DAS driving strategy optimization method driven by real-time data of subway lines

Abstract

This invention relates to the technical field of subway train driving strategy optimization and discloses a method for optimizing subway line driving strategies using real-time data driven by DAS. The method comprises: acquiring multidimensional operational data, including real-time train position sequences, station passenger flow data, and the status of trackside signaling equipment; extracting line operational characteristics using a convolutional spatiotemporal encoder, generating a passenger flow fluctuation prediction map using a time series decomposition algorithm, and performing feature fusion on the trackside signaling equipment status to generate an equipment health assessment matrix; inputting the multidimensional data into a driving strategy model to generate an initial driving instruction sequence; assigning operational control priorities using a regional clustering algorithm; and constructing a dynamic optimization strategy network in conjunction with a genetic algorithm to generate a final driving strategy solution; and finally, collecting execution status logs in real time and updating the strategy network using an abnormal decision detection model. This method achieves dynamic optimization of subway driving strategies, improving operational safety, efficiency, and comfort.

Description

Technical Field

[0001] The present invention relates to the technical field of subway train driving strategy optimization, and in particular to a DAS driving strategy optimization method driven by real-time data of subway lines. Background Art

[0002] With the rapid development of urban rail transit, subways have become a core component of modern urban public transportation. However, with the continuous increase in passenger volume, the increasing complexity of routes, and the dynamic changes in equipment operating conditions, traditional subway driving strategies have gradually exposed many shortcomings and are unable to meet the higher requirements of modern subway operations for safety, efficiency, comfort, and economy.

[0003] Traditional driving strategies are mostly based on fixed schedules and preset operating parameters, lacking the ability to effectively utilize real-time dynamic data and adaptively adjust. Regarding passenger flow, traditional methods are unable to accurately perceive changes in station passenger flow, such as inbound passenger density and transfer channel congestion, in real time. This can lead to insufficient train capacity during peak hours or wasted capacity during off-peak hours, impacting passenger experience and operational efficiency. For example, during peak hours in the morning and evening, transfer stations often face enormous passenger pressure. Traditional strategies make it difficult to adjust train stop times or departure intervals based on real-time congestion, which can easily lead to platform congestion and even safety hazards.

[0004] Regarding train operating status, traditional strategies lack sufficient precision in processing key information from real-time train position sequences, such as speed gradients and braking distance parameters. This makes it impossible to accurately and timely extract line operating characteristics, such as section delay coefficients, signal response intervals, and track occupancy status tags. This can prevent trains from responding to unexpected line conditions, such as signal delays caused by track occupancy, in a timely manner, impacting operational efficiency and safety.

[0005] Traditional methods for managing the status of trackside signaling equipment lack multi-dimensional feature integration and health assessment, making it difficult to detect potential equipment failures in advance. This can lead to sudden equipment failures, disrupting normal train operations and even causing operational accidents. For example, if minor signaling equipment failures are not detected and addressed promptly, they can trigger a chain reaction under certain conditions, leading to widespread operational delays.

[0006] Furthermore, the optimization process for traditional driving strategies often relies on manual experience and offline simulation, lacking dynamic, real-time optimization mechanisms. Faced with complex and changing operating environments, these strategies are unable to quickly generate driving strategies that adapt to real-time conditions, resulting in insufficient flexibility and responsiveness in train scheduling. For example, when a temporary fault occurs in a section, traditional strategies struggle to quickly adjust train routes and speeds, significantly reducing the operational efficiency of the entire line. Summary of the Invention

[0007] The purpose of the present invention is to provide a DAS driving strategy optimization method driven by real-time data of subway lines to solve the problems raised in the above background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: a DAS driving strategy optimization method driven by real-time data of subway lines, the method comprising: obtaining a multi-dimensional operation data set; the multi-dimensional operation data comprising a real-time train position sequence, station passenger flow data and trackside signal equipment status; the real-time train position sequence comprises a speed change gradient and a braking distance parameter, the station passenger flow data comprises an in-station passenger flow density and a transfer channel congestion degree; based on the real-time train position sequence, a convolutional spatiotemporal encoder is used to extract line operation characteristics, the line operation characteristics comprising an interval travel delay coefficient, a signal response interval time and a track occupancy status label; according to the station passenger flow data ... Passenger flow data is used to generate a passenger flow fluctuation prediction map using a time series decomposition algorithm. The map includes peak period distribution intervals and an operation conflict probability heat map. The trackside signal equipment status is subjected to multi-dimensional feature fusion to generate an equipment health assessment matrix. The line operation characteristics, passenger flow fluctuation prediction map, and equipment health assessment matrix are input into a driving strategy model to generate an initial driving instruction sequence. Based on the initial driving instruction sequence, the operation control priority is divided using a regional clustering algorithm, and a train scheduling weight distribution table is output. The train scheduling weight distribution table is integrated with the initial driving instruction sequence, and a dynamic optimization strategy network is constructed using a genetic algorithm to generate a final driving strategy solution.

[0009] Preferably, the extraction of line operation characteristics through a convolutional space-time encoder includes: performing a sliding window interception on the real-time position sequence of the train to generate a space-time slice data block; dividing the space-time slice data block into multiple operation intervals based on regional clustering rules, and calculating the standard deviation of the signal response delay in each interval; using a residual connection algorithm to extract the track occupancy status label, and generating a signal system response relationship in combination with the interval travel delay coefficient; encoding the signal response interval time, track occupancy status label and signal system response relationship as line operation characteristics.

[0010] Preferably, the generation of the passenger flow fluctuation prediction map by the time series decomposition algorithm includes: separating the trend components of the passenger flow data of the station and extracting the periodic fluctuation characteristics; predicting the future passenger flow distribution based on the autoregressive model and calculating the confidence of the peak period distribution interval; generating a risk probability model according to the historical conflict event database, and outputting an operation conflict probability heat map in combination with the confidence; mapping the peak period distribution interval and the operation conflict probability heat map into the passenger flow fluctuation prediction map.

[0011] Preferably, the driving strategy model includes a data fusion module and an instruction generation module, and the data fusion module includes: standardizing the signal response interval time in the line operation characteristics to obtain a first fusion tensor; performing morphological filtering on the operation conflict probability heat map in the passenger flow fluctuation prediction map to generate a second fusion tensor; performing eigenvalue decomposition on the equipment health assessment matrix to extract the equipment state degradation gradient to obtain a third fusion tensor; and merging the first fusion tensor, the second fusion tensor and the third fusion tensor into a strategy input sequence through a bidirectional long short-term memory network.

[0012] Preferably, the instruction generation module includes: aligning the strategy input sequence on the time axis to generate an instruction association vector; extracting key control node features through an attention mechanism to generate an instruction node association matrix; performing a tensor splicing operation on the instruction association vector and the instruction node association matrix to generate a candidate instruction set; and selecting an initial driving instruction sequence from the candidate instruction set through a branch and bound algorithm.

[0013] Preferably, the dynamic optimization strategy network is constructed by genetic algorithm, including: initializing the network population according to the train scheduling weight distribution table, and generating a fitness evaluation function based on the priority; using the initial driving instruction sequence as the chromosome encoding, and the fitness evaluation function is composed of the operating energy consumption index and the scheduling priority weight; updating the fitness score of each chromosome through the crossover mutation operator, and adjusting the population evolution direction; generating the optimal driving strategy solution that meets the multi-objective constraints according to the optimized fitness score.

[0014] Preferably, the parameter optimization method of the regional clustering rule includes: calculating the initial clustering radius and the minimum signal delay threshold based on the distribution of historical operation data; traversing the parameter combination through the grid verification algorithm, and selecting the parameters with the highest matching degree between the division result and the actual operation log; dynamically adjusting the clustering radius and the minimum signal delay threshold according to the matching degree, and optimizing the division accuracy of the signal system response relationship.

[0015] Preferably, the method for constructing the risk probability model includes: collecting historical operation conflict data of multiple lines, extracting spatiotemporal correlation features and equipment failure influencing factors; fitting the spatiotemporal correlation features with a Gaussian mixture model to generate a basic risk probability distribution table;

[0016] The basic risk probability distribution table is modified by Bayesian method according to the factors affecting equipment failure, and the risk probability model is output.

[0017] Preferably, the parameter setting method of the crossover mutation operator includes: defining the crossover probability between chromosomes as a balance factor between the operating energy consumption index and the scheduling priority weight; initializing the fitness score of each chromosome to a baseline value, and the fitness of the starting chromosome to a preset initial value; selecting the chromosome with the highest fitness score in the current population as the parent through an elite retention strategy; and performing multi-point crossover and random bit mutation operations based on the parent chromosome to generate a new generation population.

[0018] Preferably, the method also includes: real-time collection of the execution status log of the train control system, generating a strategy correction instruction set through an abnormal decision detection model, and updating the dynamic optimization strategy network; the construction method of the abnormal decision detection model includes: collecting normal operation samples in the historical execution status log, extracting the acceleration instruction change rate and the braking response time series; constructing a normal operation benchmark model based on a single-class support vector machine, and calculating the deviation index between the real-time data and the benchmark model; using a time decay function to count the cumulative abnormal value of the deviation index, and generating a strategy correction instruction set when it exceeds the dynamic threshold; and updating the chromosome encoding parameters of the dynamic optimization strategy network in combination with the correction instruction set.

[0019] Compared with the existing technology, the beneficial effects of the present invention are: in terms of data integration and feature extraction, by obtaining multi-dimensional operating data such as the real-time position sequence of trains, station passenger flow data and the status of trackside signal equipment, and using technologies such as convolutional spatiotemporal encoders, time series decomposition algorithms and multi-dimensional feature fusion, it is possible to accurately extract line operation characteristics, generate passenger flow fluctuation prediction maps and equipment health assessment matrices. This enables the system to fully and deeply understand the dynamic changes in the subway operating environment, and provides a solid data foundation for the optimization of driving strategies. For example, by performing sliding window interception and regional clustering on the real-time position sequence of trains, it is possible to accurately calculate the standard deviation of the signal response delay in each interval, and combine the residual connection algorithm to extract the track occupancy status label, thereby generating an accurate signal system response relationship, providing a key basis for real-time scheduling of trains.

[0020] In terms of driving strategy generation and optimization, the extracted multi-dimensional features are input into the driving strategy model. The data fusion module and the command generation module generate an initial driving command sequence. A regional clustering algorithm is then used to prioritize operational control and output a train dispatch weight distribution table. Based on this, a dynamic optimization strategy network is constructed using a genetic algorithm. This network, combined with fitness evaluation functions such as operational energy consumption indicators and dispatch priority weights, optimizes the initial driving command sequence and generates an optimal driving strategy that meets multiple objective constraints. This full-process dynamic optimization mechanism, from data to strategy, enables driving strategies to adapt in real time to changes in passenger flow, line operating conditions, and equipment status, significantly improving the flexibility and responsiveness of train operations. For example, in the face of sudden passenger peaks, the system can promptly adjust train stop times and departure intervals based on passenger flow fluctuation predictions to alleviate platform congestion. When the health of trackside signaling equipment deteriorates, the system can proactively adjust train speeds and routes to avoid operational accidents caused by equipment failures.

[0021] In terms of exception handling and policy correction, the system collects the train control system's execution status logs in real time, utilizes an anomaly decision detection model to generate a policy correction instruction set, and updates the dynamic optimization policy network. This model, based on a single-class support vector machine, constructs a normal operation baseline model. This model monitors the degree of deviation from normal operation in real time and, when an anomaly is detected, promptly generates correction instructions and adjusts the driving strategy. This real-time anomaly detection and policy correction mechanism can effectively improve the safety and reliability of subway operations and reduce operational delays and accident risks caused by unexpected anomalies. For example, if a train's braking response time exhibits an anomaly, the system can quickly detect and generate correction instructions, adjusting the braking strategy to ensure a safe train stop.

[0022] In terms of overall performance improvement, the proposed method achieves multi-objective optimization of subway operations through a multi-dimensional data-driven and dynamic optimization mechanism. This not only reduces train operating energy consumption and improves energy efficiency, but also optimizes train dispatching priorities, enhancing overall operational efficiency. Furthermore, through accurate passenger flow forecasting and equipment status management, passenger comfort and safety can be enhanced. For example, optimizing driving strategies through genetic algorithms can effectively reduce operating energy consumption while ensuring train safety, achieving a win-win situation for both economic and environmental benefits. By rationally adjusting train operating speeds and stop times, passenger waiting times at platforms and congestion in carriages can be reduced, improving the passenger experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a working principle diagram of the DAS driving strategy optimization method driven by real-time subway line data according to the present invention;

[0024] Figure 2Flowchart for generating passenger flow fluctuation prediction graph for time series decomposition algorithm;

[0025] Figure 3 This is the flowchart of the driving strategy model data fusion module;

[0026] Figure 4 Flowchart of the instruction generation module for the driving policy model. DETAILED DESCRIPTION

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0028] See also Figure 1-Figure 4 The present invention relates to a method for optimizing a DAS driving strategy for a subway line driven by real-time data. The method comprises the following steps: obtaining multi-dimensional operational data, including real-time train position sequences, station passenger flow data, and the status of trackside signaling equipment. The real-time train position sequence includes speed gradient and braking distance parameters, reflecting the dynamic operational status of the train on the line; the station passenger flow data includes inbound passenger flow density and transfer channel congestion, characterizing the passenger flow distribution within the station; and the trackside signaling equipment status data is used to monitor the real-time operating status of the signaling equipment.

[0029] Based on the real-time train position sequence, a convolutional spatiotemporal encoder is used to extract line operation characteristics. The specific process is as follows: the real-time train position sequence is intercepted by a sliding window to generate spatiotemporal slice data blocks; the spatiotemporal slice data blocks are divided into multiple operating intervals based on regional clustering rules, and the standard deviation of the signal response delay within each interval is calculated; the residual connection algorithm is used to extract the track occupancy status label, and the signal system response relationship is generated by combining the interval travel delay coefficient. The signal response interval time, track occupancy status label, and signal system response relationship are encoded into line operation characteristics, including the interval travel delay coefficient, signal response interval time, and track occupancy status label.

[0030] Based on station passenger flow data, a time series decomposition algorithm is used to generate a passenger flow fluctuation prediction map. The specific steps are: separating the trend components of station passenger flow data to extract periodic fluctuation characteristics; predicting future passenger flow distribution based on an autoregressive model and calculating the confidence level of the peak-hour distribution interval; generating a risk probability model based on a historical conflict event database and outputting a heat map of operational conflict probability based on the confidence level; and mapping the peak-hour distribution interval and operational conflict probability heat map into a passenger flow fluctuation prediction map, which includes the peak-hour distribution interval and operational conflict probability heat map.

[0031] The multi-dimensional features of the trackside signal equipment status are fused, and through operations such as data cleaning, feature extraction, and normalization, an equipment health assessment matrix is ​​generated to comprehensively reflect the health status of the equipment.

[0032] The driving strategy model, which includes a data fusion module and an instruction generation module, is fed with route operation characteristics, a passenger flow fluctuation prediction map, and an equipment health assessment matrix. The data fusion module normalizes the signal response intervals in the route operation characteristics to generate a first fused tensor. The module applies morphological filtering to the operational conflict probability heatmap in the passenger flow fluctuation prediction map to generate a second fused tensor. The equipment health assessment matrix is ​​subjected to eigenvalue decomposition to extract equipment state degradation gradients, generating a third fused tensor. The first, second, and third fused tensors are combined into a strategy input sequence using a bidirectional long short-term memory network. The instruction generation module aligns the strategy input sequence time-axis to generate an instruction association vector. Key control node features are extracted using an attention mechanism to generate an instruction node association matrix. The instruction association vector and the instruction node association matrix are then concatenated to generate a candidate instruction set. The initial driving instruction sequence is selected from the candidate instruction set using a branch-and-bound algorithm.

[0033] Based on the initial driving instruction sequence, the operation control priority is divided through the regional clustering algorithm. According to the operation characteristics, passenger flow conditions and equipment status of different areas, the priority weight of each area is calculated, and the train scheduling weight distribution table is output to clarify the scheduling priority of different areas.

[0034] The train dispatch weight distribution table is integrated with the initial driving instruction sequence, and a dynamic optimization strategy network is constructed using a genetic algorithm. The specific process is as follows: the network population is initialized according to the train dispatch weight distribution table, and a fitness evaluation function is generated based on the priority. This function is composed of operating energy consumption indicators and dispatch priority weights. The initial driving instruction sequence is used as the chromosome encoding, and the fitness score of each chromosome is updated through the crossover mutation operator to adjust the population evolution direction. The optimized fitness score is used to generate the optimal driving strategy solution that meets multiple objective constraints.

[0035] The present invention is further described below in conjunction with Examples 1 to 5: Example 1: In the process of extracting line operation characteristics, when processing the real-time position sequence of the train, it is necessary to first determine the interception method of the sliding window. The size of the sliding window is not fixed, but is dynamically adjusted according to the length of the line section and the train running speed. For example, when the train runs on a line section with a long section length and a high design speed, the size of the sliding window is increased accordingly to ensure that enough position data points can be included in one window to accurately reflect the changes in the train's running status in the section; in areas with short sections and lower speed limits, the sliding window is appropriately reduced to avoid data redundancy or feature extraction deviation due to excessively large windows. Through this dynamic adjustment mechanism, the generated spatiotemporal slice data blocks can more accurately correspond to the actual operation scenario, providing a reliable data basis for subsequent feature extraction.

[0036] When dividing a spatiotemporal slice data block into multiple operating intervals based on regional clustering rules, the parameters of the regional clustering rules must first be determined. Specifically, the initial clustering radius and minimum signal delay threshold are calculated based on the distribution of historical operating data. Historical operating data covers the operation records of trains in different time periods and under different line conditions. Through statistical analysis of this data, the distribution pattern of train operating status parameters can be obtained, thereby preliminarily determining the reasonable range of the clustering radius and signal delay threshold. For example, by analyzing the distribution of signal response delays in historical data, an initial minimum signal delay threshold is determined, and areas with signal response delays less than this threshold are considered as the basis for preliminary division into the same type of operating intervals.

[0037] Next, a grid validation algorithm is used to traverse the parameter combinations. This algorithm generates different combinations of cluster radius and minimum signal delay thresholds within a certain range, divides each combination into operating intervals, and compares the division results with the actual operation logs. The actual operation logs record the actual operation of the train in each interval in detail, including key information such as signal response time and track occupancy status. The effectiveness of different parameter combinations is evaluated by calculating the degree of match between the division results and the actual operation logs. For example, determining whether the standard deviation of the signal response delay within the divided operating interval matches the fluctuations in the actual log for that interval is consistent with the fluctuations in the actual log. During the traversal process, a large number of parameter combinations need to be tested to ensure that the parameters with the highest degree of match are selected.

[0038] After selecting the parameters with the highest matching degree, it does not mean that the parameter optimization process is over. Instead, the clustering radius and the minimum signal delay threshold need to be dynamically adjusted based on the matching results. For example, if it is found that the division results under the current parameter combination have a low matching degree with the actual operation log during certain special time periods (such as morning and evening peak hours), it may be due to factors such as increased train operation density and increased signal system load during peak hours. At this time, it is necessary to appropriately adjust the clustering radius or the minimum signal delay threshold according to the specific matching degree differences to improve the division accuracy in different scenarios. Through this dynamic adjustment mechanism, the division of the signal system response relationship can be continuously optimized to make it more in line with the actual operation situation, laying the foundation for the subsequent extraction of accurate line operation characteristics.

[0039] When using the residual connection algorithm to extract track occupancy status labels, a suitable neural network structure must be constructed. Specifically, a multi-layer convolutional neural network (CNN) is used to extract features from spatiotemporal slice data blocks. Convolutional neural networks have powerful spatial feature extraction capabilities and can capture the spatial distribution characteristics of track occupancy status from spatiotemporal slice data. In a multi-layer convolutional neural network, each convolutional layer is responsible for extracting features at a different level. Shallow convolutional layers extract more basic features, such as local speed change patterns, while deeper convolutional layers extract more complex and abstract features, such as track occupancy trends across the entire section.

[0040] To alleviate the vanishing gradient problem that may occur in deep networks, a residual connection structure is introduced into convolutional neural networks. The core idea of ​​residual connections is to establish an identity mapping path in the network, allowing gradients to backpropagate more smoothly, thereby allowing the training of deeper neural networks. Through residual connections, the network can more easily optimize the complex feature extraction process during the learning process, ensuring that the track occupancy status label can be accurately extracted. For example, in a certain convolutional layer, the input data is processed by convolution operations and activation functions, and then directly added to the input data of the layer to obtain the output data of the layer. This method enables the network to more effectively capture subtle changes in the track occupancy status during the learning process.

[0041] After extracting the track occupancy status label, it is necessary to combine it with the section travel delay coefficient to generate a signal system response relationship. The section travel delay coefficient reflects the train's travel efficiency within the section and is calculated based on the ratio of the train's actual operating time in the section to the theoretical minimum operating time. By correlating the track occupancy status label with the section travel delay coefficient, the response characteristics of the signal system under different track occupancy states can be determined. For example, if the track occupancy status label indicates that a section is busy, if the travel delay coefficient for that section is large, it indicates that the signal system's response in this state may be delayed, and further optimization of the signal system's scheduling strategy is required.

[0042] A multidimensional vector encoding approach is used to encode the signal response interval, track occupancy status label, and signal system response relationship as line operation characteristics. Specifically, the signal response interval is converted into a single dimension in a vector, the value of which directly reflects the signal system's response speed. The track occupancy status label is converted into multiple dimensions using one-hot encoding, with each dimension corresponding to a specific track occupancy state (e.g., idle, occupied, about to be occupied). The signal system response relationship is converted into other dimensions in the vector by extracting key characteristic parameters (e.g., response delay mean and variance). This multidimensional vector encoding approach integrates different types of characteristic parameters into a unified vector space, facilitating subsequent data fusion processing and input into driving strategy models.

[0043] Throughout the entire process of extracting line operation features, each step is closely linked logically and through data transfer. The spatiotemporal slice data blocks generated by sliding window extraction serve as the foundation for all subsequent processing. Parameter optimization of regional clustering rules ensures the rationality of operating interval divisions. The residual connection algorithm ensures the accuracy of track occupancy status label extraction, and multidimensional vector encoding facilitates the effective utilization of features. This rigorous processing flow enables comprehensive and accurate line operation features to be extracted from real-time train position sequences, providing critical input for optimizing subway driving strategies, enabling them to better adapt to the actual operating conditions of the line and improving the safety and efficiency of subway operations.

[0044] Example 2: Generating a passenger flow fluctuation prediction map requires multi-stage processing of station passenger flow data to extract effective features and complete the prediction mapping. First, to separate the trend component from the data, the seasonal decomposition algorithm (STL) is used to process the raw flow data. This algorithm iteratively decomposes the flow series into three components: a trend term, a seasonal term, and a residual term. The trend term reflects the overall trend of passenger flow over a longer period, such as the annual increase in passenger flow at a station in a certain area as a city develops. The seasonal term captures periodic fluctuations, such as the difference in passenger flow between weekdays and weekends and the fixed time distribution of morning and evening peaks. The residual term encompasses random fluctuations in the data that cannot be explained by the trend and seasonal terms. This decomposition process clearly extracts the periodic fluctuation characteristics of passenger flow data, such as the fixed peak time patterns of weekday morning peaks (7:30-9:00) and evening peaks (17:30-19:00), as well as the relatively flat distribution of weekend passenger flow throughout the day, with a possible small peak in the afternoon.

[0045] When predicting the future passenger flow distribution based on the autoregressive model, the order of the model must first be determined based on historical data. By analyzing the passenger flow data series over a period of time (such as the passenger flow data every 15 minutes in the past 30 days), the partial autocorrelation function (PACF) is used to determine the reasonable order of the autoregressive model. For example, if the PACF is truncated after the lag of 3 orders, the 3rd order autoregressive model (AR(3)) is selected as the prediction model. The model parameters are estimated using the least squares method. By minimizing the sum of squares of the error between the predicted value and the actual value, the coefficients of each lag term in the model are determined, thereby establishing a quantitative relationship between the passenger flow data and the historical value. For example, establish a model of the form The prediction model (this is just for illustration, not involving formula derivation), where is the passenger flow value at the current moment, 、 、 The passenger flow value at the historical moment, 、 、 is the model coefficient, The model predicts the passenger flow distribution for each future time period and calculates the confidence level of the peak-hour distribution interval based on historical prediction errors. For example, by calculating the standard deviation of the predicted value, a confidence interval (e.g., 90% confidence level) is determined to quantify the reliability of the prediction results and clarify the possible time range and probability level of the peak period.

[0046] When building a risk probability model, historical operational conflict data from multiple lines must first be collected. This data covers spatiotemporal correlation characteristics such as the time and location of the conflict, the speed of the trains involved, and the status of signal equipment, as well as factors influencing equipment failures, such as the type of equipment failure (e.g., signal machine failure, track circuit failure) and duration. A Gaussian mixture model (GMM) is used to fit these spatiotemporal correlation characteristics. The GMM assumes that the data is composed of a mixture of multiple Gaussian distributions. The expectation-maximization (EM) algorithm is used to estimate the mean, variance, and weight of each Gaussian component, thereby generating a basic risk probability distribution table. For example, if analysis reveals that the probability of operational conflicts on a certain line is higher during peak hours in the morning and evening, or near transfer stations, these areas and time periods can be classified as high-risk areas and assigned higher initial probability values ​​in the basic risk probability distribution table.

[0047] Subsequently, the basic risk probability distribution table is Bayesian-corrected based on the factors affecting equipment failure. Bayesian correction adjusts the basic risk probability by introducing prior knowledge of equipment failure. For example, if it is known that the probability of operational conflict in the area will increase significantly after a certain type of signal equipment failure (such as an abnormal signal display) occurs, then based on the basic risk probability, the Bayesian formula is used to calculate the posterior risk probability when the equipment failure occurs, thereby obtaining a risk probability model that is more in line with the actual situation. In specific operations, the conditional probability relationship between equipment failure and operational conflict is first determined, and then the risk probability of the corresponding area and time period is dynamically adjusted based on the real-time monitored equipment status (such as whether a failure has occurred), so that the risk probability model can reflect the impact of the equipment status on the operational conflict in real time.

[0048] When mapping peak-hour distribution intervals and conflict probability heat maps into a passenger flow fluctuation forecast map, Geographic Information System (GIS) technology was used to visualize spatiotemporal information. First, the subway line map and station locations were digitally modeled to establish a spatial geographic coordinate system. For peak-hour distribution intervals, the time axis was used as the dimension, and the peak hour time range and passenger flow density level (e.g., high, medium, or low) were labeled at the corresponding station locations. For example, during the morning rush hour, transfer stations were colored red (representing high passenger flow) and standard stations were colored orange (representing medium passenger flow). For the conflict probability heat map, the conflict probability values ​​for each area calculated based on the risk probability model were visualized using a color gradient: areas with low conflict probability are displayed in green, areas with medium probability are displayed in yellow, and areas with high probability are displayed in red. Darker colors indicate higher conflict probability, providing a clearer picture of the risk level of different areas.

[0049] During the specific implementation process, it is necessary to ensure that the peak period distribution intervals are fully aligned with the time and space dimensions of the operation conflict probability heat map. For example, for the morning peak period (7:30-9:00) of a certain station, the operation conflict probability distribution of the station and the surrounding line sections during this period is synchronously displayed in the heat map, so that the two form an interconnected overall map. Through this mapping method, the driving strategy model can intuitively obtain the spatiotemporal correlation characteristics between passenger flow fluctuations and operation risks. For example, it can identify that during the morning peak period and in areas near transfer stations, not only is the passenger flow density high, but the probability of operation conflicts increases significantly. Therefore, when formulating driving strategies, higher attention is paid to this area, and corresponding scheduling measures (such as increasing train stop time and adjusting departure frequency) are taken to reduce conflict risks and improve operational efficiency.

[0050] The entire process of generating a passenger flow fluctuation prediction map, from data decomposition and model prediction to risk modeling and visualization mapping, is closely interconnected and mutually verified. A seasonal decomposition algorithm ensures accurate extraction of cyclical characteristics of passenger flow, an autoregressive model enables quantitative prediction of future passenger flow, a Gaussian mixture model combined with Bayesian correction constructs a dynamically adjustable risk probability model, and geographic information system technology transforms abstract data analysis results into intuitive and easy-to-understand graphical representations. This rigorous processing pipeline generates a passenger flow fluctuation prediction map that includes peak hour distribution intervals and operational conflict probability heat maps. This provides comprehensive passenger flow information support for optimizing subway driving strategies, enabling driving strategies to proactively respond to passenger flow fluctuations and potential risks, resulting in smarter and safer train scheduling and operational control.

[0051] Example 3: In the data fusion module of the driving strategy model, the processing of multi-source data needs to go through multiple steps such as standardization, filtering, feature decomposition, and time series modeling to achieve effective fusion of data of different dimensions. First, the signal response interval in the line operation characteristics is standardized using the Z-score standardization method. The core idea is to convert the original data into a standard normal distribution with a mean of 0 and a standard deviation of 1, eliminating the influence of different feature dimensions. The specific formula is: in, Indicates the original signal response interval time data, is the mean of the data, is the standard deviation, This is the first fused tensor after normalization. Through this process, the signal response intervals across different lines and time periods can be unified to the same dimension, facilitating subsequent tensor operations and feature fusion with other data.

[0052] For the heat map of the probability of running conflicts in the passenger flow fluctuation prediction map, morphological filtering is required to remove noise and smooth the edges. Morphological filtering is based on the theory of mathematical morphology. It performs operations such as corrosion and expansion on the pixel matrix of the heat map by designing structural elements. For example, using a size of The heatmap is first eroded to remove isolated noise points at the edges, and then dilated to fill tiny internal holes, generating a smoother and more reliable second fusion tensor. This process effectively reduces outliers caused by sensor errors or data acquisition fluctuations, making the spatial distribution of operational conflict probabilities clearer and providing more accurate input for subsequent data fusion.

[0053] For the processing of the equipment health evaluation matrix, the eigenvalue decomposition method is used to extract the equipment state degradation gradient. The equipment health evaluation matrix is ​​a square matrix composed of multi-dimensional features (such as voltage, current, temperature, working time, etc. of signal equipment), which is recorded as By using the singular value decomposition (SVD) algorithm, the matrix Decompose into ,in and is an orthogonal matrix, is a diagonal matrix whose diagonal elements are matrices The singular values ​​are arranged from largest to smallest, reflecting the importance and energy distribution of each feature in the matrix. By analyzing the temporal trend of the singular values, the device state degradation gradient—the rate of decline or slope of change of the singular values—can be extracted as the third fused tensor. For example, if the singular values ​​of a device have recently shown a significant downward trend, this indicates that the device's health may be degrading, and its potential impact on the signal system's response needs to be considered in the driving strategy.

[0054] After the three fusion tensors are generated, the data is merged through a bidirectional long short-term memory network (Bi-LSTM). The Bi-LSTM network consists of two LSTM layers, one forward and one backward, which can simultaneously capture the forward and backward temporal dependencies in the data and is suitable for processing multi-dimensional data with time series characteristics. In specific implementation, the first fusion tensor (normalized signal response interval), the second fusion tensor (filtered operation conflict probability heat map), and the third fusion tensor (device state degradation gradient) are aligned according to the time dimension to form the input sequence. , where each time step Contains the corresponding dimension features of the three tensors.

[0055] Each LSTM unit in a Bi-LSTM network contains a forget gate, an input gate, and an output gate, using a gating mechanism to control information retention and forgetting. The forward LSTM layer processes data sequentially from the start time step to the end time step of the sequence, capturing contextual information from past moments. The backward LSTM layer processes data from the end time step to the start time step, capturing contextual information from future moments. The outputs of the two layers are concatenated in the hidden layer to generate a policy input sequence containing bidirectional temporal information. For example, if the signal response interval at a certain moment is abnormally prolonged, the forward LSTM layer can combine previous signal delay history to determine whether it is a persistent issue. The backward LSTM layer can then predict whether the anomaly will trigger a chain reaction based on subsequent changes in device health status. This ensures that the policy input sequence fully reflects the temporal correlations and causal relationships in the data.

[0056] In the actual operation of data fusion, attention should be paid to the dimensionality matching of different tensors. For example, the first fused tensor for signal response intervals may be a one-dimensional vector (each time step corresponds to a scalar value), the second fused tensor for the run conflict probability heat map may be a two-dimensional matrix (corresponding to the probability distribution of spatial positions), and the third fused tensor for the device state degradation gradient may be a multi-dimensional vector (containing degradation characteristics of multiple devices). To achieve effective merging, it is necessary to perform a dimensional expansion operation to increase the dimensionality of the lower-dimensional tensor to the same number as the highest-dimensional tensor. For example, a one-dimensional tensor can be expanded into a two-dimensional or three-dimensional tensor to align with the other tensors along the time and space axes. Specifically, dimensional expansion can be performed by padding with zero values ​​or repeating eigenvalues ​​to ensure that the dimensional structure of each tensor is consistent when input into the Bi-LSTM network, avoiding computational errors caused by dimensional mismatch.

[0057] Furthermore, the data fusion process must consider the differences in temporal resolution between different data sources. For example, real-time train position data may be collected once per second, while station passenger flow data may be collected once per minute. To address this issue, the low-temporal-resolution data must first be interpolated, such as using linear interpolation or cubic spline interpolation, to increase its temporal resolution to match that of the high-frequency data, ensuring accurate alignment on the time axis. For example, passenger flow data collected once per minute can be interpolated to data points once per second, ensuring that they fully match the timestamps of the train position data. This ensures that the Bi-LSTM network can accurately capture the correlation between multiple data sources at the same time.

[0058] The processing flow of the entire data fusion module closely revolves around the characteristics of multi-source data. Data quality is enhanced through preprocessing steps such as standardization, filtering, and feature decomposition, and the Bi-LSTM network enables in-depth extraction and fusion of time series features. Standardization resolves the issue of dimensional differences, morphological filtering enhances the reliability of spatial features, eigenvalue decomposition explores the underlying patterns of changes in device status, and the Bi-LSTM network captures long-range dependencies in the data through bidirectional time series modeling. These steps work together to transform line operation characteristics, passenger flow fluctuation prediction maps, and equipment health assessment matrices into a strategy input sequence containing rich spatiotemporal information and time series correlations. This provides comprehensive and accurate input data for the subsequent command generation module, ensuring that the driving strategy model can make scientific and reasonable decisions based on multi-dimensional information and achieve optimized control of subway train operations.

[0059] Example 4: In the command generation module of the driving strategy model, the processing of the strategy input sequence requires multiple steps, including timeline alignment, key feature extraction, tensor concatenation, and command selection, to generate a reasonable initial driving command sequence. First, to address the timeline alignment of the strategy input sequence, the dynamic time warping (DTW) algorithm is used to address the inconsistency of sequence lengths due to potential differences in the acquisition frequency or timestamps of route operation characteristics, passenger flow fluctuation prediction maps, and equipment health assessment matrices. For example, assume that signal response interval data is collected once per second, while operation conflict probability heatmap data is updated every 5 seconds. The two time series have different lengths and their time points are not fully aligned. In this case, the DTW algorithm calculates the similarity distance between the two sequences to find the optimal alignment path on the timeline. It then stretches or compresses the short sequence to align it one-to-one with the time points of the long sequence, generating a time-aligned command association vector. This process ensures consistency in the temporal dimension of data from different sources, facilitating subsequent analysis of the mutual influence of various feature parameters at the same time.

[0060] The attention mechanism uses a scaled dot-product attention (ScaledDot-ProductAttention) mechanism to extract key control node features. Its core concept is to assign different attention weights to each feature node in the policy input sequence, highlighting key nodes with a significant impact on the driving strategy. For example, the policy input sequence may include features such as increased signal response intervals during a certain period, increased passenger density at transfer stations, and degraded health of certain signal equipment. The attention mechanism automatically calculates the weights of these feature nodes: if an increased signal response interval poses a risk of signal delays for subsequent trains, the weight of this feature will be significantly increased; if increased passenger density at transfer stations may affect train stop times, the weight of this feature will also increase accordingly. This method generates a command node association matrix, in which each element represents the attention weight of the corresponding feature node, intuitively reflecting the importance of each node in driving control. For example, in the transfer station area during the morning rush hour, nodes related to passenger density and signal response delay may have higher weights than other common nodes, indicating that these nodes are control factors that require special attention.

[0061] When performing a tensor concatenation operation on the instruction association vector and the instruction node association matrix, the dimensional structure of both must be clarified. The instruction association vector is a one-dimensional sequence aligned along the time axis, with each element corresponding to the comprehensive feature value of a time point. The instruction node association matrix is ​​a two-dimensional matrix, with rows representing time points, columns representing different feature nodes, and elements representing the attention weights of each node. The tensor concatenation operation combines the one-dimensional vector and the two-dimensional matrix into a three-dimensional tensor by expanding and merging them along the dimensions. For example, the instruction association vector is expanded into a two-dimensional matrix (with the vector values ​​at each time point repeated across multiple rows) and then concatenated with the instruction node association matrix along the third dimension, forming a three-dimensional "time-node-feature" structure. This concatenation method preserves both global features (the overall operating status reflected by the instruction association vector) and local key node features (the key control factors highlighted by the instruction node association matrix), providing a richer feature representation for generating candidate instruction sets.

[0062] When selecting an initial driving instruction sequence from a candidate instruction set using a branch-and-bound algorithm, a multi-objective optimization function must be defined, such as operating energy consumption, punctuality, and passenger comfort. The candidate instruction set contains a variety of possible driving instruction combinations, such as different acceleration / braking strategies and stop time adjustment schemes. The branch-and-bound algorithm decomposes the problem into multiple subproblems (a branching process) and calculates upper or lower bounds for the objective function for each subproblem (a bounding process), gradually eliminating subproblems that are unlikely to contain optimal solutions and narrowing the search scope. For example, in a certain operating range, the candidate instruction set includes three options: "Maintain current speed," "Accelerate and pass," and "Slow down and avoid." The algorithm first calculates the estimated energy consumption and impact on punctuality for each option: "Accelerate and pass" may reduce energy consumption but increase punctuality risk, while "Slow down and avoid" may improve safety but increase energy consumption. By comparing the objective function values ​​of each option, the option with the best overall performance is selected as part of the initial driving instruction sequence. During the search, the algorithm prioritizes subproblems with better objective function values ​​to ensure that a feasible solution close to the optimal solution is found within a limited time.

[0063] In practical applications, the processing of the instruction generation module needs to be dynamically adjusted based on specific scenarios. For example, when a significant increase in congestion is detected in the transfer channel at a certain station, the corresponding passenger flow feature value in the instruction association vector will increase, and the attention mechanism will automatically increase the weight of this feature node, thereby increasing the weight of this node in the instruction node association matrix. After tensor splicing, the candidate instruction set may generate instructions such as extending the stop time at the station and slowing down the entry speed to assist station staff in directing passenger flow. When selecting instructions, the branch and bound algorithm comprehensively considers the impact of extending the stop time on the operation schedule of subsequent trains (such as the possibility of causing subsequent train delays) and the impact of slowing down the entry speed on energy consumption, and selects the optimal instruction combination through multi-objective trade-offs.

[0064] Furthermore, the command generation module must handle data noise and uncertainty. For example, a characteristic parameter in the device health assessment matrix may exhibit an outlier value due to a temporary sensor failure, resulting in interference in the policy input sequence. In this case, the attention mechanism mitigates the impact of this outlier by assigning weights to the node, thus preventing the generation of erroneous driving commands. Furthermore, the branch-and-bound algorithm incorporates robustness constraints when calculating the objective function, ensuring that the selected command sequence maintains stable performance despite a certain degree of data fluctuation.

[0065] The entire command generation module's processing flow is based on timeline alignment, using an attention mechanism to achieve feature weighting, tensor splicing to fuse multi-dimensional information, and ultimately achieving command optimization through a branch-and-bound algorithm. Timeline alignment ensures the temporal and spatial consistency of data, the attention mechanism simulates the focus process in human decision-making, tensor splicing organically combines global and local features, and the branch-and-bound algorithm provides an efficient multi-objective optimization method. These links work together to generate a highly targeted and comprehensive initial driving command sequence based on real-time multi-dimensional data. This provides a reliable decision-making basis for subsequent driving strategy optimization, enabling subway trains to make intelligent and reasonable control responses in complex operating environments, balancing multiple requirements such as safety, efficiency, and passenger experience.

[0066] Example 5: The process of constructing a dynamic optimization strategy network using a genetic algorithm involves key steps such as population initialization, fitness evaluation, crossover and mutation, and population updates. Real-time feedback is also used to dynamically modify the strategy network. Taking the optimization of driving strategies during the morning rush hour on a subway line as an example, the initial population is constructed based on a train scheduling weight distribution table. This table divides the line into high-priority scheduling areas (such as transfer station areas) and normal scheduling areas based on factors such as passenger flow density and signal equipment load during the morning rush hour. Each chromosome corresponds to a set of initial driving instruction sequences. For example, a chromosome encodes "Limit train speed to 50 km / h in high-priority areas and extend stop times to 40 seconds; maintain 60 km / h in normal areas with stop times of 25 seconds." Parameters such as speed and stop time constitute the chromosome's gene segments.

[0067] During population initialization, the crossover probability between chromosomes is defined as a balancing factor between the operating energy consumption index and the dispatch priority weight. For example, during the morning rush hour, when the dispatch priority weight is higher (e.g., set to 0.6) and the operating energy consumption weight is lower (e.g., 0.4), the crossover probability can be set to 0.7. This promotes gene exchange in the high-priority area's dispatching strategy and accelerates population evolution to adapt to the complex scenarios of the morning rush hour. The fitness score of each chromosome is initialized to a baseline value (e.g., 100 points). The fitness of the starting chromosome (i.e., the initial driving instruction sequence) is determined based on a preset initial value (e.g., 85 points based on historical strategies), providing an initial reference standard for population evolution.

[0068] When selecting parent chromosomes using the elite retention strategy, the individuals with the highest fitness scores are selected from the current population. For example, if a chromosome, during simulated operation during the morning rush hour, meets the scheduling requirements of high-priority areas (e.g., no departure delays due to passenger congestion) while maintaining low energy consumption, and its fitness score is 95, higher than the population average of 88, it is selected as the parent. The parent chromosome generates the next generation of the population through multi-point crossover and random bit mutation. Multi-point crossover selects multiple gene segment exchange points. For example, crossover points are set at the speed control gene and the stop time gene, swapping the high-priority area speed gene (50 km / h) of parent 1 with the normal area speed gene (65 km / h) of parent 2 to generate a new chromosome combination. Random bit mutation randomly modifies a specific gene position within the chromosome, such as mutating the normal area stop time gene of a chromosome from 25 seconds to 28 seconds. This increases the genetic diversity of the population and avoids falling into local optimal solutions.

[0069] During the fitness evaluation process, the fitness function consists of an operating energy consumption index and a dispatch priority weight. The operating energy consumption index is determined by calculating the train's energy consumption model at different speeds and accelerations. For example, when the speed increases from 50 km / h to 60 km / h, the energy consumption per unit distance increases by approximately 15%. The dispatch priority weight is scored based on whether the dispatch tasks in high-priority areas are completed on time (such as whether the train departs on time during peak hours). Failure to complete the tasks on time results in a corresponding deduction. Through iterative evolution, the fitness scores of the chromosomes in the population gradually improve. For example, after five generations of evolution, the fitness score of the optimal chromosome increased from an initial 85 to 98 points. The corresponding driving strategy maintains dispatch efficiency in high-priority areas while reducing energy consumption by approximately 10% compared to the initial strategy (this is a logical description only, no actual data is available).

[0070] The real-time correction mechanism is implemented through an abnormal decision detection model. For example, if a train experiences a sudden signaling failure during operation, causing the actual braking command to be executed later than planned in the initial driving command sequence, the train control system's execution status log will record this abnormal event, including the acceleration command change rate (e.g., actual acceleration drops sharply from 0.3 m / s² to 0.1 m / s²) and the braking response time series (e.g., planned braking time 10:05:00, actual execution time 10:05:03). The abnormal decision detection model constructs a single-class support vector machine baseline model based on historical normal operation samples (e.g., command execution data without faults within the past week) and calculates the deviation index between the real-time data and the baseline model. If the cumulative abnormal value of the deviation index exceeds a dynamic threshold (e.g., three consecutive braking response delays exceeding two seconds), a policy correction instruction set is generated, such as "reduce train speed by 10 km / h in the next three sections and increase braking distance margin."

[0071] The correction instruction set implements strategy adjustments by updating the chromosome-encoded parameters of the dynamic optimization strategy network. For example, the speed reduction parameter in the correction instruction is encoded as a new gene segment, replacing the corresponding gene position in the original chromosome, forming a new chromosome population. This process enables the genetic algorithm to respond to sudden abnormalities in real time and adjust the optimization direction. For example, during the period before a signal equipment failure is repaired, the chromosomes in the population will generally carry the "reduce speed, increase braking margin" gene. Until the abnormal event is resolved, the baseline model relearns normal operating data and gradually phases out the correction gene.

[0072] In practical applications, the parameters of the crossover and mutation operators need to be dynamically adjusted based on the operating characteristics of different time periods. For example, during off-peak hours, the scheduling priority weight is reduced (e.g., 0.4) and the operating energy consumption weight is increased (e.g., 0.6). The crossover probability can be adjusted to 0.5 to balance energy optimization with regular scheduling needs; the mutation probability is appropriately increased (e.g., from 0.01 to 0.03) to explore more low-energy driving strategy combinations. The application ratio of the elite retention strategy can also be adjusted based on population diversity. For example, when multiple high-fitness chromosomes appear in the population, the top 20% of elite individuals are retained; if the population diversity is insufficient, the elite retention range is expanded to 30% to avoid the loss of high-quality genes.

[0073] The entire implementation process achieves dynamic optimization of driving strategies by combining the global search capabilities of genetic algorithms with the local correction capabilities of real-time feedback mechanisms. Population initialization and fitness assessment ensure the goal-oriented nature of strategy optimization, crossover and mutation operations maintain population diversity, elite retention strategies accelerate the evolutionary process, and abnormal decision detection and correction mechanisms empower the system with adaptive capabilities to cope with emergencies. Taking the morning rush hour as an example, this mechanism, through continuous iterative evolution and real-time correction, can generate optimal driving strategies that meet high-priority scheduling requirements while balancing energy efficiency in complex scenarios such as sudden changes in passenger flow and temporary equipment failures, ensuring the safe and efficient operation of subway trains in a dynamically changing operating environment.

[0074] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0075] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A DAS driving strategy optimization method driven by real-time data of subway lines, characterized by: include: Acquire a multi-dimensional operational data set; the multi-dimensional operational data includes a real-time train position sequence, station passenger flow data, and trackside signal equipment status; the real-time train position sequence includes speed gradient and braking distance parameters, and the station passenger flow data includes inbound passenger flow density and transfer channel congestion; based on the real-time train position sequence, extract line operational characteristics using a convolutional spatiotemporal encoder, the line operational characteristics including section travel delay coefficient, signal response interval, and track occupancy status label; Based on the passenger flow data of the station, a passenger flow fluctuation prediction map is generated by a time series decomposition algorithm, wherein the map includes a peak period distribution interval and an operation conflict probability heat map; The multi-dimensional features of the trackside signal equipment status are fused to generate an equipment health assessment matrix; the line operation characteristics, passenger flow fluctuation prediction map and equipment health assessment matrix are input into the driving strategy model to generate an initial driving instruction sequence; based on the initial driving instruction sequence, the operation control priority is divided through a regional clustering algorithm, and a train scheduling weight distribution table is output; the train scheduling weight distribution table is fused with the initial driving instruction sequence, and a dynamic optimization strategy network is constructed through a genetic algorithm to generate a final driving strategy solution.

2. The method for optimizing DAS driving strategy based on real-time subway line data according to claim 1, characterized in that: The method of extracting line operation characteristics through a convolutional spatiotemporal encoder includes: performing a sliding window capture on the real-time train position sequence to generate a spatiotemporal slice data block; dividing the spatiotemporal slice data block into multiple operation intervals based on regional clustering rules, and calculating the standard deviation of the signal response delay in each interval; extracting the track occupancy status label using a residual connection algorithm, and generating a signal system response relationship in combination with the interval travel delay coefficient; and encoding the signal response interval time, track occupancy status label, and signal system response relationship into line operation characteristics.

3. The method for optimizing DAS driving strategy based on real-time subway line data according to claim 1, characterized in that: The method of generating a passenger flow fluctuation prediction map through a time series decomposition algorithm includes: separating the trend components of the passenger flow data at the station and extracting periodic fluctuation characteristics; predicting future passenger flow distribution based on an autoregressive model and calculating the confidence level of the peak-hour distribution interval; generating a risk probability model based on a historical conflict event database and outputting an operation conflict probability heat map in combination with the confidence level; and mapping the peak-hour distribution interval and the operation conflict probability heat map into a passenger flow fluctuation prediction map.

4. The method for optimizing DAS driving strategy based on real-time subway line data according to claim 1, characterized in that: The driving strategy model includes a data fusion module and an instruction generation module. The data fusion module includes: standardizing the signal response interval time in the line operation characteristics to obtain a first fusion tensor; performing morphological filtering on the operation conflict probability heat map in the passenger flow fluctuation prediction map to generate a second fusion tensor; performing eigenvalue decomposition on the equipment health assessment matrix to extract the equipment state degradation gradient to obtain a third fusion tensor; and merging the first fusion tensor, the second fusion tensor, and the third fusion tensor into a strategy input sequence through a bidirectional long short-term memory network.

5. The method for optimizing DAS driving strategy based on real-time subway line data according to claim 4, characterized in that: The instruction generation module includes: aligning the strategy input sequence on the time axis to generate an instruction association vector; extracting key control node features through an attention mechanism to generate an instruction node association matrix; performing a tensor splicing operation on the instruction association vector and the instruction node association matrix to generate a candidate instruction set; and selecting an initial driving instruction sequence from the candidate instruction set through a branch and bound algorithm.

6. The method for optimizing DAS driving strategy driven by real-time subway line data according to claim 1, characterized in that: The method of constructing a dynamic optimization strategy network through a genetic algorithm includes: initializing a network population according to a train dispatch weight distribution table and generating a fitness evaluation function based on priority; using the initial driving instruction sequence as a chromosome encoding, wherein the fitness evaluation function is composed of an operating energy consumption index and a dispatch priority weight; updating the fitness score of each chromosome through a crossover mutation operator and adjusting the population evolution direction; and generating an optimal driving strategy solution that meets multiple objective constraints based on the optimized fitness score.

7. The method for optimizing DAS driving strategy based on real-time subway line data according to claim 2, characterized in that: The parameter optimization method of the regional clustering rule includes: calculating the initial clustering radius and the minimum signal delay threshold based on the distribution of historical operation data; traversing the parameter combination through the grid verification algorithm to select the parameter with the highest matching degree between the division result and the actual operation log; dynamically adjusting the clustering radius and the minimum signal delay threshold based on the matching degree to optimize the division accuracy of the signal system response relationship.

8. The method for optimizing DAS driving strategy based on real-time subway line data according to claim 3, characterized in that: The method for constructing the risk probability model includes: collecting historical operation conflict data of multiple lines, extracting spatiotemporal correlation characteristics and equipment failure influencing factors; fitting the spatiotemporal correlation characteristics with a Gaussian mixture model to generate a basic risk probability distribution table; and performing Bayesian correction on the basic risk probability distribution table according to the equipment failure influencing factors to output a risk probability model.

9. The method for optimizing DAS driving strategy based on real-time subway line data according to claim 6, characterized in that: The parameter setting method of the crossover mutation operator includes: defining the crossover probability between chromosomes as a balance factor between the operating energy consumption index and the scheduling priority weight; initializing the fitness score of each chromosome to a baseline value, and the fitness of the starting chromosome to a preset initial value; selecting the chromosome with the highest fitness score in the current population as the parent generation through an elite retention strategy; and performing multi-point crossover and random bit mutation operations based on the parent chromosome to generate a new generation population.

10. The method for optimizing DAS driving strategy based on real-time subway line data according to claim 9, characterized in that: The method also includes: real-time collection of the execution status log of the train control system, generating a strategy correction instruction set through an abnormal decision detection model, and updating the dynamic optimization strategy network; the construction method of the abnormal decision detection model includes: collecting normal operation samples in the historical execution status log, extracting the acceleration instruction change rate and the braking response time series; constructing a normal operation benchmark model based on a single-class support vector machine, and calculating the deviation index between the real-time data and the benchmark model; using a time decay function to count the cumulative abnormal value of the deviation index, and generating a strategy correction instruction set when it exceeds a dynamic threshold; and updating the chromosome encoding parameters of the dynamic optimization strategy network in combination with the correction instruction set.

Images