A short-circuit simulation test method and system for a subway DC power supply system

Abstract

The present invention belongs to the field of power system protection and control technology, and specifically relates to a short-circuit simulation test method and system for a subway DC power supply system, aiming to solve the technical problems of insufficient flexibility of the prior art, difficulty in adapting to complex actual working conditions, and lack of adaptability. The present invention includes generating short-circuit working conditions by presetting and adjusting a variety of short-circuit fault parameters; determining transient fluctuation separation results based on short-circuit working conditions; using graph neural networks to build a system topology model, combining the optimized double Q learning strategy to generate a long-term protection plan, and using physical information-based neural networks and Monte Carlo simulations to generate initial risk assessment results; combining Bayesian networks and fuzzy logic to build a hybrid risk assessment model to obtain target risk assessment results, and based on the target risk assessment results, using natural language generation technology and planning algorithms to generate emergency response plans; the present invention improves the accuracy and reliability of short-circuit simulation tests of subway DC power supply systems.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of power system protection and control, and in particular to a short-circuit simulation test method and system for a subway direct current power supply system. Background Art

[0002] The DC power supply system of the subway is a core component of urban rail transit. Its safety and reliability are directly related to the safety of passengers and the normal operation of the system. In actual operation, short circuit failure is one of the common electrical problems, which may lead to serious safety accidents and long-term downtime. Therefore, conducting short-circuit simulation tests on the DC power supply system of the subway, identifying potential risks in advance and formulating effective protective measures have become important means to ensure the safety of the system.

[0003] At present, the short-circuit simulation test of the subway DC power supply system mainly adopts traditional simulation software and rule-based protection strategy. The simulation software uses commercial simulation tools to simulate short-circuit conditions, but these tools often rely on preset models and parameters, lack flexibility, and are difficult to adapt to complex actual conditions; the rule-based protection strategy detects and handles short-circuit faults by setting fixed thresholds and logical rules, which lacks adaptability. Summary of the invention

[0004] The embodiment of the present invention provides a short-circuit simulation test method and system for a subway DC power supply system, which are used to solve the problems of insufficient flexibility, difficulty in adapting to complex actual working conditions, and lack of adaptability in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a short-circuit simulation test method for a subway DC power supply system, comprising:

[0006] Preset a plurality of short-circuit fault parameters, adjust the short-circuit fault parameters to simulate short-circuit faults at different locations and with different intensities, and generate a short-circuit working condition based on the short-circuit faults;

[0007] Based on the short-circuit working condition, a convolutional neural network and a variational mode decomposition technique are applied to decompose the short-circuit signal into a plurality of intrinsic mode functions, a long short-term memory network is used to predict a change trend of the intrinsic mode function, and a transient fluctuation separation result is determined based on the intrinsic mode function and the change trend;

[0008] Based on the transient fluctuation separation results, a system topology model is constructed using a graph neural network. Based on the system topology model, changes in the power transmission path under different short-circuit conditions are simulated. Based on the changes, the optimized dual-Q learning strategy is used to evaluate the application effects of different short-circuit countermeasures, and effect evaluation results corresponding to different short-circuit countermeasures are obtained. Based on the effect evaluation results, a long-term protection plan is generated;

[0009] Based on the long-term protection plan, a finite element analysis is performed using a neural network based on physical information to obtain a thermal effect distribution simulation result, and the thermal effect distribution simulation result is evaluated using a Monte Carlo simulation to generate an initial risk assessment result;

[0010] Based on the initial risk assessment results, a hybrid risk assessment model is constructed in combination with Bayesian networks and fuzzy logic. Based on the hybrid risk assessment model, a target risk assessment result is obtained. Based on the target risk assessment result, an emergency response plan is generated using natural language generation technology and planning algorithms. The emergency response plan is used to resolve short circuit faults.

[0011] Optionally, based on the transient fluctuation separation result, a system topology model is constructed using a graph neural network, and based on the system topology model, changes in the power transmission path under different short-circuit conditions are simulated. Based on the changes, an optimized dual-Q learning strategy is used to evaluate the application effects of different short-circuit countermeasures, and effect evaluation results corresponding to different short-circuit countermeasures are obtained. Based on the effect evaluation results, a long-term protection plan is generated, including:

[0012] Based on the transient fluctuation separation results, a system topology model is constructed using graph neural network and attention mechanism. Based on the system topology model, multi-scale simulation technology is used to simulate the changes in the power transmission path under different short-circuit conditions.

[0013] A multi-condition adversarial generative network is introduced to conduct adversarial training on the double Q learning strategy in reinforcement learning to obtain an optimized double Q learning strategy. Based on the change, the optimized double Q learning strategy is used to select short-circuit countermeasures, and the application effect of the short-circuit countermeasures is evaluated to obtain an effect evaluation result. In the evaluation process, a Bayesian optimization algorithm and a causal reasoning mechanism are introduced to optimize the evaluation process.

[0014] Based on the effect evaluation results, explanatory artificial intelligence technology is introduced to generate a long-term protection plan, which includes an equipment configuration adjustment plan, an operating procedure optimization plan, and a maintenance plan.

[0015] Optionally, a multi-condition adversarial generative network is introduced to perform adversarial training on the double Q learning strategy in reinforcement learning to obtain an optimized double Q learning strategy, and based on the change situation, a short-circuit countermeasure is selected using the optimized double Q learning strategy, and the application effect of the short-circuit countermeasure is evaluated to obtain an effect evaluation result, and a Bayesian optimization algorithm and a causal reasoning mechanism are introduced in the evaluation process to optimize the evaluation process, including:

[0016] A multi-condition adversarial generative network is introduced to generate specific short-circuit conditions through a generator, and the specific short-circuit conditions are evaluated through a discriminator to obtain a condition evaluation result. The specific short-circuit conditions include chain reactions, multiple fault combinations, and extreme environmental conditions.

[0017] Based on the working condition evaluation result, adversarial training is performed on the double Q learning strategy in reinforcement learning to obtain an optimized double Q learning strategy;

[0018] Based on the changes, short-circuit countermeasures are selected using the optimized dual-Q learning strategy, and the application effects of the short-circuit countermeasures are evaluated to obtain effect evaluation results. A Bayesian optimization algorithm and a causal reasoning mechanism are introduced into the evaluation process, and the optimal hyperparameter combination is searched through the Bayesian optimization algorithm. The causal relationship between the short-circuit countermeasures and the performance of the subway DC power supply system is analyzed through the causal reasoning mechanism. Based on the causal relationship and the optimal hyperparameter combination, the evaluation process is optimized.

[0019] Optionally, based on the short-circuit condition, a convolutional neural network and a variational mode decomposition technique are applied to decompose the short-circuit signal into a plurality of intrinsic mode functions, a long short-term memory network is used to predict a change trend of the intrinsic mode function, and based on the intrinsic mode function and the change trend, a transient fluctuation separation result is determined, including:

[0020] The current and voltage data streams of the subway DC power supply system are collected in real time, and a convolutional neural network is used to extract key features of the current and voltage data streams to obtain short-circuit signal features, wherein the key features include mutations and peak values ​​when a short circuit occurs;

[0021] Based on the short-circuit signal characteristics, the short-circuit signal is decomposed into a plurality of intrinsic mode functions using variational mode decomposition technology to extract transient fluctuation characteristics at different time scales, wherein the intrinsic mode functions characterize specific frequency components of the short-circuit signal;

[0022] The time series of the intrinsic mode function is modeled using a long short-term memory network to predict the change trend of the intrinsic mode function, and the transient fluctuation separation result is determined by combining the transient fluctuation characteristics and the change trend.

[0023] Optionally, based on the long-term protection plan, a finite element analysis is performed using a neural network based on physical information to obtain a thermal effect distribution simulation result, and the thermal effect distribution simulation result is evaluated using a Monte Carlo simulation to generate an initial risk assessment result, including:

[0024] Based on the long-term protection plan, combined with the physical laws and boundary conditions of the subway DC power supply system, a finite element analysis is performed on the subway DC power supply system using a neural network based on physical information to generate a thermal effect distribution simulation result. In the finite element analysis process, a multi-scale modeling technology is introduced to extract the thermal effect distribution characteristics at different time scales and different spatial scales. The boundary conditions include a power input point, a load connection point, and a grounding position.

[0025] The probability density function in the thermal effect distribution simulation result is evaluated using a Monte Carlo simulation method to obtain the statistical characteristics of the thermal effect distribution, and an initial risk assessment result is generated based on the statistical characteristics.

[0026] Optionally, based on the initial risk assessment result, a hybrid risk assessment model is constructed in combination with Bayesian network and fuzzy logic, and a target risk assessment result is obtained based on the hybrid risk assessment model. Based on the target risk assessment result, an emergency response plan is generated using natural language generation technology and planning algorithms, including:

[0027] Based on the initial risk assessment results, using a Bayesian network to define probabilistic dependencies between nodes, and based on the nodes and the probabilistic dependencies, processing uncertainties and probabilistic dependencies to obtain posterior probability distributions corresponding to different nodes, wherein the nodes include physical variables and equipment states, and the uncertainties include random uncertainty, epistemic uncertainty, and fuzzy uncertainty;

[0028] Introducing fuzzy logic to process the fuzzy uncertainty so as to convert qualitative information into quantitative indicators, combining the quantitative indicators with the posterior probability distribution to construct a hybrid risk assessment model, and obtaining a target risk assessment result based on the hybrid risk assessment model, wherein the qualitative information includes expert experience and qualitative descriptions in the rule base;

[0029] Based on the target risk assessment results, natural language generation technology and context awareness mechanism are used to generate target text descriptions corresponding to different risk levels, a path planning algorithm and a multi-agent system are used to determine the shortest rescue path, and an emergency response plan is generated based on the target text description and the shortest rescue path.

[0030] Optionally, based on the target risk assessment result, using natural language generation technology and context awareness mechanism, generating target text descriptions corresponding to different risk levels, using a path planning algorithm and a multi-agent system to determine the shortest rescue path, and generating an emergency response plan based on the target text description and the shortest rescue path, including:

[0031] Based on the target risk assessment results, identifying potential risk points of different risk levels, and based on the potential risk points, generating a preliminary text description using natural language generation technology;

[0032] optimizing the preliminary text description to obtain a target text description through a context-aware mechanism and domain-specific data, wherein the domain-specific data includes historical fault records;

[0033] A path planning algorithm is used to determine a preliminary rescue path. In the process of determining the preliminary rescue path, actual environmental factors of the subway DC power supply system are comprehensively considered, including equipment layout, channel width, and personnel distribution;

[0034] A multi-agent system and a reinforcement learning algorithm are introduced to dynamically optimize the preliminary rescue path by simulating the behavior of multiple rescue teams to obtain the shortest rescue path. An emergency response plan is generated by combining the target text description and the shortest rescue path.

[0035] In a second aspect, an embodiment of the present invention provides a short-circuit simulation test system for a subway DC power supply system, comprising:

[0036] An adjustment module, used for presetting a plurality of short-circuit fault parameters, adjusting the short-circuit fault parameters to simulate short-circuit faults at different locations and with different intensities, and generating a short-circuit working condition based on the short-circuit faults;

[0037] An analysis module is used to decompose the short-circuit signal into multiple intrinsic modal functions based on the short-circuit working condition by applying convolutional neural network and variational mode decomposition technology, predict the change trend of the intrinsic modal function by using long short-term memory network, and determine the transient fluctuation separation result based on the intrinsic modal function and the change trend;

[0038] A construction module is used to construct a system topology model using a graph neural network based on the transient fluctuation separation result, simulate the change of the power transmission path under different short-circuit conditions based on the system topology model, and use the optimized double Q learning strategy to evaluate the application effects of different short-circuit countermeasures based on the change, so as to obtain the effect evaluation results corresponding to different short-circuit countermeasures, and generate a long-term protection plan based on the effect evaluation results;

[0039] An evaluation module is used to perform finite element analysis based on the long-term protection plan using a neural network based on physical information to obtain a thermal effect distribution simulation result, evaluate the thermal effect distribution simulation result using Monte Carlo simulation, and generate an initial risk assessment result;

[0040] A generation module is used to construct a hybrid risk assessment model based on the initial risk assessment result in combination with Bayesian network and fuzzy logic, obtain a target risk assessment result based on the hybrid risk assessment model, and generate an emergency response plan based on the target risk assessment result using natural language generation technology and planning algorithm. The emergency response plan is used to resolve short circuit faults.

[0041] In a third aspect, an embodiment of the present invention provides a computing device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a short-circuit simulation test method for a subway DC power supply system as described in any one of the first aspects.

[0042] In a fourth aspect, an embodiment of the present invention provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a short-circuit simulation test method for a subway DC power supply system as described in any one of the first aspects.

[0043] In an embodiment of the present invention, a plurality of short-circuit fault parameters are preset, and the short-circuit fault parameters are adjusted to simulate short-circuit faults at different locations and with different intensities, and a short-circuit operating condition is generated based on the short-circuit fault; based on the short-circuit operating condition, a convolutional neural network and a variational mode decomposition technique are applied to decompose the short-circuit signal into a plurality of intrinsic mode functions, and a long short-term memory network is used to predict the change trend of the intrinsic mode function, and a transient fluctuation separation result is determined based on the intrinsic mode function and the change trend; based on the transient fluctuation separation result, a system topology model is constructed using a graph neural network, and changes in the power transmission path under different short-circuit operating conditions are simulated based on the system topology model, and based on the changes, an optimized double Q learning strategy is used to evaluate different The application effect of short-circuit countermeasures is used to obtain effect evaluation results corresponding to different short-circuit countermeasures, and a long-term protection plan is generated based on the effect evaluation results; based on the long-term protection plan, a finite element analysis is performed using a neural network based on physical information to obtain a thermal effect distribution simulation result, and a Monte Carlo simulation is used to evaluate the thermal effect distribution simulation result to generate an initial risk assessment result; based on the initial risk assessment result, a hybrid risk assessment model is constructed in combination with Bayesian network and fuzzy logic, and a target risk assessment result is obtained based on the hybrid risk assessment model, and an emergency response plan is generated based on the target risk assessment result using natural language generation technology and planning algorithms, and the emergency response plan is used to solve short-circuit faults. The technical solution provided by the present invention can perform comprehensive and accurate short-circuit fault simulation, ensure that the test covers various possible fault conditions, and improve the authenticity and comprehensiveness of the simulation; through the convolutional neural network, the real-time current and voltage data streams are feature extracted to quickly and accurately capture the short-circuit characteristics, thereby improving the sensitivity and response speed of monitoring; through the system topology model, the changes in the power transmission path can be accurately reflected, and the adaptability and flexibility of the model are enhanced; after multi-dimensional risk assessment and simulation, an emergency response plan is generated, which not only solves the short-circuit fault, but also provides a long-term protection plan for the system, ensuring that it can respond quickly and effectively in the face of emergencies. Among them, the generator can generate specific short-circuit conditions including chain reactions, multiple fault combinations, and extreme environmental conditions, which expands the scope and depth of the simulation test, and the discriminator is used to evaluate these specific short-circuit conditions, ensuring that the identification of potential risks and system vulnerabilities is more accurate and comprehensive. Adversarial training improves the stability and adaptability of the dual Q learning strategy in the face of different short-circuit conditions, making it more effective in dealing with unknown or extreme situations; the optimized dual Q learning strategy obtained through adversarial training can make faster and more accurate decisions to minimize the losses caused by short circuits.The optimized dual-Q learning strategy can quickly select the most effective short-circuit countermeasures under changing circumstances, thereby enhancing the emergency response capability of the system. By introducing the Bayesian optimization algorithm to search for the optimal hyperparameter combination, the evaluation process is made more scientific and reasonable. At the same time, the causal relationship between the countermeasures and system performance is analyzed through the causal reasoning mechanism, thereby improving the reliability of the evaluation.

[0044] These and other aspects of the present invention will become more apparent from the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 A flow chart of a short-circuit simulation test method for a subway DC power supply system provided by an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of the structure of a short-circuit simulation test system for a subway DC power supply system provided by an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of the structure of a computing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0050] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0052] Figure 1 A flowchart of a short-circuit simulation test method for a subway DC power supply system is provided in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0053] In view of the shortcomings of the existing subway DC power supply system short-circuit simulation test method, such as insufficient flexibility and poor adaptability, a more accurate, efficient and comprehensive solution is urgently needed to respond quickly and effectively when facing complex working conditions. Based on this, the present invention provides a subway DC power supply system short-circuit simulation test method, such as Figure 1 ,include:

[0054] Step 101: Preset a plurality of short-circuit fault parameters, adjust the short-circuit fault parameters to simulate short-circuit faults at different locations and with different intensities, and generate a short-circuit working condition based on the short-circuit faults;

[0055] In this step, the short-circuit fault parameters include fault type, fault location and fault intensity. Fault type includes single-phase, two-phase and three-phase short circuits, fault location includes specific nodes or areas, and fault intensity includes current changes. Adjustment refers to simulating different short-circuit scenarios by changing the values ​​of the above parameters. Short-circuit conditions refer to the short-circuit state under specific conditions, including detailed information such as the time, location and intensity of the fault.

[0056] For example, the following short-circuit fault parameters are preset: the fault type is a three-phase short circuit, the fault location is the transformer at subway station A, and the fault intensity is the instantaneous increase of the current from the normal value of 100A to 500A. By adjusting these parameters, short-circuit faults of different locations and intensities can be simulated. For example, the fault location is changed to the cable at subway station B, and the fault intensity is adjusted to the current increasing from 100A to 800A. Based on these parameters, the corresponding short-circuit condition file is generated to record the specific situation of each simulation.

[0057] Step 102: Based on the short-circuit condition, a convolutional neural network and a variational mode decomposition technique are applied to decompose the short-circuit signal into a plurality of intrinsic mode functions, a long short-term memory network is used to predict a change trend of the intrinsic mode function, and a transient fluctuation separation result is determined based on the intrinsic mode function and the change trend;

[0058] In this step, the transient fluctuation separation result refers to the identification of the transient fluctuation part in the short circuit event through the analysis of the IMF and its changing trend.

[0059] In this step, for the generated short-circuit condition, a convolutional neural network is used to extract the key features of the short-circuit signal, and then it is decomposed into multiple intrinsic mode functions through variational mode decomposition technology. Assume that three intrinsic mode functions are obtained, representing low-frequency, medium-frequency and high-frequency fluctuations respectively. Then, the long short-term memory network is used to predict the change trend of each intrinsic mode function over time. Finally, the transient fluctuation separation result is determined by combining the intrinsic mode function and its predicted trend, that is, the significant transient fluctuations in the short-circuit event are identified.

[0060] Step 103: Based on the transient fluctuation separation result, a system topology model is constructed using a graph neural network, and changes in the power transmission path under different short-circuit conditions are simulated based on the system topology model. Based on the changes, the optimized dual-Q learning strategy is used to evaluate the application effects of different short-circuit countermeasures, and effect evaluation results corresponding to different short-circuit countermeasures are obtained. Based on the effect evaluation results, a long-term protection plan is generated;

[0061] In this step, the power transmission path refers to the specific path along which power flows in the system, which changes due to the short-circuit condition.

[0062] Based on the results of transient fluctuation separation, this step uses graph neural networks to build a topological model of the subway DC power supply system to describe the connection relationship between each circuit breaker, transformer, and cable. Simulate the changes in the power transmission path under different short-circuit conditions. For example, when a short circuit occurs in the transformer of subway station A, how the power transmission path is redistributed. Then, the optimized dual-Q learning strategy is used to evaluate the effects of different response measures, such as immediately cutting off the power supply and starting the backup power supply. Finally, a long-term protection plan is generated based on the effect evaluation results, such as installing more circuit breakers at key nodes or regularly checking the health of specific cables.

[0063] Step 104: Based on the long-term protection plan, a finite element analysis is performed using a neural network based on physical information to obtain a thermal effect distribution simulation result, and the thermal effect distribution simulation result is evaluated using a Monte Carlo simulation to generate an initial risk assessment result;

[0064] In this step, the thermal effect distribution simulation result refers to the temperature field distribution diagram obtained through finite element analysis, which shows the heat distribution in the short circuit event.

[0065] This step uses a finite element analysis based on a neural network based on physical information to simulate the distribution of thermal effects in a short circuit event according to the long-term protection plan. For example, assuming that the temperature of a certain cable section rises sharply during a short circuit, the temperature field distribution diagram is obtained through finite element analysis. Then, the uncertainty of the simulation results is evaluated using Monte Carlo simulation to generate initial risk assessment results. For example, the probability of overheating of a certain cable section under certain conditions is 5%.

[0066] Step 105: Based on the initial risk assessment result, a hybrid risk assessment model is constructed in combination with Bayesian network and fuzzy logic. Based on the hybrid risk assessment model, a target risk assessment result is obtained. Based on the target risk assessment result, an emergency response plan is generated using natural language generation technology and planning algorithm. The emergency response plan is used to solve the short circuit fault.

[0067] This step is based on the initial risk assessment results, combined with Bayesian networks and fuzzy logic to build a hybrid risk assessment model. For example, considering the impact of different factors such as temperature, humidity, and equipment aging on short-circuit risk, the causal relationship between them is analyzed through Bayesian networks, and fuzzy logic is used to process uncertain information. Finally, the target risk assessment result is obtained. For example, the probability of a serious short circuit in a certain cable section in the next year is 10%. Based on this result, natural language generation technology and planning algorithms are used to generate emergency response plans. For example, if an abnormal increase in the temperature of a certain cable section is detected, the backup power supply is immediately started and the maintenance personnel are notified to check.

[0068] The embodiments of the present invention can achieve the following beneficial effects through the above steps:

[0069] By adjusting the short-circuit fault parameters at different locations and intensities, all possible short-circuit scenarios can be fully covered;

[0070] Efficiently extract and analyze transient fluctuation characteristics, improving the understanding of the dynamic characteristics of short-circuit events;

[0071] Optimize the selection and evaluation of short-circuit countermeasures to ensure the effectiveness of the countermeasures and the rationality of long-term protection;

[0072] The hybrid risk assessment model can handle different types of uncertainties and obtain more accurate target risk assessment results. Based on the target risk assessment results, it can automatically generate detailed and actionable emergency response plans to ensure that actions can be taken quickly and effectively when facing short-circuit faults.

[0073] In order to improve the evaluation accuracy of short-circuit countermeasures and the long-term protection capability of the system, based on this, the present invention provides a specific embodiment, step 103, based on the transient fluctuation separation result, using the graph neural network to build a system topology model, based on the system topology model to simulate the change of the power transmission path under different short-circuit conditions, based on the change, using the optimized double Q learning strategy to evaluate the application effect of different short-circuit countermeasures, to obtain the effect evaluation results corresponding to different short-circuit countermeasures, based on the effect evaluation results, generate a long-term protection plan, specifically including the following steps:

[0074] Step 301: Based on the transient fluctuation separation result, a system topology model is constructed using a graph neural network and an attention mechanism, and based on the system topology model, a multi-scale simulation technology is used to simulate the change of the power transmission path under different short-circuit conditions;

[0075] Assuming that the transient fluctuation separation results have been obtained, next, the graph neural network combined with the attention mechanism is used to construct a topological model of the subway DC power supply system. For example, the graph neural network can describe the connection relationship between each circuit breaker, transformer, and cable, while the attention mechanism can highlight those nodes that play a key role in short-circuit events, such as transformers with high current changes or cables with frequent faults; based on the constructed system topology model, multi-scale simulation technology is used to simulate the changes in power transmission paths under different short-circuit conditions. For example, when a section of cable is short-circuited, the multi-scale simulation technology can simulate the changes in the flow of current on the cable at the micro level, and simulate the redistribution of power in the entire power supply system at the macro level. This helps to fully understand the impact of short-circuit events on the system.

[0076] Step 302: introducing a multi-condition adversarial generative network to perform adversarial training on the double Q learning strategy in reinforcement learning to obtain an optimized double Q learning strategy, selecting short-circuit countermeasures based on the change situation using the optimized double Q learning strategy, and evaluating the application effect of the short-circuit countermeasures to obtain an effect evaluation result, and introducing a Bayesian optimization algorithm and a causal reasoning mechanism in the evaluation process to optimize the evaluation process;

[0077] In this step, short-circuit countermeasures refer to specific measures taken for different short-circuit conditions, such as cutting off power supply, starting backup power supply, etc. Effect evaluation results refer to the evaluation of the effects of different short-circuit countermeasures, including response speed, system recovery capability, etc.

[0078] This step introduces a multi-condition adversarial generative network. The generator generates specific short-circuit conditions, such as chain reactions, multiple fault combinations, and extreme environmental conditions, and the discriminator evaluates the authenticity and complexity of these conditions. Based on these evaluation results, the dual-Q learning strategy in reinforcement learning is trained adversarially to obtain the optimized dual-Q learning strategy. For example, the optimized dual-Q learning strategy can more accurately select countermeasures, such as immediately starting the backup power supply when an abnormal increase in the temperature of a certain cable section is detected; based on the changes in the power transmission path obtained by multi-scale simulation technology, the optimized dual-Q learning strategy is used to select the most appropriate short-circuit countermeasures and evaluate its application effect. For example, the recovery time and stability of the system after power is cut off are evaluated. During the evaluation process, the Bayesian optimization algorithm is introduced to automatically search for the optimal hyperparameter combination, and the causal relationship between short-circuit countermeasures and system performance is analyzed through the causal reasoning mechanism. For example, it is analyzed whether cutting off the power supply does reduce the incidence of subsequent failures.

[0079] Step 303: Based on the effect evaluation result, introduce explanatory artificial intelligence technology to generate a long-term protection plan, wherein the long-term protection plan includes an equipment configuration adjustment plan, an operating procedure optimization plan, and a maintenance plan;

[0080] In this step, equipment configuration adjustment planning refers to adjusting the configuration of existing equipment, such as adding circuit breakers or replacing aging cables. Operation procedure optimization planning refers to optimizing daily operation procedures, such as regularly checking the health of specific equipment. Maintenance planning refers to developing detailed maintenance plans, such as regular inspections and maintenance.

[0081] Based on the above effect evaluation results, this step introduces explanatory artificial intelligence technology (XAI) to generate a long-term protection plan. For example, XAI technology can help explain why some countermeasures are more effective than others and provide specific improvement suggestions. Based on these suggestions, the following long-term protection plans are formulated: the equipment configuration adjustment plan includes installing more circuit breakers at key nodes, or replacing severely aged cables to improve the overall safety of the system; the operating procedure optimization plan includes optimizing daily operating procedures, such as regularly checking the health of specific cables to ensure that the equipment is in the best working condition; the maintenance plan includes formulating a detailed maintenance plan, such as conducting a comprehensive overhaul every quarter to ensure the normal operation of all equipment.

[0082] The embodiments of the present invention can achieve the following beneficial effects through the above steps:

[0083] Accurately construct system topology models and power transmission path simulations to ensure a comprehensive understanding and accurate prediction of system behavior;

[0084] Optimize the selection and evaluation of short-circuit countermeasures, improve the effectiveness of countermeasures and the scientific nature of the evaluation process;

[0085] Generate a scientific and reasonable long-term protection plan, ensure the transparency and operability of protection measures, and improve the overall security and stability of the system.

[0086] In the short-circuit simulation test of the subway DC power supply system, the existing technology faces multiple difficulties. Complex fault modes are difficult to simulate comprehensively and accurately; there are challenges in the selection and evaluation of countermeasures, and the traditional reinforcement learning method is unstable and has poor convergence. To solve these problems, the present invention introduces a multi-condition adversarial generative network (GAN) to conduct adversarial training on the double Q learning strategy in reinforcement learning. The generator generates a specific short-circuit condition and the discriminator evaluates it. After optimization, the double Q learning strategy can select countermeasures more stably and efficiently. At the same time, the Bayesian optimization algorithm is combined to automatically search for the optimal hyperparameter combination, and the causal relationship between short-circuit countermeasures and system performance is analyzed using the causal reasoning mechanism, thereby optimizing the evaluation process and ensuring the reliability and accuracy of the evaluation results. This method not only improves the accuracy and diversity of short-circuit condition simulation, but also enhances the effectiveness of countermeasures and the adaptive ability of the system, ensuring the safe operation and rapid response capability of the subway DC power supply system. Based on this, the present invention provides a specific embodiment, step 302, introducing a multi-condition adversarial generation network to perform adversarial training on the double Q learning strategy in reinforcement learning, to obtain an optimized double Q learning strategy, based on the change situation, using the optimized double Q learning strategy to select short-circuit countermeasures, and evaluating the application effect of the short-circuit countermeasures to obtain an effect evaluation result, introducing a Bayesian optimization algorithm and a causal reasoning mechanism in the evaluation process to optimize the evaluation process, specifically comprising the following steps:

[0087] Step 311: introducing a multi-condition adversarial generation network, generating a specific short-circuit working condition through a generator, evaluating the specific short-circuit working condition through a discriminator, and obtaining a working condition evaluation result, wherein the specific short-circuit working condition includes a chain reaction, a combination of multiple faults, and an extreme environmental condition;

[0088] In this step, the specific short-circuit condition refers to a short-circuit event that occurs under specific conditions, such as a chain reaction, a combination of multiple faults, and a short circuit under extreme environmental conditions. The condition evaluation result refers to the result obtained after the discriminator evaluates the generated short-circuit condition, which reflects the authenticity and complexity of these conditions.

[0089] This step introduces a multi-condition adversarial generative network to simulate different types of short-circuit conditions in the subway DC power supply system. The generator is used to generate a series of specific short-circuit conditions. For example, a chain reaction refers to a short circuit in multiple devices one after another, resulting in a larger power outage. A combination of multiple faults refers to the simultaneous occurrence of multiple faults of different types, such as a short circuit in a certain cable section and an adjacent transformer. Extreme environmental conditions, such as a short circuit in a high temperature or high humidity environment, may affect the heat dissipation performance of the system. The discriminator evaluates each generated short-circuit condition to determine its authenticity. For example, for the generated chain reaction condition, the discriminator may give a higher score, indicating that the condition is very realistic; while for some unrealistic conditions, it gives a lower score. Finally, the generator is continuously improved based on the feedback from the discriminator to generate more realistic short-circuit conditions. The evaluation results can be used for subsequent training and optimization.

[0090] Short-circuit faults in subway DC power supply systems may involve faults of various types and intensities, such as chain reactions, multiple fault combinations, and extreme environmental conditions. Traditional multi-condition adversarial generative networks with a single generator and discriminator are difficult to fully simulate these complex short-circuit conditions. In addition, in standard multi-condition adversarial generative networks, adversarial training between the generator and the discriminator may lead to training instability and mode collapse. Therefore, the present invention provides a formula for optimizing the objective function of a multi-condition adversarial generative network, which more comprehensively simulates different types of short-circuit scenarios by introducing multiple generators and conditional variables. Each generator focuses on generating a specific type of short-circuit data, which can reduce the pressure on a single generator and improve the stability of the training process. In addition, the introduction of conditional variables enables the generator to generate diversified samples according to different conditions, further avoiding mode collapse; by optimizing the formula of the objective function of the multi-condition adversarial generative network, not only can realistic short-circuit condition data be generated, but also effective predictions can be made under unseen conditions, so that it can cope with more diverse short-circuit fault scenarios in practical applications. The specific calculation formula is as follows:

[0091] ;

[0092] in, Represents the optimization objective function of the multi-conditional adversarial generation network; Represents multiple generators Perform minimization operation to minimize the optimization objective function; Represents the discriminator Perform maximization operation to maximize the optimization objective function; Represents the optimization objective function, which is used to measure the performance of the generator and the discriminator;

[0093]

[0094] ;

[0095] in, represents the optimization objective function; represents the discriminator, which is used to distinguish between real data and generated data; Represents multiple generators, each of which is responsible for generating data of different types or characteristics; Represents the distribution of real data The sample is the actual observed data point; Represents the probability distribution of real data, representing the actual short-circuit condition data occurring in the subway DC power supply system; Represents conditional variables, indicating different types of short-circuit conditions, such as chain reactions, multiple fault combinations, etc.; Express Sample below Find the expected value; Represents the discriminator for the sample In the condition variable The output under is logarithmic. The larger the value, the more the discriminator believes that the sample is real. It is the first influencing parameter of the optimization objective function; represents the accumulation operation from the 1st generator to the nth generator, where n is the number of generators; Express Sample below Find the expected value; Represents the random noise vector of the i-th generator, which is used as the input of the generator to generate new data samples; represents a random noise vector The probability distribution of , usually a Gaussian distribution or other simple distribution, is used to initialize the input of the generator; Representation Generator According to the random noise vector and condition variables Generated data samples; Represents the data sample In the condition variable Below, the expected value of the log probability of the discriminator output is calculated for each generator separately, and the results of all generators are summarized by the sum symbol.

[0096] This formula aims to enable the generator to generate realistic short-circuit condition data through a minimization and maximization optimization process, while the discriminator can effectively distinguish between real data and generated data. Through the combination of multiple generators and conditional variables, this formula can generate more diverse and realistic short-circuit condition data, and can more accurately evaluate the quality of generated data; the discriminator can evaluate whether these generated short-circuit conditions are realistic enough, thereby improving the quality of the data generated by the generator; through adversarial training, the generator's strategy is continuously optimized to make the data it generates closer to the real data distribution, thereby better simulating actual short-circuit events; by simulating different types of short-circuit conditions, a comprehensive risk assessment can be conducted, and a long-term protection plan can be formulated to ensure the safe operation of the subway DC power supply system.

[0097] Step 312: performing adversarial training on the dual-Q learning strategy in reinforcement learning based on the working condition evaluation result to obtain an optimized dual-Q learning strategy;

[0098] This step conducts adversarial training on the dual-Q learning strategy in reinforcement learning based on the short-circuit conditions and condition evaluation results. The short-circuit conditions generated by the generator are used as input, and the dual-Q learning strategy, as an intelligent agent, tries to select the best response measures. The discriminator evaluates whether the agent's choice is reasonable and provides feedback. For example, if the agent chooses to cut off the power supply immediately and start the backup power supply, the discriminator will evaluate the effect of this measure. The generator continuously adjusts the generated short-circuit conditions based on the feedback from the discriminator, and the dual-Q learning strategy continuously optimizes its decision-making ability based on these conditions. After multiple rounds of adversarial training, the dual-Q learning strategy gradually learns how to make the best decisions under complex short-circuit conditions. For example, it learns to prioritize the protection of key nodes when facing chain reactions, and to take more conservative response measures under extreme environmental conditions.

[0099] Step 313: Based on the change, the optimized dual-Q learning strategy is used to select short-circuit countermeasures, and the application effect of the short-circuit countermeasures is evaluated to obtain an effect evaluation result. The Bayesian optimization algorithm and the causal reasoning mechanism are introduced in the evaluation process, and the optimal hyperparameter combination is searched by the Bayesian optimization algorithm. The causal relationship between the short-circuit countermeasures and the performance of the subway DC power supply system is analyzed by the causal reasoning mechanism. Based on the causal relationship and the optimal hyperparameter combination, the evaluation process is optimized;

[0100] In this step, the change refers to the change of the power transmission path under the short-circuit condition, such as current redistribution, voltage fluctuation, etc.

[0101] In this step, short-circuit countermeasures are selected and evaluated based on the changes using the optimized dual-Q learning strategy. Based on the current short-circuit condition and its changes, the dual-Q learning strategy selects the most appropriate countermeasure. For example, when an abnormal increase in the temperature of a certain cable section is detected, the dual-Q learning strategy recommends immediately cutting off the power supply and starting the backup power supply. The application effect of the selected countermeasure is evaluated. For example, the recovery time and stability of the system after the power supply is cut off are evaluated. The Bayesian optimization algorithm and causal reasoning mechanism are introduced in the evaluation process: The Bayesian optimization algorithm is used to automatically search for the optimal hyperparameter combination to ensure the efficiency and accuracy of the evaluation process. The causal reasoning mechanism analyzes the causal relationship between short-circuit countermeasures and system performance. For example, it is analyzed whether cutting off the power supply has indeed reduced the incidence of subsequent failures, or whether starting the backup power supply has significantly improved the recovery speed of the system. Based on the causal relationship and the optimal hyperparameter combination, the evaluation process is further optimized. For example, through the causal reasoning mechanism, it is found that although starting the backup power supply can quickly restore power supply, long-term operation may cause accelerated equipment aging. Therefore, the evaluation indicators are adjusted to increase the equipment life as an important consideration.

[0102] The embodiments of the present invention can achieve the following beneficial effects through the above steps:

[0103] Simulate complex short-circuit conditions to enhance the ability to identify potential risks and the accuracy of assessment;

[0104] Improved the robustness and decision-making efficiency of the dual-Q learning strategy, and enhanced the system's ability to cope with unknown or extreme conditions;

[0105] Scientifically evaluating the application effects of short-circuit countermeasures ensures the reliability and scientific nature of the evaluation process and enhances the system's emergency response capabilities.

[0106] In order to accurately capture and analyze the transient characteristics of short-circuit faults, the complex short-circuit signal must be carefully decomposed and predicted. Based on this, the present invention provides a specific embodiment, step 102, based on the short-circuit working condition, applying convolutional neural network and variational mode decomposition technology, decomposing the short-circuit signal into multiple intrinsic mode functions, using long short-term memory network to predict the change trend of the intrinsic mode function, based on the intrinsic mode function and the change trend, determining the transient fluctuation separation result, specifically including the following steps:

[0107] Step 201: collecting current and voltage data streams of a subway DC power supply system in real time, applying a convolutional neural network to extract key features of the current and voltage data streams to obtain short-circuit signal features, wherein the key features include mutations and peak values ​​when a short circuit occurs;

[0108] In this step, the current and voltage data streams refer to the current and voltage measurements transmitted in real time from the subway DC power supply system. The short-circuit signal features refer to the key information extracted from the current and voltage data that can characterize the short-circuit event, such as mutations and peaks.

[0109] For example, high-precision current and voltage sensors installed at each key node can collect current and voltage data of the subway DC power supply system once a second. These data are stored in the database in the form of time series to form current and voltage data streams. For example, current sensors and voltage sensors are installed at the transformer of subway station A to record data once a second. Convolutional neural networks (CNNs) are used to process the collected current and voltage data streams. Convolutional neural networks can identify mutations and peaks in the data, such as sudden increases in current or voltage, and peaks such as the maximum value reached by current or voltage. For example, when a short circuit occurs in a certain section of cable, the current will increase instantly, and the voltage may drop sharply. CNN can capture these mutations and peaks and extract them as short-circuit signal features.

[0110] Step 202: Based on the short-circuit signal characteristics, the short-circuit signal is decomposed into a plurality of intrinsic mode functions using a variational mode decomposition technique to extract transient fluctuation characteristics at different time scales, wherein the intrinsic mode functions characterize specific frequency components of the short-circuit signal;

[0111] In this step, the intrinsic mode function (IMF) refers to the single frequency component signal decomposed by the variational mode decomposition technique, which represents the characteristics of the short-circuit signal at different time scales. The transient fluctuation characteristics refer to the rapid change characteristics that occur in a short-circuit event in a short period of time, which usually contain important fault information.

[0112] In this step, the extracted short-circuit signal features are input into the variational mode decomposition (VMD) technique. Variational mode decomposition decomposes the original signal into multiple intrinsic mode functions, each of which characterizes the characteristics of the short-circuit signal at different time scales. For example, the variational mode decomposition technique may decompose three intrinsic mode functions: low-frequency IMF, which reflects slow changes on a long time scale, such as background noise or normal operating conditions; medium-frequency IMF, which reflects fluctuations on a medium time scale, such as changes in the power transmission path; and high-frequency IMF, which reflects transient fluctuations on a short time scale, such as mutations and peaks when a short circuit occurs. High-frequency IMF pays special attention to transient fluctuation features on a short time scale, which contain important information about short-circuit events. For example, high-frequency IMF can capture sudden changes in current and sudden drops in voltage, helping to more accurately identify the time and location of short circuits.

[0113] Step 203: Modeling the time series of the intrinsic mode function using a long short-term memory network to predict the change trend of the intrinsic mode function, and determining a transient fluctuation separation result by combining the transient fluctuation characteristics and the change trend;

[0114] In this step, the time series of each intrinsic mode function is input into the long short-term memory network (LSTM) for modeling. The LSTM network can capture the changing trend of the intrinsic mode function over time and predict future fluctuations. For example, for high-frequency IMF, the LSTM network can predict its changing trend in the next few seconds, helping to warn of potential short-circuit risks in advance. The LSTM network can not only predict the changing trend of the intrinsic mode function, but also identify abnormal fluctuations. For example, if the high-frequency IMF shows an abnormally increasing trend in a certain period of time, the LSTM network can issue an alarm to indicate that there may be an impending short-circuit event. The transient fluctuation separation result is finally determined by combining the transient fluctuation characteristics of the intrinsic mode function and the changing trend predicted by the LSTM network. For example, by analyzing the mutation of the high-frequency IMF and the trend predicted by the LSTM network, the specific time and intensity of the short-circuit event can be identified, thereby providing a basis for subsequent countermeasures.

[0115] The embodiments of the present invention can achieve the following beneficial effects through the above steps:

[0116] Real-time acquisition and application of convolutional neural network to extract short-circuit signal features ensures timely discovery and high-precision detection of short-circuit events;

[0117] The variational mode decomposition technique is used to extract transient fluctuation characteristics at different time scales, which enhances the understanding and analysis capabilities of the dynamic characteristics of short-circuit events;

[0118] The long short-term memory network is used to predict the changing trend of the intrinsic mode function, and combined with the transient fluctuation characteristics, scientific transient fluctuation separation results are provided, providing reliable support for early warning and fault diagnosis.

[0119] In order to further ensure the safety and reliability of the system, the thermal effect distribution and its potential risks are deeply analyzed. Based on this, the present invention provides a specific embodiment, step 104, based on the long-term protection plan, a finite element analysis is performed using a neural network based on physical information to obtain a thermal effect distribution simulation result, and the thermal effect distribution simulation result is evaluated using Monte Carlo simulation to generate an initial risk assessment result, which specifically includes the following steps:

[0120] Step 401: Based on the long-term protection plan, combined with the physical laws and boundary conditions of the subway DC power supply system, a finite element analysis is performed on the subway DC power supply system using a neural network based on physical information to generate a thermal effect distribution simulation result. In the finite element analysis process, a multi-scale modeling technology is introduced to extract the thermal effect distribution characteristics at different time scales and different spatial scales. The boundary conditions include a power input point, a load connection point, and a grounding position.

[0121] In this step, boundary conditions refer to defining constraints on the system boundaries, such as power input points, load connection points, and grounding locations.

[0122] This step is based on long-term protection planning, and uses a physical information-based neural network (PINN) to perform finite element analysis (FEA) on the subway DC power supply system to generate thermal effect distribution simulation results. The specific process is as follows: According to the physical laws of the subway DC power supply system, such as the heat conduction equation, current density formula, etc., reasonable boundary conditions are set, such as power input points, load connection points and grounding positions. The power input point is like a voltage source, the load connection point is like a current consumption point, and the grounding position is like a zero potential point. These boundary conditions ensure the accuracy of the simulation results. In the finite element analysis process, multi-scale modeling technology is introduced to extract the distribution characteristics of thermal effects at different spatial and temporal scales. For example, at the microscale, the temperature changes inside the cable are analyzed to identify key parts that may overheat. At the mesoscale, the heat transfer paths between different devices are studied to evaluate how heat is transferred from one device to another. At the macroscale, the temperature field distribution of the entire subway station or line section is simulated to understand the global thermal effect situation. Transient response captures the transient temperature rise when a short circuit event occurs and evaluates its impact on the subway DC power supply system. Steady-state response predicts the stable temperature distribution after long-term operation and evaluates the long-term health of the equipment. Dynamic response, analyze the temperature change process over time, and evaluate the thermal effect evolution of the subway DC power supply system under different working conditions. Finally, through the neural network and multi-scale modeling technology based on physical information, detailed thermal effect distribution simulation results are generated to show the temperature changes at different spatial and temporal scales.

[0123] Step 402: using a Monte Carlo simulation method to evaluate the probability density function in the thermal effect distribution simulation result to obtain the statistical characteristics of the thermal effect distribution, and generating an initial risk assessment result based on the statistical characteristics;

[0124] This step is based on the simulation results of thermal effect distribution. Monte Carlo simulation method is used to perform multiple random sampling on the simulation results of thermal effect distribution, calculate the probability of occurrence of each temperature point, and generate a probability density function (PDF). For example, simulate the probability of a certain cable section reaching a specific temperature in a short circuit event; extract statistical characteristics such as average temperature, standard deviation, maximum temperature, etc. from the probability density function. These statistical characteristics reflect the overall situation of thermal effect distribution and its uncertainty; based on the extracted statistical characteristics, generate initial risk assessment results. For example, if a certain cable section has a high probability of reaching a high temperature in a short period of time, then this cable section has a high risk of overheating and needs to be paid special attention and protected.

[0125] The embodiments of the present invention can achieve the following beneficial effects through the above steps:

[0126] Combining physical laws and boundary conditions, finite element analysis is performed using a neural network based on physical information to ensure the accuracy and reliability of thermal effect distribution simulation. At the same time, multi-scale modeling technology is used to fully extract thermal effect characteristics at different scales.

[0127] Generate detailed thermal effect distribution simulation results, providing a comprehensive understanding of the system's thermal behavior and laying a solid foundation for subsequent risk assessment and protection planning;

[0128] The statistical characteristics of thermal effect distribution are evaluated through Monte Carlo simulation method to generate scientific initial risk assessment results, ensuring the objectivity and effectiveness of the assessment process.

[0129] After completing the detailed analysis of the short-circuit condition and the separation of transient fluctuations, in order to further improve the accuracy and reliability of risk assessment, it is ensured that countermeasures can be taken quickly and effectively when facing a short-circuit fault. Based on this, the present invention provides a specific embodiment, step 105, based on the initial risk assessment result, combined with Bayesian network and fuzzy logic, a hybrid risk assessment model is constructed, based on the hybrid risk assessment model, a target risk assessment result is obtained, based on the target risk assessment result, natural language generation technology and planning algorithm are used to generate an emergency response plan, which specifically includes the following steps:

[0130] Step 501: Based on the initial risk assessment result, a Bayesian network is used to define a probabilistic dependency relationship between nodes, and based on the nodes and the probabilistic dependency relationship, uncertainties and probabilistic dependencies are processed to obtain a posterior probability distribution corresponding to different nodes, wherein the nodes include physical variables and equipment states, and the uncertainties include random uncertainty, epistemic uncertainty, and fuzzy uncertainty;

[0131] In this step, uncertainty includes random uncertainty, epistemic uncertainty and fuzzy uncertainty. Random uncertainty refers to the uncertainty caused by random factors, such as fluctuations in power transmission; epistemic uncertainty refers to the uncertainty caused by incomplete knowledge, such as insufficient understanding of equipment performance; fuzzy uncertainty refers to the uncertainty caused by fuzzy information, such as expert experience and qualitative descriptions in the rule base.

[0132] This step uses Bayesian networks to define nodes and their probabilistic dependencies based on the initial risk assessment results. For example, physical variable nodes such as temperature, current, voltage, etc.; equipment status nodes such as transformers, circuit breakers, and cables are normal, overheated, and faulty; probabilistic dependencies such as high temperature may cause transformer failure, high current may cause circuit breakers to trip, etc. Bayesian networks can handle different types of uncertainty. Random uncertainty considers random fluctuations in the power transmission process, such as instantaneous changes in current. Epistemic uncertainty combines existing knowledge to assess the uncertainty of equipment health. Fuzzy uncertainty handles qualitative descriptions in expert experience and rule bases, such as severe cable aging. Based on the initial risk assessment results, the probability distribution of each node in the Bayesian network is updated to obtain the posterior probability distribution. For example, the probability of a certain section of cable reaching high temperature in a short circuit event is updated from 5% to 10%, indicating that its risk has increased.

[0133] Step 502: Introduce fuzzy logic to process the fuzzy uncertainty to convert qualitative information into quantitative indicators, combine the quantitative indicators and the posterior probability distribution to build a hybrid risk assessment model, and obtain a target risk assessment result based on the hybrid risk assessment model, wherein the qualitative information includes expert experience and qualitative descriptions in the rule base;

[0134] In this step, quantitative indicators refer to the conversion of qualitative information into specific numerical values ​​through fuzzy logic. Qualitative information refers to the qualitative descriptions in expert experience and rule bases, such as serious cable aging and good transformer performance. Qualitative descriptions in expert experience refer to the opinions and experiences from experts in the field, which usually exist in qualitative form. For example, in a high temperature environment, equipment is more prone to failure. Qualitative descriptions in the rule base refer to pre-defined rules. For example, if the current intensity exceeds a certain threshold, the risk of short circuit increases. Fuzzy logic can determine the degree of risk under different conditions and quantify it based on these rules.

[0135] This step is based on the posterior probability distribution, and introduces fuzzy logic to process fuzzy uncertainty and convert qualitative information into quantitative indicators. For example, through fuzzy logic, the serious aging of a certain section of cable in expert experience is converted into a numerical value of the aging degree, such as 0.8. The good performance of the transformer described in the rule base is converted into a performance score, such as 0.9. Combining the posterior probability distribution in the Bayesian network and the quantitative indicators converted by fuzzy logic, a hybrid risk assessment model is constructed. For example, the aging degree of the cable and the performance score of the transformer are included in the model to evaluate the risk of the entire system. Based on the hybrid risk assessment model, the target risk assessment result is obtained. For example, the probability of a certain section of cable suffering a serious short circuit in the next year is 15%, and the probability of the transformer failing in the next year is 5%.

[0136] Step 503: Based on the target risk assessment result, using natural language generation technology and context awareness mechanism, generate target text descriptions corresponding to different risk levels, use path planning algorithm and multi-agent system to determine the shortest rescue path, and generate an emergency response plan based on the target text description and the shortest rescue path;

[0137] Based on the target risk assessment results, this step uses natural language generation technology and context-aware mechanisms to generate target text descriptions corresponding to different risk levels. For example, low risk means that the cable is in good health and does not require special attention; medium risk means that the cable is seriously aged, and it is recommended to check regularly and maintain it in time; high risk means that the cable is seriously aged, and the backup power supply should be started immediately and the maintenance personnel should be notified; the path planning algorithm and multi-agent system are used to determine the shortest rescue path. For example, when a short circuit occurs at subway station A, the shortest path from the nearest emergency response center to the fault point is calculated, and multiple agents, such as robots and maintenance personnel, are dispatched to work together. Combined with the target text description and the shortest rescue path, a detailed emergency response plan is generated. For example, the emergency response plan includes response measures, action guidelines, and path guidance. Response measures include immediately cutting off the power supply and starting the backup power supply; action guidelines include notifying maintenance personnel to go to the designated location for inspection and repair; path guidance includes providing a specific route map and time estimate for the shortest rescue path.

[0138] The embodiments of the present invention can achieve the following beneficial effects through the above steps:

[0139] Using Bayesian networks to define nodes and probability dependencies, dealing with various uncertainties, and obtaining posterior probability distributions provides a solid mathematical foundation for subsequent risk assessment and decision-making;

[0140] By processing fuzzy uncertainty through fuzzy logic, qualitative information is converted into quantitative indicators, and a hybrid risk assessment model is constructed to generate reliable target risk assessment results;

[0141] Based on the target risk assessment results, a detailed emergency response plan is generated using natural language generation technology and path planning algorithms, ensuring the operability of the response measures and the efficiency of the rescue process.

[0142] After obtaining detailed target risk assessment results, in order to effectively deal with short circuit risks of different levels, based on this, the present invention provides a specific embodiment, step 503, based on the target risk assessment results, using natural language generation technology and context perception mechanism, generating target text descriptions corresponding to different risk levels, using path planning algorithms and multi-agent systems to determine the shortest rescue path, based on the target text description and the shortest rescue path, generating an emergency response plan, specifically including the following steps:

[0143] Step 511: Based on the target risk assessment result, identify potential risk points of different risk levels, and generate a preliminary text description based on the potential risk points using natural language generation technology;

[0144] In this step, potential risk points refer to high-risk areas or equipment determined based on risk assessment results, which are key points that may cause failures or accidents.

[0145] This step identifies potential risk points of different risk levels based on the target risk assessment results. For example, a high-risk point is a section of aging cable, with a 20% probability of a serious short circuit in the next year; a medium-risk point is a transformer, with a 10% probability of failure in the next year; and a low-risk point is other equipment that is operating normally. Using natural language generation technology, a preliminary text description is generated for each potential risk point. For example, a high-risk point is a section of cable that is seriously aged and has a high risk of short circuit. Please start the backup power supply immediately and notify the maintenance personnel to check; a medium-risk point is a transformer that performs well, but it is recommended to monitor its operating status regularly; a low-risk point is that the equipment is in good health and does not require special attention.

[0146] Step 512: Optimizing the preliminary text description through a context-aware mechanism and domain-specific data to obtain a target text description, wherein the domain-specific data includes historical fault records;

[0147] In this step, domain-specific data refers to relevant data in a specific field, such as historical fault records and operating procedures of the subway DC power supply system.

[0148] This step adjusts the content of the preliminary text description according to the current context, such as time, location, and environmental conditions. For example, if it is currently nighttime operation, add reminders for nighttime operation safety, that is, please ensure sufficient lighting at night and pay attention to safety precautions; use domain-specific data, such as combining historical fault records, to optimize text descriptions. For example, if this section of cable has had overheating problems many times in the past, add detailed historical information. This cable has had three overheating incidents in the past 6 months, please give priority to it. Combine context-aware mechanisms and domain-specific data to generate more accurate target text descriptions. For example, a section of cable is severely aged and has had three overheating incidents in the past 6 months. There is a high risk of short circuit. Please immediately start the backup power supply and notify maintenance personnel for emergency inspections, ensure sufficient lighting at night, and pay attention to safety precautions.

[0149] Step 513: using a path planning algorithm to determine a preliminary rescue path, and in the process of determining the preliminary rescue path, comprehensively considering actual environmental factors of the subway DC power supply system, the actual environmental factors including equipment layout, channel width, and personnel distribution;

[0150] In this step, actual environmental factors refer to specific environmental features that affect path selection, such as equipment layout, channel width, personnel distribution, etc.

[0151] This step is based on the target text description, uses the path planning algorithm, and combines the actual environmental factors of the subway DC power supply system to determine the preliminary rescue path. For example, the shortest path from the nearest emergency response center to a section of aging cable. When determining the preliminary rescue path, consider the following actual environmental factors: equipment layout to avoid crossing dense equipment areas and reduce interference from obstacles. Channel width to ensure that the rescue team can pass smoothly, especially when large equipment is being moved; personnel distribution, such as avoiding crowded areas during peak hours to ensure the safety of rescuers. Finally, determine a preliminary rescue path that comprehensively considers equipment layout, channel width, and personnel distribution. For example, choose a spacious and barrier-free channel to avoid crowded areas during peak hours to ensure that the rescue team can reach the fault point quickly.

[0152] Step 514: Introduce a multi-agent system and a reinforcement learning algorithm, dynamically optimize the preliminary rescue path by simulating the behaviors of multiple rescue teams, obtain the shortest rescue path, and generate an emergency response plan by combining the target text description and the shortest rescue path;

[0153] This step is based on the preliminary rescue path. The multi-agent system is used to simulate the behavior of multiple rescue teams, considering the collaboration and conflict between different teams. For example, it simulates how maintenance personnel, robots, and emergency teams work together. The preliminary rescue path is dynamically optimized through the reinforcement learning algorithm. For example, the agent continuously adjusts the path based on real-time feedback to ensure that the rescue team can quickly reach the fault point. The reinforcement learning algorithm can evaluate the pros and cons of different paths and select the best path. After optimization by the multi-agent system and reinforcement learning algorithm, the dynamically optimized shortest rescue path is finally obtained. For example, through multiple simulations and optimizations, a rescue path that is not only the shortest but also the safest and most efficient is determined. Combined with the target text description and the shortest rescue path, a detailed emergency response plan is generated. For example, the plan includes countermeasures, immediately cutting off the power supply and starting the backup power supply; action guide, notifying maintenance personnel to go to the designated location for inspection and repair; path guidance, providing a specific route map and time estimate for the shortest rescue path; precautions, ensuring adequate lighting at night, paying attention to safety protection measures, and avoiding crowded areas during peak hours.

[0154] The embodiments of the present invention can achieve the following beneficial effects through the above steps:

[0155] Based on the target risk assessment results, potential risk points are accurately identified and a preliminary text description is generated using natural language generation technology, ensuring the comprehensiveness and professionalism of the response measures;

[0156] By optimizing text descriptions through context-aware mechanisms and domain-specific data, the accuracy and practicality of text descriptions are enhanced, and their authority and operability are improved;

[0157] A path planning algorithm is used to determine the initial rescue path, and the path is dynamically optimized through a multi-agent system and reinforcement learning algorithm to generate an efficient emergency response plan, ensuring the efficiency and safety of the rescue process.

[0158] Figure 2 The present invention provides a schematic diagram of a short-circuit simulation test system for a subway DC power supply system. Figure 2 As shown, the system includes:

[0159] An adjustment module 21, used for presetting a plurality of short-circuit fault parameters, adjusting the short-circuit fault parameters to simulate short-circuit faults at different locations and with different intensities, and generating a short-circuit working condition based on the short-circuit faults;

[0160] The analysis module 22 is used to decompose the short-circuit signal into multiple intrinsic modal functions based on the short-circuit working condition by applying convolutional neural network and variational mode decomposition technology, predict the change trend of the intrinsic modal function by using long short-term memory network, and determine the transient fluctuation separation result based on the intrinsic modal function and the change trend;

[0161] A construction module 23 is used to construct a system topology model using a graph neural network based on the transient fluctuation separation result, simulate the change of the power transmission path under different short-circuit conditions based on the system topology model, and use the optimized double Q learning strategy to evaluate the application effects of different short-circuit countermeasures based on the change, so as to obtain effect evaluation results corresponding to different short-circuit countermeasures, and generate a long-term protection plan based on the effect evaluation results;

[0162] An evaluation module 24 is used to perform finite element analysis based on the long-term protection plan using a neural network based on physical information to obtain a thermal effect distribution simulation result, and to evaluate the thermal effect distribution simulation result using Monte Carlo simulation to generate an initial risk assessment result;

[0163] The generation module 25 is used to construct a hybrid risk assessment model based on the initial risk assessment results, combined with Bayesian network and fuzzy logic, obtain the target risk assessment results based on the hybrid risk assessment model, and generate an emergency response plan based on the target risk assessment results using natural language generation technology and planning algorithms. The emergency response plan is used to resolve short circuit faults.

[0164] Figure 2 The short-circuit simulation test system of a subway DC power supply system can perform Figure 1 The implementation principle and technical effect of the short-circuit simulation test method for a subway DC power supply system described in the embodiment shown are not repeated here. The specific way in which each module and unit performs operations in the short-circuit simulation test system for a subway DC power supply system in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0165] In one possible design, Figure 2 The short-circuit simulation test system of a subway DC power supply system in the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0166] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0167] The processing component 32 is used to: preset a variety of short-circuit fault parameters, adjust the short-circuit fault parameters to simulate short-circuit faults at different locations and different intensities, and generate a short-circuit working condition based on the short-circuit fault; based on the short-circuit working condition, apply convolutional neural network and variational mode decomposition technology to decompose the short-circuit signal into multiple intrinsic mode functions, use long short-term memory network to predict the change trend of the intrinsic mode function, and determine the transient fluctuation separation result based on the intrinsic mode function and the change trend; based on the transient fluctuation separation result, use graph neural network to build a system topology model, simulate the change of power transmission path under different short-circuit working conditions based on the system topology model, and use the optimized double Q learning strategy to evaluate based on the change. The application effects of different short-circuit countermeasures are obtained to obtain effect evaluation results corresponding to different short-circuit countermeasures, and a long-term protection plan is generated based on the effect evaluation results; based on the long-term protection plan, a finite element analysis is performed using a neural network based on physical information to obtain a thermal effect distribution simulation result, and the thermal effect distribution simulation result is evaluated using Monte Carlo simulation to generate an initial risk assessment result; based on the initial risk assessment result, a hybrid risk assessment model is constructed in combination with Bayesian network and fuzzy logic, and a target risk assessment result is obtained based on the hybrid risk assessment model, and an emergency response plan is generated based on the target risk assessment result using natural language generation technology and planning algorithms, and the emergency response plan is used to resolve short-circuit faults.

[0168] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0169] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0170] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0171] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0172] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0173] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0174] The embodiment of the present invention further provides a computer storage medium storing a computer program, which can achieve the above-mentioned Figure 1 A short-circuit simulation test method for a subway DC power supply system according to the illustrated embodiment.

[0175] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0176] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0177] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A short-circuit simulation test method for a subway DC power supply system, characterized in that: include: Preset a plurality of short-circuit fault parameters, adjust the short-circuit fault parameters to simulate short-circuit faults at different locations and with different intensities, and generate a short-circuit working condition based on the short-circuit faults; Based on the short-circuit working condition, a convolutional neural network and a variational mode decomposition technique are applied to decompose the short-circuit signal into a plurality of intrinsic mode functions, a long short-term memory network is used to predict a change trend of the intrinsic mode function, and a transient fluctuation separation result is determined based on the intrinsic mode function and the change trend; Based on the transient fluctuation separation results, a system topology model is constructed using a graph neural network. Based on the system topology model, changes in the power transmission path under different short-circuit conditions are simulated. Based on the changes, an optimized dual-Q learning strategy is used to evaluate the application effects of different short-circuit countermeasures, and effect evaluation results corresponding to different short-circuit countermeasures are obtained. Based on the effect evaluation results, a long-term protection plan is generated. The optimized dual-Q learning strategy is obtained by introducing a multi-condition adversarial generation network to conduct adversarial training on the dual-Q learning strategy in reinforcement learning. In the evaluation process, a Bayesian optimization algorithm and a causal reasoning mechanism are introduced to optimize the evaluation process. Based on the long-term protection plan, a finite element analysis is performed using a neural network based on physical information to obtain a thermal effect distribution simulation result, and the thermal effect distribution simulation result is evaluated using a Monte Carlo simulation to generate an initial risk assessment result; Based on the initial risk assessment results, a hybrid risk assessment model is constructed in combination with Bayesian networks and fuzzy logic. Based on the hybrid risk assessment model, a target risk assessment result is obtained. Based on the target risk assessment result, an emergency response plan is generated using natural language generation technology and planning algorithms. The emergency response plan is used to resolve short circuit faults.

2. The method according to claim 1, characterized in that Based on the transient fluctuation separation results, a system topology model is constructed using a graph neural network. Based on the system topology model, the changes in the power transmission path under different short-circuit conditions are simulated. Based on the changes, the optimized dual-Q learning strategy is used to evaluate the application effects of different short-circuit countermeasures, and the effect evaluation results corresponding to different short-circuit countermeasures are obtained. Based on the effect evaluation results, a long-term protection plan is generated, including: Based on the transient fluctuation separation results, a system topology model is constructed using graph neural network and attention mechanism. Based on the system topology model, multi-scale simulation technology is used to simulate the changes in the power transmission path under different short-circuit conditions. A multi-condition adversarial generative network is introduced to conduct adversarial training on the double Q learning strategy in reinforcement learning to obtain an optimized double Q learning strategy. Based on the change, the optimized double Q learning strategy is used to select short-circuit countermeasures, and the application effect of the short-circuit countermeasures is evaluated to obtain an effect evaluation result. In the evaluation process, a Bayesian optimization algorithm and a causal reasoning mechanism are introduced to optimize the evaluation process. Based on the effect evaluation results, explanatory artificial intelligence technology is introduced to generate a long-term protection plan, which includes an equipment configuration adjustment plan, an operating procedure optimization plan, and a maintenance plan.

3. The method according to claim 2, characterized in that A multi-condition adversarial generative network is introduced to conduct adversarial training on the double Q learning strategy in reinforcement learning to obtain an optimized double Q learning strategy. Based on the change, the optimized double Q learning strategy is used to select short-circuit countermeasures, and the application effect of the short-circuit countermeasures is evaluated to obtain an effect evaluation result. In the evaluation process, a Bayesian optimization algorithm and a causal reasoning mechanism are introduced to optimize the evaluation process, including: A multi-condition adversarial generative network is introduced to generate specific short-circuit conditions through a generator, and the specific short-circuit conditions are evaluated through a discriminator to obtain a condition evaluation result. The specific short-circuit conditions include chain reactions, multiple fault combinations, and extreme environmental conditions. Based on the working condition evaluation result, adversarial training is performed on the double Q learning strategy in reinforcement learning to obtain an optimized double Q learning strategy; Based on the changes, short-circuit countermeasures are selected using the optimized dual-Q learning strategy, and the application effects of the short-circuit countermeasures are evaluated to obtain effect evaluation results. A Bayesian optimization algorithm and a causal reasoning mechanism are introduced into the evaluation process, and the optimal hyperparameter combination is searched through the Bayesian optimization algorithm. The causal relationship between the short-circuit countermeasures and the performance of the subway DC power supply system is analyzed through the causal reasoning mechanism. Based on the causal relationship and the optimal hyperparameter combination, the evaluation process is optimized.

4. The method according to claim 1, characterized in that: Based on the short-circuit working condition, the short-circuit signal is decomposed into multiple intrinsic modal functions by applying convolutional neural network and variational mode decomposition technology, and the change trend of the intrinsic modal function is predicted by using long short-term memory network. Based on the intrinsic modal function and the change trend, the transient fluctuation separation result is determined, including: The current and voltage data streams of the subway DC power supply system are collected in real time, and a convolutional neural network is used to extract key features of the current and voltage data streams to obtain short-circuit signal features, wherein the key features include mutations and peak values ​​when a short circuit occurs; Based on the short-circuit signal characteristics, the short-circuit signal is decomposed into a plurality of intrinsic mode functions using variational mode decomposition technology to extract transient fluctuation characteristics at different time scales, wherein the intrinsic mode functions characterize specific frequency components of the short-circuit signal; The time series of the intrinsic mode function is modeled using a long short-term memory network to predict the change trend of the intrinsic mode function, and the transient fluctuation separation result is determined by combining the transient fluctuation characteristics and the change trend.

5. The method according to claim 1, characterized in that Based on the long-term protection plan, a finite element analysis is performed using a neural network based on physical information to obtain a thermal effect distribution simulation result, and the thermal effect distribution simulation result is evaluated using Monte Carlo simulation to generate an initial risk assessment result, including Based on the long-term protection plan, combined with the physical laws and boundary conditions of the subway DC power supply system, a finite element analysis is performed on the subway DC power supply system using a neural network based on physical information to generate a thermal effect distribution simulation result. In the finite element analysis process, a multi-scale modeling technology is introduced to extract the thermal effect distribution characteristics at different time scales and different spatial scales. The boundary conditions include a power input point, a load connection point, and a grounding position. The probability density function in the thermal effect distribution simulation result is evaluated using a Monte Carlo simulation method to obtain the statistical characteristics of the thermal effect distribution, and an initial risk assessment result is generated based on the statistical characteristics.

6. The method according to claim 1, characterized in that Based on the initial risk assessment results, a hybrid risk assessment model is constructed by combining Bayesian network and fuzzy logic. Based on the hybrid risk assessment model, a target risk assessment result is obtained. Based on the target risk assessment result, an emergency response plan is generated by using natural language generation technology and planning algorithms, including: Based on the initial risk assessment results, using a Bayesian network to define probabilistic dependencies between nodes, and based on the nodes and the probabilistic dependencies, processing uncertainties and probabilistic dependencies to obtain posterior probability distributions corresponding to different nodes, wherein the nodes include physical variables and equipment states, and the uncertainties include random uncertainty, epistemic uncertainty, and fuzzy uncertainty; Introducing fuzzy logic to process the fuzzy uncertainty so as to convert qualitative information into quantitative indicators, combining the quantitative indicators with the posterior probability distribution to construct a hybrid risk assessment model, and obtaining a target risk assessment result based on the hybrid risk assessment model, wherein the qualitative information includes expert experience and qualitative descriptions in the rule base; Based on the target risk assessment results, natural language generation technology and context awareness mechanism are used to generate target text descriptions corresponding to different risk levels, a path planning algorithm and a multi-agent system are used to determine the shortest rescue path, and an emergency response plan is generated based on the target text description and the shortest rescue path.

7. The method according to claim 6, characterized in that Based on the target risk assessment results, natural language generation technology and context perception mechanism are used to generate target text descriptions corresponding to different risk levels, a path planning algorithm and a multi-agent system are used to determine the shortest rescue path, and based on the target text description and the shortest rescue path, an emergency response plan is generated, including: Based on the target risk assessment results, identifying potential risk points of different risk levels, and based on the potential risk points, generating a preliminary text description using natural language generation technology; optimizing the preliminary text description to obtain a target text description through a context-aware mechanism and domain-specific data, wherein the domain-specific data includes historical fault records; A path planning algorithm is used to determine a preliminary rescue path. In the process of determining the preliminary rescue path, actual environmental factors of the subway DC power supply system are comprehensively considered, including equipment layout, channel width, and personnel distribution; A multi-agent system and a reinforcement learning algorithm are introduced to dynamically optimize the preliminary rescue path by simulating the behavior of multiple rescue teams to obtain the shortest rescue path. An emergency response plan is generated by combining the target text description and the shortest rescue path.

8. A short-circuit simulation test system for a subway DC power supply system, characterized in that: include: An adjustment module, used for presetting a plurality of short-circuit fault parameters, adjusting the short-circuit fault parameters to simulate short-circuit faults at different locations and with different intensities, and generating a short-circuit working condition based on the short-circuit faults; An analysis module is used to decompose the short-circuit signal into multiple intrinsic modal functions based on the short-circuit working condition by applying convolutional neural network and variational mode decomposition technology, predict the change trend of the intrinsic modal function by using long short-term memory network, and determine the transient fluctuation separation result based on the intrinsic modal function and the change trend; A construction module is used to construct a system topology model using a graph neural network based on the transient fluctuation separation result, simulate the changes in the power transmission path under different short-circuit conditions based on the system topology model, and use the optimized dual-Q learning strategy to evaluate the application effects of different short-circuit countermeasures based on the changes, and obtain the effect evaluation results corresponding to different short-circuit countermeasures. Based on the effect evaluation results, a long-term protection plan is generated. The optimized dual-Q learning strategy is obtained by introducing a multi-condition adversarial generation network to conduct adversarial training on the dual-Q learning strategy in reinforcement learning. In the evaluation process, a Bayesian optimization algorithm and a causal reasoning mechanism are introduced to optimize the evaluation process; An evaluation module is used to perform finite element analysis based on the long-term protection plan using a neural network based on physical information to obtain a thermal effect distribution simulation result, evaluate the thermal effect distribution simulation result using Monte Carlo simulation, and generate an initial risk assessment result; A generation module is used to construct a hybrid risk assessment model based on the initial risk assessment result in combination with Bayesian network and fuzzy logic, obtain a target risk assessment result based on the hybrid risk assessment model, and generate an emergency response plan based on the target risk assessment result using natural language generation technology and planning algorithm. The emergency response plan is used to resolve short circuit faults.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a short-circuit simulation test method for a subway DC power supply system as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a short-circuit simulation test method for a subway DC power supply system as described in any one of claims 1 to 7 is implemented.

Images