Voltage fluctuation suppression method and equipment for subway direct-current power supply system

Abstract

The invention provides a voltage fluctuation suppression method and device for a subway direct-current power supply system, and the method comprises the steps: obtaining the voltage amplitude change information of the subway direct-current power supply system at different sampling moments, and determining a real-time voltage fluctuation sequence and the operation state characteristics of each power supply node according to the voltage amplitude change information; space-time correlation mapping processing is carried out on the real-time voltage fluctuation sequence and the operation state features, a fluctuation correlation feature set is generated, and the fluctuation correlation feature set comprises voltage fluctuation space-time propagation path features and inter-node coupling influence features; calling a pre-trained fluctuation suppression decision model to perform dynamic suppression strategy generation processing on the fluctuation association feature set to obtain a voltage regulation parameter sequence for each power supply node; according to the voltage regulation parameter sequence and the operation state characteristics of the power supply nodes, generating a node-coordinated fluctuation suppression execution instruction set; and according to the fluctuation suppression execution instruction set, triggering voltage regulation operation of each power supply node to suppress voltage fluctuation.

Description

Technical Field

[0001] This disclosure relates to the field of subway power supply technology, and in particular to a method and device for suppressing voltage fluctuations in a subway DC power supply system. Background Technology

[0002] In subway operations, the DC power supply system is a critical infrastructure for ensuring the normal operation of trains. However, due to factors such as dynamic load changes, equipment start-up and shutdown, and the characteristics of the power grid itself, subway DC power supply systems often experience voltage fluctuations. Voltage fluctuations not only affect the normal operation of train equipment and reduce its lifespan, but can also negatively impact the passenger experience and, in severe cases, even lead to safety accidents.

[0003] Currently, methods for suppressing voltage fluctuations in subway DC power supply systems have the following shortcomings. Firstly, most existing methods focus only on the real-time values ​​of voltage fluctuations, neglecting the spatiotemporal correlation between different power supply nodes. Subway power supply systems are complex networks where voltage fluctuations propagate and influence each other across nodes. Suppression based solely on voltage fluctuation information from a single node is insufficient to comprehensively and accurately grasp the essential characteristics of voltage fluctuations, thus failing to achieve effective suppression. Secondly, traditional suppression strategies are often static and cannot be dynamically adjusted according to the real-time operating status of the power supply system. During subway operation, load changes are frequent and complex; static suppression strategies cannot adapt to these dynamic changes, resulting in poor suppression effects and failing to meet the requirements for safe and stable subway operation. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, an embodiment of the present invention provides a method for suppressing voltage fluctuations in a subway DC power supply system, comprising: The system acquires voltage amplitude variation information of the subway DC power supply system at different sampling times, and determines the real-time voltage fluctuation sequence and the operating status characteristics of each corresponding power supply node based on the voltage amplitude variation information. It then performs spatiotemporal correlation mapping processing on the real-time voltage fluctuation sequence and the operating status characteristics to generate a fluctuation correlation feature set with node correlation, including the spatiotemporal propagation path characteristics of voltage fluctuations and the coupling influence characteristics between nodes. A pre-trained fluctuation suppression decision model is invoked to perform dynamic suppression strategy generation processing on the fluctuation correlation feature set, resulting in a voltage adjustment parameter sequence for each power supply node. Based on the voltage adjustment parameter sequence and the operating status characteristics of each power supply node, a set of node-coordinated fluctuation suppression execution instructions is generated. Finally, based on the fluctuation suppression execution instruction set, the voltage adjustment operation of each power supply node is triggered to suppress voltage fluctuations.

[0005] In one possible implementation, the operating status features include the load change trends and equipment operating mode identifiers of each power supply node. The step of performing spatiotemporal correlation mapping on the real-time voltage fluctuation sequence and the operating status features to generate a set of fluctuation correlation features with node correlation includes: partitioning the real-time voltage fluctuation sequence according to the power supply nodes to obtain local voltage fluctuation sub-sequences corresponding to each power supply node; calculating a similarity index between local voltage fluctuation sub-sequences of adjacent power supply nodes based on the local voltage fluctuation sub-sequences corresponding to each power supply node, where the similarity index characterizes the degree of consistency in fluctuation patterns; and constructing a wave correlation feature set between the power supply nodes based on the similarity index and the topological connection relationship of the power supply network in the subway DC power supply system. The system generates a dynamic propagation adjacency matrix. Based on the propagation adjacency matrix and the real-time voltage fluctuation sequence, it analyzes the diffusion speed and attenuation characteristics of voltage fluctuations from the starting node to surrounding nodes, and generates spatiotemporal propagation path features of voltage fluctuations according to the diffusion speed and attenuation characteristics. It calculates the cross-influence coefficient of the load change trends of different power supply nodes, which describes the degree of influence of the load change of any power supply node on the voltage fluctuation of another power supply node. It combines the cross-influence coefficient corresponding to the power supply node with the matching degree of the device operation mode identifier to generate inter-node coupling influence features. Finally, it concatenates the spatiotemporal propagation path features and the inter-node coupling influence features along the time dimension to form a set of fluctuation correlation features with node correlation.

[0006] In one possible implementation, calculating the similarity index between local voltage fluctuation subsequences of adjacent power supply nodes based on the local voltage fluctuation subsequences corresponding to each power supply node includes: performing length normalization processing on the local voltage fluctuation subsequences of adjacent power supply nodes, and extracting the fluctuation extreme point distribution information of each length-normalized local voltage fluctuation subsequence; calculating the voltage amplitude difference between extreme points at corresponding positions in the two local voltage fluctuation subsequences based on the position of the maximum point and the first voltage amplitude corresponding to the maximum point, the position of the minimum point and the second voltage amplitude corresponding to the minimum point; calculating the amplitude similarity component based on the cumulative sum of the voltage amplitude differences; analyzing the fluctuation period characteristics of the two local voltage fluctuation subsequences, extracting the target fluctuation frequency component through Fourier transform, and calculating the matching degree of the target fluctuation frequency component as the period similarity component; and weighting and summing the amplitude similarity component and the period similarity component according to a preset weight ratio to obtain the similarity index between the local voltage fluctuation subsequences of adjacent power supply nodes.

[0007] In one possible implementation, the step of analyzing the diffusion speed and attenuation characteristics of voltage fluctuations from the starting node to surrounding nodes based on the fluctuation propagation adjacency matrix and the real-time voltage fluctuation sequence, and generating spatiotemporal propagation path characteristics of voltage fluctuations based on the diffusion speed and attenuation characteristics, includes: determining the starting node and corresponding starting time of the voltage fluctuation from the real-time voltage fluctuation sequence using a fluctuation initiation identification algorithm; expanding the search for adjacent power supply nodes layer by layer using a multi-order neighbor node search model constructed based on the fluctuation propagation adjacency matrix with the starting node as the center; recording the fluctuation time when each power supply node first encounters a voltage fluctuation that meets preset requirements, and calculating the relationship between each fluctuation time and the starting node. The time difference between the initial and final moments is taken as the arrival time of the fluctuation at the corresponding power supply node; the fluctuation propagation speed is calculated based on the arrival time of the fluctuation at each power supply node and the physical distance between the power supply nodes, wherein the physical distance is determined based on the node coordinates in the power supply network topology; the peak voltage amplitude of each power supply node during the fluctuation propagation process is extracted, and the ratio of the peak voltage amplitude to the peak voltage amplitude of the starting node is taken as the attenuation coefficient of the corresponding power supply node; the arrival time of the fluctuation, the fluctuation propagation speed, and the attenuation coefficient of each power supply node are arranged in the spatial distribution order of the nodes to generate a spatiotemporal propagation path feature characterizing the propagation sequence and intensity change of the voltage fluctuation.

[0008] In one possible implementation, generating inter-node coupling influence features by combining the matching degree of the cross-influence coefficient corresponding to the power supply node and the equipment operation mode identifier includes: extracting the equipment operation mode identifier of each power supply node and constructing a mode identifier matrix; determining the mode matching degree between any two power supply nodes based on whether their equipment operation mode identifiers are the same; determining the mode influence coefficient of any two power supply nodes based on their mode matching degree; multiplying the cross-influence coefficient and the mode influence coefficient to obtain a comprehensive influence coefficient; arranging the comprehensive influence coefficient according to the node connection relationship based on the topology of the power supply network to form a coupling influence matrix between power supply nodes; deleting elements in the coupling influence matrix between power supply nodes whose comprehensive influence coefficient is less than a preset threshold, and retaining elements whose comprehensive influence coefficient is greater than the preset threshold to generate inter-node coupling influence features.

[0009] In one possible implementation, the step of calling a pre-trained fluctuation suppression decision model to perform dynamic suppression strategy generation processing on the fluctuation-related feature set to obtain a voltage regulation parameter sequence for each power supply node includes: inputting the fluctuation-related feature set into the feature encoding layer of the fluctuation suppression decision model, performing high-dimensional feature mapping processing to generate a time-dependent encoded feature vector; performing node importance weight allocation processing on the encoded feature vector through the node attention module of the fluctuation suppression decision model to generate attention coefficients for each power supply node; performing weighted aggregation processing on the encoded feature vector based on the attention coefficients to obtain an aggregated feature vector; inputting the aggregated feature vector into the time-series prediction layer of the fluctuation suppression decision model to perform multi-step voltage fluctuation prediction processing to generate a voltage fluctuation trend prediction sequence for predicting future periods; determining the dynamic change curve of the regulation amount for voltage regulation for each power supply node based on the voltage fluctuation trend prediction sequence and the inter-node coupling influence features in the fluctuation-related feature set; and discretizing the dynamic change curve of the regulation amount at preset time intervals to generate a voltage regulation parameter sequence for each power supply node.

[0010] In one possible implementation, the node attention module of the fluctuation suppression decision model performs node importance weight allocation processing on the encoded feature vector to generate attention coefficients for each power supply node, including: parsing the feature components of each power supply node in the encoded feature vector, extracting the fluctuation sensitivity index of each feature component, wherein the fluctuation sensitivity index is positively correlated with the rate of change of voltage fluctuation amplitude; constructing an initial attention vector based on the fluctuation sensitivity index, wherein each element in the initial attention vector corresponds to an initial attention value of a power supply node; calculating the degree of deviation between the load change trend in the operating state characteristics of each power supply node and the average load change trend of the subway DC power supply system, generating a load deviation coefficient; introducing the load deviation coefficient into the attention calculation function to dynamically adjust the initial attention vector, and combining the cross-influence coefficient in the inter-node coupling influence characteristics to perform neighborhood propagation processing on the adjusted attention vector; and normalizing the attention vector after neighborhood propagation processing to generate attention coefficients for each power supply node.

[0011] In one possible implementation, determining the dynamic change curve of the regulation amount for voltage regulation of each power supply node based on the voltage fluctuation trend prediction sequence and the inter-node coupling influence characteristics in the fluctuation correlation feature set includes: determining the fluctuation range of voltage regulation based on the time points in the voltage fluctuation trend prediction sequence where the predicted voltage value is greater than a preset voltage stability threshold; calculating the deviation between the predicted voltage value and the preset voltage stability threshold at each time point in each fluctuation range to obtain the basic regulation amount corresponding to each time point; determining the set of associated nodes that have a significant impact on the voltage fluctuation of each power supply node based on the inter-node coupling influence characteristics in the fluctuation correlation feature set; extracting the regulation amount from the voltage regulation parameter sequence of the associated node set of each power supply node and calculating its coupling regulation contribution value to the corresponding power supply node; superimposing the basic regulation amount at the time point with the corresponding coupling regulation contribution value to obtain the total regulation amount of the corresponding power supply node; and determining the dynamic change curve of the regulation amount for voltage regulation of each power supply node based on the total regulation amount using a curve fitting algorithm.

[0012] In one possible implementation, generating a set of node-coordinated fluctuation suppression execution instructions based on the voltage regulation parameter sequence and the operating status characteristics of each power supply node includes: parsing the voltage regulation parameter sequence of each power supply node, extracting the voltage regulation amount and regulation duration parameters at each time point; determining the regulation authority level of each power supply node based on the equipment operating mode identifier in the operating status characteristics of each power supply node, wherein the regulation authority level is used to limit the maximum range of the voltage regulation amount; performing a rationality verification process on the voltage regulation amount based on the regulation authority level, and truncating and correcting the regulation amount that exceeds the authority range; and analyzing the differences between the voltage regulation amounts of different power supply nodes at the same time point. The coordination mechanism calculates a consistency coefficient for the adjustment direction, which is used to determine the degree of cooperation between the adjustment operations of the power supply nodes. When the consistency coefficient is lower than a preset threshold, a collaborative optimization algorithm is invoked to generate a node collaborative adjustment scheme for adjusting the voltage regulation amount of the target power supply node. The node collaborative adjustment scheme makes the consistency coefficient between the target power supply nodes greater than the preset threshold. The target power supply node includes at least one of the power supply nodes with a consistency coefficient lower than the preset threshold. The corrected and adjusted voltage regulation amount and adjustment duration parameters are associated and encapsulated with the corresponding power supply node identifier to generate a set of node collaborative fluctuation suppression execution instructions.

[0013] In conjunction with a second aspect of the present invention, an embodiment of the present invention provides a voltage fluctuation suppression device for a subway DC power supply system, the device comprising: a memory storing a computer program thereon; and a processor for executing the computer program stored in the memory to implement the voltage fluctuation suppression method for the subway DC power supply system according to any one of the first aspects.

[0014] The aforementioned technical solution comprehensively grasps the dynamic information of voltage fluctuations and the real-time operation of the power supply system by acquiring the real-time voltage fluctuation sequence and the corresponding operating status characteristics of the metro DC power supply system. It performs spatiotemporal correlation mapping on the real-time voltage fluctuation sequence and operating status characteristics to generate a fluctuation correlation feature set with node correlation. This deeply reveals the spatiotemporal propagation path of voltage fluctuations between different nodes and the coupling influence characteristics between nodes, overcoming the limitations of traditional methods that only focus on voltage fluctuations of a single node, and grasping the laws of voltage fluctuations from a global perspective. A pre-trained fluctuation suppression decision model is invoked to dynamically generate suppression strategies for the fluctuation correlation feature set, resulting in a voltage adjustment parameter sequence for each power supply node. This enables dynamic adjustment of the suppression strategy, adapting to the dynamic changes in the load of the metro power supply system and improving the targeting and effectiveness of suppression. Based on the voltage adjustment parameter sequence and the operating status characteristics of each power supply node, a set of node-coordinated fluctuation suppression execution instructions is generated and sent to the corresponding power supply control module to trigger voltage adjustment operations. Through node coordination, comprehensive and effective suppression of voltage fluctuations is achieved, ensuring the stable operation of the metro DC power supply system and improving the safety and reliability of train operation.

[0015] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the execution flow of the voltage fluctuation suppression method for a subway DC power supply system provided in an embodiment of the present invention.

[0017] Figure 2 This is one implementation provided by an embodiment of the present invention. Figure 1 A flowchart illustrating step S12.

[0018] Figure 3 This is one implementation provided by an embodiment of the present invention. Figure 1 A flowchart of step S13.

[0019] Figure 4 This is one implementation provided by an embodiment of the present invention. Figure 1 A flowchart illustrating step S14.

[0020] Figure 5 This is a schematic diagram of exemplary hardware and software components of the voltage fluctuation suppression device for a subway DC power supply system provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0023] This invention provides a method for suppressing voltage fluctuations in a subway DC power supply system. (See also...) Figure 1 As shown, the method includes: In step S11, the voltage amplitude change information of the subway DC power supply system at different sampling times is obtained, and based on the voltage amplitude change information, the real-time voltage fluctuation sequence of the subway DC power supply system and the operating status characteristics of each power supply node in the subway DC power supply system are determined.

[0024] Among these, voltage amplitude variation information refers to the deviation data between the instantaneous voltage value and the reference value (such as the rated voltage) collected in real time by voltage sensors in the subway DC power supply system, used to reflect the dynamic characteristics of voltage changes over time. The real-time voltage fluctuation sequence is a discrete-time series formed by arranging the voltage amplitude variation information at the sampling time in chronological order, used to quantify the amplitude, frequency, and duration of voltage fluctuations. The operating status characteristics of the power supply node can include parameters such as node current, load power, temperature, switch status, and harmonic content, used to describe the real-time operating conditions of the power supply node in the power supply network.

[0025] In this embodiment, a distributed voltage sensor network can be used to acquire the voltage amplitude of the DC bus and each power supply node in the subway DC power supply system at a high-frequency sampling rate (e.g., millisecond level), while simultaneously collecting auxiliary parameters such as current and power. The sampled voltage amplitude change information is then filtered (e.g., using moving average filtering) to eliminate noise. The voltage deviation (ΔV = V_sample - V_rated) is then calculated and arranged in chronological order to generate a real-time voltage time series.

[0026] Furthermore, based on the collected harmonic content, current and power data, combined with the node topology location, the load rate, power factor and other characteristics are calculated, and abnormal states (such as overload and short circuit) are identified through threshold judgment or machine learning models (such as isolated forest), thus obtaining the operating status characteristics of each power supply node.

[0027] In step S12, the real-time voltage fluctuation sequence and the operating state features are subjected to spatiotemporal correlation mapping processing to generate a fluctuation correlation feature set with node correlation. The fluctuation correlation feature set includes the spatiotemporal propagation path features of voltage fluctuations and the coupling influence features between nodes.

[0028] The spatiotemporal correlation mapping combines the time series of voltage fluctuations with the topology (spatial dimension) of the power supply network to analyze the propagation path and coupling relationship of fluctuations between nodes. The spatiotemporal propagation path features in the fluctuation correlation feature set describe the propagation direction, delay, and attenuation of fluctuations from the source node to other nodes. The inter-node coupling influence features quantify the degree of mutual influence of voltage fluctuations between nodes (e.g., correlation coefficient, causal relationship strength). The fluctuation correlation feature set is a multi-dimensional feature set that integrates spatiotemporal propagation path features and inter-node coupling influence features, used to characterize node correlation.

[0029] In this embodiment, a graph model of the power supply network is constructed, where nodes represent power supply equipment (such as traction substations and overhead contact line segments), and edges represent physical connections. Based on the graph model of the power supply network, the time difference of voltage fluctuations at different power supply nodes can be calculated using a cross-correlation function, thereby determining the propagation direction of voltage fluctuations. Then, based on the voltage fluctuations of each power supply node represented by the real-time voltage fluctuation sequence along the propagation direction, a decay curve of the voltage fluctuation amplitude with propagation distance (such as an exponential decay model) is fitted, thereby generating the spatiotemporal propagation path characteristics of voltage fluctuations based on the time difference, propagation direction, and decay curve. Next, it is determined whether the fluctuation of node A has predictive power for node B, thus establishing a causal relationship. Finally, based on partial least squares regression (PLSR) or a neural network, a nonlinear mapping relationship of voltage fluctuations between nodes is established, generating inter-node coupling influence characteristics.

[0030] In step S13, a pre-trained fluctuation suppression decision model is invoked to perform dynamic suppression strategy generation processing on the fluctuation-related feature set, thereby obtaining a voltage regulation parameter sequence for each power supply node.

[0031] The pre-trained fluctuation suppression decision model is a deep learning model (such as an LSTM-Attention network) trained on historical data. Its input is fluctuation-related features, and its output is the voltage regulation parameters (such as regulation amplitude and duration) of each power supply node. The dynamic suppression strategy dynamically adjusts the regulation parameters according to real-time fluctuation features to achieve adaptive suppression.

[0032] In this embodiment, historical voltage fluctuation events and corresponding suppression measures (such as adjusting transformer taps and switching capacitors) are collected, and the optimal adjustment parameters are labeled. An LSTM network is used to capture time dependencies, and an attention mechanism is used to focus on key node features for model pre-training, resulting in a pre-trained fluctuation suppression decision model.

[0033] In this embodiment of the disclosure, the training steps of the fluctuation suppression decision model are as follows: Step S210: Collect historical voltage fluctuation data of the subway DC power supply system, the corresponding power supply node operating status characteristics and historical fluctuation suppression strategy data, and construct a model training dataset.

[0034] To train a decision model capable of accurately generating fluctuation suppression strategies, a large amount of historical data can be collected. Historical voltage fluctuation data includes voltage amplitude change sequences for each power supply node under different time periods and operating conditions; power supply node operating status characteristics include historical load change trends and equipment operating mode identifiers; historical fluctuation suppression strategy data includes the sequence of adjustment parameters adopted for historical voltage fluctuations and actual suppression effect data.

[0035] The collected data undergoes preprocessing, including data cleaning, missing value imputation, and outlier handling, to ensure data quality and consistency. The preprocessed data is then divided into training, validation, and test sets according to a specific ratio to construct a complete model training dataset.

[0036] Step S220: Perform the same spatiotemporal correlation mapping process as in step S120 on the historical voltage fluctuation data and power supply node operating status features in the model training dataset to generate a set of historical fluctuation correlation features.

[0037] Using a method identical to step S120, spatiotemporal correlation mapping is performed on the historical voltage fluctuation data and power supply node operating status characteristics in the model training dataset. This includes steps such as dividing local voltage fluctuation subsequences by power supply node, calculating the similarity index of adjacent node subsequences, constructing a fluctuation propagation adjacency matrix, analyzing fluctuation propagation path characteristics, calculating cross-influence coefficients, generating inter-node coupling influence characteristics, and splicing them to form a set of historical fluctuation correlation features.

[0038] This processing enables historical data to be transformed into feature data consistent with the format of fluctuation-related feature sets in practical applications, providing suitable input for model training.

[0039] Step S230: Construct the network structure of the fluctuation suppression decision model, wherein the network structure includes a feature encoding layer, a node attention module, a temporal prediction layer, and a policy generation layer.

[0040] The network structure of the fluctuation suppression decision model is constructed using a deep learning framework. The feature encoding layer employs a combination of convolutional neural networks and recurrent neural networks to perform high-dimensional feature mapping and capture temporal dependencies on the input set of historical fluctuation-related features. The node attention module uses a self-attention mechanism to assign node importance weights to the encoded feature vectors. The temporal prediction layer uses a long short-term memory network to predict future voltage fluctuation trends based on the aggregated feature vectors. The policy generation layer uses a fully connected neural network to generate a voltage regulation parameter sequence based on the predicted fluctuation trends and the coupling effects between nodes.

[0041] The layers are connected by neurons to form a complete network structure, and the initial values ​​of the network parameters are set by random initialization.

[0042] Step S240: Using the historical fluctuation correlation feature set as input and the voltage regulation parameter sequence in the corresponding historical fluctuation suppression strategy data as the expected output, train the fluctuation suppression decision model.

[0043] The set of historical volatility-related features is input into the constructed volatility suppression decision model, which calculates the predicted voltage regulation parameter sequence through forward propagation. The predicted result is compared with the expected output (i.e., the actual voltage regulation parameter sequence) in the historical volatility suppression strategy data, and the loss value between the two is calculated.

[0044] The loss function is the mean squared error function, which measures the difference between the predicted value and the expected output. The parameters of each layer of the model are adjusted based on the loss value using the backpropagation algorithm to reduce the prediction error.

[0045] During training, the model is iteratively trained using the training set. After a certain number of iterations, the model's performance is evaluated using the validation set. If the loss value on the validation set no longer decreases or shows an upward trend, training is stopped to avoid overfitting.

[0046] Step S250: Use the test set to evaluate the performance of the trained fluctuation suppression decision model. When the prediction accuracy of the model reaches the preset standard, the model training is considered complete.

[0047] After training, the performance of the fluctuation suppression decision model was evaluated using a test set. Evaluation metrics included the similarity between the predicted voltage regulation parameter sequence and the actual sequence, and the fluctuation suppression effect after voltage regulation based on the predicted sequence.

[0048] The model is considered complete when its prediction accuracy reaches a preset standard, such as the similarity between the predicted sequence and the actual sequence being above a preset threshold, and the fluctuation suppression effect based on the predicted sequence meeting system requirements. The trained model can then be deployed to a real-world subway DC power supply system to generate real-time fluctuation suppression strategies.

[0049] Step S260: Regularly update and optimize the trained fluctuation suppression decision model, and adjust the model parameters to adapt to changes in the system's operating state based on new operating data and suppression effect feedback.

[0050] The operating status of the subway DC power supply system will change over time, such as equipment aging and changes in load mode. Therefore, the fluctuation suppression decision model can be updated and optimized regularly.

[0051] Regularly collect new operational data and feedback on fluctuation suppression effectiveness, add this data to the model training dataset, and retrain the model according to steps S220 to S240, adjusting the model parameters. In this way, the model can continuously adapt to changes in the system's operating state and maintain a good ability to generate fluctuation suppression strategies.

[0052] During model updates and optimizations, incremental training can be used to avoid completely retraining the model, thereby improving update efficiency. Simultaneously, the updated model's performance will be evaluated to ensure that its performance is not lower than that of the original model.

[0053] Furthermore, the generated spatiotemporal propagation path features and inter-node coupling influence features are input into a pre-trained fluctuation suppression decision model, outputting a sequence of adjustment parameters for each power supply node (e.g., ΔV_i(t), t=1,2,…,T). The model parameters are then optimized online using reinforcement learning (e.g., the PPO algorithm) to adapt to newly emerging fluctuation patterns. This enables an end-to-end mapping from fluctuation features to suppression strategies, improving decision-making efficiency.

[0054] In step S14, a set of node-coordinated fluctuation suppression execution instructions is generated based on the voltage regulation parameter sequence and the operating status characteristics of each power supply node.

[0055] Node coordination involves using coordinated control of multiple power supply nodes to achieve global voltage stability and prevent new fluctuations caused by local adjustments. The fluctuation suppression execution instruction set includes instruction parameters such as the adjustment target value, execution time, and priority of each power supply node.

[0056] In this embodiment of the disclosure, power balance constraints ensure that the total power demand after adjustment matches the power supply capacity, and device limiting constraints limit the adjustment range (e.g., ΔV) of individual power supply nodes. i (t) ≤ 5% V_rated) to avoid equipment overload, ΔV i (t) represents the adjustment range of the i-th power supply node, and V_rated is the preset range threshold.

[0057] Furthermore, a distributed optimization algorithm (such as ADMM) is employed to decompose the global problem into node-level subproblems. Each node solves these subproblems independently and then coordinates through communication to reach a consensus. Command priorities can be dynamically allocated based on node criticality (e.g., traction substations have higher priority than ordinary load nodes). This ensures the executability and global optimality of the control commands.

[0058] In one possible implementation, the target voltage adjustment ΔV of each node within the time window t∈[1,T] i (t). The operating status characteristics of the power supply node may include the real-time load rate λ. i Power factor cosϕ i Equipment health status (such as temperature, degree of aging), etc.

[0059] Preprocessing operation: Normalization, which normalizes ΔV i (t), λ i Features are mapped to the [0,1] interval to eliminate dimensional differences.

[0060] Status label mapping converts device health status into weight coefficients (e.g., weight = 1 when the health status is "good" and weight = 0.5 when the health status is "fault warning"), which are used for subsequent constraint reinforcement.

[0061] Furthermore, minimize the global adjustment cost while satisfying all constraints:

[0062] Among them, C i (ΔV i (t) represents the adjustment cost of power supply node i (such as equipment loss and number of operations), and ρ is the penalty coefficient.

[0063] Where, ω i ω is the priority weight coefficient of power supply node i, used to quantify the relative importance of different nodes in global optimization. Different nodes have different degrees of impact on system stability. For example, voltage fluctuations in traction substations may directly affect train operation safety, while load fluctuations in ordinary sections have a smaller impact. iBy assigning higher weights to critical nodes, you can ensure that their adjustment commands are executed preferentially. For example, the weights can be adjusted based on the node's real-time status (such as load rate λ). i ω changes dynamically. For example, when the load rate of node i exceeds a threshold, ω... i It can be increased in real time, enhancing its contribution to global optimization.

[0064] Here, Penalty(constraints) is a scalar function used to quantify the degree of constraint violation during the optimization process. By transforming hard constraints (such as device limiting and power balancing) into penalty terms, it ensures that the optimization solution satisfies physical feasibility. For example, if the node voltage regulation ΔV i (t) If the deviation exceeds the device limit (e.g., ±5%), the penalty term will significantly increase the objective function value, forcing the optimizer to adjust the solution. Soft constraint handling: For non-critical constraints (e.g., adjustment rate limits), a certain degree of violation can be allowed by adjusting the penalty coefficient ρ to balance optimization efficiency and constraint strictness.

[0065] Furthermore, the global optimization is broken down into N node-level subproblems, with each node independently solving its local objective function (e.g., minimizing its own adjustment cost). Inter-node coupling information (e.g., power balance) is exchanged using Lagrange multipliers, and the process is iteratively updated until convergence. For example, a convergence condition could be that the change in the global objective function value between two consecutive iterations is less than a threshold ϵ (e.g., ϵ=10). −4 ), to ensure coordination and consistency.

[0066] In step S15, voltage regulation operations of each power supply node are triggered according to the set of fluctuation suppression execution instructions to suppress voltage fluctuations.

[0067] Voltage regulation can be achieved by adjusting node voltage through power electronic devices (such as DC / DC converters, SVG) or mechanical devices (such as on-load tap changers).

[0068] In this embodiment, communication with the regulating device is achieved via, for example, the IEC 61850 protocol or Modbus, to issue commands (such as setting the reactive power output reference value of the SVG). The regulated node voltage is monitored in real time; if the deviation exceeds a threshold (e.g., ±2%V_rated), secondary regulation is triggered. Alternatively, a PID controller or model predictive control (MPC) can be used to dynamically adjust the regulation amount to eliminate steady-state errors.

[0069] The aforementioned technical solution comprehensively grasps the dynamic information of voltage fluctuations and the real-time operation of the power supply system by acquiring the real-time voltage fluctuation sequence and the corresponding operating status characteristics of the metro DC power supply system. It performs spatiotemporal correlation mapping on the real-time voltage fluctuation sequence and operating status characteristics to generate a fluctuation correlation feature set with node correlation. This deeply reveals the spatiotemporal propagation path of voltage fluctuations between different nodes and the coupling influence characteristics between nodes, overcoming the limitations of traditional methods that only focus on voltage fluctuations of a single node, and grasping the laws of voltage fluctuations from a global perspective. A pre-trained fluctuation suppression decision model is invoked to dynamically generate suppression strategies for the fluctuation correlation feature set, resulting in a voltage adjustment parameter sequence for each power supply node. This enables dynamic adjustment of the suppression strategy, adapting to the dynamic changes in the load of the metro power supply system and improving the targeting and effectiveness of suppression. Based on the voltage adjustment parameter sequence and the operating status characteristics of each power supply node, a set of node-coordinated fluctuation suppression execution instructions is generated and sent to the corresponding power supply control module to trigger voltage adjustment operations. Through node coordination, comprehensive and effective suppression of voltage fluctuations is achieved, ensuring the stable operation of the metro DC power supply system and improving the safety and reliability of train operation.

[0070] In one possible implementation, the operating status characteristics include the load change trend of each power supply node and the equipment operating mode identifier, wherein the load change trend is used to characterize the load power change pattern of the power supply node in the time dimension, typically expressed as a time series (e.g., P). i (t) or a trend function (such as linear, periodic). Equipment operating mode identifiers are discrete labels used to characterize the current operating state of power supply equipment (such as "normal mode", "derating mode", "fault mode"), and are used to quantify the differences in the equipment's response characteristics to voltage fluctuations.

[0071] See Figure 2 As shown, in step S12, the spatiotemporal correlation mapping processing of the real-time voltage fluctuation sequence and the operating state features to generate a set of fluctuation correlation features with node correlation includes: In step S121, the real-time voltage fluctuation sequence is divided into partitions according to the power supply nodes to obtain local voltage fluctuation sub-sequences corresponding to each power supply node.

[0072] Among them, the local voltage fluctuation subsequence is a time subsequence that contains only the voltage fluctuation data of a single node after the global voltage fluctuation sequence is divided according to the power supply node.

[0073] In this embodiment of the disclosure, the real-time voltage fluctuation sequence V(t) is aligned to the sampling time of each power supply node according to the timestamp to ensure time synchronization.

[0074] Node partitioning can be based on the power supply network topology, dividing V(t) into N subsequences, each containing only voltage fluctuation data of node i. A sliding window processing method can then be applied, for example, by applying a sliding window (e.g., window length Tw = 10 seconds) to each subsequence to extract local fluctuation segments and reduce non-stationary interference.

[0075] In this embodiment, since the real-time voltage fluctuation sequence is a set of voltage amplitudes of the entire power supply system at different sampling times, it can be partitioned according to the power supply node to analyze the voltage fluctuation of each power supply node. Based on the number and distribution range of the power supply node, the voltage amplitude data belonging to the same power supply node in the real-time voltage fluctuation sequence are extracted to form the local voltage fluctuation subsequence corresponding to that node.

[0076] For example, suppose a power supply system has multiple power supply nodes, numbered A, B, C, etc. From the real-time voltage fluctuation sequence, the voltage amplitude at all sampling times marked as node A is extracted and arranged in chronological order to obtain the local voltage fluctuation subsequence of node A. Similarly, the local voltage fluctuation subsequences of nodes B, C, etc., can be obtained. Each local voltage fluctuation subsequence completely records the change in voltage amplitude of the corresponding node over time during the monitoring period.

[0077] In step S122, based on the local voltage fluctuation subsequence corresponding to each power supply node, a similarity index is calculated between the local voltage fluctuation subsequences of adjacent power supply nodes. The similarity index is used to characterize the degree of consistency of the fluctuation pattern.

[0078] Among them, the similarity index is an indicator used to quantify the degree of morphological consistency between two local voltage fluctuation subsequences (such as dynamic time warping distance and Pearson correlation coefficient).

[0079] In this embodiment of the disclosure, the local subsequence V of adjacent nodes (such as nodes i and j) i (t) and V j (t), the similarity index S is calculated using the Dynamic Time Warping (DTW) algorithm. ij :

[0080] Where DTW is the dynamic time warp distance, the smaller the value, the more similar the shapes, len(V i V represents the voltage time series of power supply node i. i The length of (t), len(V) j ) represents the voltage time series V of power supply node j. j The length of (t).

[0081] Furthermore, the similarity index S ij Mapped to the [0,1] interval, 1 represents complete similarity, and 0 represents complete irrelevance. The similarity index reflects the consistency of fluctuation patterns; high similarity indicates that there may be direct or indirect fluctuation propagation paths between nodes.

[0082] In step S123, a wave propagation adjacency matrix between the power supply nodes is constructed based on the similarity index and the topological connection relationship of the power supply network in the subway DC power supply system.

[0083] The wave propagation adjacency matrix is ​​a matrix used to describe the relationship of voltage wave propagation between power supply nodes, and the element values ​​represent the intensity or probability of wave propagation.

[0084] In this embodiment, considering the power supply network topology (such as a traction substation-overhead contact network-train node chain), the propagation relationship is calculated only for physically connected node pairs. An N×N matrix A is then constructed, where the elements A in matrix A are... ij For: If nodes i and j are physically connected, then A ij It is the product of the similarity index corresponding to nodes i and j and the reciprocal of the electrical distance from node i to j.

[0085] The wave propagation adjacency matrix can quantify the propagation intensity of waves between nodes. The closer the electrical distance between nodes and the more similar their shapes, the higher the propagation intensity.

[0086] In step S124, based on the wave propagation adjacency matrix and the real-time voltage wave sequence, the diffusion speed and attenuation characteristics of the voltage wave from the starting node to the surrounding nodes are analyzed, and the spatiotemporal propagation path characteristics of the voltage wave are generated according to the diffusion speed and the attenuation characteristics.

[0087] The diffusion velocity describes the time delay required for a voltage fluctuation to propagate from the starting node to an adjacent node. The attenuation characteristic describes how the amplitude of a voltage fluctuation decreases with increasing distance during propagation (e.g., exponential attenuation, linear attenuation).

[0088] In this embodiment of the disclosure, for each node i, the voltage fluctuation start time t is identified by a mutation detection algorithm. i,start Then, the time delay Δt for the wave to propagate to the adjacent node j is calculated. ij =t j,start -t i,start .

[0089] Furthermore, the diffusion rate v ij =d ij / Δt ij , where d ijIt is the electrical distance from node i to j. The attenuation characteristic can be fitted as the fluctuation amplitude attenuation curve A(d) = A0 × e −αd Where A0 is the initial node amplitude and α is the attenuation coefficient. α is estimated using the least squares method to quantize the attenuation rate.

[0090] Spatiotemporal propagation path features are obtained through path feature encoding. For example, the diffusion velocity and attenuation coefficient can be concatenated into a feature vector pi=[vi1,…,viN,αi] to describe the wave propagation characteristics of node i. Spatiotemporal propagation path features are used to describe the spatial diffusion law and energy attenuation mechanism of waves.

[0091] In step S125, the cross-influence coefficient of the load change trends of different power supply nodes is calculated. The cross-influence coefficient is used to describe the degree of influence of the load change of any power supply node on the voltage fluctuation of another power supply node.

[0092] The cross-influence coefficient is a scalar value used to quantify the degree to which a load change at one node affects the voltage fluctuation at another node.

[0093] In this embodiment of the disclosure, load change trends are an important factor affecting voltage fluctuations. Load changes at different power supply nodes may influence each other, thereby affecting each other's voltage fluctuations. Therefore, the cross-influence coefficient of load change trends at different nodes can be calculated.

[0094] First, the load variation trends are extracted from the operating status characteristics of each power supply node. These load variation trends can be represented by load variation curves over a period of time. Then, for any two nodes, the relationship between the load variation trend of one node and the voltage fluctuation of the other node is analyzed.

[0095] Statistical analysis is used to determine how the voltage fluctuation of one node changes when the load of another node changes, and then the cross-influence coefficient is calculated. If the load change of one node has a significant impact on the voltage fluctuation of another node, the cross-influence coefficient is large; conversely, the cross-influence coefficient is small.

[0096] In step S126, the coupling influence characteristics between nodes are generated by combining the cross-influence coefficient corresponding to the power supply node and the matching degree of the equipment operation mode identifier.

[0097] Among them, the cross-influence coefficient reflects the degree of influence of load changes on voltage fluctuations, while the matching degree of equipment operation mode identifier reflects the similarity of different nodes in equipment operation characteristics. Combining these two can more comprehensively describe the coupling influence relationship between nodes and generate inter-node coupling influence characteristics.

[0098] In step S127, the spatiotemporal propagation path features and the inter-node coupling influence features are spliced ​​together according to the time dimension to form a set of fluctuation correlation features with node correlation.

[0099] Among them, the spatiotemporal propagation path feature describes the propagation of voltage fluctuations in time and space, and the inter-node coupling influence feature describes the mutual influence relationship between nodes. By splicing these two features along the time dimension, they can be integrated into a unified feature set, namely the fluctuation correlation feature set.

[0100] In this embodiment of the disclosure, splicing by time dimension means combining the spatiotemporal propagation path features and inter-node coupling influence features within the same time segment, so that the feature set can reflect both the voltage fluctuation propagation state at different time points and the inter-node coupling influence at the corresponding time points, thereby fully presenting the voltage fluctuation features with node correlation.

[0101] In one possible implementation, step S122, calculating the similarity index between local voltage fluctuation subsequences of adjacent power supply nodes based on the local voltage fluctuation subsequences corresponding to each power supply node, includes: In step S1221, the local voltage fluctuation subsequences of adjacent power supply nodes are length normalized, and the distribution information of the fluctuation extreme points of each length normalized local voltage fluctuation subsequence is extracted.

[0102] Since the local voltage fluctuation subsequences of different adjacent power supply nodes may differ in sampling duration or number of sampling points, to ensure the accuracy and comparability of subsequent calculations, these subsequences can first be normalized in length. By interpolation or truncation, subsequences of different lengths can be adjusted to the same length.

[0103] After length normalization, the distribution information of extreme points of fluctuations in each subsequence is extracted. Extreme points include maxima and minima. A maxima is the point where the voltage amplitude reaches its relative highest value within a certain time period, and a minima is the point where the voltage amplitude reaches its relative lowest value. For each local voltage fluctuation subsequence, the change in voltage amplitude is analyzed segment by segment to determine the locations of all maxima and minima (i.e., the corresponding sampling times) and the voltage amplitude corresponding to each location. This information together constitutes the distribution information of extreme points of fluctuations.

[0104] In step S1222, the voltage amplitude difference between the extreme points at corresponding positions in the two local voltage fluctuation subsequences is calculated based on the position of the maximum point in the fluctuation extreme point distribution information and the first voltage amplitude corresponding to the maximum point, the position of the minimum point in the fluctuation extreme point distribution information and the second voltage amplitude corresponding to the minimum point.

[0105] In this embodiment of the disclosure, for the local voltage fluctuation subsequences of two adjacent power supply nodes, after extracting the extreme point distribution information, the corresponding extreme point positions in time are found. Here, the corresponding positions refer to extreme points that are at the same or similar positions on the normalized time axis.

[0106] In step S1223, the amplitude similarity component is calculated based on the cumulative sum of the voltage amplitude differences.

[0107] In this embodiment of the disclosure, the voltage amplitude difference between these corresponding extreme points is calculated, that is, the voltage amplitude at the extreme point of one subsequence is subtracted from the voltage amplitude at the corresponding extreme point of another subsequence, resulting in a series of difference quantities. These difference quantities are then accumulated and summed, and the amplitude similarity component is calculated based on the magnitude of the accumulated sum. The smaller the accumulated sum, the closer the voltage amplitudes of the two subsequences are at their extreme points, and the larger the amplitude similarity component; conversely, the larger the accumulated sum, the smaller the amplitude similarity component.

[0108] In step S1224, the fluctuation period characteristics of the two local voltage fluctuation subsequences are analyzed, the target fluctuation frequency component is extracted by Fourier transform, and the matching degree of the target fluctuation frequency component is calculated as the period similarity component.

[0109] The target fluctuation frequency component can be the main fluctuation frequency component. By analyzing the fluctuation period of the local voltage fluctuation subsequence, the repetitive pattern of voltage fluctuation can be understood. Fourier transforms are performed on the local voltage fluctuation subsequences of two adjacent power supply nodes to convert the time-domain signals into frequency-domain signals, thereby extracting their respective main fluctuation frequency components.

[0110] In this embodiment of the disclosure, the main fluctuation frequency component refers to the frequency value with relatively high energy in the frequency domain. Then, the matching degree of the main fluctuation frequency components of the two subsequences is calculated, that is, comparing whether their main frequency components are the same or similar. The higher the matching degree, the more similar the fluctuation periodic characteristics of the two subsequences, and the larger the periodic similarity component; conversely, the smaller the periodic similarity component.

[0111] In step S1225, the amplitude similarity component and the period similarity component are weighted and summed according to a preset weight ratio to obtain the similarity index between the local voltage fluctuation subsequences of adjacent power supply nodes.

[0112] In this embodiment of the disclosure, after obtaining the amplitude similarity component and the period similarity component, they are weighted and summed according to a preset weight ratio. The preset weight ratio is determined based on the degree of influence of amplitude changes and period changes on the consistency of wave pattern in the actual power supply system. For example, if the amplitude change has a greater impact on the consistency of wave pattern in the subway DC power supply system, then a higher weight is set for the amplitude similarity component; conversely, a higher weight is set for the period similarity component.

[0113] The result obtained after weighted summation is the similarity index between the local voltage fluctuation subsequences of adjacent power supply nodes. To facilitate subsequent calculations and analysis, the similarity index is normalized, adjusting its value range to a preset characteristic interval, such as standardizing it to between 0 and 1. Normalization eliminates dimensional differences between different similarity indices, making them comparable.

[0114] In one possible implementation, in step S124, the step of analyzing the diffusion rate and attenuation characteristics of voltage fluctuations from the starting node to surrounding nodes based on the fluctuation propagation adjacency matrix and the real-time voltage fluctuation sequence, and generating spatiotemporal propagation path characteristics of voltage fluctuations according to the diffusion rate and the attenuation characteristics, includes: In step S1241, the starting node and corresponding starting time of the voltage fluctuation are determined from the real-time voltage fluctuation sequence using a fluctuation initiation identification algorithm.

[0115] In this embodiment, the fluctuation initiation identification algorithm is specifically designed to detect the starting position and time of voltage fluctuations. The real-time voltage fluctuation sequence is analyzed node-by-node and time-by-time. When the voltage amplitude of a certain node deviates from the normal stable range and this deviation persists for a certain period, that node is determined as the starting node of the voltage fluctuation, and the corresponding deviation start time is the initiation time.

[0116] For example, in a real-time voltage fluctuation sequence, if the voltage amplitude of node A suddenly exceeds the normal range at a certain moment and continues to deviate thereafter, then node A is determined as the starting node, and the moment when it suddenly exceeds the range is the starting moment.

[0117] In step S1242, with the starting node as the center, the multi-order neighbor node search model constructed based on the wave propagation adjacency matrix is ​​used to expand the search for adjacent power supply nodes layer by layer.

[0118] In this embodiment, after determining the starting node, a multi-order neighbor node search model is constructed using the wave propagation adjacency matrix, centered on this node. First-order neighbor nodes are those directly connected to the starting node, i.e., nodes in the wave propagation adjacency matrix whose corresponding row or column element value is not 0. Second-order neighbor nodes are those directly connected to first-order neighbor nodes but not directly connected to the starting node, and so on, expanding the search for neighboring nodes layer by layer. This determines the range of all nodes that voltage fluctuations may propagate to.

[0119] In step S1243, the fluctuation time when the voltage fluctuation that first meets the preset requirements of each power supply node is recorded, and the time difference between each fluctuation time and the starting time is calculated as the fluctuation arrival time of the corresponding power supply node.

[0120] In this embodiment of the disclosure, after determining the range of nodes that may be affected by fluctuations, the local voltage fluctuation subsequences of these nodes are monitored, and the moment when a significant voltage fluctuation first occurs at each node is recorded. A significant voltage fluctuation refers to a voltage amplitude deviating from the normal range and reaching a certain fluctuation magnitude.

[0121] Then, the fluctuation arrival time for each node is calculated, which is the time difference between the moment when a significant voltage fluctuation first occurs at that node and the starting moment at the starting node. The fluctuation arrival time reflects the time required for the voltage fluctuation to propagate from the starting node to that node.

[0122] In step S1244, the fluctuation propagation speed is calculated based on the fluctuation arrival time of each power supply node and the physical distance between the power supply nodes, wherein the physical distance is determined based on the node coordinates in the power supply network topology diagram.

[0123] Among them, the fluctuation propagation speed reflects how fast voltage fluctuations propagate in space. The fluctuation propagation speed of different nodes may vary due to factors such as the connection method between nodes and line impedance.

[0124] In this embodiment of the disclosure, the physical distance between nodes can be calculated based on the coordinates of each node in the power supply network topology diagram. The coordinate information is a fixed parameter determined during the design of the power supply network. For each node, the diffusion speed of the wave from the starting node to that node is obtained by dividing the physical distance between that node and the starting node by the wave arrival time.

[0125] In step S1245, the peak voltage amplitude of each power supply node during the wave propagation process is extracted, and the ratio of the peak voltage amplitude to the peak voltage amplitude of the starting node is used as the attenuation coefficient of the corresponding power supply node.

[0126] In this embodiment of the disclosure, during the propagation of voltage fluctuations to each node, the voltage amplitude at that node will exhibit a peak value, which is the maximum voltage amplitude during the fluctuation process. The peak voltage amplitude of each node is extracted and compared with the peak voltage amplitude of the starting node.

[0127] The ratio of the peak voltage amplitude at each node to the peak voltage amplitude at the starting node is calculated; this ratio is the attenuation coefficient. The attenuation coefficient reflects the intensity attenuation of the voltage fluctuation during propagation; the smaller the attenuation coefficient, the more significant the attenuation of the fluctuation intensity.

[0128] In step S1246, the arrival time of the fluctuation, the diffusion speed of the fluctuation, and the attenuation coefficient of each power supply node are arranged in the spatial distribution order of the nodes to generate spatiotemporal propagation path features that characterize the propagation timing and intensity changes of the voltage fluctuation.

[0129] In this embodiment of the disclosure, the arrival time, diffusion rate and attenuation coefficient of each node are arranged according to the spatial distribution order of the nodes to form a multi-dimensional feature set, which is the spatiotemporal propagation path feature of voltage fluctuation.

[0130] For example, by arranging the nodes in order of their distance from the starting node to the nearest, the arrival time, diffusion rate, and attenuation coefficient of the fluctuations corresponding to each node are listed in sequence. This clearly shows the propagation sequence of voltage fluctuations in time and the intensity changes in space, and fully describes the propagation path characteristics of voltage fluctuations.

[0131] In one possible implementation, step S126, generating inter-node coupling influence characteristics by combining the cross-influence coefficient corresponding to the power supply node and the matching degree of the device operating mode identifier, includes: In step S1261, the device operation mode identifier of each power supply node is extracted, and a mode identifier matrix is ​​constructed.

[0132] In this embodiment of the disclosure, the device operation mode identifier is extracted from the operating status characteristics of each power supply node. The device operation mode identifier of each node is represented by a specific symbol or code, which represents the device operation type and parameter settings of that node.

[0133] These device operating mode identifiers are arranged in order of node number to form a mode identifier matrix. The rows and columns of the matrix correspond to each power supply node, and the element values ​​in the matrix are the device operating mode identifiers corresponding to two nodes.

[0134] In step S1262, the mode matching degree between any two power supply nodes is determined based on whether the device operation mode identifiers of any two power supply nodes in the mode identifier matrix are the same.

[0135] In this embodiment of the disclosure, for any two elements corresponding to nodes in the pattern identifier matrix, their device operating mode identifiers are compared to see if they are the same. If the device operating mode identifiers of the two nodes are the same, their matching degree is high; if they are different, their matching degree is low.

[0136] The degree of matching reflects the similarity between two nodes in terms of equipment operating characteristics. The same equipment operating mode means that they may have more similarities in terms of power supply capacity, regulation methods, etc.

[0137] In step S1263, the mode influence coefficient of any two power supply nodes is determined based on the mode matching degree of the any two power supply nodes.

[0138] In this embodiment of the disclosure, to facilitate subsequent calculations and analysis, the matching degree is converted into a specific pattern influence coefficient value. A higher matching degree corresponds to a larger pattern influence coefficient value, while a lower matching degree corresponds to a smaller pattern influence coefficient value.

[0139] For example, the pattern influence coefficient corresponding to a high degree of matching can be set to a larger numerical range, and the pattern influence coefficient corresponding to a low degree of matching can be set to a smaller numerical range. The specific numerical range is determined according to the actual situation, so as to effectively distinguish the degree of influence of different matching degrees.

[0140] In step S1264, the cross-influence coefficient and the mode influence coefficient are multiplied to obtain the comprehensive influence coefficient.

[0141] In this embodiment of the disclosure, the cross-influence coefficient reflects the impact of load changes on voltage fluctuations, and the mode influence coefficient reflects the degree of influence of the similarity of equipment operation modes on the nodes. By multiplying these two coefficients, a comprehensive influence coefficient that takes into account both factors can be obtained.

[0142] The larger the overall influence coefficient, the more significant the coupling effect between the two nodes; conversely, the smaller the overall influence coefficient, the weaker the coupling effect between the two nodes.

[0143] In step S1265, the comprehensive influence coefficients are arranged according to the node connection relationship based on the topology of the power supply network to form a coupling influence matrix between power supply nodes.

[0144] In this embodiment of the disclosure, the topology of the power supply network shows the connection relationship between each node. Based on this connection relationship, the calculated comprehensive influence coefficients are arranged according to the corresponding node pairs to form a coupling influence matrix between nodes.

[0145] The rows and columns of the matrix correspond to the respective power supply nodes, and the element values ​​are the combined influence coefficients between the corresponding two nodes. This matrix provides a clear visual indication of which nodes have strong coupling effects.

[0146] In step S1266, elements in the coupling influence matrix between the power supply nodes whose comprehensive influence coefficient is less than a preset threshold are deleted, while elements whose comprehensive influence coefficient is greater than the preset threshold are retained, thereby generating inter-node coupling influence features.

[0147] In this embodiment, the coupling influence matrix may contain a large number of elements and information. To simplify subsequent processing, feature compression can be performed on it. Feature compression can be performed using methods such as principal component analysis to extract the main influence relationship information in the matrix and ignore the secondary and less influential information.

[0148] After feature compression, the result is the inter-node coupling influence feature, which is a refined representation of the inter-node coupling influence relationship. It can reduce the amount of data while retaining key information, making it easier for subsequent model processing and analysis.

[0149] In one possible implementation, see Figure 3 As shown, in step S13, the pre-trained fluctuation suppression decision model is invoked to perform dynamic suppression strategy generation processing on the fluctuation-related feature set, resulting in a voltage regulation parameter sequence for each power supply node, including: In step S131, the set of fluctuation-related features is input into the feature encoding layer of the fluctuation suppression decision model for high-dimensional feature mapping processing to generate a time-dependent encoded feature vector.

[0150] In this embodiment, the feature encoding layer of the fluctuation suppression decision model is responsible for processing the input fluctuation-related feature set, transforming it into a high-dimensional feature form more suitable for subsequent model processing. The fluctuation-related feature set contains multi-dimensional and multi-time-point information, which the feature encoding layer integrates and maps through a series of nonlinear transformations and feature extraction operations.

[0151] During processing, the feature encoding layer focuses on the temporal relationships between features, associating feature information from different time points to generate temporally dependent encoded feature vectors. These encoded feature vectors can more effectively express key information within the fluctuating feature set. For example, the feature encoding layer can extract feature patterns within a local time range through convolutional operations, and then capture the dependencies between different time segments through recurrent connections, thereby generating high-dimensional encoded feature vectors containing temporal information.

[0152] In step S132, the node attention module of the fluctuation suppression decision model performs node importance weight allocation processing on the encoded feature vector to generate the attention coefficient of each power supply node.

[0153] The node attention module determines the importance of different nodes in the current voltage fluctuation suppression decision based on the feature information of each power supply node in the encoded feature vector, and then assigns them corresponding attention coefficients. Nodes with higher importance are assigned higher attention coefficients. In step S133, the encoded feature vector is weighted and aggregated based on the attention coefficients to obtain an aggregated feature vector.

[0154] In this embodiment of the disclosure, the generated attention coefficients are used to perform weighted aggregation processing on the encoded feature vector. Specifically, the feature components of each power supply node in the encoded feature vector are multiplied by their corresponding attention coefficients, and then these weighted feature components are concatenated to form an aggregated feature vector.

[0155] Through this process, the feature components of nodes with high attention coefficients occupy a more important position in the aggregated feature vector, and their information is strengthened, while the influence of the feature components of nodes with low attention coefficients is relatively weakened. The aggregated feature vector can focus on the information of key nodes.

[0156] In step S134, the aggregated feature vector is input into the time-series prediction layer of the fluctuation suppression decision model to perform multi-step voltage fluctuation prediction processing and generate a voltage fluctuation trend prediction sequence for predicting future periods.

[0157] The main function of the time-series prediction layer is to predict the voltage fluctuations of each power supply node over a future period based on the input aggregated feature vector. Multi-step prediction means predicting not only the voltage fluctuations at a single future time point, but also the voltage fluctuations at multiple consecutive future time points, forming a sequence.

[0158] In this embodiment of the disclosure, the timing prediction layer can adopt a structure such as a recurrent neural network, which can be used to capture the timing dependencies contained in the aggregated feature vector, thereby predicting the future voltage fluctuation trend.

[0159] In step S135, based on the voltage fluctuation trend prediction sequence and the inter-node coupling influence characteristics in the fluctuation correlation feature set, the dynamic change curve of the adjustment amount for voltage regulation of each power supply node is determined.

[0160] Among them, the voltage fluctuation trend prediction sequence indicates the possible changes in the voltage of each node in the future. Combined with the inter-node coupling influence characteristics in the fluctuation correlation feature set, that is, the mutual influence relationship between each node, the change of the adjustment amount required by each power supply node to suppress voltage fluctuation over time can be calculated, that is, the dynamic change curve of voltage adjustment amount.

[0161] In step S136, the dynamic change curve of the adjustment amount is discretized according to a preset time interval to generate a voltage adjustment parameter sequence for each power supply node.

[0162] The voltage regulation dynamic change curve is continuous, while the actual voltage regulation operation can be performed at certain time intervals. Therefore, the curve can be discretized. According to the preset time interval, the corresponding regulation value is extracted from the dynamic change curve. These values ​​are arranged in chronological order to form a voltage regulation parameter sequence for each power supply node.

[0163] In this embodiment of the disclosure, the preset time interval can be determined according to the requirements of the power supply system such as adjustment accuracy and response speed. The smaller the time interval, the more refined the adjustment parameter sequence, but at the same time, it will increase the amount of data and processing complexity.

[0164] In one possible implementation, in step S132, the node attention module of the fluctuation suppression decision model performs node importance weight allocation processing on the encoded feature vector to generate attention coefficients for each power supply node, including: The feature components of each power supply node in the encoded feature vector are analyzed, and the fluctuation sensitivity index of each feature component is extracted. The fluctuation sensitivity index is positively correlated with the rate of change of voltage fluctuation amplitude.

[0165] In this embodiment, the encoded feature vector is first parsed to separate the feature components of each power supply node. Each feature component contains various information about the corresponding node regarding voltage fluctuations. A fluctuation sensitivity index is extracted from these feature components, which reflects the node's sensitivity to voltage fluctuations.

[0166] The fluctuation sensitivity index is positively correlated with the rate of change of voltage fluctuation amplitude. That is, when the rate of change of voltage fluctuation amplitude of a node is large, the value of its fluctuation sensitivity index is also large, indicating that the node is more sensitive to voltage fluctuations and can be given more attention in suppression decisions.

[0167] An initial attention vector is constructed based on the fluctuation sensitivity index, and each element in the initial attention vector corresponds to an initial attention value of the power supply node.

[0168] In this embodiment of the disclosure, after obtaining the fluctuation sensitivity index of each power supply node, these indexes are used as initial attention values ​​and arranged in order of node number to construct an initial attention vector. Each element in the initial attention vector corresponds to a power supply node, and the larger the value of the element, the higher the importance of the node is considered in the initial stage.

[0169] The load deviation coefficient is generated by calculating the degree of deviation between the load change trend in the operating status characteristics of each power supply node and the average load change trend of the subway DC power supply system.

[0170] In this embodiment of the disclosure, the system average load change trend is obtained by calculating the average of the load change trends of all power supply nodes. For each power supply node, the degree of deviation between its load change trend and the system average load change trend is calculated. The greater the deviation, the more significant the difference between the load change of that node and the load change of the overall system.

[0171] The load deviation coefficient is generated based on this degree of deviation. Nodes with a large degree of deviation correspond to a larger load deviation coefficient, while nodes with a small degree of deviation correspond to a smaller load deviation coefficient.

[0172] The load deviation coefficient is introduced into the attention calculation function to dynamically adjust the initial attention vector. Combined with the cross-influence coefficient in the inter-node coupling influence feature, the adjusted attention vector is subjected to neighborhood propagation processing.

[0173] In this embodiment, the attention calculation function is a function used in the node attention module to adjust the attention value. It dynamically adjusts the initial attention vector based on the input load deviation coefficient. For nodes with a large load deviation coefficient, i.e., nodes with large load fluctuations, the attention calculation function increases their attention value in the initial attention vector to highlight the importance of these nodes in voltage fluctuation suppression. The adjusted attention vector can better reflect the differences in the importance of different nodes under the current load conditions.

[0174] The cross-influence coefficient in the coupling influence feature between nodes reflects the degree of mutual influence between nodes, with nodes with strong coupling having a larger cross-influence coefficient. Based on the adjusted attention vector, neighborhood propagation processing is performed in conjunction with the cross-influence coefficient, which allows the attention values ​​of nodes with strong coupling to influence each other and tend to be similar.

[0175] For example, if nodes A and B have a strong coupling relationship, when node A's attention value is high, the attention value of node B will also increase accordingly through neighborhood propagation, and vice versa. This ensures that nodes with significant mutual influence receive coordinated attention during the decision-making process.

[0176] The attention vector after neighborhood propagation is normalized to generate the attention coefficient of each power supply node.

[0177] In this embodiment of the disclosure, the element values ​​in the attention vector after propagation processing may have large differences. In order to facilitate subsequent weighted aggregation processing, they can be normalized to adjust each element value to a preset range, such as between 0 and 1.

[0178] Each element in the normalized attention vector is the final attention coefficient of the corresponding power supply node. These coefficients can accurately reflect the importance of each node in the current voltage fluctuation suppression decision.

[0179] In one possible implementation, step S135, determining the dynamic change curve of the adjustment amount for voltage regulation of each power supply node based on the voltage fluctuation trend prediction sequence and the inter-node coupling influence characteristics in the fluctuation correlation feature set, includes: The fluctuation range of voltage regulation is determined based on the time points in the voltage fluctuation trend prediction sequence where the predicted voltage value is greater than the preset voltage stability threshold.

[0180] In this embodiment, the preset voltage stability threshold is determined according to the safety operation standards of the subway DC power supply system and serves as the basis for determining whether voltage adjustment is necessary. The predicted voltage value at each time point in the voltage fluctuation trend prediction sequence is compared with the voltage stability threshold. If the predicted value exceeds the range of the stability threshold, the interval containing that time point is determined as the fluctuation interval for adjustment. The fluctuation interval for adjustment may be a continuous time period or multiple discontinuous time periods, depending on the comparison result between the voltage fluctuation trend prediction sequence and the stability threshold.

[0181] Calculate the deviation between the predicted voltage value and the preset voltage stability threshold at each time point in each fluctuation range to obtain the basic adjustment amount corresponding to each time point.

[0182] In this embodiment of the disclosure, for each defined fluctuation range to be regulated, the difference between the predicted voltage value and the voltage stability threshold at each time point within that range is calculated. This difference is the base regulation amount. The base regulation amount reflects the amount of regulation performed by the node itself in order to bring the voltage back to the stability threshold range, without considering the influence of other nodes.

[0183] For example, if the predicted voltage value is higher than the upper limit of the stability threshold within a certain fluctuation range, the base adjustment amount is positive, indicating that the voltage needs to be reduced; if the predicted voltage value is lower than the lower limit of the stability threshold, the base adjustment amount is negative, indicating that the voltage needs to be increased.

[0184] Based on the inter-node coupling influence characteristics in the fluctuation correlation feature set, a set of associated nodes that have a significant impact on the voltage fluctuation of each supply node is determined.

[0185] In this embodiment of the disclosure, the information such as the comprehensive influence coefficient in the inter-node coupling influence characteristics reflects the degree of influence between nodes. By analyzing these characteristics, other nodes that have a significant impact on the voltage fluctuation of the current node can be identified. These nodes together constitute the associated node set.

[0186] The determination of significant influence can be achieved by setting an influence threshold. When the combined influence coefficient between two nodes is greater than the threshold, they are considered to have a significant influence on each other.

[0187] Extract the regulation amount from the voltage regulation parameter sequence of the associated node set of each supply node, and calculate the coupling regulation contribution value of the corresponding power supply node.

[0188] In this embodiment, the voltage regulation parameter sequence of the associated node set may have been generated in previous processing or calculated synchronously in this processing. The regulation amounts of these associated nodes within the corresponding fluctuation range are extracted, and then, based on information such as the cross-influence coefficient in the coupling influence characteristics between nodes, the coupling regulation contribution value of these regulation amounts to the current node is calculated.

[0189] The calculation of the coupling regulation contribution value can be the sum of the products of the regulation amount of the associated node and the corresponding cross-influence coefficient. It reflects the degree of influence of the regulation operation of the associated node on the voltage fluctuation of the current node.

[0190] The base adjustment amount at the specified time point is superimposed with the corresponding coupling adjustment contribution value to obtain the total adjustment amount for the corresponding power supply node.

[0191] In this embodiment of the disclosure, the basic regulation amount is the regulation amount required by the current node itself, and the coupled regulation contribution value is the influence of the regulation of the associated node on the current node. The two are superimposed to obtain the total regulation requirement required by the current node to achieve voltage stability.

[0192] The total regulation takes into account its own voltage deviation and the influence of associated nodes, and can more accurately reflect the actual regulation required by the current node.

[0193] Based on the total adjustment amount, the dynamic change curve of the adjustment amount used for voltage regulation at each power supply node is determined by a curve fitting algorithm.

[0194] In this embodiment of the disclosure, the curve fitting algorithm can fit a smooth curve based on the value of the total regulation demand at each time point within the fluctuation range. This curve is the dynamic change curve of the voltage regulation of the power supply node. This dynamic change curve of the voltage regulation can clearly show the continuous change of the regulation over time within the future fluctuation range.

[0195] In one possible implementation, see Figure 4 As shown, in step S14, the step of generating a set of node-coordinated fluctuation suppression execution instructions based on the voltage regulation parameter sequence and the operating status characteristics of each power supply node includes: In step S141, the voltage regulation parameter sequence of each power supply node is parsed, and the voltage regulation amount and regulation duration parameters at each time point are extracted.

[0196] In this embodiment of the disclosure, the voltage regulation parameter sequence of each power supply node is analyzed to extract the voltage regulation amount that needs to be performed at each time point and the duration that the regulation amount needs to last, i.e., the regulation duration parameter.

[0197] These parameters form the basis for generating execution instructions, specifying when, how much, and for how long each node should be adjusted.

[0198] In step S142, the adjustment authority level of each power supply node is determined according to the equipment operation mode identifier in the operation status characteristics of each power supply node. The adjustment authority level is used to limit the maximum range of voltage adjustment.

[0199] Different equipment operating modes correspond to different regulation capabilities and safety limitations. Based on the equipment operating mode identifier of each power supply node, its corresponding regulation authority level can be determined. The regulation authority level specifies the maximum allowable regulation range for that node during voltage regulation, preventing damage to equipment or the system due to over-regulation.

[0200] In this embodiment of the disclosure, a node in standby mode may have a lower level of adjustment authority and a smaller maximum adjustment range; while a node in normal operating mode with better device performance may have a higher level of adjustment authority and allow for a wider range of adjustments.

[0201] In step S143, the voltage adjustment amount is checked for reasonableness based on the adjustment authority level, and the adjustment amount that exceeds the authority range is truncated and corrected.

[0202] In this embodiment of the disclosure, the reasonableness of the voltage adjustment amount extracted at each time point is verified according to the determined adjustment permission level. If the adjustment amount is within the maximum range allowed by the permission level, the adjustment amount is considered reasonable; if the adjustment amount exceeds the permission range, it can be truncated and corrected to the maximum or minimum value within the permission range.

[0203] This verification and correction process ensures that the adjustment operations of each node are carried out within a safe and feasible range.

[0204] In step S144, the coordination between the voltage regulation amounts of different power supply nodes at the same time point is analyzed, and the consistency coefficient of the regulation direction is calculated. The consistency coefficient is used to determine the degree of cooperation between the regulation operations of the power supply nodes.

[0205] In this process, the voltage regulation operations of different power supply nodes can coordinate with each other to achieve the best fluctuation suppression effect at the same time point. The consistency coefficient of the regulation direction is calculated by analyzing the regulation direction of each node at the same time point, i.e., whether the voltage is increased or decreased.

[0206] In this embodiment of the disclosure, the consistency coefficient can be calculated by statistically analyzing the proportion of nodes with the same adjustment direction to the total number of nodes. The higher the proportion, the larger the consistency coefficient, indicating a higher degree of cooperation in the adjustment operations between nodes; conversely, the smaller the consistency coefficient, the lower the degree of cooperation.

[0207] In step S145, when the consistency coefficient is lower than a preset threshold, a collaborative optimization algorithm is invoked to generate a node collaborative adjustment scheme for adjusting the voltage regulation amount of the target power supply node. The node collaborative adjustment scheme makes the consistency coefficient between the target power supply nodes greater than the preset threshold. The target power supply node includes at least one of the power supply nodes whose consistency coefficient is lower than the preset threshold.

[0208] In this embodiment, when the consistency coefficient is lower than a preset threshold, it indicates poor coordination of the adjustment operations between nodes, which may affect the fluctuation suppression effect or even cause mutual interference. In this case, a collaborative optimization algorithm can be invoked to synchronously adjust the voltage regulation of the relevant nodes to improve coordination.

[0209] In step S146, the corrected and adjusted voltage regulation amount and regulation duration parameters are associated and encapsulated with the corresponding power supply node identifier to generate a set of node-coordinated fluctuation suppression execution instructions.

[0210] In this embodiment, the node coordinated adjustment scheme determines the adjustment amount and execution time of each node. Based on this, the adjustment duration parameter of each power supply node is updated according to the magnitude of the adjustment amount and the system's response characteristics. The adjustment duration parameter can be matched with the adjustment amount and execution time to ensure that the adjustment operation can play a full role and stabilize the voltage within the target range.

[0211] The updated voltage regulation parameter sequence and regulation duration parameters are encapsulated according to a preset format. This encapsulated information constitutes the core content of the node-coordinated fluctuation suppression execution instruction set. During the encapsulation process, the parameters are structured to facilitate parsing and execution by the power supply control module.

[0212] In one possible implementation, in step S145, when the consistency coefficient is lower than a preset threshold, the collaborative optimization algorithm is invoked to generate a node collaborative adjustment scheme for synchronously adjusting the voltage regulation of the target power supply node, including: Power supply nodes with a consistency coefficient lower than a preset threshold are added to the set of nodes to be optimized as power supply nodes to be adjusted.

[0213] In this embodiment, based on the calculation results of the consistency coefficient, nodes whose adjustment directions are inconsistent at the same time point, resulting in a consistency coefficient below a preset threshold, are selected to form a set of nodes to be optimized. The adjustment operations of these nodes need to be adjusted to improve coordination.

[0214] An optimization function is constructed with the goal of the complementary regulation between power supply nodes. The optimization function is positively correlated with the synergistic effect of the voltage regulation of the power supply nodes.

[0215] In this embodiment, the goal of the optimization function is to improve the complementarity between the adjustment amounts of each node in the set of nodes to be optimized. That is, by adjusting the adjustment amounts, the adjustment operations of each node can cooperate with each other to form a better synergistic effect. The value of the optimization function is positively correlated with the synergistic effect of the node adjustment amounts; the better the synergistic effect, the larger the value of the optimization function.

[0216] For example, the optimization function can include a coordination term between the adjustment amounts of each node. When the adjustment directions of the nodes are consistent or can complement each other, the value of the coordination term is larger, thereby increasing the value of the optimization function.

[0217] Set constraints for the optimization variables, including upper and lower limits for voltage regulation, regulation rate limits, and node load capacity limits.

[0218] In this embodiment of the disclosure, to ensure the feasibility and safety of the optimized adjustment amount, constraints can be set for the optimization variables. The upper and lower limits of the voltage adjustment amount are determined by the adjustment authority level of each node; the adjustment rate limit refers to the maximum change range of the adjustment amount per unit time to prevent excessively rapid adjustment from impacting the system; the node load capacity limit considers the load bearing capacity of the node to ensure that the adjusted voltage does not cause the load to exceed its capacity range.

[0219] These constraints together constitute the feasible region of the optimization problem, and the optimization process must be carried out within this feasible region.

[0220] The particle swarm optimization algorithm is invoked to solve the optimization function, thereby obtaining the optimal combination of adjustment values ​​for the set of nodes to be optimized.

[0221] Among them, the particle swarm optimization algorithm is an optimization algorithm based on swarm intelligence. It continuously updates the position and velocity of particles by simulating the motion of particles in the search space to find the optimal solution of the optimization function.

[0222] In this embodiment, the adjustment amount of the set of nodes to be optimized is used as the position variable of the particles. Under the constraints, the optimization function is solved, and the optimal combination of adjustment amounts that maximizes the value of the optimization function is finally obtained. This optimal combination of adjustment amounts enables the adjustment operations of the set of nodes to be optimized to have the best synergistic effect.

[0223] The optimal adjustment combination is compared with the voltage adjustment amount of each power supply node before optimization to obtain the adjustment range of each power supply node to be adjusted.

[0224] In this embodiment, the optimal combination of regulation values ​​obtained through particle swarm optimization is compared with the original voltage regulation values ​​to calculate the change range of the regulation value at each node, i.e., the adjustment range. The adjustment range reflects the degree to which the regulation values ​​at each node can be changed in order to achieve the optimal synergistic effect.

[0225] Based on the adjustment range of the power supply node to be adjusted, the voltage regulation amount of the target power supply node is synchronously corrected to generate an adjusted voltage regulation parameter sequence.

[0226] In this embodiment, the voltage regulation of relevant nodes is synchronously corrected based on the calculated adjustment range. This means that the original regulation amount is changed according to the adjustment range to form a new voltage regulation parameter sequence. Synchronous correction ensures that the changes in regulation amounts of each node are coordinated, avoiding disruption of overall synergy due to adjusting a single node. For example, if the adjustment range of node A is an increase of a certain amount, and the adjustment range of node B is a decrease of the corresponding amount, synchronous correction ensures that the changes in regulation amounts of these two nodes occur simultaneously to achieve the expected synergistic effect.

[0227] The synergy of the synchronously corrected voltage regulation parameter sequence is verified, and the consistency coefficient of the regulation direction is recalculated until the consistency coefficient reaches or exceeds the preset threshold.

[0228] In this embodiment, after synchronous correction, the voltage regulation parameter sequence needs to be verified for synergy again. The consistency coefficient of the adjustment direction of each node at the same time point is recalculated, and it is checked whether it reaches a preset threshold. If the consistency coefficient is still below the threshold, the synchronous correction process can be repeated: the optimization function is reconstructed, constraints are set, the particle swarm optimization algorithm is called to solve the problem, and synchronous correction is performed until the consistency coefficient meets the requirements. This process of multiple verifications and adjustments ensures that the final voltage regulation parameter sequence has good synergy.

[0229] The voltage regulation parameter sequence that has passed the coordination verification is associated with the corresponding timestamp to generate a node coordination regulation scheme for synchronously adjusting the voltage regulation amount of the target power supply node.

[0230] In this embodiment, after the voltage regulation parameter sequence passes the coordination verification, each regulation amount is associated with a corresponding timestamp. The timestamp clarifies the specific time point at which each regulation amount is executed, giving the regulation scheme a clear arrangement in the time dimension. Through this association, a final node coordinated regulation scheme is formed, which specifies in detail the regulation amount of each power supply node at different time points.

[0231] This invention provides a voltage fluctuation suppression device for a subway DC power supply system. The device includes: a memory storing a computer program; and a processor for executing the computer program stored in the memory to implement the voltage fluctuation suppression method for the subway DC power supply system described in any of the foregoing embodiments.

[0232] Figure 5 The voltage fluctuation suppression device 100 for the subway DC power supply system shown includes a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, for example, via a bus 1002. Optionally, the voltage fluctuation suppression device 100 for the subway DC power supply system may further include a communication component 1004, which can be used for data interaction between the device 100 and other devices, such as data transmission and / or data reception. It should be noted that in actual scheduling, the communication component 1004 is not limited to one, and the structure of the voltage fluctuation suppression device 100 for the subway DC power supply system does not constitute a limitation on the embodiments of this application.

[0233] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0234] Bus 1002 may include a pathway for transmitting information between the aforementioned components. Bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 1002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0235] The memory 1003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing program code and capable of being read by a computer, without limitation herein.

[0236] The memory 1003 is used to store program code for executing the embodiments of this disclosure, and its execution is controlled by the processor 1001. The processor 1001 is used to execute the program code stored in the memory 1003 to implement the steps shown in the aforementioned embodiments of the voltage fluctuation suppression method for the subway DC power supply system.

[0237] The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present disclosure, various changes, modifications, substitutions and variations can be made to these embodiments, and all such changes, modifications, substitutions and variations fall within the protection scope of the present disclosure.

[0238] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction, and such combinations should also be considered as part of this disclosure. To avoid unnecessary repetition, this disclosure will not further describe the various possible combinations. The technical scope of this application is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for suppressing voltage fluctuations in a subway DC power supply system, characterized in that, The method includes: The voltage amplitude change information of the subway DC power supply system at different sampling times is obtained, and based on the voltage amplitude change information, the real-time voltage fluctuation sequence of the subway DC power supply system and the operating status characteristics of each power supply node in the subway DC power supply system are determined. The real-time voltage fluctuation sequence and the operating state characteristics are subjected to spatiotemporal correlation mapping to generate a fluctuation correlation feature set with node correlation. The fluctuation correlation feature set includes the spatiotemporal propagation path characteristics of voltage fluctuations and the coupling influence characteristics between nodes. The pre-trained fluctuation suppression decision model is invoked to perform dynamic suppression strategy generation processing on the fluctuation-related feature set, thereby obtaining a voltage regulation parameter sequence for each power supply node; Based on the voltage regulation parameter sequence and the operating status characteristics of each power supply node, a set of node-coordinated fluctuation suppression execution instructions is generated. According to the set of fluctuation suppression execution instructions, the voltage regulation operation of each of the power supply nodes is triggered to suppress voltage fluctuations.

2. The voltage fluctuation suppression method for a subway DC power supply system as described in claim 1, characterized in that, The operational status characteristics include the load change trends and equipment operation mode identifiers of each power supply node. The spatiotemporal correlation mapping processing of the real-time voltage fluctuation sequence and the operational status characteristics generates a set of fluctuation correlation characteristics with node correlation, including: The real-time voltage fluctuation sequence is divided into partitions according to the power supply nodes to obtain local voltage fluctuation sub-sequences corresponding to each power supply node. Based on the local voltage fluctuation subsequence corresponding to each power supply node, a similarity index is calculated between the local voltage fluctuation subsequences of adjacent power supply nodes. The similarity index is used to characterize the degree of consistency of the fluctuation pattern. Based on the similarity index and the topological connection relationship of the power supply network in the subway DC power supply system, a wave propagation adjacency matrix is ​​constructed between the power supply nodes; Based on the wave propagation adjacency matrix and the real-time voltage wave sequence, the diffusion speed and attenuation characteristics of voltage waves from the starting node to the surrounding nodes are analyzed, and the spatiotemporal propagation path characteristics of voltage waves are generated according to the diffusion speed and the attenuation characteristics. Calculate the cross-influence coefficient of the load change trends of different power supply nodes, the cross-influence coefficient being used to describe the degree of influence of the load change of any power supply node on the voltage fluctuation of another power supply node; By combining the cross-influence coefficient corresponding to the power supply node and the matching degree of the equipment operation mode identifier, the coupling influence characteristics between nodes are generated; The spatiotemporal propagation path features and the inter-node coupling influence features are spliced ​​together along the time dimension to form a set of fluctuation correlation features with node correlation.

3. The voltage fluctuation suppression method for a subway DC power supply system as described in claim 2, characterized in that, The step of calculating the similarity index between local voltage fluctuation subsequences of adjacent power supply nodes based on the local voltage fluctuation subsequences corresponding to each power supply node includes: The local voltage fluctuation subsequences of adjacent power supply nodes are length normalized, and the distribution information of the fluctuation extreme points of each length normalized local voltage fluctuation subsequence is extracted. Based on the location of the maximum point in the fluctuation extreme point distribution information and the first voltage amplitude corresponding to the maximum point, the location of the minimum point in the fluctuation extreme point distribution information and the second voltage amplitude corresponding to the minimum point, calculate the voltage amplitude difference of the extreme points at corresponding positions in the two local voltage fluctuation subsequences; Based on the cumulative sum of the voltage amplitude differences, the amplitude similarity component is calculated; The fluctuation period characteristics of the two local voltage fluctuation subsequences are analyzed, the target fluctuation frequency components are extracted by Fourier transform, and the matching degree of the target fluctuation frequency components is calculated as the period similarity component. The amplitude similarity component and the period similarity component are weighted and summed according to a preset weight ratio to obtain the similarity index between the local voltage fluctuation subsequences of adjacent power supply nodes.

4. The voltage fluctuation suppression method for a subway DC power supply system as described in claim 2, characterized in that, The method described above, based on the wave propagation adjacency matrix and the real-time voltage wave sequence, analyzes the diffusion speed and attenuation characteristics of voltage waves from the starting node to surrounding nodes, and generates spatiotemporal propagation path characteristics of voltage waves according to the diffusion speed and attenuation characteristics, including: The starting node and corresponding starting time of the voltage fluctuation are determined from the real-time voltage fluctuation sequence using a fluctuation initiation identification algorithm. Centered on the starting node, a multi-order neighbor node search model is constructed based on the wave propagation adjacency matrix to expand the search for adjacent power supply nodes layer by layer; Record the first time when the voltage fluctuation that meets the preset requirements occurs at each power supply node, and calculate the time difference between each fluctuation time and the start time as the fluctuation arrival time of the corresponding power supply node; The fluctuation propagation speed is calculated based on the fluctuation arrival time of each power supply node and the physical distance between the power supply nodes, wherein the physical distance is determined based on the node coordinates in the power supply network topology diagram; Extract the peak voltage amplitude of each power supply node during the wave propagation process, and use the ratio of the peak voltage amplitude to the peak voltage amplitude of the starting node as the attenuation coefficient of the corresponding power supply node. The arrival time, diffusion rate, and attenuation coefficient of the fluctuations at each power supply node are arranged in the spatial distribution order of the nodes to generate spatiotemporal propagation path characteristics that characterize the propagation timing and intensity changes of the voltage fluctuations.

5. The voltage fluctuation suppression method for a subway DC power supply system as described in claim 2, characterized in that, The process of generating inter-node coupling influence characteristics by combining the cross-influence coefficient corresponding to the power supply node and the matching degree of the equipment operation mode identifier includes: Extract the device operation mode identifiers of each power supply node and construct a mode identifier matrix; The pattern matching degree between any two power supply nodes is determined based on whether the device operation mode identifiers of any two power supply nodes in the pattern identifier matrix are the same. Based on the mode matching degree of any two power supply nodes, determine the mode influence coefficient of any two power supply nodes; The cross-influence coefficient and the mode influence coefficient are multiplied to obtain the comprehensive influence coefficient; Based on the topology of the power supply network, the comprehensive influence coefficients are arranged according to the node connection relationship to form a coupling influence matrix between power supply nodes; Elements in the coupling influence matrix between the power supply nodes whose comprehensive influence coefficient is less than a preset threshold are deleted, while elements whose comprehensive influence coefficient is greater than the preset threshold are retained, thus generating inter-node coupling influence features.

6. The voltage fluctuation suppression method for a subway DC power supply system as described in claim 1, characterized in that, The process of calling the pre-trained fluctuation suppression decision model to perform dynamic suppression strategy generation on the fluctuation-related feature set yields a voltage regulation parameter sequence for each power supply node, including: The set of fluctuation-related features is input into the feature encoding layer of the fluctuation suppression decision model, and high-dimensional feature mapping is performed to generate a time-dependent encoded feature vector. The node attention module of the fluctuation suppression decision model performs node importance weight allocation on the encoded feature vector to generate the attention coefficient of each power supply node. The encoded feature vector is weighted and aggregated based on the attention coefficient to obtain an aggregated feature vector. The aggregated feature vector is input into the time-series prediction layer of the fluctuation suppression decision model to perform multi-step voltage fluctuation prediction processing and generate a voltage fluctuation trend prediction sequence for predicting future periods. Based on the voltage fluctuation trend prediction sequence and the inter-node coupling influence characteristics in the fluctuation correlation feature set, determine the dynamic change curve of the adjustment amount used for voltage regulation of each power supply node; The dynamic change curve of the adjustment amount is discretized according to a preset time interval to generate a voltage adjustment parameter sequence for each power supply node.

7. The voltage fluctuation suppression method for a subway DC power supply system as described in claim 6, characterized in that, The node attention module of the fluctuation suppression decision model performs node importance weight allocation on the encoded feature vector to generate attention coefficients for each power supply node, including: The feature components of each power supply node in the encoded feature vector are analyzed, and the fluctuation sensitivity index of each feature component is extracted. The fluctuation sensitivity index is positively correlated with the rate of change of voltage fluctuation amplitude. An initial attention vector is constructed based on the fluctuation sensitivity index, and each element in the initial attention vector corresponds to an initial attention value of the power supply node. Calculate the degree of deviation between the load change trend in the operating status characteristics of each power supply node and the average load change trend of the subway DC power supply system, and generate a load deviation coefficient; The load deviation coefficient is introduced into the attention calculation function to dynamically adjust the initial attention vector. Combined with the cross-influence coefficient in the inter-node coupling influence feature, the adjusted attention vector is subjected to neighborhood propagation processing. The attention vector after neighborhood propagation is normalized to generate the attention coefficient of each power supply node.

8. The voltage fluctuation suppression method for a subway DC power supply system as described in claim 6, characterized in that, The step of determining the dynamic change curve of the adjustment amount for voltage regulation at each power supply node based on the voltage fluctuation trend prediction sequence and the inter-node coupling influence characteristics in the fluctuation correlation feature set includes: The fluctuation range of voltage regulation is determined based on the time points in the voltage fluctuation trend prediction sequence where the predicted voltage value is greater than the preset voltage stability threshold. Calculate the deviation between the predicted voltage value and the preset voltage stability threshold at each time point in each fluctuation interval to obtain the basic adjustment amount corresponding to each time point; Based on the inter-node coupling influence characteristics in the fluctuation correlation feature set, determine the set of associated nodes that have a significant impact on the voltage fluctuation of each supply node; Extract the regulation amount from the voltage regulation parameter sequence of the associated node set of each supply node, and calculate the coupling regulation contribution value of the corresponding power supply node; The basic adjustment amount at the time point is superimposed with the corresponding coupling adjustment contribution value to obtain the total adjustment amount for the corresponding power supply node; Based on the total adjustment amount, the dynamic change curve of the adjustment amount used for voltage regulation at each power supply node is determined by a curve fitting algorithm.

9. The method for suppressing voltage fluctuations in a subway DC power supply system as described in any one of claims 1-8, characterized in that, The step of generating a set of node-coordinated fluctuation suppression execution instructions based on the voltage regulation parameter sequence and the operating status characteristics of each power supply node includes: Analyze the voltage regulation parameter sequence of each power supply node and extract the voltage regulation amount and regulation duration parameters at each time point; Based on the equipment operation mode identifier in the operation status characteristics of each power supply node, the adjustment authority level of each power supply node is determined, and the adjustment authority level is used to limit the maximum range of voltage adjustment. The voltage adjustment amount is checked for rationality based on the adjustment authority level, and the adjustment amount that exceeds the authority range is truncated and corrected. The coordination between voltage regulation amounts of different power supply nodes at the same time point is analyzed, and the consistency coefficient of regulation direction is calculated. The consistency coefficient is used to determine the degree of cooperation between the regulation operations of the power supply nodes. When the consistency coefficient is lower than a preset threshold, a collaborative optimization algorithm is invoked to generate a node collaborative adjustment scheme for adjusting the voltage regulation of the target power supply node. The node collaborative adjustment scheme makes the consistency coefficient between the target power supply nodes greater than the preset threshold. The target power supply node includes at least one of the power supply nodes with a consistency coefficient lower than the preset threshold. The corrected and adjusted voltage regulation amount and regulation duration parameters are associated and encapsulated with the corresponding power supply node identifier to generate a set of node-coordinated fluctuation suppression execution instructions.

10. A voltage fluctuation suppression device for a subway DC power supply system, characterized in that, The device includes: A memory on which computer programs are stored; A processor is configured to execute the computer program stored in the memory to implement the voltage fluctuation suppression method for the subway DC power supply system according to any one of claims 1-9.

Images