A method for identifying key causal chains in public transportation carbon efficiency, electronic equipment, and media.

Abstract

This invention discloses a method, electronic device, and medium for identifying key causal chains of public transport carbon efficiency, comprising: constructing a route-level public transport carbon efficiency variable set, wherein the route-level public transport carbon efficiency variable set includes control variables, covariates, and carbon emissions per unit passenger mileage; the control variables are public transport operation strategies; the covariates are public transport carbon efficiency influencing factors; assuming that the route-level public transport carbon efficiency variable set follows a directed acyclic graph structure composed of a system of linear equations, solving for the causal effect matrix; establishing a causal directed graph of public transport carbon efficiency with route-level public transport carbon efficiency variables as nodes and causal effects as edge weights; performing post-nonlinear causal relationship tests and conditional independence causal tests on the causal directed graph of public transport carbon efficiency to obtain an optimized causal directed graph of public transport carbon efficiency, thereby obtaining the key causal links of public transport carbon efficiency.

Description

Technical Field

[0001] This invention belongs to the field of urban transportation carbon efficiency technology, and particularly relates to a method for identifying key causal chains in public transportation carbon efficiency, electronic equipment, and media. Background Technology

[0002] With the accelerating pace of urbanization and the deepening of the green and low-carbon development concept, urban public transport systems are playing an increasingly important role in carbon emission reduction and energy utilization optimization in the transportation sector. Carbon efficiency has become a core indicator for measuring the green operation level and sustainable development of public transport systems. Carbon efficiency is affected by multiple factors, including road infrastructure configuration, multimodal transport transfers, public transport operation strategies, population density, and land use structure, and there are complex multi-level linkages and causal coupling relationships among these variables.

[0003] Existing causal analysis methods for carbon efficiency mainly rely on traditional statistical inference tools such as structural equation modeling and regression analysis. These methods typically require subjective assumptions about the causal order or structure between variables, limiting the scope of analysis to local causal relationships between variable pairs. They struggle to comprehensively reveal the multivariate, multi-dimensional global causal structure of urban public transport systems. Furthermore, when faced with complex scenarios such as the fusion of multi-source heterogeneous data, nonlinearity among variables, and dynamic changes in public transport systems, the applicability and modeling accuracy of traditional methods are significantly insufficient, making it difficult to accurately identify key causal links affecting carbon efficiency improvement.

[0004] Therefore, there is an urgent need to develop technical means that can systematically characterize the complex multi-level causal relationships between variables, dynamically track the evolution of causal chains, and adapt to the characteristics of multi-source heterogeneous data. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for identifying key causal chains for public transportation carbon efficiency, an electronic device, and a medium.

[0006] In a first aspect, embodiments of the present invention provide a method for identifying key causal chains for public transportation carbon efficiency, the method comprising the following steps:

[0007] A set of bus carbon efficiency variables at the route level is constructed, which includes control variables, covariates, and carbon emissions per unit passenger mileage; the control variables are bus operation strategies; and the covariates are factors influencing bus carbon efficiency.

[0008] Assuming that the set of variables for bus carbon efficiency at the route level follows a directed acyclic graph structure composed of a system of linear equations, the causal effect matrix is ​​solved; and a causal directed graph of bus carbon efficiency is constructed with variables as nodes and causal effects as edge weights.

[0009] By performing post-nonlinear causal relationship tests and / or conditional independence causal tests on the causal directed graph of bus carbon efficiency, an optimized causal directed graph of bus carbon efficiency is obtained, thereby revealing the key causal links of bus carbon efficiency.

[0010] Calculating the global causal chain evolution index, the strong causal chain stability coefficient, and the local causal effect variation coefficient reveals the spatiotemporal evolution law of key causal chains;

[0011] Time-series variables were selected from the bus carbon efficiency variables at the route level, and a full-input model and a variable-removed model were established using a time-series neural network. The time-series causal effect was quantified by the difference in prediction error between the two models.

[0012] In a second aspect, embodiments of the present invention provide an electronic device, comprising:

[0013] At least one processor; and

[0014] A memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores one or more computer programs that can be executed by the at least one processor, and the one or more computer programs are executed by the at least one processor to enable the at least one processor to perform the above-described method for identifying key causal chains for public transportation carbon efficiency.

[0016] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the above-mentioned method for identifying key causal chains for public transportation carbon efficiency.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] This invention provides a method for identifying key causal chains in public transportation carbon efficiency. It constructs a route-level public transportation carbon efficiency variable set, which includes control variables, covariates, and carbon emissions per unit passenger mileage. The control variables are public transportation operation strategies; the covariates are factors influencing public transportation carbon efficiency. Assuming the route-level public transportation carbon efficiency variable set follows a directed acyclic graph structure composed of linear equations, the causal effect matrix is ​​solved. A causal directed graph of public transportation carbon efficiency is established, with variables as nodes and causal effects as edge weights. Post-nonlinear causal relationship tests and / or conditional independence causal tests are performed on the causal directed graph of public transportation carbon efficiency to obtain an optimized causal directed graph of public transportation carbon efficiency, thereby accurately identifying key causal links in public transportation carbon efficiency. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the key causal chain identification method for public transportation carbon efficiency provided in this embodiment of the invention;

[0021] Figure 2 This is a technical roadmap of the key causal chain identification method for public transportation carbon efficiency provided in this embodiment of the invention;

[0022] Figure 3 This is the key causal chain identification result for public transportation carbon efficiency provided in the embodiments of the present invention;

[0023] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It should be noted that, unless otherwise specified, the features in the following embodiments and implementation methods can be combined with each other.

[0026] like Figure 1 and Figure 2 As shown, this embodiment of the invention provides a method for identifying key causal chains for public transportation carbon efficiency, the method comprising the following steps:

[0027] Step S1: Construct a set of bus carbon efficiency variables at the route level. The set of bus carbon efficiency variables at the route level includes control variables, covariates, and carbon emissions per unit passenger mileage. The control variables are bus operation strategies. The covariates are factors affecting bus carbon efficiency.

[0028] In this example, by integrating multi-source heterogeneous data (including but not limited to bus smart cards, mobile phone signaling, road network and POI data), and taking bus routes as the analysis unit, a route-level bus carbon efficiency variable set S={F,T,BEP} is established.

[0029] Among them, Bus CO Emissions per Passenger-Kilometer (BEP) is a carbon efficiency indicator. Control variables T include, but are not limited to, bus operation strategies, such as dedicated bus lanes, bus signal priority strategies, priority electrification strategies, construction of transfer hubs and seamless connection facilities, and bus fare discounts. The covariate set F = {f1, f2, ..., f...} NF Factors affecting the carbon efficiency of public transport include, but are not limited to, station density, population density, POI density (work, residence), operating speed, passenger load factor, road type ratio (highway, trunk road, primary road, tertiary road), fleet size, vehicle type, departure interval, bus-subway connection index, and turnover.

[0030] Step S2: Assuming that the set of variables for bus carbon efficiency at the route level follows a directed acyclic graph structure composed of a system of linear equations, solve for the causal effect matrix; and establish a causal directed graph of bus carbon efficiency with variables as nodes and causal effects as edge weights.

[0031] Specifically, step S2 includes the following sub-steps:

[0032] Step S201: Assuming that the set of bus carbon efficiency variables S at the route level follows a directed acyclic graph (DAG) structure composed of a system of linear equations, and that each noise term is non-Gaussian and independent, a variable prediction model is constructed. The variable prediction model is used to represent that each variable is composed of the weighted sum of the variables of its parent nodes plus non-Gaussian independent noise.

[0033] The expression for the variable prediction model is as follows:

[0034]

[0035] In the formula, Pa(i) is the variable s i The set of parent nodes (i.e., the direct dependent variable), s j Let b represent the j-th variable. i,j It is the variable s i With variable s j The causal effect coefficient between them, e i It is the variable s i Non-Gaussian independent noise.

[0036] Step S202: The prediction model is converted into a matrix equation form, where the variable matrix is ​​equal to the product of the inverse matrix of the identity matrix minus the causal effect matrix and the non-Gaussian independent noise matrix.

[0037] Specifically, the prediction model is transformed into matrix form and rearranged to obtain:

[0038] S=(IB) -1 e

[0039] In the formula, S is the variable matrix, B is the causal effect matrix, I is the identity matrix, and e is the non-Gaussian independent noise matrix.

[0040] Step S203: Decompose the variable matrix so that the non-Gaussian independent noise matrix is ​​the product of the unmixing matrix and the variable matrix.

[0041] To solve for the causal effect matrix B, this example uses an improved Fast Independent Component Analysis (FastICA) to decompose the variable matrix S into:

[0042] e=WS

[0043] Where W is the unmixing matrix, i.e., IB.

[0044] Step S204: Calculate the unmixing matrix. Subtract the unmixing matrix from the identity matrix to obtain the causal effect matrix B, i.e., B=IW. Based on the causal effect matrix, establish a causal directed graph of public transport carbon efficiency with variables as nodes and causal effects as edge weights.

[0045] Specifically, the causal effect matrix B is used as the adjacency matrix of the causal directed graph of public transport carbon efficiency, and the variables are used as nodes and the causal effects are used as edge weights to establish the causal directed graph of public transport carbon efficiency.

[0046] Furthermore, in this example, to improve the reliability of the causal relationship and eliminate noise interference, a boundary weight threshold parameter β is introduced. Specifically, only when |b i,j When |≥β, variable i is considered to have a direct causal effect on variable j; otherwise, it is considered noise or a weak influence and is removed. In this example, the edge weight threshold parameter β is taken as the 5th percentile of the absolute value of all causal effect coefficients based on their distribution characteristics, thus ensuring that only representative causal relationship edges are retained in the causal directed graph of public transport carbon efficiency.

[0047] Furthermore, in this example, the process of calculating the unmixing matrix includes:

[0048] Step S20401 involves multi-scale centering (mean removal), standardization (unit variance), and higher-order centering (skewness and kurtosis to zero) of the bus route-level carbon efficiency variable set S to proactively eliminate possible variable bias, scale imbalance, and anomalies in higher-order statistical features, thereby enhancing the robustness and generalization ability of subsequent matrix decomposition.

[0049] Step S20402, randomly initialize an NS ×N S The orthogonal matrix is ​​taken as the unmixing matrix W. For each component i = 1, 2, ..., N of the unmixing matrix W. S N S Let S be the number of samples in the set of bus carbon efficiency variables at the route level, and let the parallel iteration continue until convergence.

[0050] An adaptive weighted combination of various nonlinear functions (such as tanh, cubic, exponential, ReLU, etc.) is introduced, defined as:

[0051]

[0052] Among them, each weight It can be dynamically adjusted based on the current variable characteristics to achieve adaptive characterization of different variable distributions. It is a non-linear function index. Indicates the first A nonlinear function, This represents an adaptive weighted combination nonlinear function constructed for the i-th variable.

[0053] Update the i-th component w of the unmixing matrix W according to the following formula. i The formula:

[0054]

[0055] in, It is the updated unmixed vector. Let S be the k-th sample in the set of bus route-level carbon efficiency variables, and E represent the mean of all samples. It is the total number of samples. Let i be the derivative of the nonlinear function of variable i.

[0056] Step S20403: After each update, Gram-Schmidt orthogonalization with an adaptive momentum factor is applied to orthogonalize the unmixed matrix W to prevent components from collapsing to the same direction. Adaptive momentum is also introduced to improve convergence speed and global optimum robustness. The momentum update formula is as follows:

[0057]

[0058] Where β is the momentum coefficient, which is dynamically adjusted using the Adam optimizer, and τ is the number of iterations.

[0059] Step S20404: Repeat the above operation until all components w are processed. i If the changes are very small (e.g., the inner product of two adjacent iterations is close to 1), then stop iterating; otherwise, continue.

[0060] Step S3 involves performing post-nonlinear causal relationship tests and / or conditional independence causal tests on the causal directed graph of bus carbon efficiency to obtain the optimized causal directed graph of bus carbon efficiency, thereby obtaining the key causal links of bus carbon efficiency.

[0061] In this case, if there is a directed edge e between node i and node j in the causal directed graph of public transport carbon efficiency... i,j Then, a post-nonlinear causality test needs to be performed on this edge. The specific steps are as follows:

[0062] In step S301A, for any directed edge to be tested in the causal directed graph of bus carbon efficiency, construct a forward nonlinear causal model and a reverse nonlinear causal model respectively; in the forward nonlinear causal model, use reversible nonlinear mapping to map the causal variable and the noise term to the result variable; in the reverse nonlinear causal model, use reversible nonlinear mapping to map the result variable and the noise term to the causal variable.

[0063] In a positive nonlinear causal model, if we assume the existence of a positive nonlinear causal relationship i→j, then there exist invertible nonlinear functions h1, h2, and a noise term E such that:

[0064]

[0065] Wherein, the noise term E and the input variable s i Independent, s i s represents the causal variable. j Represents the result variable.

[0066] In a reverse nonlinear causal model, if we assume the existence of a reverse nonlinear causal relationship j→i, then there exists an invertible nonlinear function. , and noise term E', such that:

[0067]

[0068] Wherein, the noise term E' and the input variable s j Independent, s i s represents the causal variable. j Represents the result variable.

[0069] Step S302A: The reversible nonlinear function is modeled using a neural network. The noise term is trained to approximate a Gaussian distribution and is independent of the input variables.

[0070] Furthermore, in this example, the neural network can be a multilayer perceptron.

[0071] In a forward nonlinear causal model, the noise term E is estimated as follows:

[0072]

[0073] H1 and H2 are approximated by a neural network. The training objective is to make the noise term E approximate a Gaussian distribution and correlate it with the input variable s. i independent.

[0074] The loss function is designed as follows:

[0075]

[0076] In the formula, N sample It represents the total number of samples, and k represents the k-th sample. Let s represent the noise term of the k-th sample. j,k s represents the k-th sample j Observed values ​​of the variable.

[0077] In the inverse nonlinear causal model, the noise term E' is estimated as follows:

[0078]

[0079] G1 and G2 are approximated by a neural network. The training objective is to make the noise term E' approximate a Gaussian distribution and correlate it with the input variable s. j independent.

[0080] The loss function is designed as follows:

[0081]

[0082] In the formula, s represents the noise estimate of the k-th sample in the inverse nonlinear causal model. i,k s represents the k-th sample i Variable value.

[0083] In step S303A, the independence between the input variables and the estimated noise term of the forward nonlinear causal model and the reverse nonlinear causal model are statistically tested respectively. If the forward model passes the independence test but the reverse model fails, the directed edge is retained; otherwise, the directed edge is deleted from the causal directed graph of bus carbon efficiency.

[0084] The neural network parameters are iteratively updated using the Adam optimization algorithm, and the optimal hyperparameters are determined using a grid search method. After the network training is complete, the positive s i →s j With reverse s j →s i The independence between the input variables and the estimated noise term is statistically tested. The test method can be either mutual information test or nonparametric test. In this example, the mutual information test is used as an example, with the positive s... i →s jp-value The calculation expression is as follows:

[0085]

[0086]

[0087] In the formula, The expression represents the mutual information independence test function used to test whether two variables are independent.

[0088] reverse s j →s i p-value The calculation expression is as follows:

[0089]

[0090]

[0091] Compare the p-values ​​of the forward and reverse tests to determine the direction of causality: if (e.g., 0.05), and , believing that s exists i →s j The significant nonlinear causal relationship preserves the edge e in the causal directed graph of public transport carbon efficiency. i,j Otherwise, if the causal relationship is deemed invalid or its direction unclear, edge e should be removed from the causal directed graph of public transport carbon efficiency. i,j That is, b i,j =0.

[0092] Step S304A: Traverse each directed edge to be tested in the causal directed graph of bus carbon efficiency, and output the route-level causal graph of bus carbon efficiency after the nonlinear causal relationship test.

[0093] In this case, if there exists a directed edge e between node i and node j in the causal directed graph of public transport carbon efficiency. i,j A conditional independence causality test needs to be performed on this side, which includes the following steps:

[0094] Step S301B: For nodes i and j connected by any directed edge to be tested in the causal directed graph of bus carbon efficiency, extract all parent nodes of these two nodes in the causal directed graph of bus carbon efficiency, merge the two sets of parent nodes, and remove the two nodes themselves to form the condition set C. i,j .

[0095] Specifically, for any directed edge e to be tested in the causal directed graph of public transport carbon efficiency... i,jFor the two connected nodes i and j, extract all parent nodes (i.e., nodes that directly influence them) of nodes i and j in the causal directed graph of public transport carbon efficiency. Remove nodes i and j themselves from the union of their parent node sets to form the condition set C. i,j The expression is as follows:

[0096]

[0097] In the formula, Pa(i) is the variable s i The set of parent nodes, Pa(j) is the variable s. j The set of parent nodes.

[0098] Step S302B: For each pair of samples, calculate variable s. i Variable s j Condition set C i,j The corresponding Gaussian kernel matrix.

[0099] Specifically, for each pair of samples k and k', the variable s is calculated respectively. i , variable s j and condition set C i,j The elements of the corresponding Gaussian kernel matrix , and The expression is as follows:

[0100]

[0101]

[0102]

[0103] in, , and s are variables respectively i Variable s j and condition set C i,j Gaussian kernel width, s i,k The variable s represents the k-th sample. i The observed value, s i,k' The variable s represents the k'th sample. j The observed value, C i,j,k This indicates that the k-th sample is in the condition set C. i,j The vector of observed values ​​for all variables, C i,j,k' This indicates that the k'-th sample is in the condition set C. i,j The observed vector of all variables. To improve computational efficiency, the Gaussian kernel width is the median of the Euclidean distance between each sample pair.

[0104] Step S303B, based on condition set Ci,j The corresponding Gaussian kernel matrix Calculate the condition set C i,j Corresponding kernel projection matrix Used to remove condition set pairs of variables s in the kernel space. i Variable s j The influence of the nuclear projection matrix The expression is as follows:

[0105]

[0106] in, It is a regularization parameter (to prevent the matrix from becoming non-invertible; it is usually taken as a small positive number, such as 10). -3 ), where I represents the identity matrix.

[0107] Subsequently, the kernel ridge regression method was used to refactor the variable s. i The corresponding residual kernel matrix M i and variable s j The corresponding residual kernel matrix M j Projection to condition set C i,j In the space where it is located, we obtain the variable s. i Variable s j The residual kernel matrix R i and R j The expression is as follows:

[0108]

[0109]

[0110] Step S304B, based on variable s i The corresponding residual kernel matrix and variable s j The corresponding residual kernel matrix is ​​used to calculate the variable s. i and variable s j The conditional independence test statistic is obtained; the p-value corresponding to the conditional independence test statistic is obtained through a permutation test; if the p-value is less than the threshold, it is considered that the conditional independence is true in the condition set C. i,j Below, variable s i and variable s j There are significant conditional dependencies between them, and the directed edges connecting nodes i and j in the causal directed graph of public transport carbon efficiency are preserved.

[0111] Based on variable s i The corresponding residual kernel matrix R i Variable s i The corresponding residual kernel matrix R j Calculate variables and variable s j Conditional independence test statistic Ti,j The expression is as follows:

[0112]

[0113] In the formula, N sample Indicates the sample size, if If T approaches 0, then T approaches 0.

[0114] The conditional independence test statistic T is obtained through the substitution test. i,j The corresponding p-value The expression is as follows:

[0115]

[0116] In the formula, Null hypothesis (in condition set C) i,j Below, variable s i and variable s j Distribution of statistics under conditional independence.

[0117] like (If we take 0.05), then we reject the null hypothesis, assuming that given the condition set C... i,j Below, variable s i With variable s j There are significant conditional dependencies between them, preserving the edge e in the causal directed graph of public transport carbon efficiency. i,j .like Then it is impossible to reject the null hypothesis, and it is believed that given the condition set C i,j Below, variable s i With variable s j Conditionally independent, edge e should be removed from the causal directed graph of public transport carbon efficiency. i,j (i.e. b) i,j =0).

[0118] Step S4: Based on the optimized causal directed graph of bus carbon efficiency, extract the key causal links of bus carbon efficiency and visualize them.

[0119] In this example, the causal directed graph of bus carbon efficiency after passing the post-nonlinear causality test and / or conditional independence causality test is visualized. To highlight the different causal strengths, the K-means clustering algorithm is first used to cluster all edge weights b. i,j Clustering was performed to determine the threshold for high causal effects. .when When, a bold solid line is used to connect node i and node j to highlight a strong causal relationship; when When b, a dashed line is used to represent a moderately strong causal relationship between node i and node j; when b i,jWhen =0, do not draw the connection line between node i and node j, and eliminate invalid causal paths.

[0120] Analysis of the carbon efficiency causal graph identified key causal links that can significantly improve the carbon efficiency of the public transportation system. Taking the morning rush hour (7:00–8:00) as an example, [the following is a reference to...] Figure 3 The causal relationship diagram shown reveals the key causal link for improving public transport carbon efficiency: "Bus lane setup → Metro connection index → ​​Bus connection index → ​​Fleet size → Turnover → Carbon efficiency". Specifically, the scientific configuration of bus lanes can effectively improve the connection level between multi-modal transportation and significantly optimize the seamless transfer capability between buses and rail transit. This process, by improving the metro connection index and bus connection index, further guides the reasonable adjustment of fleet size and turnover, forming a multi-level mediating mechanism, thereby indirectly and significantly positively impacting the core indicator of public transport carbon efficiency, BEP. This causal chain clearly reveals the path of comprehensive regulation of carbon efficiency through multi-level and complex coupling relationships between variables in urban traffic intervention measures. Based on the above analysis, this invention suggests: prioritizing the core links represented by mediating variables and formulating a systematic and comprehensive intervention policy. Specifically, this includes: (1) systematically improving the multi-modal transportation connection system and optimizing the convenience and connection efficiency of transfers between buses and rail transit; (2) scientifically planning the layout of bus lanes, improving bus operating speed and service level, and stimulating positive linkage of subsequent causal chain links.

[0121] It is worth emphasizing that the quantitative visualization of the causal directed graph of public transport carbon efficiency not only provides a theoretical basis for the targeted and precise implementation and prioritization of urban traffic intervention measures, but also serves as an important decision-making tool for policy simulation, intervention effect evaluation and dynamic optimization, thereby significantly improving the scientific, systematic and sustainable nature of public transport system carbon efficiency improvement strategies.

[0122] Step S5: Calculate the Global Causal Chain Evolution Index (GCCEI), the High Causal Chain Stability Coefficient (HCCSC), and / or the Local Causal Effect Variation (LCEV) to analyze the key causal links in public transport carbon efficiency.

[0123] The calculation process of the global causal chain evolution index includes: calculating the average difference between the edge weights of each pair of nodes during peak and off-peak periods; the expression is as follows:

[0124]

[0125] In the formula, N S For the total number of variables, and These represent the edge weights of the peak and flat peak adjacency matrices, respectively. This index reflects the average variation of the overall causal structure; the larger the value, the stronger the dynamics of the causal network.

[0126] For a causal chain identified as having a "strong causal relationship" at any given time period, the overlap ratio between the two time periods is defined as the strong causal chain stability coefficient (HCCSC). The calculation process for the strong causal chain stability coefficient includes: obtaining the intersection of the strong causal chain of bus carbon efficiency during peak hours and the strong causal chain of bus carbon efficiency during off-peak hours; obtaining the union of the strong causal chain of bus carbon efficiency during peak hours and the strong causal chain of bus carbon efficiency during off-peak hours; and using the ratio of the intersection to the union as the strong causal chain stability coefficient. The expression is as follows:

[0127]

[0128] In the formula, , These represent the sets of causal chains with "strong causal relationships" during peak and off-peak periods, respectively. A higher HCCSC indicates better cross-period stability of the strong causal path structure.

[0129] For each directed edge The calculation process of the local causal effect variation coefficient includes: obtaining the difference between the edge weights between node i and node j during peak hours and between the edge weights between node i and node j during off-peak hours; obtaining the maximum values ​​of the edge weights between node i and node j during peak hours and between the edge weights between node i and node j during off-peak hours; and using the ratio of the difference to the maximum value as the local causal effect variation coefficient; the expression is as follows:

[0130]

[0131] In the formula, To prevent tiny positive numbers with a denominator of zero, This represents the edge weight between node i and node j corresponding to the peak period. The edge weights between nodes i and j represent the off-peak periods. LCEV reflects the sensitivity and magnitude of variation of a specific causal relationship across different time periods. Causal chain evolution and stability indices such as GCCEI and HCCSC are calculated for each city and each major analysis period (e.g., peak and off-peak). The index values ​​for each city are spatially mapped to their geographic coordinate system, forming a spatial distribution layer at the city scale. Spatial autocorrelation methods such as Moran's I and Local Moran's I (LISA) are used to assess the clustering or dispersion characteristics of GCCEI and HCCSC in geographic space, determining whether significant spatial positive or negative correlation blocks exist. Spatial hotspot detection methods such as Getis-Ord Gi* are employed to identify significantly high-value areas (hotspots) and low-value areas (cold spots) of GCCEI and HCCSC, thereby locating the cities or regions with the most active or stable carbon efficiency causal chain evolution.

[0132] Step S6: Analyze the Granger causal relationship of the bus route-level carbon efficiency variable.

[0133] Specifically, addressing the limitations of traditional Granger causality tests in multivariate, multi-route bus systems (such as their applicability only to single-threaded or variable pairs, and their difficulty in accurately revealing nonlinear and complex interactions between variables in the system), this invention proposes a Granger causality analysis method for analyzing the carbon efficiency variables of buses at the route level. This example can automatically model the nonlinearity, higher-order causality, and time-series structure between variables, improving the ability to discriminate causal relationships in multivariate, multi-route systems and broadening its applicability. The specific implementation steps are as follows:

[0134] Step S601: Select the set of time-series variables of interest from the route-level bus carbon efficiency variable set S. ,in, This represents the total number of time series variables. Each time series variable... All have The data structures represent the number of lines respectively. and number of time periods The time-series variables include, but are not limited to, indicators related to public transport operation and carbon efficiency, such as population density, operating speed, passenger load factor, departure interval, and turnover.

[0135] For each time series variable Design and train a neural network prediction model. The input includes the values ​​of all relevant variables on all lines at different historical times, and the output is the predicted value of the target variable for all lines at the current time. The neural network architecture can be selected based on the data characteristics, choosing an architecture suitable for time-series modeling, such as Long Short-Term Memory (LSTM), Temporal Convolutional Network (TCN), or Multilayer Perceptron (MLP). Its prediction relationship can be expressed as:

[0136]

[0137] in, For the first The predicted value of the i-th variable at time t for each line. For the neural network mapping, z is the maximum lag order.

[0138] To determine the j-th time series variable For the i-th time series variable To determine whether Granger causality exists, two types of neural network models were built for comparison:

[0139] The input consists of historical lag information for all time series variables; a full-input model is then established. :

[0140]

[0141] In the formula, This represents the predicted value of the i-th variable for the l-th line at time t using a fully input neural network model that includes historical lag information of all time-series variables. It represents the set of values ​​for all time-series variables across all lines and historical moments.

[0142] Input is the variable to be tested after removing it. Historical information of all variables following the historical lag term is used to build a variable-removed model. :

[0143]

[0144] In the formula, This means that the variable being tested is not included. Given historical lag information, the neural network predicts the value of the l-th line and the i-th variable at time t. Indicates the variable being tested The set of all historically lagging information.

[0145] Both types of models use mean squared error (MSE) as the loss function, as follows:

[0146]

[0147]

[0148] By comparing the difference in prediction errors between the two types of models, a Granger causality quantification index is defined:

[0149]

[0150] This indicator reflects the introduction Historical information The degree of improvement in prediction accuracy. To verify... The significance of the hypothesis, setting the null hypothesis. : Introduction Historical values ​​will not increase significantly. The prediction accuracy. The bootstrap method is used to... To perform a statistical test, the distribution under the null hypothesis and the corresponding p-value are obtained through repeated sampling, denoted as . Set a statistical significance threshold. (e.g., 0.05). If Then the variable is considered to be right A significant Granger causal relationship exists across all bus routes; otherwise, a significant Granger causal relationship is determined to exist between the two.

[0151] In summary, this invention provides a method for identifying key causal chains in public transportation carbon efficiency. It constructs a route-level public transportation carbon efficiency variable set, which includes control variables, covariates, and carbon emissions per unit passenger mileage. The control variables are public transportation operation strategies; the covariates are factors influencing public transportation carbon efficiency. Assuming the route-level public transportation carbon efficiency variable set follows a directed acyclic graph structure composed of linear equations, the causal effect matrix is ​​solved. A causal directed graph of public transportation carbon efficiency is established, with variables as nodes and causal effects as edge weights. Post-nonlinear causal relationship tests and / or conditional independence causal tests are performed on the causal directed graph of public transportation carbon efficiency to obtain an optimized causal directed graph of public transportation carbon efficiency, thereby accurately identifying key causal links in public transportation carbon efficiency.

[0152] Accordingly, this application also provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the above-described method for identifying key causal chains for public transportation carbon efficiency. Figure 4 The diagram shown is a hardware structure diagram of any device with data processing capabilities in which the key causal chain identification method for public transportation carbon efficiency provided in this embodiment of the invention is located, except... Figure 4 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0153] Accordingly, this application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the aforementioned method for identifying key causal chains for public transportation carbon efficiency. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0154] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method for identifying key causal chains for public transportation carbon efficiency, characterized in that, The method includes: A set of bus carbon efficiency variables at the route level is constructed, which includes control variables, covariates, and carbon emissions per unit passenger mileage; the control variables are bus operation strategies; and the covariates are factors influencing bus carbon efficiency. Assuming that the set of variables for bus carbon efficiency at the route level follows a directed acyclic graph structure composed of a system of linear equations, the causal effect matrix is ​​solved; and a causal directed graph of bus carbon efficiency is constructed with variables as nodes and causal effects as edge weights. By performing post-nonlinear causal relationship tests and / or conditional independence causal tests on the causal directed graph of bus carbon efficiency, an optimized causal directed graph of bus carbon efficiency is obtained, thereby revealing the key causal links of bus carbon efficiency. The process of establishing a causal directed graph for public transport carbon efficiency includes: Assuming that the set of bus carbon efficiency variables at the route level follows a directed acyclic graph structure composed of a system of linear equations, a variable prediction model is constructed. The variable prediction model is used to represent that each variable is composed of the weighted sum of the variables of its parent nodes plus non-Gaussian independent noise. The prediction model is transformed into a matrix equation form, where the variable matrix is ​​equal to the product of the inverse matrix of the identity matrix minus the causal effect matrix and the non-Gaussian independent noise matrix. The variable matrix is ​​decomposed such that the non-Gaussian independent noise matrix is ​​the product of the unmixing matrix and the variable matrix; Calculate the unmixing matrix, and obtain the causal effect matrix by subtracting the unmixing matrix from the identity matrix; based on the causal effect matrix, construct a causal directed graph of public transport carbon efficiency with variables as nodes and causal effects as edge weights; The process of solving the unmixing matrix includes: initializing an orthogonal matrix as the unmixing matrix; selecting a nonlinear function for weighted combination based on the line-level bus carbon efficiency variable; updating each component in the unmixing matrix based on the weighted combination nonlinear function; and forming new components through momentum updates; determining whether the update magnitude of each component in the unmixing matrix is ​​lower than a preset threshold. If the threshold is met, the iteration stops and the final unmixing matrix is ​​output.

2. The method for identifying key causal chains for public transportation carbon efficiency according to claim 1, characterized in that, The process of performing a post-nonlinear causal relationship test on the causal directed graph of public transport carbon efficiency includes: For any directed edge to be tested in the causal directed graph of public transport carbon efficiency, a forward nonlinear causal model and a reverse nonlinear causal model are constructed respectively. In the forward nonlinear causal model, an invertible nonlinear mapping is used to map the causal variable and the noise term to the result variable. In the reverse nonlinear causal model, an invertible nonlinear mapping is used to map the result variable and the noise term to the causal variable. The invertible nonlinear mapping is modeled using a neural network. By training the neural network, the noise term is made to approximate a Gaussian distribution and be independent of the input variables. The independence between the input variables and the estimated noise term of the forward nonlinear causal model and the reverse nonlinear causal model are statistically tested. If the forward model passes the independence test but the reverse model fails, the current directed edge is retained; otherwise, the current directed edge is deleted from the causal directed graph of bus carbon efficiency. Traverse each directed edge to be tested in the causal directed graph of bus carbon efficiency, and output the route-level causal graph of bus carbon efficiency after the nonlinear causal relationship test.

3. The method for identifying key causal chains for public transportation carbon efficiency according to claim 1, characterized in that, The process of performing conditional independence and causality tests on the causal directed graph of public transport carbon efficiency includes: For any directed edge in the causal directed graph of public transport carbon efficiency connecting nodes i and j, extract all parent nodes of these two nodes in the causal directed graph of public transport carbon efficiency. Merge the two sets of parent nodes and remove the two nodes themselves to form the condition set C. i,j ; For each pair of samples, calculate the variable s separately. i Variable s j Condition set C i,j The corresponding Gaussian kernel matrix; Based on condition set C i,j The corresponding Gaussian kernel matrix, and the calculation condition set C i,j The corresponding kernel projection matrix; variable s i The corresponding Gaussian kernel matrix and variable s j The corresponding Gaussian kernel matrix is ​​projected onto the condition set C. i,j In the space where it is located, we obtain the variable s. i The corresponding residual kernel matrix and variable s j The corresponding residual kernel matrix; Based on variable s i The corresponding residual kernel matrix and variable s j The corresponding residual kernel matrix is ​​used to calculate the variable s. i and variable s j The conditional independence test statistic is obtained; the p-value corresponding to the conditional independence test statistic is obtained through a permutation test; if the p-value is less than the threshold, it is considered that the conditional independence is true in the condition set C. i,j Below, variable s i and variable s j There are significant conditional dependencies between them, and the directed edges connecting nodes i and j in the causal directed graph of public transport carbon efficiency are preserved. Traverse each directed edge to be tested in the causal directed graph of bus carbon efficiency, and output the route-level causal graph of bus carbon efficiency after the conditional independence causal test.

4. The method for identifying key causal chains for public transportation carbon efficiency according to claim 1, characterized in that, The method further includes: visualizing the directed graph of bus carbon efficiency causality after passing the post-nonlinear causality test and / or conditional independence causality test; including: Cluster all directed edge weights to set a high causal effect threshold; When the weight of the directed edge between node i and node j is greater than or equal to the high causal effect threshold, it indicates that there is a strong causal relationship between node i and node j in terms of public transport carbon efficiency. The first label is used to highlight the directed edge between node i and node j. When the weight of the directed edge between node i and node j is less than or equal to the high causal effect threshold, it indicates that there is a moderate causal relationship between node i and node j in terms of public transport carbon efficiency. The second label is used to highlight the directed edge between node i and node j. When the weight of the directed edge between node i and node j is 0, it means that there is no causal relationship between node i and node j regarding public transport carbon efficiency, and no directed edge is drawn between node i and node j.

5. The method for identifying key causal chains for public transportation carbon efficiency according to claim 1, characterized in that, The method further includes: calculating the global causal chain evolution index, the strong causal chain stability coefficient, and / or the local causal effect variation coefficient to analyze key causal links in public transport carbon efficiency; The calculation process of the global causal chain evolution index includes: calculating the average value of the difference between the edge weights of each pair of nodes during peak and off-peak periods; The calculation process of the strong causal chain stability coefficient includes: obtaining the intersection of the strong causal relationship link of bus carbon efficiency during peak hours and the strong causal relationship link of bus carbon efficiency during off-peak hours, obtaining the union of the strong causal relationship link of bus carbon efficiency during peak hours and the strong causal relationship link of bus carbon efficiency during off-peak hours, and taking the ratio of the intersection to the union as the strong causal chain stability coefficient. The calculation process of the local causal effect variation coefficient includes: obtaining the difference between the edge weights between node i and node j during peak hours and between the edge weights between node i and node j during off-peak hours; obtaining the maximum value of the edge weights between node i and node j during peak hours and between the edge weights between node i and node j during off-peak hours; and using the ratio of the difference to the maximum value as the local causal effect variation coefficient.

6. The method for identifying key causal chains for public transportation carbon efficiency according to claim 1, characterized in that, The method further includes: A set of time-series variables containing the time dimension was selected from the set of bus carbon efficiency variables at the route level; Determining whether a Granger causal relationship exists between the j-th time-series variable and the i-th time-series variable includes: constructing a first neural network model, the input of which is the historical lag information of all time-series variables; constructing a second neural network model, the input of which is the historical information of all time-series variables after removing the historical lag term of the j-th time-series variable; using the difference between the loss function corresponding to the first neural network model and the loss function corresponding to the second neural network model as the Granger causality effect quantification index; performing a statistical test on the Granger causality effect quantification index to obtain a p-value; if the p-value is less than the statistical significance threshold, then it is determined that a Granger causal relationship exists between the j-th time-series variable and the i-th time-series variable.

7. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores one or more computer programs that can be executed by the at least one processor, the one or more of the computer programs being executed by the at least one processor to enable the at least one processor to perform the bus carbon efficiency key causal chain identification method as described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for identifying key causal chains for public transport carbon efficiency as described in any one of claims 1-6.

Images