Test System and Method for Prompt Function of Visual Overhaul Locomotive Driving Guidance Device

Abstract

The present invention discloses a test system and method for the prompt function of a visualization heavy-haul locomotive driving guidance device, including the following steps: S1, collecting multi-dimensional operation data of the heavy-haul locomotive and constructing a data set; S2, processing the data set and generating an initial topological graph structure according to data relevance; S3, using a graph convolutional network to extract features from the initial topological graph structure and generating a high-dimensional feature vector; S4, constructing a causal relationship model through a causal inference method to identify key causal nodes and causal paths; S5, using an eagle swarm optimization algorithm to dynamically optimize the initial topological graph structure; S6, dynamically verifying the optimized prompt function logic to generate a prompt function test result and test feedback data; S7, adjusting the prompt function logic design and parameters according to the test feedback data, and iteratively optimizing the performance of the prompt function in multiple rounds. The present invention combines a graph convolutional network, causal inference and an eagle swarm optimization algorithm to realize the dynamic optimization and test of the prompt function of the heavy-haul locomotive.

Description

Technical Field

[0001] The present invention relates to the technical field of driving guidance for heavy-haul locomotives, and particularly to a test system and method for the prompting function of a visual heavy-haul locomotive driving guidance device. Background Art

[0002] As a key equipment in modern railway transportation, the driving safety and operation efficiency of heavy-haul locomotives are of great significance for ensuring the smooth completion of transportation tasks. During the operation of heavy-haul locomotives, the prompting function of the guidance device undertakes the important tasks of providing driving information in real time, optimizing operation decisions, improving operation efficiency, and ensuring driving safety. However, the existing prompting function technologies of driving guidance devices still have many limitations in aspects such as the acquisition and analysis of multi-dimensional operation data, the adaptability to complex scenarios, the optimization ability of prompting logic, and test efficiency, and cannot fully meet the actual needs of modern railway transportation.

[0003] First of all, in terms of the acquisition and analysis of multi-dimensional operation data, the existing technologies mainly rely on single sensors or simple signal fusion methods, resulting in fewer dimensions and weak correlations of operation data. The current prompting function design is usually based on fixed operation rules, and this method cannot make full use of multi-dimensional dynamic data in heavy-haul locomotive driving, such as speed, station information, line signals, cylinder pressure, speed limits, phase separation crossing, weather information, etc. Due to insufficient data acquisition and limited analysis capabilities, the prompting function is prone to information omission or misjudgment in the face of complex scenarios (such as steep slopes, sharp curves, or bad weather), thus affecting driving safety and operation efficiency.

[0004] Secondly, in terms of data processing and model construction, the existing technologies usually adopt traditional data cleaning and simple feature extraction methods, and these methods have weak correlation analysis capabilities for data and cannot accurately capture the complex relationships between multi-dimensional data. For example, the non-linear correlation between running speed and slope is not fully explored in the optimization of prompting logic, resulting in the difficulty for the prompting content to fully adapt to the actual scenario. Especially in high-dynamic scenarios, the feature extraction efficiency of the existing methods is low, and the accuracy of the processing results is insufficient, which limits the real-time response ability of the prompting function.

[0005] In addition, during the optimization process of prompting logic, the existing technologies mainly rely on rule-based decision models and are difficult to dynamically adapt to complex operation scenarios. This prompting design based on fixed logic shows obvious lag and limitations when dealing with dynamically changing operation states. Especially in the face of continuously changing slopes or sudden weather conditions, the prompting function is prone to deviate from the actual needs and cannot accurately guide the driver to adjust operation strategies. This not only reduces the driving safety of heavy-haul locomotives but also may lead to excessive energy consumption and increased transportation costs.

[0006] In the prompt function testing and verification phase, traditional technologies usually adopt manual scenario simulation and static analysis methods, which are inefficient and lack accuracy. Manual simulation is difficult to cover all complex operation scenarios during the operation of heavy-haul locomotives, and the representativeness and universality of test results are limited. In addition, due to the lack of a systematic feedback mechanism and an intelligent optimization process, existing technologies are difficult to achieve the automation and closed-loop optimization of the prompt function testing, resulting in a long test cycle and test results that are difficult to directly use for the improvement and iteration of the prompt logic.

[0007] Another important defect in the existing technology is the lack of intelligent visualization means for the performance of the prompt function. During the development and optimization process of the prompt function, test results and the optimization process are usually presented in the form of static data reports. This method not only increases the difficulty of result analysis but also limits the intuitive understanding of the dynamic changes in the prompt logic by developers. In dealing with complex operation scenarios, the lack of clear dynamic display means for the prompt function further increases the difficulty of optimizing the prompt logic.

[0008] Therefore, how to provide a test system and method for the prompt function of a visual heavy-haul locomotive driving guidance device is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0009] An object of the present invention is to provide a test system and method for the prompt function of a visual heavy-haul locomotive driving guidance device. The present invention uses a graph convolutional network, causal inference, and eagle swarm optimization algorithm, and through multi-dimensional data analysis and dynamic verification technology, realizes the optimization and efficient testing of the prompt function of the heavy-haul locomotive driving guidance device. Specifically, the present invention can accurately identify key causal nodes and paths under complex operation scenarios and dynamically adjust the prompt logic, significantly improving the accuracy, real-time performance, and adaptability of the prompt function. In addition, combined with visualization technology, it real-time displays topological changes and the optimization process, forming a closed-loop feedback mechanism, further improving the test efficiency and the optimization effect of the prompt function, and is applicable to the operation guidance of heavy-haul locomotives under various complex scenarios.

[0010] The test method for the prompt function of the visual heavy-haul locomotive driving guidance device according to the embodiment of the present invention includes the following steps:

[0011] S1. Collect multi-dimensional operation data of the heavy-haul locomotive through sensors and construct a data set;

[0012] S2. Perform noise filtering, outlier removal, and normalization processing on the data set, and generate an initial topological graph structure according to data correlation;

[0013] S3. Use a graph convolutional network to extract features from the initial topological graph structure, capture the multi-scale relationship between nodes and neighborhoods through multiple layers of convolution, and generate a high-dimensional feature vector;

[0014] S4. Based on the high-dimensional feature vectors, construct a causal relationship model through causal inference methods, analyze the causal relationships between nodes, and identify the key causal nodes and causal paths that have important impacts on the prompting function;

[0015] S5. According to the key causal nodes and causal paths, use the eagle flock optimization algorithm to dynamically optimize the initial topology graph structure. By combining global search and local optimization, adjust the node priorities and edge weights to optimize the prompting function logic;

[0016] S6. Dynamically verify the optimized prompting function logic. By visually displaying the topological changes of the dynamic graph structure, generate the prompting function test results and test feedback data;

[0017] S7. Adjust the prompting function logic design and parameters according to the test feedback data, and use the dynamically optimized topology graph structure to iteratively optimize the performance of the prompting function in multiple rounds.

[0018] Optionally, the multi-dimensional operation data specifically includes the driving speed of heavy-haul locomotives, station information, line signals, cylinder pressure, speed limits, and neutral section passing.

[0019] Optionally, the S2 specifically includes:

[0020] S21. Conduct a preliminary cleaning of the data set, identify noise data based on statistical distribution characteristics, and remove the data that does not conform to the normal operating state through filtering techniques;

[0021] S22. Eliminate outliers from the cleaned data set. By calculating the deviation values and dispersions of the data distribution, identify the extreme outliers, and use the interpolation method to fill in the missing or abnormal data;

[0022] S23. Normalize the multi-dimensional features in the data set, map the feature values to a unified range, and eliminate the influence between different feature dimensions;

[0023] S24. Based on the normalized data set, use the Pearson correlation coefficient to calculate the linear correlation between the operation features:

[0024]

[0025] where r ij represents the correlation coefficient between operation feature i and operation feature j, x ik represents the value of operation feature i in the k-th data, x jk represents the value of operation feature j in the k-th data, represents the mean value of operation feature i, represents the mean value of operation feature j, and n represents the total number of data samples;

[0026] S25. Define the relationship between nodes and edges according to the analysis results of the correlation of operation characteristics, where nodes represent operation characteristics and edges represent the correlation between operation characteristics:

[0027]

[0028] where w ij represents the weight between node i and node j, |r ij | represents the absolute value of the correlation coefficient between operation characteristic i and operation characteristic j, and τ represents the set correlation threshold;

[0029] S26. Generate an initial topological graph structure according to the definition of operation characteristic nodes and edges, and combine the correlation analysis results, where the layout of nodes and edges dynamically reflects the correlation between operation characteristics.

[0030] Optionally, the specific steps of S3 are as follows:

[0031] S31. Input the generated initial topological graph structure into the graph convolutional network and initialize the node feature matrix:

[0032]

[0033] where H (0) represents the node feature matrix, represents the initial feature vector of node i, and N represents the total number of nodes;

[0034] S32. Normalize the adjacency matrix of the initial topological graph, add self-loops to enhance the connectivity of the graph, and calculate the normalized adjacency matrix:

[0035]

[0036] where, represents the normalized adjacency matrix, A represents the original adjacency matrix, I represents the identity matrix, represents the degree matrix of A + I;

[0037] S33. In each layer of graph convolution, perform a convolution operation on the node features using the variable topology adjustment mechanism:

[0038]

[0039] where H (l+1) represents the node feature matrix of the (l + 1)-th layer, σ represents the activation function, N(i) represents the set of neighbor nodes of node i, represents the weight value in the normalized adjacency matrix, W (l) represents the training weight matrix of the l-th layer, c ij represents the normalization factor, and α represents the dynamic adjustment coefficient. Denote the feature vector of node j in the l-th layer;

[0040] S34. Perform multi-scale fusion on the output features of multiple layers of convolutions, integrate features of different levels through concatenation operations, and generate multi-scale features;

[0041] S35. Dynamically adjust the weight contribution of each layer of features according to the sparsity of the multi-scale features, and introduce a sparsity-driven weighting mechanism:

[0042]

[0043] where H (weighted) Denote the feature representation after weighted fusion, L denote the total number of layers, exp denote the exponential function, β denote the sparsity adjustment coefficient, H (l) Denote the node feature matrix of the l-th layer, ||H (l) ||1 denote the sparsity of the node feature matrix of the l-th layer, ||H (m) ||1 denote the sparsity of the node feature matrix of the m-th layer;

[0044] S36. Perform a linear transformation on the fused weighted features through a fully connected layer or a mapping matrix to generate a high-dimensional feature vector.

[0045] Optionally, the specific steps of S4 include:

[0046] S41. Based on the high-dimensional feature vector, construct a causal relationship model G = (V, E, W), where V is the set of nodes, representing the features related to the hint function; E is the set of edges, representing the causal relationships between features; W is the set of edge weights, used to characterize the strength of the causal influence between features;

[0047] S42. Quantify the causal strength of each edge in the edge set E, and calculate the mutual information entropy of the causal strength based on Bayesian network inference:

[0048]

[0049] where C pq Denote the causal strength between node p and node q, P(x p , x q ) denote the joint probability distribution of node p and node q, P(x p ) denote the marginal probability distribution of node p, P(x q ) denote the marginal probability distribution of node q;

[0050] S43. In the causal path optimization, combine the inference results of the Bayesian network to dynamically optimize the path set:

[0051]

[0052] Among them, F path represents the path optimization objective function, P represents the set of candidate causal paths, and Ω pq represents the correlation weight between node p and node q, exp represents the exponential function, γ represents the degree value attenuation coefficient, and d q represents the degree value of node q, and Δ pq represents the path length between node p and node q, and λ represents the penalty factor of the path length;

[0053] S44. Normalize the overall weight of the causal path:

[0054]

[0055] Among them, W pq represents the normalized causal weight matrix, and N(p) represents the set of neighborhood nodes of node p;

[0056] S45. Assign priorities to the input node features of the hint function according to the normalized causal weight matrix, and give priority to including high-weight nodes in the optimization process:

[0057]

[0058] Among them, P p represents the priority of node p, and Φ(x q ) represents the feature importance score of node q;

[0059] S46. Output the key causal nodes and causal paths that have an important impact on the hint function.

[0060] Optionally, the S5 specifically includes:

[0061] S51. Initialize the initial topology graph structure according to the key causal nodes and causal paths to generate a population P = {P (1) , P (2) , …, P (M)}, where m represents the individual label, M represents the number of individuals in the population, V represents the set of nodes, E represents the set of edges, represents the set of edge weights, represents the edge weight between node v p and node v q in individual m, represents the set of node priorities, represents the priority of node v p in individual m;

[0062] S52. In the global search stage, introduce the strong individuals of the eagle flock optimization algorithm and define the global objective function:

[0063]

[0064] Among them, represents the global objective function, N(v p ) represents the set of neighbor nodes of node v p , C pq represents the causal strength between node v p and node v q , represents the degree value of node v p , represents the degree value of node v q , and μ represents the degree value attenuation factor;

[0065] S53. In the local optimization stage, introduce the weak individuals of the eagle flock optimization algorithm to fine-tune the edge weights according to the global search results of the strong individuals:

[0066]

[0067] Among them, represents the updated edge weight, δ represents the edge weight update coefficient, tanh represents the hyperbolic tangent function, ζ represents the priority influence factor, represents the priority of node v q in individual m, represents the priority of node v r in individual m;

[0068] S54. Make an adaptive adjustment to the node priority and update the node priority according to the topological change situation:

[0069]

[0070] Among them, represents the updated node priority, exp represents the exponential function, η represents the priority adjustment coefficient, |N(v p )| represents the number of neighbor nodes of node v p ;

[0071] S55. Evaluate the fitness of the individuals:

[0072]

[0073] Among them, represents the fitness function, and θ1, θ2, and θ3 represent the weight parameters;

[0074] S56. Update the eagle group through multiple rounds of iteration. In each round of iteration, select the individual with the highest fitness as the new center, and transfer the node priority and edge weight distribution to the next generation of individuals. By repeating the global search and local optimization process, an optimized topological structure is finally obtained.

[0075] Optionally, the S6 specifically includes:

[0076] S61. Dynamically verify the optimized hint function logic, set different operating scenarios, including normal operation, ramp, sharp turn, and bad weather, and record the trigger time, trigger conditions, and hint content of the hint function in each scenario.

[0077] S62. In each operating scenario, collect the signal data and test data output by the hint function, including the accuracy, real-time performance, and adaptability to the operating state of the hint signal.

[0078] S63. Dynamically analyze the test data. By comparing the hint signal with the actual operating state, evaluate the performance of the hint function in different operating scenarios, and calibrate the error range and potential optimization space of the hint signal.

[0079] S64. Display the topological changes of the dynamic graph structure through visualization technology, including the process of node priority, edge weight adjustment, and causal path dynamic changes.

[0080] S65. Generate the test results of the hint function, including the accuracy of the trigger time, the adaptability of the hint content, and the real-time performance of the hint response.

[0081] S66. Output the test results and scenario analysis data as test feedback data, and record the visualization log generated during the test process.

[0082] The test system for the hint function of the visualization heavy-haul locomotive driving guidance device according to the embodiment of the present invention includes the following modules:

[0083] Multi-dimensional operation data acquisition module, used to acquire the multi-dimensional operation data of the heavy-haul locomotive.

[0084] Data preprocessing module, used to process the multi-dimensional operation data to generate an initial topological graph structure.

[0085] Feature extraction module, used to extract features from the initial topological graph structure through a graph convolutional network to generate a high-dimensional feature vector.

[0086] Causal relationship analysis module, used to construct a causal relationship model based on the high-dimensional feature vector to identify key causal nodes and causal paths.

[0087] Optimization calculation module, which is used to dynamically optimize the initial topological graph structure by using the eagle flock optimization algorithm, adjust the node priority and edge weight, and optimize the hint function logic;

[0088] Dynamic verification module, which is used to dynamically verify the optimized hint function logic, collect test data and generate test results;

[0089] Visualization display module, which is used to display the topological changes of the dynamic graph structure and generate test feedback data for the hint function;

[0090] Feedback and storage module, which is used to store and record the test feedback data and support the multi-round iterative optimization of the hint function logic.

[0091] The beneficial effects of the present invention are as follows:

[0092] First of all, in terms of the acquisition and processing of multi-dimensional operation data, the present invention realizes the comprehensive acquisition of multi-dimensional dynamic data such as speed, station information, line signal, cylinder pressure, speed limit and overrun, etc., and generates the initial topological graph structure by combining noise filtering, outlier removal and normalization processing. This method not only improves the quality of data, but also excavates the deep correlation between various operation characteristics, providing strong support for subsequent feature extraction and model construction.

[0093] Secondly, the present invention uses a graph convolutional network to extract features from the initial topological graph structure, captures the multi-scale relationship between nodes and neighborhoods through multi-layer convolution operations, and combines multi-scale feature fusion and sparsity-driven weighting mechanism to effectively solve the problem of insufficient processing of complex non-linear correlations by traditional methods. This feature extraction method significantly enhances the adaptability of the hint logic to complex operation scenarios, and at the same time improves the global perception ability and local optimization effect of multi-dimensional dynamic data, making the hint function more accurate and efficient.

[0094] In addition, the present invention constructs a causal relationship model based on the causal inference method, identifies the key causal nodes and paths that have an important impact on the hint logic, and optimizes the decision logic of the hint function by dynamically adjusting the causal weight and priority allocation. Combining with the eagle flock optimization algorithm, the combination of global search and local optimization further improves the dynamic adaptability of the hint function. Especially in complex scenarios such as steep slopes, sharp turns or bad weather conditions, the optimized hint logic can adjust the node priority and edge weight in real time to provide more practical hint content.

[0095] Finally, in terms of the testing and verification of the prompting function, the present invention sets up a variety of operating scenarios to comprehensively and dynamically verify the optimized prompting logic, and combines visualization technology to display the topological change process of the dynamic graph structure in real time, including node priority adjustment, edge weight optimization, and dynamic change of the causal path. By generating traceable test feedback data and optimization logs, a closed-loop optimization mechanism is formed, which not only improves the efficiency of the prompting function test, but also provides data support and intuitive display for the continuous improvement of the prompting logic, solving the deficiencies of low efficiency and lack of dynamic visualization in traditional testing methods. Description of the Drawings

[0096] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the accompanying drawings:

[0097] Figure 1 is the overall flowchart of the testing method for the prompting function of the visualization heavy-haul locomotive driving guidance device proposed by the present invention;

[0098] Figure 2 is the structural schematic diagram of the testing system for the prompting function of the visualization heavy-haul locomotive driving guidance device proposed by the present invention. Detailed Embodiments

[0099] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0100] Refer to Figure 1 , the testing method for the prompting function of the visualization heavy-haul locomotive driving guidance device includes the following steps:

[0101] S1. Collect multi-dimensional operation data of the heavy-haul locomotive through sensors and construct a data set;

[0102] S2. Perform noise filtering, outlier removal, and normalization processing on the data set, and generate an initial topological graph structure according to the data correlation;

[0103] S3. Use a graph convolutional network to extract features from the initial topological graph structure, capture the multi-scale relationship between nodes and neighborhoods through multiple layers of convolution, and generate a high-dimensional feature vector;

[0104] S4. Based on the high-dimensional feature vector, construct a causal relationship model through a causal inference method, analyze the causal relationship between nodes, and identify key causal nodes and causal paths that have an important impact on the prompting function;

[0105] S5. According to the key causal nodes and causal paths, use the eagle flock optimization algorithm to dynamically optimize the initial topology structure of the graph. By combining global search and local optimization, adjust the node priorities and edge weights to optimize the hint function logic;

[0106] S6. Dynamically verify the optimized hint function logic. By visually displaying the topological changes of the dynamic graph structure, generate the hint function test results and test feedback data;

[0107] S7. Adjust the hint function logic design and parameters according to the test feedback data, and use the dynamically optimized topology structure of the graph to iteratively optimize the hint function performance in multiple rounds.

[0108] In this embodiment, the multi-dimensional operation data specifically includes the running speed of the heavy-haul locomotive, station information, line signals, cylinder pressure, speed limit, and neutral section passing.

[0109] In this embodiment, S2 specifically includes:

[0110] S21. Conduct a preliminary cleaning of the data set, identify noise data based on statistical distribution characteristics, and remove the data that does not conform to the normal operating state through filtering techniques;

[0111] S22. Eliminate outliers from the cleaned data set. By calculating the deviation value and dispersion degree of the data distribution, identify the extreme outliers, and use the interpolation method to fill in the missing or abnormal data;

[0112] S23. Normalize the multi-dimensional features in the data set, map the feature values to a unified range, and eliminate the influence between different feature dimensions;

[0113] S24. Based on the normalized data set, use the Pearson correlation coefficient to calculate the linear correlation between the operation characteristics:

[0114]

[0115] where r ij represents the correlation coefficient between operation feature i and operation feature j, x ik represents the value of operation feature i in the k-th data, x jk represents the value of operation feature j in the k-th data, represents the mean value of operation feature i, represents the mean value of operation feature j, and n represents the total number of data samples;

[0116] S25. According to the results of the operation feature correlation analysis, define the relationship between nodes and edges, where nodes represent operation features and edges represent the correlation between operation features:

[0117]

[0118] Among them, w ij represents the weight between node i and node j, and |r ij | represents the absolute value of the correlation coefficient between operating feature i and operating feature j, and τ represents the set correlation threshold;

[0119] S26. Generate an initial topology graph structure according to the definition of operating feature nodes and edges, where the layout of nodes and edges dynamically reflects the association between operating features.

[0120] In this embodiment, the specific steps of S3 are as follows:

[0121] S31. Input the generated initial topology graph structure into a graph convolutional network and initialize the node feature matrix:

[0122]

[0123] Among them, H (0) represents the node feature matrix, represents the initial feature vector of node i, and N represents the total number of nodes;

[0124] S32. Normalize the adjacency matrix of the initial topology graph, add self-loops to enhance the connectivity of the graph, and calculate the normalized adjacency matrix:

[0125]

[0126] Among them, represents the normalized adjacency matrix, A represents the original adjacency matrix, I represents the identity matrix, represents the degree matrix of A + I;

[0127] S33. In each layer of graph convolution, perform a convolution operation on the node features using a variable topology adjustment mechanism:

[0128]

[0129] Among them, H (l+1) represents the node feature matrix of the (l + 1)-th layer, σ represents the activation function, N(i) represents the set of neighborhood nodes of node i, represents the weight value in the normalized adjacency matrix, W (l) represents the training weight matrix of the l-th layer, c ij represents the normalization factor, α represents the dynamic adjustment coefficient, represents the feature vector of node j in the l-th layer;

[0130] S34. Perform multi-scale fusion on the output features of multiple layers of convolution, and integrate features of different levels through a concatenation operation to generate multi-scale features;

[0131] S35. Dynamically adjust the weight contribution of each layer of features according to the sparsity of multi-scale features, and introduce a sparsity-driven weighting mechanism:

[0132]

[0133] Among them, H (weighted) represents the feature representation after weighted fusion, L represents the total number of layers, exp represents the exponential function, β represents the sparse adjustment coefficient, and H (l) represents the node feature matrix of the l-th layer, ||H (l) ||1 represents the sparsity of the node feature matrix of the l-th layer, and ||H (m) ||1 represents the sparsity of the node feature matrix of the m-th layer;

[0134] S36. Perform a linear transformation on the fused weighted features through a fully connected layer or a mapping matrix to generate a high-dimensional feature vector.

[0135] In this embodiment, the specific steps of S4 include:

[0136] S41. Based on the high-dimensional feature vector, construct a causal relationship model G=(V, E, W), where V is the set of nodes, representing the features related to the hint function; E is the set of edges, representing the causal relationship between features; W is the set of edge weights, used to characterize the strength of the causal influence between features;

[0137] S42. Quantify the causal strength of each edge in the edge set E, and calculate the mutual information entropy of the causal strength based on Bayesian network inference:

[0138]

[0139] Among them, C pq represents the causal strength between node p and node q, P(x p , x q ) represents the joint probability distribution of node p and node q, P(x p ) represents the marginal probability distribution of node p, and P(x q ) represents the marginal probability distribution of node q;

[0140] S43. In the causal path optimization, combine the inference results of the Bayesian network to dynamically optimize the path set:

[0141]

[0142] Among them, F path represents the path optimization objective function, P represents the set of candidate causal paths, and Ω pqDenote the correlation weight between node p and node q, exp represents the exponential function, γ represents the degree value attenuation coefficient, d q represents the degree value of node q, Δ pq represents the path length between node p and node q, and λ represents the penalty factor of the path length;

[0143] S44. Normalize the overall weight of the causal path:

[0144]

[0145] where, W pq represents the normalized causal weight matrix, and N(p) represents the set of neighborhood nodes of node p;

[0146] S45. Assign priorities to the input node features of the hint function according to the normalized causal weight matrix, and give priority to the nodes with high weights to be included in the optimization process:

[0147]

[0148] where, P p represents the priority of node p, and Φ(x q ) represents the feature importance score of node q;

[0149] S46. Output the key causal nodes and causal paths that have an important impact on the hint function.

[0150] In this embodiment, the S5 specifically includes:

[0151] S51. Initialize the initial topological graph structure according to the key causal nodes and causal paths to generate a population P = {P (1) , P (2) , …, P (M)}, where m represents the individual label, M represents the number of individuals in the population, V represents the set of nodes, E represents the set of edges, represents the set of edge weights, represents the edge weight between node v p and node v q in individual m, represents the set of node priorities, represents the priority of node v p in individual m;

[0152] S52. In the global search stage, introduce the strong individuals of the eagle flock optimization algorithm and define the global objective function:

[0153]

[0154] where, represents the global objective function, N(v p ) represents the set of neighbor nodes of node v p , C pq represents node v p and node v q the causal strength between them, represents the degree value of node v p , represents the degree value of node v q , μ represents the degree value attenuation factor;

[0155] S53. In the local optimization stage, the weak individuals introducing the eagle swarm optimization algorithm fine-tune the edge weights according to the global search results of the strong individuals:

[0156]

[0157] where represents the updated edge weight, δ represents the edge weight update coefficient, tanh represents the hyperbolic tangent function, ζ represents the priority influence factor, represents the priority of node v q in individual m, represents the priority of node v r in individual m;

[0158] S54. Perform adaptive adjustment on the node priorities and update the node priorities according to the topological change situation:

[0159]

[0160] where represents the updated node priority, exp represents the exponential function, η represents the priority adjustment coefficient, |N(v p )| represents the number of neighbor nodes of node v p ;

[0161] S55. Evaluate the fitness of the individuals:

[0162]

[0163] where represents the fitness function, and θ1, θ2, and θ3 represent the weight parameters;

[0164] S56. Update the eagle swarm through multiple rounds of iteration. In each round of iteration, select the individual with the highest fitness as the new center, and transfer the node priorities and edge weight distributions to the next generation of individuals. By repeating the global search and local optimization processes, finally obtain the optimized topological structure.

[0165] In this embodiment, the S6 specifically includes:

[0166] S61. Dynamically verify the optimized prompt function logic, set different operating scenarios, including normal operation, ramps, sharp turns, and bad weather, and record the trigger time, trigger conditions, and prompt content of the prompt function in each scenario;

[0167] S62. In each operating scenario, collect the signal data and test data output by the prompt function, including the accuracy, real-time performance, and adaptability to the operating state of the prompt signal;

[0168] S63. Dynamically analyze the test data, evaluate the performance of the prompt function in different operating scenarios by comparing the prompt signal with the actual operating state, and calibrate the error range and potential optimization space of the prompt signal;

[0169] S64. Use visualization technology to display the topological changes of the dynamic graph structure, including the process of node priority, edge weight adjustment, and causal path dynamic changes;

[0170] S65. Generate the test results of the prompt function, including the accuracy of the trigger time, the adaptability of the prompt content, and the real-time performance of the prompt response;

[0171] S66. Output the test results and scenario analysis data as test feedback data, and record the visualization log generated during the test process.

[0172] Reference Figure 2 , the test system for the prompt function of the visualization heavy-haul locomotive driving guidance device, includes the following modules:

[0173] Multidimensional operation data acquisition module, used to acquire the multidimensional operation data of the heavy-haul locomotive;

[0174] Data preprocessing module, used to process the multidimensional operation data and generate the initial topological graph structure;

[0175] Feature extraction module, used to extract features from the initial topological graph structure through a graph convolutional network to generate high-dimensional feature vectors;

[0176] Causal relationship analysis module, used to construct a causal relationship model based on the high-dimensional feature vectors to identify key causal nodes and causal paths;

[0177] Optimization calculation module, used to dynamically optimize the initial topological graph structure using the eagle flock optimization algorithm, adjust the node priority and edge weight, and optimize the prompt function logic;

[0178] Dynamic verification module, used to dynamically verify the optimized prompt function logic, collect test data, and generate test results;

[0179] A visualization display module for displaying the topological changes of a dynamic graph structure and generating test feedback data for the hint function;

[0180] A feedback and storage module for storing the recorded test feedback data and supporting multi-round iterative optimization of the hint function logic.

[0181] Example 1:

[0182] To verify the feasibility of the present invention in implementation, the present invention is applied to a certain heavy-haul railway transportation line with a total length of 120 kilometers. The line is complex and changeable, including 20 kilometers of continuous steep slopes, 12 sharp turns, and 4 common bad weather sections (foggy, rainy, etc.). During operation, the locomotive needs to ensure driving safety in a complex environment while optimizing traction energy consumption, and provide accurate hint information in real time to support the driver's operation decision-making. The traditional hint function shows problems of insufficient adaptability and lagging hint response in this scenario, being unable to effectively handle steep slopes and sharp turn areas, with a hint error rate as high as 18%, and the average hint delay time being 2.5 seconds, seriously affecting the driver's operation efficiency and driving safety.

[0183] In this embodiment, the test system of the present invention is used to dynamically optimize and verify the hint function. First, during the operation of the heavy-haul locomotive, data such as speed, station information, line signals, cylinder pressure, speed limits, and overrun are obtained in real time through the multi-dimensional operation data acquisition module and recorded at a millisecond-level frequency. The data preprocessing module filters out noise and removes outliers from these data, and after normalization processing, an initial topological graph structure is generated. Each operation feature is mapped to a node in the graph, and the correlation between features is defined as the edge weight. Subsequently, the generated initial topological graph is input into the feature extraction module, and a graph convolutional network is used to perform multi-level extraction on the node and its neighborhood features. The generated high-dimensional feature vectors provide high-quality input data for the optimization of the hint logic.

[0184] Based on the feature extraction, the causal relationship analysis module constructs a causal relationship model through causal inference methods and identifies the key causal nodes and paths of the hint function. For example, in the steep slope area, traction force, speed, and slope are identified as the nodes with the greatest influence on the hint content, while in the sharp turn area, the causal weights of acceleration, braking force, and weather information are relatively high. The optimization calculation module dynamically optimizes the hint logic by combining the eagle flock optimization algorithm, adjusts the priorities of the key nodes, and redistributes the edge weights to ensure that the hint function can quickly respond to changes in the operation state in complex scenarios.

[0185] To verify the optimized hint function logic, the dynamic verification module sets up various scenarios such as continuous steep slopes, sharp curves, and bad weather, and collects the trigger time, hint content, and the matching degree with the actual operating state of the hint function in these scenarios. The visualization display module dynamically displays the topological changes of the graph structure, including the priority adjustment of key nodes and the edge weight optimization process, and generates detailed hint function test feedback data.

[0186] The experimental results show that the optimized hint function performs better than the traditional method in complex scenarios. For example, in the continuous steep slope area, the accuracy of the hint content is increased to 98%, and the hint delay time is reduced to 0.8 seconds; in the sharp curve area, the hint error rate is reduced from the original 22% to 6%; in the bad weather section, the adaptability of the hint content is significantly enhanced, and the satisfaction of the driver with the hint information is increased to 92%. In addition, the optimized hint function significantly reduces the traction energy consumption, saving about 8% of the energy consumption during the whole operation process. The following Table 1 shows the specific experimental data.

[0187] Table 1 Comparative Analysis of the Performance of the Hint Function before and after Optimization and the Improvement Effect

[0188]

[0189] Through the effect analysis of this embodiment, it can be seen that the present invention has achieved remarkable results in solving the technical bottlenecks of the traditional hint function. For complex scenarios such as continuous steep slopes, sharp curves, and bad weather, the optimized hint function of the present invention exhibits extremely high adaptability and real-time performance. In the continuous steep slope scenario, the hint accuracy rate is increased from 85% of the traditional method to 98%, the hint delay time is significantly shortened to 0.8 seconds, and the hint error rate is reduced to 4%, providing more accurate operation suggestions for the driver and effectively improving the driving safety. In the sharp curve scenario, the hint error rate drops from 22% to 6%, and the hint adaptability is increased by 13 percentage points, which fully shows that the present invention can dynamically adapt to the operation requirements of complex curves. Under bad weather conditions, the satisfaction of the driver with the hint information is increased from 68% to 92%, further verifying the robustness and reliability of the present invention under complex operating conditions. In addition, the present invention reduces the energy consumption by optimizing the hint logic, saving about 8% of the traction energy, showing strong economic and environmental benefits.

[0190] This embodiment fully proves that the present invention effectively solves the problems of insufficient adaptability, high error, response lag, and lack of energy consumption optimization of the traditional hint function in complex scenarios. By comprehensively improving the accuracy, real-time performance, and adaptability of the hint function, it further provides technical support for the intelligence, economy, and safety of the heavy-haul locomotive driving guidance device, showing a broad application prospect in railway transportation.

[0191] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, shall be covered by the protection scope of the present invention.

Claims

1. A test method for the prompting function of a visualization locomotive operation guidance device for heavy-haul locomotives, characterized in that, It includes the following steps: S1. Collect multi-dimensional operation data of the heavy-haul locomotive through sensors and construct a data set; S2. Perform noise filtering, outlier removal, and normalization processing on the data set, and generate an initial topological graph structure according to data relevance; S3. Use a graph convolutional network to extract features from the initial topological graph structure, capture multi-scale relationships between nodes and neighborhoods through multi-layer convolution, and generate high-dimensional feature vectors; S4. Based on the high-dimensional feature vectors, construct a causal relationship model through a causal inference method, analyze the causal relationships between nodes, and identify key causal nodes and causal paths that have an important impact on the prompting function; S5. According to the key causal nodes and causal paths, use the eagle swarm optimization algorithm to dynamically optimize the initial topological graph structure, combine global search and local optimization, adjust node priorities and edge weights, and optimize the prompting function logic; S6. Dynamically verify the optimized prompting function logic, generate a prompting function test result and test feedback data by visualizing the topological changes of the dynamic graph structure; S7. Adjust the prompting function logic design and parameters according to the test feedback data, and use the dynamically optimized topological graph structure to iteratively optimize the performance of the prompting function in multiple rounds.

2. The test method for the prompting function of the visualization heavy-haul locomotive driving guidance device according to claim 1, characterized in that, The multi-dimensional operation data specifically includes the running speed of the heavy-haul locomotive, station information, line signals, cylinder pressure, speed limits, and neutral section passing.

3. The test method for the prompting function of the visualization heavy-haul locomotive driving guidance device according to claim 1, characterized in that, The S2 specifically includes: S21. Perform preliminary cleaning on the data set, identify noise data based on statistical distribution characteristics, and remove data that does not conform to the normal operating state through filtering techniques; S22. Remove outliers from the cleaned data set, identify extreme outliers by calculating the deviation value and dispersion degree of the data distribution, and use interpolation methods to fill in missing or abnormal data; S23. Perform normalization processing on the multi-dimensional features in the data set, map the feature values to a unified range, and eliminate the influence between different feature dimensions; S24. Based on the normalized data set, calculate the linear correlation between operation features using the Pearson correlation coefficient: where r ij represents the correlation coefficient between operating feature i and operating feature j, x ik represents the value of operating feature i in the k-th data, x jk represents the value of operating feature j in the k-th data, represents the mean value of operating feature i, represents the mean value of operating feature j, and n represents the total number of data samples; S25. According to the analysis results of the operation feature correlation, define the relationship between nodes and edges, where nodes represent operation features and edges represent the association between operation features: where w ij represents the weight between node i and node j, and |r ij | represents the absolute value of the correlation coefficient between operating feature i and operating feature j, and τ represents the set correlation threshold; S26. According to the definition of operation feature nodes and edges, combined with the correlation analysis results, generate an initial topological graph structure, where the layout of nodes and edges dynamically reflects the association between operation features.

4. The test method for the prompt function of the visualization heavy-haul locomotive driving guidance device according to claim 1, characterized in that, The S3 specifically includes: S31. Input the generated initial topological graph structure into the graph convolutional network and initialize the node feature matrix: Among them, H (0) represents the node feature matrix, represents the initial feature vector of node i, and N represents the total number of nodes; S32. Normalize the adjacency matrix of the initial topological graph, add self-loops to enhance the connectivity of the graph, and calculate the normalized adjacency matrix: Among them, represents the normalized adjacency matrix, A represents the original adjacency matrix, and I represents the identity matrix. represents the degree matrix of A + I; S33. In each layer of graph convolution, perform a convolution operation on the node features using a variable topology adjustment mechanism: Among them, H (l+1) represents the node feature matrix of the (l + 1)-th layer, σ represents the activation function, and N(i) represents the set of neighboring nodes of node i. represents the weight value in the normalized adjacency matrix, and W (l) represents the training weight matrix of the l-th layer, and c ij represents the normalization factor, and α represents the dynamic adjustment coefficient. represents the feature vector of node j in the l-th layer; S34. Perform multi-scale fusion on the output features of multi-layer convolution, integrate features at different levels through a splicing operation, and generate multi-scale features; S35. Dynamically adjust the weight contribution of each layer of features according to the sparsity of the multi-scale features, and introduce a sparsity-driven weighting mechanism: Among them, H (weighted) represents the feature representation after weighted fusion, L represents the total number of layers, exp represents the exponential function, β represents the sparse adjustment coefficient, and H (l) represents the node feature matrix of the l-th layer, ||H (l) ||1 represents the sparsity of the node feature matrix of the l-th layer, and ||H (m) ||1 represents the sparsity of the node feature matrix of the m-th layer; S36. Linearly transform the fused weighted features through a fully connected layer or a mapping matrix to generate a high-dimensional feature vector.

5. The test method for the prompting function of the visualization heavy-haul locomotive driving guidance device according to claim 1, characterized in that, The specific steps of S4 are as follows: S41. Based on the high-dimensional feature vector, construct a causal relationship model G = (V, E, W), where V is the set of nodes representing features related to the prompting function; E is the set of edges representing the causal relationships between features; W is the set of edge weights used to characterize the strength of causal influence between features. S42. Quantify the causal strength of each edge in the edge set E, and calculate the mutual information entropy of the causal strength based on Bayesian network inference: Among them, C pq represents the causal strength between node p and node q, P(x p , x q ) represents the joint probability distribution of node p and node q, P(x p ) represents the marginal probability distribution of node p, and P(x q ) represents the marginal probability distribution of node q; S43. In the causal path optimization, combine the inference results of the Bayesian network to dynamically optimize the path set: Among them, F path represents the path optimization objective function, P represents the set of candidate causal paths, and Ω pq represents the correlation weight between nodes p and q, exp represents the exponential function, γ represents the degree value attenuation coefficient, and d q represents the degree value of node q, and Δ pq represents the path length between nodes p and q, and λ represents the penalty factor of the path length; S44. Normalize the overall weight of the causal path: Among them, W pq represents the normalized causal weight matrix, and N(p) represents the set of neighboring nodes of node p; S45. According to the normalized causal weight matrix, assign priorities to the input node features of the prompting function, and give priority to the high-weight nodes in the optimization process: Among them, P p represents the priority of node p, and Φ(x q ) represents the feature importance score of node q; S46. Output the key causal nodes and causal paths that have an important impact on the prompting function.

6. The test method for the prompting function of the visualization heavy-haul locomotive operation guidance device according to claim 1, wherein The specific steps of S5 are as follows: S51. Initialize the initial topology graph structure based on key causal nodes and causal paths to generate a population P = {P (1) , P (2) , …, P (M)}, where m represents the individual label, M represents the number of individuals in the population, V represents the node set, E represents the edge set, represents the edge weight set, represents the edge weight between node v p and node v q in individual m, represents the node priority set, represents the priority of node v p in individual m; S52. In the global search stage, introduce the strong individuals of the eagle flock optimization algorithm and define the global objective function: Among them, represents the global objective function, N(v p ) represents the set of neighbor nodes of node v p , C pq represents the causal strength between node v p and node v q ; represents the degree value of node v p ; represents the degree value of node v q ; μ represents the degree value attenuation factor. S53. In the local optimization stage, introduce the weak individuals of the eagle flock optimization algorithm to fine-tune the edge weights according to the global search results of the strong individuals: Among them, represents the updated edge weight, δ represents the edge weight update coefficient, tanh represents the hyperbolic tangent function, and ζ represents the priority influence factor. represents the priority of node v in individual m q . represents the priority of node v in individual m r , and C pq represents the causal strength between node p and node q. S54. Adaptively adjust the node priorities and update the node priorities according to the topological changes: Among them, represents the updated node priority, exp represents the exponential function, η represents the priority adjustment coefficient, |N(v p )| represents the number of neighbor nodes of node v p ; S55. Evaluate the fitness of the individuals: Among them, represents the fitness function, and θ1, θ2, and θ3 represent weight parameters; S56. Through multiple rounds of iteration, update the eagle flock. In each round of iteration, select the individual with the highest fitness as the new center, and transfer the node priorities and edge weight distributions to the next generation of individuals. By repeating the global search and local optimization processes, finally obtain the optimized topological structure.

7. The test method for the prompting function of the visual heavy-haul locomotive operation guidance device according to claim 1, characterized in that, The specific steps of S6 are as follows: S61. Dynamically verify the optimized prompting function logic, set different operating scenarios, including normal operation, slopes, sharp turns, and bad weather, and record the trigger time, trigger conditions, and prompt content of the prompting function in each scenario; S62. In each operating scenario, collect the signal data and test data output by the prompting function, including the accuracy, real-time performance, and adaptability to the operating state of the prompt signal; S63. Dynamically analyze the test data. By comparing the prompt signal with the actual operating state, evaluate the performance of the prompting function in different operating scenarios, and calibrate the error range and potential optimization space of the prompt signal; S64. Use visualization technology to display the topological changes of the dynamic graph structure, including the processes of node priority adjustment, edge weight adjustment, and causal path dynamic changes; S65. Generate the test results of the prompting function, including the accuracy of the trigger time, the adaptability of the prompt content, and the real-time performance of the prompt response; S66. Output the test results and scenario analysis data as test feedback data, and record the visualization logs generated during the test process.

8. A test system for the prompting function of the visual heavy-haul locomotive driving guidance device, which executes the test method for the prompting function of the visual heavy-haul locomotive driving guidance device according to any one of claims 1 to 7, characterized in that, It includes the following modules: A multi-dimensional operation data acquisition module for acquiring multi-dimensional operation data of a heavy-haul locomotive; A data preprocessing module for processing the multi-dimensional operation data to generate an initial topological graph structure; A feature extraction module for extracting features from the initial topological graph structure through a graph convolutional network to generate high-dimensional feature vectors; A causal relationship analysis module for constructing a causal relationship model based on the high-dimensional feature vectors to identify key causal nodes and causal paths; An optimization calculation module for dynamically optimizing the initial topological graph structure using the eagle flock optimization algorithm, adjusting node priorities and edge weights, and optimizing the hint function logic; A dynamic verification module for dynamically verifying the optimized hint function logic, collecting test data and generating test results; A visualization display module for displaying the topological changes of the dynamic graph structure and generating test feedback data for the hint function; A feedback and storage module for storing and recording the test feedback data and supporting multi-round iterative optimization of the hint function logic.

Images