy Process (AHP), entropy weight method, and comprehensive weighting method is as follows:

[0161] Step S31: Calculate the weights of the green travel indicators initially screened in step S2 using the SHAP method by the Analytic Hierarchy Process.

[0162] Suppose we now want to compare n factors X = {x 1 , …, xn To determine the influence magnitude of a certain factor Z, and to provide relatively reliable data, a method of pairwise comparison of factors to establish a pairwise comparison matrix is adopted; that is, each time two different factors x i and x j are taken, and a ij represents the ratio of the influence degrees of x i and x j on the target. Among them, all comparison results are represented by the matrix C = (a ij ), n*n where C is the judgment matrix between Z and X; if the ratio of the response degrees of x i and x j to Z is a ij , then the ratio of the influence of x i and x j on Z should be a ji = 1 / a ij . The value of a ij is scaled by referring to the numbers from 1 to 9 and their reciprocals;

[0163] Table 1 Assignment criteria for elements in the judgment matrix

[0164] <![CDATA[Judgment scale (a ij )]]> Definition 1 <![CDATA[x i and x j are equally important]]> 3 <![CDATA[x i more important than x j slightly important]]> 5 <![CDATA[x i is more important than x j obviously important]]> 7 <![CDATA[x i is more important than x j strongly important]]> 9 <![CDATA[x i than x j extremely important]]> 2、4、6、8 At the intermediate value between the above two adjacent comparison scales

[0165] Perform n(n - 1) / 2 pairwise judgments; performing n(n - 1) / 2 judgments can obtain more information at various levels. In this way, through repeated comparisons of various indicators, an accurate ranking can be derived.

[0166] For the single - ranking judgment matrix C of the hierarchical indicators corresponding to its largest eigenvalue λ max , the obtained characteristic quantity K, after normalization, gives the ranking weight value of the corresponding factors at the same level relative to a certain factor at the upper level, which is the single - ranking of the hierarchy.

[0167] The steps for consistency checking of the judgment matrix are as follows:

[0168] Since the judgment matrix is the basis for finally calculating the weights, it is required that the matrix generally has consistency to avoid judgments that violate common sense such as "A is significantly more important than B, B is significantly more important than C, and C is significantly more important than A", which violates the principle of authenticity of evaluation. Analyze the error and compatibility of the judgment matrix;

[0169] Let the consistency index be C.I (Consistency Index), that is:

[0170]

[0171] And check the corresponding random consistency index R.I (Random Index), as shown in Table 2:

[0172] Table 2 R.I. values corresponding to different matrix orders n

[0173] n 1 2 3 4 5 6 7 8 9 R.I 0 0 0.58 0.9 1.12 1.24 1.32 1.41 1.45

[0174] The consistency ratio is determined by the following method:

[0175]

[0176] Normally, if C.R (Consistency Ratio) < 0.1, the matrix is judged as a consistent matrix and the consistency value meets the expectation; otherwise, the matrix needs to be modified again.

[0177] The method for the total sorting of hierarchical indicators is as follows: Suppose the first upper layer, layer A, contains A 1 , …, A m , a total of m factors, and the weight values of the final total sorting of the hierarchy are a 1 , …, a m ; Also assume that the layer B behind the first layer A contains n elements B 1 , …, B n , and their weight values of the final single sorting hierarchy regarding A j are b 1j , …, b nj respectively. When B i is not relevant to A j , b ij = 0; It is required to find the weights of each element in layer B for the overall goal layer A, that is, to find the weight values b 1 , …, b n of the total sorting of each element in layer B. The calculation method is as follows:

[0178]

[0179] Perform a consistency test on the total sorting of the hierarchy. The test method is still the same as the previous single sorting of the hierarchy, and it is carried out step by step from the high level to the low level for all levels; Suppose the elements in layer B of the criterion layer that are relevant to A j have passed the consistency test in the final single sorting test of each relative comparison judgment matrix, and the consistency hierarchical indicators C.I(j) of the single sorting are obtained, where j = 1, …, m, and the corresponding randomly selected average consistency indicators are R.I(j), C.I(j); R.I(j) has also been obtained during the single sorting of the hierarchy. Then the random consistency ratio of the final total sorting of layer B is:

[0180]

[0181] When C.R < 0.1, it is considered that the total hierarchical sorting result meets the expectation, and the weight value under the Analytic Hierarchy Process (AHP) is calculated.

[0182] Step S32: Calculate the weights of the green travel indicators initially screened in Step S2 using the SHAP method by the entropy weight method.

[0183] In different systems, the probability of each indicator appearing in each state is P i , when i = 1, 2,..., m =, the entropy of the system is:

[0184]

[0185] P i = 1 / m, i = 1, 2,..., m, that is, when the effects generated in multiple states are exactly the same, take the highest value; that is

[0186] e = -lnm

[0187] If there are now m evaluation objects and n evaluation indicators, and the initial decision matrix R = (r ij ) m×n , for a certain indicator r j :

[0188]

[0189] In the formula:

[0190]

[0191] The magnitude of the entropy value is closely related to the change of the indicator. The smaller the entropy value of a specific indicator, the greater the variability of the indicator. If the entropy value of a specific indicator is larger, the change of the indicator is smaller, and the basic data information is less sufficient. After calculating the entropy value, calculate the entropy weight for evaluation.

[0192] The entropy weight method includes the following steps:

[0193] Suppose there are n evaluation objects and m evaluation indicators, and the initial decision matrix R = (r ij ) n×mm ;

[0194]

[0195] r ij represents the evaluation value of the jth indicator of the ith evaluation object;

[0196] Step S321, standardize the matrix;

[0197] Since the meanings represented by each indicator are different, the indicator evaluation criteria vary, and the indicator data are different, it is necessary to standardize the indicators. For safety evaluation, the larger the indicator value, the more representative it is. Therefore, indicators with larger values do not need to be modified.

[0198] Step S322, the proportion P of the indicator eigenvalue of the i-th evaluation object under the j-th indicator ij :

[0199]

[0200] Step S323, the entropy value e under the j-th indicator j :

[0201]

[0202] where: k = 1 / ln n;

[0203] Step S324, calculate the entropy weight value W of each indicator j :

[0204]

[0205] The more uniform the distribution of the indicators or data, the larger the entropy value. Conversely, the smaller the entropy value, that is, the greater the distribution difference between the indicators. The difference coefficient of the indicators is calculated as: d j = 1 - e j ;

[0206] Step S325, calculate the comprehensive evaluation value of each evaluation object:

[0207]

[0208] Obtain the green travel weight value based on the entropy weight method;

[0209] Step S33: Calculate the weights of the green travel indicators initially screened in Step S2 using the SHAP method by the comprehensive weighting method.

[0210] First, the subjective weight B based on AHP i and the objective weight W of the entropy weight method i , introduce two linear coefficients δ 1 and δ 2 to construct the possible comprehensive weight V i , that is:

[0211] V i = δ 1 B i + δ 2 W i

[0212] Secondly, construct the objective function and constraints, find the equilibrium point, and optimize the linear coefficient δ 1 and δ 2 to minimize the deviation between the combined comprehensive weight V i and B i and W i , that is

[0213] min(||V i -B i || 2 +||V i -W i || 2 ) = min(||δ 1 B i +δ 2 W i -B i || 2 +||δ 1 B i +δ 2 W i -W i || 2 )

[0214]

[0215] According to the properties of matrix differentiation, the condition for the optimal first derivative of the objective function to obtain the minimum value is

[0216]

[0217] Finally, normalize the optimized linear coefficients δ 1 and δ 2 to determine the optimal comprehensive weight of the green traffic evaluation index of the central city, that is

[0218]

[0219] where is the normalized subjective weight coefficient; is the normalized objective weight coefficient; is the comprehensive weight of the evaluation index i in the evaluation index system.

[0220] Step S34: Calculate the mean value of the green travel index weights calculated by the three methods in Step S31, Step S32, and Step S33 to obtain the comprehensive index weight.

[0221] In Step S4, the method for normalizing and calculating the expected output of the sample data is as follows:

[0222] S41: Normalize the original data using the min-max normalization function:

[0223]

[0224] x i represents the original value of the index, and y i represents the normalized index value, x max and x min respectively represent the maximum and minimum values obtained for the same index;

[0225] S42: After normalizing the sample data in S41, multiply the normalized sample data by the weights obtained in step S3 to calculate the expected output of the sample data.

[0226] The method for constructing the green travel evaluation model in step S5 is as follows:

[0227] S51: Data preprocessing, dividing the training samples and test samples;

[0228] S52: Determine the number of input layer neurons and output layer neurons of the evaluation model neural network according to the number of input features and the number of evaluation features;

[0229] S53: Set the initial weights and thresholds of the neural network;

[0230] S54: Use the training set as the input vector of the evaluation model, introduce the objective function, train the evaluation model, and introduce the white shark optimization algorithm to optimize the weights of the evaluation model;

[0231] The BP neural network propagates forward layer by layer, compares the final output obtained with the expected output, and adjusts the weights and thresholds through the gradient descent strategy to finally establish a neural network model with the output result within the allowable error range:

[0232]

[0233] Among them, w(m) represents the weight after the mth iteration, and η is the iteration step size, that is, the learning rate; the value of η largely determines the accuracy of the iteration result. If the value of η is too large or too small, the probability of finding the optimal solution will be reduced. To avoid the error caused by inappropriate selection of the η value, the white shark optimization algorithm is introduced to optimize the weights and thresholds of the evaluation model;

[0234] S541: Initialize the white shark population and randomly generate the individual positions of the white shark population uniformly That is, the weights of the evaluation model:

[0235]

[0236] In the formula is the position of the $i$-th great white shark in the $j$-th dimension, where $i = 1, 2, \ldots, n$, and $n$ is the size of the great white shark population; $u$ j , $l$ j are the upper and lower limits of the $j$-th dimension search space respectively; $R$ is a random number between $0$ and $1$;

[0237] S542: Velocity update. Great white sharks spend most of their time hunting and tracking prey. In the WSO algorithm, great white sharks sense the position of prey based on its movement and update their own velocity:

[0238]

[0239] In the formula: is the velocity of the $i$-th great white shark at the $(k + 1)$-th iteration; is the velocity of the $i$-th great white shark at the $k$-th iteration; is the global optimal position obtained by the great white shark in the $k$-th iteration so far; is the position of the $i$-th great white shark at the $k$-th iteration; The optimal position corresponding to the velocity of the $i$-th great white shark at the $k$-th iteration; $c$ 1 , $c$ 2 are random numbers between $0$ and $1$; $p$ 1 , $p$ 2 are respectively control coefficients, described as $k$ and $K$ are the current iteration number and the maximum iteration number respectively; $p$ max , $p$ min are the maximum and minimum values of the control coefficient respectively; $\mu$ represents the contraction factor;

[0240] S543: Position update: (1) Update based on the position of the optimal prey. In the great white shark optimization algorithm, great white sharks update their own position by moving towards the optimal prey to find the optimal or sub-optimal prey:

[0241]

[0242] In the formula: is the position of the $i$-th great white shark at the $(k + 1)$-th iteration; is the negation operator; is the bitwise operator; $a$ and $b$ are one-dimensional binary vectors, described as $\text{sgn}(-)$ is the sign return function; $\mu$ and $l$ are the upper and lower limits of the search space respectively; $\omega$ 0 is a logical vector, described as $f$ is the great white shark wave motion frequency; $m$ v is the power of the great white shark increasing with the iteration number when approaching the prey; The meanings of other parameters are the same as above;

[0243] (2) Update based on the position of the best great white shark. In the great white shark optimization algorithm, the great white shark moves towards the position of the optimal great white shark to approach the optimal position of the prey:

[0244]

[0245] In the formula: is the new position of the i-th great white shark relative to the prey at the (k + 1)-th iteration; r 1 、r 2 、r 3 are random numbers between 0 and 1; D k is the distance between the prey and the great white shark; S s is the olfactory and visual intensity parameter of the great white shark when approaching the best prey; the meanings of other parameters are the same as above.

[0246] S544: By simulating the feeding behavior of the great white shark group, the first two optimal solutions are retained, and the positions of other great white shark individuals are updated according to these optimal positions:

[0247]

[0248] S545: Through the iterative calculation of the great white shark optimization algorithm, the candidate solution corresponding to the optimal great white shark position is the optimal weight of the evaluation model, which is substituted into the BP neural network to complete the training of the model;

[0249] S546: Input the green travel indicators to be evaluated into the trained model to obtain the corresponding evaluation values of the green travel indicators.

[0250] The above-described embodiments of the present application do not constitute a limitation on the protection scope of the present application.

Claims

1. An index system construction and evaluation method for urban green travel level, characterized in that: The method comprises: Step S1: Extract green travel indicators of each city and build a green travel indicator system; select multiple indicators from comprehensive evaluation, infrastructure, transportation equipment, operation services, and capacity building to determine the green travel indicator system; Step S2: Conduct correlation analysis on each indicator and conduct initial screening of the indicators; For the constructed indicator system, the SHAP model is used to analyze the correlation between indicators; Step S3: Calculate the relative weight of each indicator and identify the key indicators; Confirm the indicator weights and identify key indicators through the hierarchical analysis method and entropy weight method; Step S4: normalization processing, calculating the expected output of sample data; The min-max normalization function is used to normalize the original data, and the expected output value of each indicator is determined in combination with the weight in step S3; Step S5: construct a green travel evaluation model to evaluate the green travel effect; The White Shark optimization algorithm is used to optimize the weights and thresholds of the BP neural network, a green travel evaluation model is constructed, and the green travel effect is evaluated.

2. The method according to claim 1, characterized in that: The method for constructing the green travel index system in S1 is as follows: S11: Build indicator library: The existing indicators of green travel have been sorted out, including five types of indicators: infrastructure, transportation equipment, operation services, capacity building and comprehensive evaluation, which initially constitute a three-level indicator system; among them, infrastructure and operation services account for more than 81%, and public transportation facilities account for 30% of the infrastructure monitoring indicators. S12: Construction of indicator system framework: The index system is divided into a target layer and three-level indicators to study the green travel structure and influencing factors and build an index system framework; S13: Selection of statistical indicators: After completing the construction of the candidate indicator library, the specific monitoring indicators in the candidate indicator library are preliminarily associated with the secondary indicators of the green travel indicator system using the logical analysis method to build a green travel evaluation logic tree. Multiple important indicators are selected from comprehensive evaluation, infrastructure, transportation equipment, operation services, and capacity building to build a green travel indicator system: Comprehensive evaluation indicators include green travel ratio, urban public transportation motorized travel share, green travel service satisfaction rate, and urban public transportation carbon emission intensity reduction rate; Infrastructure indicators include the coverage rate of urban rail transit stations and public bus and tram stations, the mileage of bus lanes per 100 standard buses, the connection rate of transportation transfer facilities around urban rail transit stations and public bus and tram transfer stations, the rate of adjacent public transportation opening, and the proportion of sidewalks. Transport equipment indicators include the proportion of new energy and clean energy buses, the proportion of new and updated new energy and clean energy buses, and the proportion of new and updated low-floor and low-entry city buses; Operational service indicators include the average operating speed of public buses and trams during morning and evening peak hours, bus line density, the proportion of conventional bus branches and mini-lines, the proportion of customized lines, public transportation punctuality rate, real-time forecast rate of public bus and tram arrival information, application of green travel information platform, and public transportation driving accident fatality rate; Capacity building indicators include establishing a cross-departmental and cross-field green travel coordination mechanism, and organizing annual thematic publicity activities such as green travel and public transportation.

3. The method according to claim 1, characterized in that The method of analyzing the correlation between the indicators by the SHAP method in step S2 includes: The SHAP method uses the additive feature attribution method to generate an interpretable model. Assume that the input variables of the model are X = (x1, x2, ..., x L ), then the explanation model g(x') of the original model f(x) is expressed as: Where φ0 represents the constant value when there is no input, φ i represents the contribution value of feature i, x' represents the simplified input variable, N represents the number of input features, L{i} is the set of all input features, H represents all possible combinations, and |H| represents the number of non-zero items in H.

4. The method according to claim 3, characterized in that In step S3, weights are extracted by using the analytic hierarchy process, entropy weight method and comprehensive weighting method, including: Step S31: Calculate the weight of the green travel index obtained by the initial screening using the SHAP method in step S2 using the analytic hierarchy process; Step S32: Calculate the weight of the green travel index obtained by the initial screening using the SHAP method in step S2 using the entropy weight method; Step S33: Calculate the weight of the green travel index obtained by the initial screening using the SHAP method in step S2 using a comprehensive weighting method; Step S34: Calculate the average of the green travel index weights calculated by the three methods of step S31, step S32, and step S33 to obtain a comprehensive index weight.

5. The method according to claim 4, characterized in that Step S31: Calculate the weight of the green travel index obtained by the initial screening using the SHAP method in step S2 using the analytic hierarchy process, including: Suppose we compare n factors X = {x1, ..., x n } To determine the influence of a factor Z, we compare the factors pairwise to establish a pairwise comparison matrix; that is, we take two different factors x each time. i and x j , with a ij Represents x i and x j , the ratio of the impact on the target, where the comparison results are all expressed in matrix C = (a ij ) n*n Indicates that C is the judgment matrix between ZX; if x i and x j , the ratio of the reaction degree to Z is a ij , then x i and x j , the ratio of its influence on Z should be a ji =1 / a ij a ij The values ​​are scaled by referring to numbers from 1 to 9 and their reciprocals; Table 1 The standard for assigning values ​​to elements in the judgment matrix Perform n(n-1) / 2 pairwise judgments; The hierarchical index single ranking judgment matrix C corresponds to its maximum eigenvalue λ max The obtained characteristic quantity K, after normalization, is a ranking weight of the relative importance of the corresponding factors at the same level to a factor at the previous level, which is the hierarchical single ranking. The steps for consistency check of judgment matrix are as follows: Analyze the error and compatibility of the judgment matrix; Assume the consistency index is CI, then: And find out the corresponding random consistency index RI (Random Index), as shown in Table 2: Table 2 RI values ​​corresponding to different matrix order n The consistency ratio is determined by the following method: In general, if CR < 0.1, the judgment matrix is ​​a consistency matrix and the consistency value meets expectations, otherwise the judgment matrix is ​​revised; The total ranking method of the hierarchical indicators is as follows: Assume that the first level A contains A1, ..., A m , there are m factors in total, and the weight values ​​of the final hierarchical total ranking are a1,…,a m ; Assume that the first level A and the subsequent level B contain n elements B1, ..., B n , and they are about A j The final single-ranking weight values ​​are b 1j , …, b nj , when B i With A j When not relevant, b ij =0; the weight of each element in layer B for the total target layer A is required, that is, the weight value b1,…,b of the total ranking of each element in layer B is required n The calculation method is as follows: Perform consistency check on the total order of the levels, and proceed step by step from high to low levels for all levels; assume that the criteria level B and A j The judgment matrices of the relative comparison of the relevant elements have undergone consistency testing in the final single sorting test, and the consistency level index CI(j) of the single sorting is obtained, j = 1, ..., m, and the corresponding average random extraction consistency indexes are RI(j) and CI(j); RI(j) has also been obtained in the hierarchical single sorting, so the random consistency ratio of the final total sorting of the B level is: When CR is less than 0.1, it is considered that the total hierarchical ranking result meets expectations, and the weight value under the hierarchical analysis method is calculated.

6. The method according to claim 3, characterized in that: Step S32: Calculate the weight of the green travel index obtained by the initial screening using the SHAP method in step S2 using the entropy weight method; including: In different systems, the probability of each indicator appearing in each state is P i , i=1, 2, …, m=, then the entropy of the system is: P i =1 / m, i=1, 2, ..., m, that is, when the effects produced under multiple conditions are exactly the same, the highest value is taken; that is, e=-lnm If there are m evaluation objects and n evaluation indicators, the initial decision matrix R = (r ij ) m×n , for a certain indicator r j : Where: The smaller the entropy value of a specific indicator, the greater the variability of the indicator. If the entropy value of a specific indicator is larger, the variability of the indicator is smaller. The entropy weight method includes the following steps: Assuming there are n evaluation objects and m evaluation indicators, the initial decision matrix R = (r ij ) n×mm ; r ij It represents the evaluation value of the jth indicator of the i-th evaluation object; Step S321, normalizing the matrix; Step S322: the index characteristic value proportion P of the i-th evaluation object under the j-th index ij : Step S323: entropy value e under the jth index j : Where: k = 1 / ln n; Step S324, calculate the entropy weight W of each indicator j : The more evenly the indicators or data are distributed, the greater the entropy value is. Conversely, the smaller the entropy value is, that is, the greater the distribution difference between indicators is. The difference coefficient of the indicator is calculated as: d j =1-e j ; Step S325, calculating the comprehensive evaluation value of each evaluation object: The green travel weight value based on the entropy weight method is obtained.

7. The method according to claim 3, characterized in that Step S33: Calculate the weight of the green travel index obtained by the initial screening using the SHAP method in step S2 using a comprehensive weighting method, including: First, based on the subjective weight B of AHP i And the objective weight W of the entropy weight method i , introduce two linear coefficients δ1 and δ2 to construct possible comprehensive weights V i ,Right now: V i =δ1B i +δ2W i Secondly, construct the objective function and constraints, find the equilibrium point, and optimize the linear coefficients δ1 and δ2 to make the combined comprehensive weight V i With B i and W i The minimum deviation between min(||V i -B i ||2+||V i -W i ||2)=min(||δ1B i +δ2W i -B i ||2+||δ1B i +δ2W i -W i ||2) According to the properties of matrix differentials, the condition for the optimal first-order derivative of the objective function to achieve the minimum value is Finally, the optimized linear coefficients δ1 and δ2 are normalized to determine the optimal comprehensive weight of the central city green transportation evaluation index, namely: in, is the normalized subjective weight coefficient; is the normalized objective weight coefficient; is the comprehensive weight of evaluation index i in the evaluation index system.

8. The method according to claim 1, characterized in that In step S4, the normalization process is performed to calculate the expected output of the sample data as follows: S41: Use min-max normalization function to normalize the raw data: x i Indicates the original value of the indicator, y i Represents the normalized index value, x max and x min Respectively represent the maximum and minimum values ​​obtained for the same indicator; S42: After the sample data is normalized in S41, the normalized sample data is multiplied by the weight obtained in step S3 to calculate the expected output of the sample data.

9. The method according to claim 1, characterized in that: The method for constructing the green travel evaluation model in step S5 is: S51: data preprocessing, dividing training samples and test samples; S52: Determine the number of input layer neurons and the number of output layer neurons of the evaluation model neural network according to the number of input features and the number of evaluation features; S53: Setting the initial weights and thresholds of the neural network; S54: taking the training set as the input vector of the evaluation model, introducing the objective function, training the evaluation model, and introducing the White Shark optimization algorithm to optimize the weight of the evaluation model; The BP neural network propagates forward through layers, compares the final output with the expected output, adjusts the weights and thresholds through the gradient descent strategy, and finally establishes a neural network model with output results within the allowable error range: Wherein, w(m) represents the weight after the mth iteration, η is the iteration step length, i.e., the learning rate; the White Shark optimization algorithm is introduced to optimize the weight and threshold of the evaluation model; S541: Initialize the white shark population and randomly generate uniformly distributed individual positions of the white shark population That is, the weight of the evaluation model: In the formula is the j-dimensional position of the ith white shark, i = 1, 2, ..., n, where n is the size of the white shark population; u j , l j are the upper and lower limits of the j-th dimension search space respectively; R is a random number between 0 and 1; S542: Speed ​​update; In the WSO algorithm, the white shark senses the position of the prey based on its movement and updates its own speed: Where: is the (k+1)th iteration speed of the i-th white shark; is the k-th iteration speed of the i-th white shark; is the global optimal position obtained by the white shark in the kth iteration so far; is the k-th iteration position of the i-th white shark; The optimal position corresponding to the kth iteration speed of the i-th white shark; c1 and c2 are random numbers between 0 and 1; p1 and p2 are The control coefficient is described as k and K are the current number of iterations and the maximum number of iterations respectively; p max 、p min are the maximum and minimum values ​​of the control coefficient respectively; μ represents the contraction factor; S543: Position update: 1) Based on the position update of the optimal prey, in the white shark optimization algorithm, the white shark updates its own position by moving towards the optimal prey to find the optimal or suboptimal prey: Where: is the (k+1)th iteration position of the i-th white shark; is the negation operator; is a bitwise operator; a and b are one-dimensional binary vectors, described as sgn(-) is a sign-returning function; μ and l are the upper and lower limits of the search space respectively; ω0 is a logical vector, described as f is the frequency of white shark wave motion; m v It is the power that increases with the number of iterations when the white shark approaches the prey; the other parameters have the same meaning as above; 2) Update based on the best white shark position. In the white shark optimization algorithm, the white shark moves towards the optimal white shark position to get closer to the optimal position of the prey: Where: is the new position of the i-th white shark relative to the prey in the (k+1)th iteration; r1, r2, r3 are random numbers between 0 and 1; D k is the distance between the prey and the white shark; S s are the parameters of the white shark's sense of smell and vision when approaching the best prey; other parameters have the same meanings as above. S544: The first two optimal solutions are retained by simulating the feeding behavior of white sharks, and the positions of other white sharks are updated according to these optimal positions: S545: The candidate solution corresponding to the optimal white shark position obtained by iterative calculation of the white shark optimization algorithm, that is, the optimal weight of the evaluation model, is substituted into the BP neural network to complete the model training; S546: Input the green travel index to be evaluated into the trained model to obtain the corresponding green travel index evaluation value.

Images