An intelligent rail transit signal system and its automatic dispatching method

Abstract

The present invention discloses an intelligent rail transit signal system and an automatic dispatching method, which relate to the technical field of rail transit. The method includes data acquisition, machine learning modeling, a dispatching optimization algorithm based on machine learning to generate a dispatching strategy, real-time signal control, automatic dispatching decision-making, system monitoring and optimization. The present invention solves the problem that the rail transit data generation samples do not conform to the actual physical laws by adding physical constraints to guide the generation process; solves problems such as insufficient data generation diversity and generation mode collapse through methods such as optimizing the generator training by a dynamic memory mechanism. The negative entropy incentive function is used to solve problems such as high-dimensional redundancy of rail transit data, non-linear structural features, and poor model generalization ability.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit, and in particular to an intelligent rail transit signal system and its automatic dispatching method. Background Art

[0002] The rail transit system is an important part of the modern urban transportation network, and its operation efficiency and safety have an important impact on the development of the city and the travel quality of residents. However, with the acceleration of the urbanization process, the challenges faced by rail transit are becoming increasingly severe, including the rapid growth of passenger flow, the complexity of the operation environment, and the frequent occurrence of emergencies. These problems pose higher requirements for the automatic dispatching and signal control of the rail transit system.

[0003] In the existing rail transit automatic dispatching methods, the data expansion method fails to effectively consider the physical constraints of rail transit data, resulting in poor quality of generated rail transit data, class imbalance in the distribution of rail transit data, the traditional dimensionality reduction method being unable to effectively handle the high-dimensional redundancy of rail transit data, failing to capture the important topological features in rail transit data, and the dimensionality reduction model being vulnerable to the local optimum problem, making it difficult to achieve stable performance in complex rail transit scenarios. Summary of the Invention

[0004] In order to solve the technical problems existing in the existing rail transit automatic dispatching, the present invention provides an intelligent rail transit signal system and its automatic dispatching method.

[0005] The present invention is achieved through the following technical solutions:

[0006] An intelligent rail transit signal automatic dispatching method, comprising:

[0007] S1: Data collection, real-time collection of data during rail transit operation, and annotation and preprocessing of the data;

[0008] S2: Machine learning modeling, using historical data and real-time data to establish a prediction model, and generating a dispatching strategy based on a machine learning-based dispatching optimization algorithm;

[0009] S3: Real-time signal control, dynamically adjusting the signal state according to the system decision;

[0010] S4: Automatic dispatching decision-making, generating an optimized dispatching plan for trains by integrating system information and the results of machine learning modeling;

[0011] S5: System monitoring and optimization, real-time monitoring of the operation state of the entire rail transit system, and continuously optimizing the system performance according to the feedback.

[0012] Further, step S1 further includes performing small sample recognition on the collected data set by using statistical analysis and machine learning algorithms. If it is detected that there are problems such as class imbalance or insufficient samples, a generative adversarial network algorithm based on an augmentation strategy is used as a sample augmentation model to augment rail transit data;

[0013] The training process of the generative adversarial network algorithm based on the augmentation strategy includes initializing the standard structure of the generative adversarial network, including the network architectures of the generator and the discriminator; the generator is used to generate samples close to the real rail transit data distribution, and the discriminator is used to judge whether the rail transit data is a real sample or a false sample generated by the generator.

[0014] Further, during the training process of the generative adversarial network, a physical constraint function is also added to the loss function of the generator.

[0015] Further, during the training process of the generative adversarial network, a dynamic memory mechanism is also used to store and update effective sample features and the output information of the generator in the historical training process.

[0016] Further, the game relationship between the generator and the discriminator is dynamically balanced through an adaptive step size adjustment mechanism, and the learning rates of the generator and the discriminator are dynamically adjusted according to the changes in the current training state and the target distribution.

[0017] Further, step S2 further includes: using an autoencoder based on adaptive gradient correction as a feature dimensionality reduction model, and optimizing the autoencoder network by using a dynamic adjustment mechanism and an adaptive gradient correction method based on data topology analysis.

[0018] Further, an activation function driven by negative entropy is used for the output of the encoding layer.

[0019] Further, the entropy weight of the rail transit data features is dynamically adjusted, and a method based on topological rail transit data analysis is used to capture the non-linear structure features of the rail transit data.

[0020] The present invention also provides an intelligent rail transit system, based on the automatic scheduling method as described above, which includes:

[0021] The intelligent rail transit signal system includes a data acquisition unit, a machine learning modeling unit, a real-time signal control unit, an automatic scheduling decision-making unit, and a system monitoring and optimization unit;

[0022] The data acquisition unit is used to collect key data during the operation of the rail transit in real time, and perform annotation and augmentation;

[0023] The machine learning modeling unit is used to establish a prediction model by using historical data and real-time data, develop a scheduling optimization algorithm based on machine learning, and generate a scheduling strategy;

[0024] The real-time signal control unit is used to receive the predictions and instructions provided by the machine learning modeling unit and the automatic scheduling decision unit, and dynamically adjust the signal state according to the system decision;

[0025] The automatic scheduling decision unit is used to generate an optimized scheduling plan for the train by integrating system information and the results of machine learning modeling;

[0026] The system monitoring and optimization unit is used to monitor the operating status of the entire rail transit system in real time, and continuously optimize the system performance according to the feedback.

[0027] In addition, to achieve the above object, the present invention also provides a computer-readable storage medium, on which program instructions of an intelligent rail transit signal automatic scheduling method are stored, and the program instructions of the intelligent rail transit signal automatic scheduling method can be executed by one or more processors to implement the steps of the intelligent rail transit signal automatic scheduling method as described above.

[0028] Compared with the prior art, the beneficial effects of the present invention are:

[0029] (1) By adding physical constraints to guide the generation process, the present invention solves the problem that the generated samples of rail transit data do not conform to the actual physical laws.

[0030] (2) The present invention performs data augmentation, optimizes the training of the generator through a dynamic memory mechanism, solves the problem of insufficient diversity in data generation, and balances the adversarial training of the generator and the discriminator through adaptive learning rate adjustment to avoid the problem of generation mode collapse.

[0031] (3) The present invention adopts an autoencoder based on adaptive gradient correction, uses a negative entropy activation function to enhance the preference for high-value features, solves the problem of high-dimensional redundancy of rail transit data, combines topological data analysis methods to capture the non-linear structural features in rail transit data, solves the problem that traditional dimensionality reduction methods are difficult to retain key topological information, and optimizes the dimensionality reduction process through a dynamic entropy weight adjustment mechanism to solve the problem of poor model generalization ability caused by data distribution heterogeneity. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0033] Figure 1Schematic diagram of the intelligent rail transit signal automatic scheduling method according to an embodiment of the present application;

[0034] Figure 2 Correlation analysis diagram between performance and physical constraint intensity according to an embodiment of the present application;

[0035] Figure 3 Data diversity comparison diagram between the augmented algorithm and the conventional generative adversarial network algorithm under different training rounds according to an embodiment of the present application;

[0036] Figure 4 Comparison diagram of the decline trend of the loss value during the training process between the adaptive gradient correction autoencoder and the conventional method according to an embodiment of the present application;

[0037] Figure 5 Distributions of the original data, the reconstructed data by the conventional technology, and the reconstructed data by the adaptive gradient correction autoencoder according to an embodiment of the present application. Detailed implementation manners

[0038] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0039] The following uses specific specific examples to illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. The present invention can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention.

[0040] It should also be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. The drawings only show the components related to the present invention, rather than being drawn according to the number, shape, and size of the components in actual implementation. The forms, numbers, and proportions of the components in actual implementation can be arbitrarily changed, and the component layout form may also be more complex.

[0041] Refer to Figure 1 , an intelligent rail transit signal automatic scheduling method, including the following steps:

[0042] S1: Data collection, including:

[0043] S11: Collect the data during the operation of rail transit in real time, and after annotation, use it as the training data set for the machine learning model;

[0044] The collected data comes from various sensors in the rail transit operation environment, including but not limited to track sensors, camera monitoring devices, on-vehicle devices, and infrastructure monitoring devices;

[0045] The data is collected through real-time dynamic capture to ensure that the information collected can reflect the immediate state of the train and the real-time changes in the track environment;

[0046] The collected data is first preliminarily formatted by the local processing node and then stored in a high-performance data server in a structured data format. The main storage format is a time series database, which is convenient for subsequent real-time query and historical data analysis.

[0047] In one embodiment, the collected data includes:

[0048] Location information , recording the specific location of the train on the track; Speed , the real-time running speed of the train; Acceleration , the real-time acceleration of the train; Train number information , the train number; Station stop time , the stop time of the train at each station; Signal status , the status information of the signal lights on the track; Number of passengers , the real-time passenger number data; Energy consumption data , the energy consumption situation during the train operation; Equipment status , the operation status of on-vehicle and track equipment; Environmental factors , such as environmental parameters like temperature and humidity.

[0049] It should be noted that this embodiment is only to illustrate a data format and type of the present invention. In actual applications, the attributes of the data are usually more than 10 attributes, and the number of data attributes may reach dozens or even hundreds.

[0050] S12: Annotate the collected data. Optionally, the annotation method is manual annotation;

[0051] In one embodiment, the annotation categories include:

[0052] 0 - Normal operation: The train operates normally according to the timetable;

[0053] 1 - Delay: The train operation time exceeds the predetermined timetable;

[0054] 2 - Emergency stop: The train stops due to a fault or an emergency;

[0055] 3 - Maintenance: The train undergoes daily or unscheduled maintenance;

[0056] 4 - Abnormal behavior: The train exhibits behaviors in abnormal operating modes, such as abnormal speed.

[0057] S13: Data preprocessing; cleaning and screening the collected data, removing outliers and noise, and providing accurate data input for the machine learning modeling unit.

[0058] S14: Small - sample data identification; performing small - sample identification on the collected dataset by using statistical analysis and machine learning algorithms. For example, using distribution detection algorithms (such as based on Kullback - Leibler divergence or distribution deviation metrics) to evaluate whether there are problems of class imbalance or insufficient samples in the data distribution.

[0059] Since the acquisition, annotation, and preprocessing of training data are time - consuming and labor - intensive, and insufficient training samples easily lead to poor model generalization ability and affect the model's accuracy; detecting possible small - sample data problems in the small - sample data identification functional detection system, and ensuring the sufficiency and reliability of model training data through data augmentation techniques;

[0060] S15: If it is detected that there are problems of class imbalance or insufficient samples, then use the generative adversarial network algorithm based on the augmentation strategy as a sample augmentation model for rail transit data augmentation. Through an adaptive data augmentation strategy and discriminator quality control method, enhance the diversity and stability of the generated rail transit data. At the same time, adopt an adaptive learning rate adjustment mechanism to balance the game between the generator and the discriminator, and improve the training efficiency and stability.

[0061] Specifically, the training process of the generative adversarial network algorithm based on the augmentation strategy is as follows:

[0062] (1) Initialize the standard structure of the generative adversarial network, including the network architectures of the generator and the discriminator; the task of the generator is to generate samples as close as possible to the real rail transit data distribution, and the discriminator is responsible for judging whether the rail transit data is a real sample or a fake sample generated by the generator. The way the generator generates rail transit data is expressed as:

[0063]

[0064] In the formula, is the generated sample, is the generator function; is the input noise vector, representing randomness; is the parameter of the generator, representing the weight of the generator. Preferably, is set to the initial random value.

[0065] Furthermore, the loss function of the generator is defined as:

[0066]

[0067] wherein, is the loss function of the generator, is the expectation operation; is the noise vector, representing the input of the generator; represents being subject to a specific distribution; is the distribution of the noise vector, representing the randomness of the input; is the weight hyperparameter of the physical constraint, controlling the influence of the physical constraint on the loss function; is the physical constraint function, a function indicating whether the generated rail transit data conforms to the physical constraints; is the discriminator function. Preferably, is set to 0.2.

[0068] Such as Figure 2 shown is the result graph of the correlation analysis between the performance and the physical constraint strength verified by experiments.

[0069] (2) Add the physical constraint function to the loss function of the generator;

[0070] The generation of rail transit data is not just a pure random noise transformation. It is also necessary to consider the physical laws of rail transit data in the real world. The generated rail transit data should conform to certain physical limitations or constraints. The present invention adopts a physical scene perception method, which not only focuses on the quality of the generated rail transit data, but also ensures that the generated rail transit data conforms to the physical laws or prior knowledge in a specific field. By adding a physical constraint term to guide the training of the generation network, the physical constraint function is added to the loss function of the generator to ensure that the generated samples are not only similar to the real rail transit data, but also satisfy the constraints of the physical laws. By adopting the physical constraint term, restrictions conforming to the physical laws in the rail transit field are imposed on the generated samples. For example, when expanding the rail transit flow data, it is ensured that the generated samples conform to the smooth law of the actual flow distribution over time or space; the calculation method of the physical constraint function is expressed as:

[0071]

[0072] wherein, is the Laplacian operator of the generated rail transit data, representing the spatial smoothness or change rate of the rail transit data; is a predefined physical constant or constraint value. Preferably, is defined as the variance value of the real rail transit data input to the generative adversarial network in the current batch.

[0073] (3) During the training process of the generative adversarial network, a dynamic memory mechanism is adopted to store and update the effective sample features and the output information of the generator in the historical training process;

[0074] The dynamic memory mechanism adaptively selects and updates the historical rail transit data according to the loss function and the quality of the generated samples during the training process, provides high-quality sample memories for the subsequent training of the generator, and ensures that the generator continuously optimizes the quality of the generated rail transit data. The dynamic memory mechanism is implemented through a memory module. For example, when processing the passenger flow data of different stations, the dynamic memory mechanism can help the generator more effectively capture and reproduce the characteristic differences between different stations, and avoid generating single or repetitive data. The update and implementation method of the memory module is expressed as:

[0075]

[0076] In the formula, is the sample feature of the -th iteration memory, representing the state of the memory module in the -th iteration; is the sample feature of the -th iteration memory, representing the state of the memory module in the -th iteration; is the updated weight coefficient, representing the balance factor of memory update; is the weighted factor based on the quality of the generated samples, representing the importance of the samples. Preferably, is set to 0.3.

[0077] Furthermore, the weighted factor based on the quality of the generated samples enables the memory mechanism to dynamically retain the generated samples that are closest to the distribution of the real rail transit data, thereby improving the training quality of the generator. The calculation method is expressed as:

[0078]

[0079] In the formula, is the L2 norm; is the real rail transit data, representing the real sample; is the total number of generated samples in the current batch; represents the -th real rail transit data; is the smoothing factor, a small constant to avoid the denominator being zero; is the feature weight matrix of the samples. Preferably, is set to .

[0080] Furthermore, to achieve adaptive non-local feature interaction, non-local operations are adopted to enhance global feature interaction. The mutual relationship between different positions of the generated samples is represented by a feature weight matrix, which helps capture long-range dependencies in rail transit data. For example, the connections between different stations or the correlation of passenger flow volumes at different time periods, making the generated samples more in line with the characteristics of the actual rail transit network. Especially, it has a significant effect when simulating the collaborative flow changes of multiple stations. The calculation method is expressed as:

[0081]

[0082]

[0083] In the formula, represents the dot product between features and and characterizes the similarity between features; is the feature value at the -th position of the generated rail transit data feature vector; is the feature value at the -th position of the generated rail transit data feature vector; is the feature dimension of the generated sample; is the weight of the -th feature of the generated sample relative to the -th feature.

[0084] (4) Conduct adversarial training of the generative adversarial network. The generator randomly generates rail transit data samples, and the discriminator judges their authenticity. The discriminator adjusts its parameters according to the feedback, making the rail transit data generated by the generator closer to the real distribution. Different from the traditional generative adversarial network, in the present invention, the generator combines the historical sample features stored in the dynamic memory mechanism during each training, adjusts the generation strategy according to the past training information, avoids falling into local optimal solutions, and improves the diversity and quality of the generated rail transit data. The discriminator loss function of the generative adversarial network is expressed as:

[0085]

[0086] In the formula, is the loss function of the discriminator; is the distribution of real rail transit data, characterizing the real distribution of rail transit data.

[0087] (5) During the training process, an adaptive data augmentation strategy is adopted to transform and enhance each batch of generated samples, including operations such as rotation, scaling, and translation, to simulate the diversity and complexity in the distribution of rail transit data. The augmentation strategy is dynamically adjusted according to the feedback of the discriminator, with a focus on strengthening the performance of the generated rail transit data in specific patterns, ensuring that the generator learns the diversity and regularity of rail transit data. For example, for the fluctuations in the passenger flow during the morning and evening rush hours, more samples during the peak hours are generated through the augmentation operation to improve the balance of the training data distribution. The implementation method of the augmentation strategy is expressed as:

[0088]

[0089] In the formula, is the augmentation strategy factor, representing the influence degree of the data augmentation operation; is the augmentation control factor, representing the weight of the discriminator feedback in the augmentation strategy; is the gradient of the augmentation operation, representing the adjustment direction of the augmentation strategy according to the discriminator feedback. Preferably, is set to 0.5.

[0090] Furthermore, the gradient of the augmentation operation is represented by the derivative of the difference between the generator output and the real rail transit data, and the calculation method is expressed as:

[0091]

[0092] In the formula, is the augmented sample, representing the rail transit data after the augmentation operation; is the partial derivative symbol.

[0093] Furthermore, the augmentation operation includes transformations such as translation, rotation, and scaling, and acts on the generated samples through the transformation function to generate new samples, expressed as:

[0094]

[0095] In the formula; is the transformation function, representing the specific augmentation operation; is the augmentation strategy factor, is the sample before augmentation.

[0096] In one embodiment, the implementation method of the transformation function includes the following specific transformations:

[0097] 1 - Translation operation, expressed as:

[0098]

[0099] In the formula, is the translation operation function, representing the translation transformation of the rail transit data; is the translation vector, representing the amplitude and direction of translation. Preferably, is set to a fixed value, such as 3.

[0100] 2 - Rotation operation, expressed as:

[0101]

[0102] In the formula, is the rotation operation function, representing the rotation transformation of rail transit data; is the rotation matrix, representing the rotation angle; is the rotation angle, representing the rotation amplitude. Preferably, is set to 30 degrees.

[0103] 3 - Scaling operation, expressed as:

[0104]

[0105] In the formula, is the scaling operation function, representing the scaling transformation of rail transit data; is the scaling factor, representing the scaling ratio. Preferably, is set to 1.2.

[0106] (6) During the training process, the discriminator optimizes its classification ability by combining the similarity information between the generated rail transit data and the historical rail transit data to distinguish between real rail transit data and generated rail transit data. At the same time, by combining the sensitivity of the generator to the sample transformation, fine - grained feature information of the generated rail transit data is added to the feedback of the discriminator to help the discriminator better adapt to the changes of the generator. For example, ensuring that the generated passenger flow data is consistent with the change trend of historical data. During this process, in order to maintain the stability of the training process, the loss function of the discriminator adopts a stability adjustment mechanism based on the quality of the generated rail transit data, expressed as:

[0107]

[0108] In the formula, is the loss function of the discriminator after stability adjustment; is the stability adjustment coefficient, representing the weight of quality control; is the quality control loss of the discriminator. Preferably, is set to 0.1.

[0109] Furthermore, the quality control loss of the discriminator combines the regularization term of the rail transit data quality, indicating that the discriminator not only judges the generated samples but also needs to evaluate the quality of the samples. The calculation method is expressed as:

[0110]

[0111] In the formula, is the quality control regularization coefficient, which characterizes the weight of the regularization term. Preferably, is set to 0.1.

[0112] (7) During the training process, the game relationship between the generator and the discriminator is dynamically balanced through an adaptive step size adjustment mechanism. The learning rates of the generator and the discriminator are dynamically adjusted according to the current training state and the change of the target distribution, so as to avoid excessive adjustment of the generator or the discriminator being too powerful, improve the stability and efficiency of training. In the rail transit data augmentation, by balancing the training, the mode collapse caused by improper parameter settings is reduced, and the diversity of the generated samples is ensured. The way of adaptive learning rate adjustment is expressed as:

[0113]

[0114] In the formula, is the learning rate of the generator and the discriminator, which characterizes the step size of parameter update; is the initial learning rate of the generator and the discriminator, which characterizes the starting step size of training; is the learning rate adjustment factor of the generator and the discriminator, which characterizes the sensitivity of step size adjustment. Preferably, is set to 0.1.

[0115] (8) Update the parameters of the generator and the discriminator according to the gradient descent method, and repeat the above steps until the preset stop iteration condition is met, which means the model training is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0116] After the rail transit data augmentation model training is completed, the trained rail transit data augmentation model is used to increase the number of samples. In one embodiment, if the original collected samples are 800, and the rail transit data augmentation model generates 200 samples, then the augmented rail transit data set contains 1000 samples.

[0117] As Figure 3 shown is the data diversity comparison chart of the augmentation algorithm and the conventional generative adversarial network algorithm under different training rounds.

[0118] S2: Machine learning modeling; It is realized to provide prediction models and optimization algorithm support for the intelligent decision-making of the system, including the following steps:

[0119] S21: Establish a prediction model using historical data and real-time data, including train running duration prediction, passenger flow prediction, and prediction of the impact of emergencies on traffic;

[0120] S22: Develop a scheduling optimization algorithm based on machine learning to generate a scheduling strategy that meets the actual requirements;

[0121] This includes: using an autoencoder machine learning model to perform feature dimensionality reduction on the augmented training dataset, and using a preset Softmax function to classify the dimensionality-reduced feature vectors to obtain classification prediction categories.

[0122] Aiming at problems such as information redundancy and the model being prone to falling into local optima during the dimensionality reduction process of high-dimensional rail transit data, the present invention uses an autoencoder based on adaptive gradient correction as the feature dimensionality reduction model, and uses a dynamic adjustment mechanism and an adaptive gradient correction method based on data topology analysis to optimize the performance of the autoencoder network in the complex rail transit data distribution and high-dimensional feature space, realizing more efficient feature compression and reconstruction.

[0123] Specifically, the training process of the autoencoder algorithm based on adaptive gradient correction is as follows:

[0124] a. Initialize the weights and biases of the autoencoder network. The initialization method is random initialization, and the initialized parameters follow a normal distribution with a mean of 0 and a variance of the identity matrix.

[0125] b. Compress the input rail transit data features through the encoder to convert them into a low-dimensional coding representation, so as to obtain the main information of the rail transit data and ignore redundancy, which helps to deal with the common high-dimensional redundancy problems in rail transit data. For example, compress complex trajectory data into key path features to reduce computational overhead. The calculation method of the encoder output is expressed as:

[0126]

[0127] In the formula, is the output of the encoding layer, is the activation function of the encoder, is the weight of the encoder, is the input rail transit data of the encoder, is the bias of the encoder. Preferably, the activation function of the encoder uses the Sigmoid activation function.

[0128] c. Use a negative entropy-driven activation function for the output of the encoding layer. Use negative entropy to measure the information content in the features, thereby increasing the model's preference for key features. By measuring the information content and focusing on processing high-value features, the model can more effectively select important variables in the rail transit data that affect the overall operation efficiency. For example, extract traffic anomaly points from the traffic flow matrix between stations. The calculation method is expressed as:

[0129]

[0130] In the formula, is the negative entropy, is the th element in the output of the encoding layer, is the th element's probability density in the output of the encoding layer.

[0131] d. Restore the encoded rail transit data back to the original dimension through the decoder, and compare the differences between the original rail transit data and the reconstructed rail transit data to guide the learning of the autoencoder network. The calculation method is expressed as:

[0132]

[0133] In the formula, is the reconstructed rail transit data, are the weights of the decoder, is the bias of the decoder, is the activation function of the decoder. Preferably, the activation function of the decoder adopts the ReLU activation function.

[0134] e. Dynamically adjust the entropy weights of the rail transit data features, and adopt a method based on topological rail transit data analysis to capture the non-linear structural features of the rail transit data, which can adaptively identify and retain important topological information in high-dimensional rail transit data. For example, in passenger flow prediction, it can maintain the important topological information of core stations, thereby effectively improving the performance in classification tasks after dimensionality reduction. The calculation method is expressed as:

[0135]

[0136] In the formula, is the entropy weight value of the th sample, is the basic adjustment coefficient, is the penalty degree coefficient, is the feature dimension after dimensionality reduction, is the coefficient affected by topological features; is the th sample's th feature's topological influence degree, which is calculated through the topological rail transit data analysis method. The topological rail transit data analysis (Topological DataAnalysis, TDA) is a method that combines algebraic topology and computational geometry, used to extract geometric and topological characteristics from complex rail transit datasets. Its main goal is to analyze the shape and structure of rail transit data to reveal potential patterns and relationships; is the th feature's weight factor, which is a training parameter and is determined by training through the gradient descent method; is the entropy value of the th feature of the th sample. Preferably, is set to 0.2, is set to 0.1, is set to 0.2.

[0137] f. Use the loss function to constrain the training process, and adopt the adaptive gradient correction algorithm in the autoencoder network update stage to alleviate local optimum and overfitting in high-dimensional training, and improve the search efficiency of the model in complex rail transit data scenarios (such as complex path classification tasks in optimized trajectory pattern recognition). The calculation method of the loss function is expressed as:

[0138]

[0139] In the formula, is the overall loss function of the autoencoder network, is the L2 norm, is the th sample's entropy value, is the regularization coefficient of the autoencoder, is the balance coefficient of the topological feature in the loss function. Preferably, is set to 0.3, is set to 0.3.

[0140] In one embodiment, by comparing the decreasing trend of the loss value during the training process, the superiority of the adaptive gradient correction autoencoder in training speed and convergence ability is proved, as Figure 4 shown.

[0141] g. Update the autoencoder network parameters in an adaptive gradient correction manner to improve the convergence speed and model robustness under complex high-dimensional rail transit data. The gradient correction and parameter update methods are expressed as:

[0142]

[0143] In the formula, is the initial gradient of the autoencoder network parameters, are the parameters of the autoencoder, including the weights of the encoder and the weights of the decoder.

[0144] Furthermore, the initial gradient is corrected according to the adaptive adjustment strategy, and the calculation method is expressed as:

[0145]

[0146] In the formula, is the gradient of the corrected autoencoder network parameters, is the learning rate corrected by the autoencoder gradient, is the gradient adjustment coefficient; is a sign function, representing the maintenance of gradient direction consistency. Preferably, Set to 0.3, Set to 0.1.

[0147] Furthermore, the network parameters of the autoencoder are updated based on the corrected gradient, and the calculation method is expressed as:

[0148]

[0149] In the formula, are the updated autoencoder network parameters, are the autoencoder network parameters before updating.

[0150] h. Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0151] In one embodiment, the data is reduced to 2D by principal component analysis, the distribution of the original data and the reconstructed data is visualized, and the performance of different methods on data reconstruction is compared. The results are as follows: Figure 5 As shown, feature 1 and feature 2 are the features of projecting data into two-dimensional space through principal component analysis, wherein feature 1 is the first principal component and feature 2 is the second principal component.

[0152] After the autoencoder model training is completed, the feature vector after dimension reduction is classified using the preset Softmax function to obtain the classification prediction category. In one embodiment, the categories include:

[0153] 0-Normal operation: The train runs normally according to the schedule;

[0154] 1- Delay: The train runs beyond the scheduled timetable;

[0155] 2- Emergency stop: The train stops due to a malfunction or emergency;

[0156] 3-Maintenance: The train undergoes routine or unplanned maintenance;

[0157] 4- Abnormal behavior: The train behaves in an abnormal operating mode, such as abnormal speed.

[0158] S23: Dynamically adjust model parameters and update the model in real time based on the latest data to improve the accuracy and robustness of the prediction.

[0159] S24: Provide a data visualization tool for analyzing the modeling results and prediction performance, and provide data support for other units.

[0160] S3: Real-time signal control; specifically, dynamically adjust the signal status according to the system decision to ensure the safety and efficiency of train operation.

[0161] The real-time signal control function includes:

[0162] Receive the predictions and instructions provided by the machine learning modeling unit and the automatic scheduling decision unit, and adjust the status of the signal lights and the track switching positions according to the optimization results.

[0163] Achieve the adaptive control of signals, support the priority handling of emergency situations, such as the priority change of signals in case of train delays or emergencies.

[0164] Realize two-way information interaction with trains through the communication system to ensure the consistency of signal control and train operation.

[0165] S4: Automatic scheduling decision-making, generate an optimized scheduling plan for trains based on the comprehensive system information and machine learning modeling results.

[0166] The automatic scheduling decision-making function includes:

[0167] Based on passenger flow prediction, operation status and line conditions, automatically plan the train operation timetable, and dynamically adjust the departure interval and stop time.

[0168] Quickly generate an emergency scheduling plan in case of emergencies (such as equipment failures, bad weather or unexpected delays), and minimize the impact on the whole network operation.

[0169] Intelligently allocate resource utilization, such as track usage rights, platform allocation, etc., to ensure the maximization of line operation efficiency.

[0170] S5: System monitoring and optimization, real-time monitor the operation status of the entire rail transit system, and continuously optimize the system performance according to the feedback.

[0171] The functions of the system monitoring and optimization unit include:

[0172] Real-time display the train operation status, signal status and key system performance indicators, and provide a comprehensive monitoring view of the system status.

[0173] Record and analyze the operation logs, identify potential problems, such as the decline in operation efficiency or signal control delays, etc., and provide a basis for subsequent optimization.

[0174] Regularly evaluate the performance of the machine learning model and scheduling algorithm, generate a performance report and feedback it to the modeling unit for improvement.

[0175] Supports a multi-level early warning mechanism to identify potential operation risks in advance and issue early warning messages for the system to adjust in a timely manner.

[0176] In this embodiment, in the data augmentation method based on the generative adversarial network, the present invention solves the problem that the generated samples of rail transit data do not conform to the actual physical laws by adding physical constraints to guide the generation process; solves the problem of insufficient diversity in data generation by optimizing the training of the generator through a dynamic memory mechanism; and balances the adversarial training of the generator and the discriminator through adaptive learning rate adjustment to avoid the problem of generation mode collapse. In the autoencoder based on adaptive gradient correction, the negative entropy activation function is used to enhance the preference for high-value features to solve the problem of high-dimensional redundancy of rail transit data; the topological data analysis method is combined to capture the non-linear structural features in rail transit data to solve the problem that traditional dimensionality reduction methods are difficult to retain key topological information; and the dimensionality reduction process is optimized through a dynamic entropy weight adjustment mechanism to solve the problem of poor model generalization ability caused by data distribution heterogeneity.

[0177] The embodiment of the present invention also proposes an intelligent rail transit signal system, which includes a data acquisition unit, a machine learning modeling unit, a real-time signal control unit, an automatic scheduling decision-making unit, and a system monitoring and optimization unit;

[0178] The data acquisition unit is used to collect key data during the operation of rail transit in real time, and after annotation, it is used as the training data set of the machine learning model; it also includes data preprocessing and small sample data identification. If it is detected that there are problems such as class imbalance or insufficient samples, the generative adversarial network algorithm based on the augmentation strategy is used as the sample augmentation model to augment rail transit data;

[0179] The machine learning modeling unit is used to establish a prediction model using historical data and real-time data, develop a scheduling optimization algorithm based on machine learning, and generate a scheduling strategy;

[0180] The real-time signal control unit is used to receive the predictions and instructions provided by the machine learning modeling unit and the automatic scheduling decision-making unit, and dynamically adjust the signal state according to the system decision;

[0181] The automatic scheduling decision-making unit is used to generate an optimized scheduling plan for the train by integrating system information and the results of machine learning modeling;

[0182] The system monitoring and optimization unit is used to monitor the operation status of the entire rail transit system in real time and continuously optimize the system performance according to the feedback.

[0183] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which program instructions of an intelligent rail transit signal automatic scheduling method are stored. The program instructions of the intelligent rail transit signal automatic scheduling method can be executed by one or more processors to implement the steps of the intelligent rail transit signal automatic scheduling method as described above.

[0184] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. An intelligent rail transit signal automatic dispatching method, characterized in that: include: S1: Data collection: real-time collection of rail transit operation data, labeling and preprocessing of data; S2: Machine learning modeling, using historical data and real-time data to build a prediction model, and generate a scheduling strategy based on a scheduling optimization algorithm based on machine learning; S3: Real-time signal control, dynamically adjusting signal status according to system decisions; S4: Automatic dispatch decision-making, generating an optimized dispatch plan for trains based on integrated system information and machine learning modeling results; S5: System monitoring and optimization, to monitor the operating status of the entire rail transit system in real time and continuously optimize system performance based on feedback; The step S1 also includes performing small sample identification on the collected data set by using statistical analysis and machine learning algorithms. If a class imbalance or insufficient sample is detected, a generative adversarial network algorithm based on an augmentation strategy is used as a sample expansion model to expand rail transit data; In the training process of the generative adversarial network, the physical constraint function is also added to the loss function of the generator; the loss function of the generator is defined as: ; In the formula, is the loss function of the generator, For the expected operation; is a noise vector, representing the input of the generator; Indicates compliance with a specific distribution; is the distribution of the noise vector, characterizing the randomness of the input; is the weight hyperparameter of the physical constraint, which controls the impact of the physical constraint on the loss function; is a physical constraint function, which indicates whether the generated rail transit data complies with the physical constraints; is the discriminator function, is the generator function, is the parameter of the generator, representing the weight of the generator; The physical constraint function is determined based on the spatial smoothness or the change rate of the rail transit data.

2. The intelligent rail transit signal automatic dispatching method according to claim 1 is characterized in that: The training process of the generative adversarial network algorithm based on the augmentation strategy includes initializing the standard structure of the generative adversarial network, including the network architecture of the generator and the discriminator; the generator is used to generate samples close to the distribution of real rail transit data, and the discriminator is used to determine whether the rail transit data is a real sample or a false sample generated by the generator.

3. The intelligent rail transit signal automatic dispatching method according to claim 2 is characterized in that: The calculation method of the physical constraint function is expressed as: ; In the formula, To generate the Laplacian operator of rail transit data, it represents the spatial smoothness or rate of change of rail transit data; are predefined physical constants or constraint values; is the generator function, is the parameter of the generator, representing the weight of the generator, is the noise vector.

4. The intelligent rail transit signal automatic dispatching method according to claim 2 is characterized in that: In the training process of the generative adversarial network, a dynamic memory mechanism is also used to store and update the effective sample features and output information of the generator in the historical training process; the update and implementation method of the memory module is expressed as: ; In the formula, For the The sample features of the iterative memory represent the memory module in the The status of the iteration; For the The sample features of the iterative memory represent the memory module in the The status of the iteration; is the updated weight coefficient, representing the balance factor of memory update; is a weighting factor based on the quality of the generated sample, characterizing the importance of the sample, The generated samples.

5. The intelligent rail transit signal automatic dispatching method according to claim 2 is characterized in that: The game relationship between the generator and the discriminator is dynamically balanced through an adaptive step size adjustment mechanism. The learning rates of the generator and the discriminator are dynamically adjusted according to the current training state and the changes in the target distribution. In the rail transit data expansion, balanced training is used to reduce the mode collapse caused by improper parameter settings and ensure the diversity of generated samples. The adaptive learning rate adjustment method is expressed as: ; In the formula, is the learning rate of the generator and the discriminator, representing the step size of parameter update; is the initial learning rate of the generator and discriminator, representing the starting step of training; is the learning rate adjustment factor of the generator and discriminator, representing the sensitivity of step size adjustment, is the loss function of the generator, is the loss function of the discriminator.

6. The intelligent rail transit signal automatic dispatching method according to claim 1 is characterized in that: The step S2 further includes: using an autoencoder based on adaptive gradient correction as a feature dimension reduction model, optimizing the autoencoder network by using a dynamic adjustment mechanism based on data topology analysis and an adaptive gradient correction method; the calculation method of the encoding output is expressed as: ; In the formula, is the output of the encoding layer, is the activation function of the encoder, is the weight of the encoder, is the input rail transit data of the encoder, The encoder offset.

7. The intelligent rail transit signal automatic dispatching method according to claim 6 is characterized in that: The output of the coding layer adopts an excitation function based on negative entropy drive.

8. The intelligent rail transit signal automatic dispatching method according to claim 7 is characterized in that: The entropy weights of rail transit data features are dynamically adjusted, and a method based on topological rail transit data analysis is used to capture the nonlinear structural characteristics of rail transit data.

9. An intelligent rail transit signal system, based on the intelligent rail transit signal automatic dispatching method according to any one of claims 1 to 8, comprising: The intelligent rail transit signal system includes a data acquisition unit, a machine learning modeling unit, a real-time signal control unit, an automatic scheduling decision unit, and a system monitoring and optimization unit; The data acquisition unit is used to collect key data in rail transit operation in real time, and to mark and expand it; The machine learning modeling unit is used to establish a prediction model using historical data and real-time data, develop a scheduling optimization algorithm based on machine learning, and generate a scheduling strategy; The real-time signal control unit is used to receive predictions and instructions provided by the machine learning modeling unit and the automatic scheduling decision unit, and dynamically adjust the signal state according to the system decision; The automatic dispatch decision unit is used to generate an optimized dispatch plan for trains by integrating system information and machine learning modeling results; The system monitoring and optimization unit is used to monitor the operating status of the entire rail transit system in real time and continuously optimize system performance based on feedback.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions of an intelligent rail transit signal automatic scheduling method, and the program instructions of the intelligent rail transit signal automatic scheduling method can be executed by one or more processors to implement the steps of the intelligent rail transit signal automatic scheduling method as described in any one of claims 1 to 8.

Images