Multi-objective ship power system fault network dynamic reconfiguration method and system based on DRL

Abstract

The application provides a DRL-based multi-target ship power system fault network dynamic reconfiguration method and system, which simplifies the circuit diagram into a directed graph, adjusts the population proportion parameter as the action selection of the deep reinforcement learning algorithm, determines the reward function according to the change of the solution distance and the entropy value, introduces the line load uniformity as an auxiliary evaluation index, and realizes the function of accelerating the convergence speed of the algorithm. The application introduces the DQN algorithm to monitor the dynamic change of the population diversity index in real time, dynamically adjusts the population proportion in iteration, and guides the evolution direction; introduces a double-population evolution strategy, splits the population to take different genetic operations, improves the evolution direction accuracy and adaptability of the NSGA-II algorithm; implements the double-population evolution strategy, breaks through the local extreme value limit while expanding the algorithm search dimension, dynamically adjusts the population segmentation proportion parameter by using the real-time monitoring of the agent on the population index, improves the reconfiguration time, and provides a solution set with better performance.

Description

Technical Field

[0001] This invention belongs to the field of marine technology, specifically relating to a dynamic reconfiguration method and system for fault networks of multi-objective marine power systems based on DRL. Background Technology

[0002] With the continuous advancement of marine technology, modern ships are gradually moving towards full electrification and automation. A major challenge facing modern ship electrical systems is ensuring the ship's survivability and operational capability in the event of a failure. Unlike land-based power systems, ship electrical systems have unique topologies, with shorter power lines and a rapid increase in current during short circuits. This can quickly affect the stability of the entire ship's electrical system, posing a significant potential threat to the lives of personnel on board. Therefore, in the event of a failure, the ship's electrical system must be able to quickly isolate the fault and reconfigure the system topology in the initial stages of the electrical failure to restore power supply to the greatest extent possible. Thus, effective fault reconfiguration strategies are urgently needed for ship electrical systems to enhance their ability to respond to emergencies and their reliability.

[0003] In recent years, with the rapid development of artificial intelligence technology, Deep Reinforcement Learning (DRL) has been increasingly applied to various service areas of power systems at an unprecedented pace. As an advanced artificial intelligence method, DRL effectively combines the powerful feature extraction capabilities of deep learning with the decision optimization capabilities of reinforcement learning—two highly valuable AI technologies. It does not passively accept knowledge but actively interacts in complex and ever-changing environments, continuously learning from environmental feedback to dynamically learn how to make the wisest and most appropriate decisions in various situations, providing strong technical support for the efficient, stable, and reliable operation of power systems. Therefore, this technology is suitable for the dynamic reconfiguration of ship power systems.

[0004] Power system-related problems can be formulated as multi-objective optimization problems, with optimal operations under different conditions obtained by solving the Pareto front. Current research in this area is limited, mainly due to the drawbacks of optimization algorithms, such as parallel optimization of multiple problems, high computational cost, and difficulty in implementation. Wang et al. (2012) used a fast non-dominated sorting genetic algorithm with an elite strategy to solve multi-objective, multi-constraint fault reconstruction problems, with the objective functions of minimizing power loss load and minimizing switching operation costs. Yang (2013) focused on ring power supply network problems, improving the classical particle swarm optimization algorithm with a two-group particle substitution strategy. He established two particle evolution groups (primary and secondary), swapped some particles during iteration, and then performed parallel optimization. However, the performance of the quantum particle swarm optimization algorithm in solving discretized problems still requires theoretical guarantees. MA et al. (2015) improved the selection strategy, preserving valuable information about infeasible solutions and thus improving convergence, but the problems of poor distribution and slow convergence speed still need to be addressed. To address the poor convergence and distribution of existing constrained multi-objective optimization algorithms in solving ship microgrid reconfiguration problems, SU et al. (2020) proposed a two-stage decomposition-based constrained multi-objective optimization method (TSMOEA / D), employing a dual-population hybrid method to handle the corresponding constraints. However, the parameter ratio adjustment of the dual population is not flexible enough, resulting in slow convergence. MA and REN (2018) proposed a differential evolution algorithm based on chaotic migration and nonparametric mutation. This algorithm has a simple structure and effectively avoids premature convergence, but it eliminates excellent infeasible solutions, leading to poor convergence and distribution when searching for optimal non-dominated solutions. Liu et al. (2024) proposed an NSGA-II algorithm that considers elite selection strategies and convergence. Although its convergence has been proven and the convergence time has been significantly shortened, it does not fully consider the constraints. In general, most of the above studies rely on parameter settings, especially in genetic algorithms where a single population is prone to losing gene diversity and getting trapped in local optima. When using a dual-population evolution strategy, the adjustment of population-related parameters is not flexible enough, resulting in slow convergence. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method and system for dynamic reconstruction of fault networks in multi-objective ship power systems based on DRL, so as to accelerate the convergence speed of the algorithm.

[0006] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL, comprising the following steps: S1: Topologicalize the circuit diagram of the power system under test and convert it into a corresponding directed graph; S2: The circuit dynamic fault reconfiguration problem is described as an optimal control problem with the goal of minimizing the weighted power loss and switching actions of the system, and a grid uniformity index is constructed as an auxiliary decision-making tool. S3: Construct a DQN-based agent to dynamically adjust the population splitting ratio; S4: Input load and fault information to perform preliminary iteration work; S5: Constructing a dual-population evolutionary strategy based on sex determination method; S6: Adjust and iterate the scaling parameters based on the results of the dual-population evolution strategy; S7: Conduct relevant tests in system scenarios to verify the effectiveness of the algorithm.

[0007] According to the above scheme, the specific steps in step S1 are as follows: The electrical system under test is divided into different electrical zones, and power is delivered to the load zone through a set of busbars arranged along the port and starboard sides; The basic network topology of the power system under test is preserved, and the load importance is divided into important loads, semi-important loads and non-important loads according to the level. The load area is powered by two buses. Important and semi-important loads have two power supply paths: normal and backup, while non-important loads have only one power supply path. The topology of the power system under test is represented by a directed graph, which includes vertices V and edges E, where vertex V represents a circuit, distribution board or power source, and edge E represents a load switch.

[0008] According to the above scheme, in step S2, the weighted power loss F1 is used as an important indicator to quantify the load recovery level after a fault:

[0009]

[0010]

[0011] In the formula, , and These represent the actual power loss of critical loads, semi-critical loads, and non-critical loads, respectively. , and These represent the weights of critical, semi-critical, and non-critical loads, respectively. , and These represent the load statuses of critical loads, semi-critical loads, and non-critical loads, where 0 indicates the load is powered on and 1 indicates the load is powered off. Switching action F2 is an important indicator for measuring the effectiveness and recovery speed of fault reconstruction schemes.

[0012] In the formula, This indicates the switching status of the power supply path and branches before the fault. This represents the switching state of the power supply path and branch at the current moment. The minimum Euclidean distance between them is used as the scalar value for the switching action.

[0013] According to the above scheme, in step S2, a decision-making method based on the power grid uniformity index is used to assist in the scheme decision of the Pareto solution set; let the load rate of the system lines be... Among them, the line load rate As shown in Equation 5: .

[0014] According to the above scheme, the specific steps in step S3 are as follows: The agent is used to monitor the dynamic changes of population diversity indicators in real time. When it is observed that the solution spacing converges and the entropy value increases, it determines that the current ratio parameter has reached the optimal configuration and achieves adaptive adjustment of the population ratio. The agent's state is divided into three states based on the decrease, constant, and increase of the population solution spacing value and the decrease, constant, and increase of the entropy value. Each state space consists of [number] states. , The size is represented by, where , They represent the first The solution spacing and entropy of the generation population and The solution spacing and entropy of the initial population are represented; The agent's adjustment of the population proportion parameter is defined as an action, including increasing, keeping it unchanged, and decreasing:

[0015] In the formula, , The first generation and first The splitting ratio parameter of the generation population; Solution spacing and entropy metric are used as evaluation indicators of solution diversity, and the agent's reward function is determined based on the changes in solution spacing and entropy metric values. ; In the solution gap metric, the number of Pareto-property-compliant frontier solutions found by the algorithm is set to... Solution spacing index Defined as:

[0016] in

[0017] In the formula It is the mean of the distances between individuals in the solution set. This represents the number of objective functions; The smaller the value, the more uniform the distribution of solutions and the better the population diversity. In entropy measurement, the population Division ,in For the number of partitions, ≤ ≤ The probability of an individual falling into the population is:

[0018] In the formula, This represents the probability that individual i falls into the i-th partition; Let N represent the number of individuals in the i-th partition; N represents the size of the entire population; the formula for calculating the population diversity entropy is:

[0019] The higher the entropy value, the more dispersed and uniform the distribution of individuals in the population, and the better the population diversity. The reward function R of the agent is determined based on the change in solution spacing and entropy value:

[0020] .

[0021] According to the above scheme, the specific steps in step S4 are as follows: After initializing the algorithm-related parameters, set the algorithm parameter to: number of iterations. Initial population proportion parameters Population size Crossover probability Probability of mutation Neural network hyperparameters Learning rate and discount rate ; Generate the initial population and the distance between solutions in the initial population, as well as the entropy metric. Perform fast non-dominated sorting and crowding calculation.

[0022] According to the above scheme, the specific steps in step S5 are as follows: The initial population is divided into two subpopulations based on the population size parameter and the sex distinction rule, and different genetic operation operators are applied to the two subpopulations. The first step is a population segmentation process based on sex determination. First, a random number of individuals of size S are generated, the reproductive capacity of each individual is calculated, and then individuals are classified into reproductive individuals with a reproductive capacity greater than the average and ordinary individuals based on their reproductive capacity. Then, the population is divided, and the first subpopulation selects individuals with strong reproductive capacity from among the reproductive individuals. If the population is insufficient, individuals with strong reproductive capacity from the ordinary population will be selected to supplement it, and the remaining individuals will form a second subpopulation. For chromosome population size, The population is divided into segments. For the first subpopulation with strong individual reproductive capacity, single-point crossover and single-point mutation are used to maintain the overall advantage of the population, and then the population is updated through an elite strategy. For the second subpopulation with weak individual reproductive capacity, multi-point crossover combined with gene sequence reversal mutation strategy is used to improve the population evolution efficiency, and then the population is updated through an elite strategy.

[0023] According to the above scheme, the specific steps in step S6 are as follows: Merge the updated population and determine if the termination condition is met; If not satisfied, calculate the inter-solution distance and entropy of the population, select an action to adjust the scaling parameter, calculate the reward, update the neural network parameters, then perform fast non-dominated sorting and crowding calculation and return to step S5 for iteration; If the conditions are met, output the Pareto optimal solution set.

[0024] Dynamic Reconstruction System for Multi-Objective Ship Power System Fault Networks Based on DRL The topology submodule is used to topologicalize the circuit diagram of the power system under test and convert it into a corresponding directed graph; The reconfiguration submodule is used to describe the circuit dynamic fault reconfiguration problem as an optimal control problem with the goal of minimizing the weighted power loss and switching actions of the system, and to construct a grid uniformity index as an auxiliary decision-making tool. The agent submodule is used to construct a DQN-based agent for dynamically adjusting the population splitting ratio. The input submodule is used to input load information and fault information for preliminary iteration work; The evolutionary submodule is used to construct a two-population evolutionary strategy based on sex determination. The parameter submodule is used to adjust and iterate the scaling parameters based on the results of the dual-population evolution strategy. The verification submodule is used to perform corresponding tests in system scenarios to verify the effectiveness of the algorithm.

[0025] A computer memory storing a computer program executable by a computer processor, the computer program executing a dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL.

[0026] The beneficial effects of this invention are as follows: 1. The present invention provides a dynamic reconstruction method and system for fault networks of multi-objective ship power systems based on DRL. By simplifying the circuit diagram into a directed graph, adjusting the population ratio parameter as the action selection of the deep reinforcement learning algorithm, determining the reward function based on the changes in solution spacing and entropy, and introducing line load uniformity as an auxiliary evaluation index, the method achieves the function of accelerating the convergence speed of the algorithm.

[0027] 2. This invention introduces the DQN (Deep Q Network) algorithm to monitor the dynamic changes of population diversity indicators in real time, dynamically adjust the population ratio and guide the direction of evolution during iteration, thereby accelerating the convergence speed of the algorithm.

[0028] 3. This invention introduces a dual-population evolutionary strategy, overcoming the problem of most algorithms relying on a single evolutionary mode. The population is split into two subpopulations with different genetic operations, improving the evolutionary guidance accuracy and adaptability of the NSGA-II algorithm. By implementing the dual-population evolutionary strategy, the algorithm preserves the advantages of the existing population while increasing the diversity of offspring, thus expanding the search dimension and overcoming local extrema limitations. Simultaneously, the real-time monitoring of population indicators by the agent dynamically adjusts the population splitting ratio parameter, accelerating the algorithm's convergence speed.

[0029] 4. In the context of dynamic fault reconstruction in an eight-zone MVDC, the present invention improves the reconstruction time and provides better and more efficient solution performance.

[0030] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of an embodiment of the present invention.

[0033] Figure 2 This is a directed graph representing a real-world scenario according to an embodiment of the present invention.

[0034] Figure 3This is a flowchart of the algorithm in an embodiment of the present invention.

[0035] Figure 4 The graph shows the convergence iteration count results of the ablation experiments for the standard NSGA-II, static dual-population NSGA-II, random proportional NSGA-II, and DRL_DPES_NSGA-II algorithms according to embodiments of the present invention.

[0036] Figure 5 This is a graph showing the ablation experiment solution spacing results of the standard NSGA-II, static dual-population NSGA-II, and DRL_DPES_NSGA-II algorithms according to embodiments of the present invention.

[0037] Figure 6 This is a graph showing the entropy measurement results of ablation experiments using the standard NSGA-II, static dual-population NSGA-II, and DRL_DPES_NSGA-II algorithms according to embodiments of the present invention.

[0038] Figure 7 This is a graph showing the weighted power loss results of the MOEA / D, NSGA-II, and DRL_DPES_NSGA-II algorithms in embodiments of the present invention.

[0039] Figure 8 This is a graph showing the convergence iteration count results of the MOEA / D, NSGA-II, and DRL_DPES_NSGA-II algorithms in embodiments of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0041] Example 1 See Figure 1 The specific steps of the dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL are as follows: S1: Topologicalize the circuit diagram of the eight-zone medium-voltage DC ship power system and convert it into a corresponding directed graph; This embodiment selects an eight-zone medium-voltage DC marine power system (MVDC SPS) as the experimental scenario. The marine power system (SPS) is divided into different electrical zones, and power is transmitted to the loads in the load zones through a set of buses arranged along the port and starboard sides. This invention retains the basic network topology of the marine power station system, while classifying loads into important loads, semi-important loads, and non-important loads according to their importance. The load zones are powered by two buses. Important and semi-important loads have two power supply paths: normal and backup, while non-important loads have only one power supply path. Please refer to Figure 2 for a practical example of a directed graph. This embodiment uses a directed graph to represent the topology of a ship's power station system. The directed graph includes vertices V and edges E, where vertex V represents a circuit, distribution board, or power source, and edge E represents a load switch.

[0042] S2: The circuit dynamic fault reconfiguration problem is described as an optimal control problem with the goal of minimizing the weighted power loss and switching actions of the system, and a grid uniformity index is constructed as an auxiliary decision-making tool. For the multi-objective function optimization problem of dynamic reconfiguration of fault network in ship power system, this embodiment, based on full consideration of load priority, formulates the dynamic fault reconfiguration problem as an optimal control problem with the objectives of minimizing the weighted power loss of the system and minimizing the switching action, and constructs a multi-objective optimization model.

[0043] Ship electrical system loads can be categorized into three types: critical loads, semi-critical loads, and non-critical loads. In different mission scenarios, the same load may experience different priority variations.

[0044] The first optimization objective in this embodiment is the weighted power loss F1, which is an important indicator used to quantify the load recovery level after a fault, as shown in Equation 1:

[0045]

[0046]

[0047] In the formula, , and These represent the actual power loss of critical loads, semi-critical loads, and non-critical loads, respectively. , and These represent the weights of critical, semi-critical, and non-critical loads, respectively. , and These represent the load states of critical loads, semi-critical loads, and non-critical loads, respectively, where 0 indicates that the load is powered on and 1 indicates that the load is powered off, as shown in Equations 2 and 3.

[0048] The second optimization objective in this embodiment is the switching operation cost F2, which is an important indicator for measuring the effectiveness and recovery speed of the fault reconstruction scheme, as shown in Equation 4:

[0049] In the formula, This indicates the switching status of the power supply path and branches before the fault. This represents the current switching state of the power supply path and branch, and the minimum Euclidean distance between them is used as a scalar value for the switching cost.

[0050] This embodiment proposes a decision-making method based on the power grid uniformity index to assist in the scheme decision problem of Pareto solution set. Let the load rate of the system lines be... Among them, the line load rate As shown in Equation 5:

[0051] S3: Construct a DQN (Deep Q Network)-based agent to dynamically adjust the population splitting ratio; The agent will monitor the dynamic changes of population diversity indicators (solution gap and entropy measurement) in real time. When it observes that the solution gap converges and the entropy value increases, it determines that the current ratio parameter has reached the optimal configuration and achieves adaptive adjustment of the population ratio.

[0052] State space: Based on the three states of decreasing, constant, and increasing population solution spacing and the three states of decreasing, constant, and increasing entropy, the state space is divided into... Each state space consists of [number] states. , The size is represented by, where , They represent the first The solution spacing and entropy of the generation population and The solution spacing and entropy of the initial population are represented.

[0053] Action Space: The agent's adjustment of the population proportion parameter is defined as an action. In this embodiment, the actions include three types: increase, remain unchanged, and decrease. The calculation formula is shown in Equation 6:

[0054] In the formula, , The first generation and first The splitting ratio parameter of the generation population.

[0055] Reward function: Two metrics, solution spacing and entropy, are used to evaluate the diversity of solutions. The reward function of the agent is determined based on the changes in solution spacing and entropy. .

[0056] In the spacing metric, the number of Pareto-type frontier solutions found by the algorithm is set to [value missing]. Solution spacing index The definition is shown in Equation 7:

[0057] in

[0058] In the formula It is the mean of the distances between individuals in the solution set. This represents the number of objective functions. The smaller the value, the more evenly the solutions are distributed, and the better the population diversity.

[0059] In entropy, the population is... Division ,in For the number of partitions, ≤ ≤ The probability of an individual falling into the population is shown in Equation 9:

[0060] In the formula, This represents the probability that individual i falls into the i-th partition. Let represent the number of individuals in the i-th partition; N represents the size of the entire population. The formula for calculating the population diversity entropy is shown in Equation 10:

[0061] The higher the entropy value, the more dispersed and uniform the distribution of individuals in the population, and the better the population diversity. The reward function R of the agent is determined based on the change in solution spacing and entropy value, as shown in Equations 11 and 12:

[0062]

[0063] S4: Input the ship's electrical system load information and fault-related information to perform preliminary iteration work; After initializing the algorithm-related parameters, set the algorithm parameter to: number of iterations. Initial population proportion parameters Population size Crossover probability Probability of mutation Neural network hyperparameters Learning rate and discount rate The initial population is generated, along with the inter-solution distance and entropy metric. Then, a fast non-dominated sort and crowding calculation are performed.

[0064] S5: Constructing a dual-population evolutionary strategy based on sex determination method; Based on population size parameters and sex differentiation rules, the initial population is divided into two subpopulations, and different genetic operations are applied to these two subpopulations. The first step is a population segmentation process based on sex determination. First, S individuals are randomly generated, and the reproductive capacity of each individual is calculated. Then, individuals are classified into reproductive individuals (those with reproductive capacity greater than the average) and ordinary individuals based on their reproductive capacity. Next, population segmentation is performed, with subpopulation 1 selecting individuals with higher reproductive capacity from among the reproductive individuals. If the population is insufficient, individuals with strong reproductive capacity from the ordinary population will be selected to supplement the population, and the remaining individuals will form subpopulation 2. For chromosome population size, The population segmentation ratio is determined. For population 1, which has a strong individual reproductive capacity, a single-point crossover and single-point mutation approach is used to maintain the overall advantage of the population, and then the population is updated through an elite strategy. For population 2, which has a weak individual reproductive capacity, a multi-point crossover combined with gene sequence reversal mutation strategy is used to improve the population evolution efficiency, and then the population is updated through an elite strategy.

[0065] S6: Adjust and iterate the scaling parameters based on the results of the dual-population evolution strategy; The updated population is merged, and the termination condition is checked. If not, the inter-solution distance and entropy of the population are calculated, and the action is selected to adjust the scaling parameters, calculate the reward, update the neural network parameters, perform fast non-dominated sorting and crowding calculation, and return to step S5 for iteration. If the condition is met, the Pareto optimal solution set is output. The overall algorithm flow is as follows: Figure 3 .

[0066] S7: The algorithm was tested and its effectiveness was verified in the context of an eight-zone medium-voltage DC ship power system.

[0067] This embodiment simplifies the circuit diagram into a directed graph, uses the adjustment of the population ratio parameter as the action selection for the deep reinforcement learning algorithm, determines the reward function based on the changes in solution spacing and entropy, and introduces line load uniformity as an auxiliary evaluation index, thereby accelerating the convergence speed of the algorithm.

[0068] Example 2 The steps in this embodiment are the same as in Embodiment 1, except that each step is applied to a specific instance. Specifically, it includes the following steps: This embodiment was tested in a typical eight-zone mid-voltage direct current (MVDC) power system to verify its effectiveness. The proposed method was implemented on a laptop equipped with a 64-bit Windows 10 operating system. This laptop was based on an AMD Ryzen 7 5800H processor with Radeon Graphics, a clock speed of 3.20 GHz, and 16 GB of memory. The algorithm language used was Python, and the PyTorch deep learning framework was used to model and iteratively train the neural network.

[0069] Regarding algorithm parameters: the population size was set to 200, the initial population ratio was 0.5, the crossover probability was 0.8, the mutation probability was 0.1, the learning rate was 0.01, and the discount factor was 0.9. The corresponding experimental fault scenario was a single-main-generator fault scenario. The number of iterations required for different algorithms to reach the standard solution, the solution spacing and entropy measurement, and the obtained Pareto optimal solution set were recorded, and the results were compared and analyzed to verify the effectiveness and superiority of the algorithms.

[0070] This embodiment introduces a dual-population evolutionary strategy to overcome the problem of most algorithms relying on a single evolutionary mode. The population is split into two subpopulations with different genetic operations, improving the evolutionary guidance accuracy and adaptability of the NSGA-II algorithm. By implementing the dual-population evolutionary strategy, the algorithm can increase the diversity of offspring while preserving the advantages of the existing population, thus expanding the search dimension and overcoming local extrema limitations. Simultaneously, the real-time monitoring of population indicators by the agent dynamically adjusts the population splitting ratio parameter, accelerating the algorithm's convergence speed.

[0071] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0072] Example 3 This embodiment is used to implement the principle of the above method embodiment to construct a dynamic reconfiguration system for fault networks of multi-objective ship power systems based on DRL, including a topology submodule, a reconfiguration submodule, an intelligent agent submodule, an input submodule, an evolution submodule, a parameter submodule, and a verification submodule; The topology submodule is used to topologicalize the circuit diagram of the power system under test and convert it into a corresponding directed graph; The reconfiguration submodule is used to describe the circuit dynamic fault reconfiguration problem as an optimal control problem with the goal of minimizing the weighted power loss and switching actions of the system, and to construct a grid uniformity index as an auxiliary decision-making tool. The agent submodule is used to construct a DQN-based agent for dynamically adjusting the population splitting ratio. The input submodule is used to input load information and fault information for preliminary iteration work; The evolutionary submodule is used to construct a two-population evolutionary strategy based on sex determination. The parameter submodule is used to adjust and iterate the scaling parameters based on the results of the dual-population evolution strategy. The verification submodule is used to perform corresponding tests in system scenarios to verify the effectiveness of the algorithm.

[0073] Each submodule is mainly used to implement the various steps of the method implementation, which will not be elaborated here.

[0074] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0075] This embodiment also includes a processor, a communication interface, a memory, and a communication bus; wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the program is executed by the processor, the processor performs the steps of the dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL.

[0076] This embodiment also provides a computer-readable storage medium storing executable instructions that, when executed by a processor, enable the processor to implement a dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL.

[0077] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0078] Furthermore, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0079] This application is described with reference to the flowchart of the method and computer program product according to Embodiment 1 and the block diagram of the device (system) according to Embodiment 3. It should be understood that each step or block in the flowchart or block diagram, as well as combinations of steps or blocks in the flowchart or block diagram, can be implemented by computer program instructions.

[0080] These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which are executable by the processor of the computer or other programmable data processing device, produce instructions for implementing the process. Figure 1 One or more processes or boxes Figure 1 A dynamic reconfiguration system for fault networks of multi-objective ship electrical systems based on DRL, specifying functions in one or more boxes.

[0081] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes or boxes Figure 1 The function specified in one or more boxes.

[0082] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes or boxes Figure 1 The steps of the dynamic reconfiguration method for fault networks of multi-objective ship electric power systems based on DRL are specified in one or more boxes.

[0083] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A dynamic reconfiguration method for fault networks in multi-objective ship electrical systems based on DRL, characterized in that: Includes the following steps: S1: Topologicalize the circuit diagram of the power system under test and convert it into a corresponding directed graph; S2: The circuit dynamic fault reconfiguration problem is described as an optimal control problem with the goal of minimizing the weighted power loss and switching actions of the system, and a grid uniformity index is constructed as an auxiliary decision-making tool. S3: Construct a DQN-based agent to dynamically adjust the population splitting ratio; S4: Input load and fault information to perform preliminary iteration work; S5: Constructing a dual-population evolutionary strategy based on sex determination method; S6: Adjust and iterate the scaling parameters based on the results of the dual-population evolution strategy; S7: Conduct relevant tests in system scenarios to verify the effectiveness of the algorithm.

2. The dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL as described in claim 1, characterized in that: The specific steps in step S1 are as follows: The electrical system under test is divided into different electrical zones, and power is delivered to the load zone through a set of busbars arranged along the port and starboard sides; The basic network topology of the power system under test is preserved, and the load importance is divided into important loads, semi-important loads and non-important loads according to the level. The load area is powered by two buses. Important and semi-important loads have two power supply paths: normal and backup, while non-important loads have only one power supply path. The topology of the power system under test is represented by a directed graph, which includes vertices V and edges E, where vertex V represents a circuit, distribution board or power source, and edge E represents a load switch.

3. The dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL as described in claim 1, characterized in that: In step S2, the weighted power loss F1 is used as an important indicator to quantify the load recovery level after a fault. In the formula, , and These represent the actual power loss of critical loads, semi-critical loads, and non-critical loads, respectively. , and These represent the weights of critical, semi-critical, and non-critical loads, respectively. , and These represent the load statuses of critical loads, semi-critical loads, and non-critical loads, where 0 indicates the load is powered on and 1 indicates the load is powered off. Switching action F2 is an important indicator for measuring the effectiveness and recovery speed of fault reconstruction schemes. In the formula, This indicates the switching status of the power supply path and branches before the fault. This represents the switching state of the power supply path and branch at the current moment. The minimum Euclidean distance between them is used as the scalar value for the switching action.

4. The dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL as described in claim 1, characterized in that: In step S2, a decision-making method based on the power grid uniformity index is used to assist in the scheme decision of the Pareto solution set; let the load rate of the system lines be... Among them, the line load rate As shown in Equation 5: 。 5. The dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL according to claim 1, characterized in that: The specific steps in step S3 are as follows: The agent is used to monitor the dynamic changes of population diversity indicators in real time. When it is observed that the solution spacing converges and the entropy value increases, it determines that the current ratio parameter has reached the optimal configuration and achieves adaptive adjustment of the population ratio. The agent's state is divided into three states based on the decrease, constant, and increase of the population solution spacing value and the decrease, constant, and increase of the entropy value. Each state space consists of [number] elements. , The size is represented by, where , They represent the first The solution spacing and entropy of the generation population and The solution spacing and entropy of the initial population are represented; The agent's adjustment of the population proportion parameter is defined as an action, including increasing, keeping it unchanged, and decreasing: In the formula, , The first generation and first The splitting ratio parameter of the generation population; Solution spacing and entropy metric are used as evaluation indicators of solution diversity, and the agent's reward function is determined based on the changes in solution spacing and entropy metric values. ; In the solution gap metric, the number of Pareto-property-compliant frontier solutions found by the algorithm is set to... Solution spacing index Defined as: in In the formula It is the mean of the distances between individuals in the solution set. This represents the number of objective functions; The smaller the value, the more uniform the distribution of solutions and the better the population diversity. In entropy measurement, the population Division ,in For the number of partitions, ≤ ≤ The probability of an individual falling into the population is: In the formula, This represents the probability that individual i falls into the i-th partition; Let N represent the number of individuals in the i-th partition; N represents the size of the entire population; the formula for calculating the population diversity entropy is: The higher the entropy value, the more dispersed and uniform the distribution of individuals in the population, and the better the population diversity. The reward function R of the agent is determined based on the change in solution spacing and entropy value: 。 6. The dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL as described in claim 1, characterized in that: The specific steps in step S4 are as follows: After initializing the algorithm-related parameters, set the algorithm parameter to: number of iterations. Initial population proportion parameters Population size Crossover probability Probability of mutation Neural network hyperparameters Learning rate and discount rate ; Generate the initial population and the distance between solutions in the initial population, as well as the entropy metric. Perform fast non-dominated sorting and crowding calculation.

7. The dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL according to claim 1, characterized in that: The specific steps in step S5 are as follows: The initial population is divided into two subpopulations based on the population size parameter and the sex distinction rule, and different genetic operation operators are applied to the two subpopulations. The first step is a population segmentation process based on sex determination. First, a random number of individuals of size S are generated, the reproductive capacity of each individual is calculated, and then individuals are classified into reproductive individuals with a reproductive capacity greater than the average and ordinary individuals based on their reproductive capacity. Then, the population is divided, and the first subpopulation selects individuals with strong reproductive capacity from among the reproductive individuals. If the population is insufficient, individuals with strong reproductive capacity from the ordinary population will be selected to supplement it, and the remaining individuals will form a second subpopulation. For chromosome population size, The population is divided into segments. For the first subpopulation with strong individual reproductive capacity, single-point crossover and single-point mutation are used to maintain the overall advantage of the population, and then the population is updated through an elite strategy. For the second subpopulation with weak individual reproductive capacity, multi-point crossover combined with gene sequence reversal mutation strategy is used to improve the population evolution efficiency, and then the population is updated through an elite strategy.

8. The dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL according to claim 1, characterized in that: The specific steps in step S6 are as follows: Merge the updated population and determine if the termination condition is met; If not satisfied, calculate the inter-solution distance and entropy of the population, select an action to adjust the scaling parameter, calculate the reward, update the neural network parameters, then perform fast non-dominated sorting and crowding calculation and return to step S5 for iteration; If the conditions are met, output the Pareto optimal solution set.

9. A dynamic reconfiguration system for fault networks of multi-objective ship electrical systems based on DRL, characterized in that: The topology submodule is used to topologicalize the circuit diagram of the power system under test and convert it into a corresponding directed graph; The reconfiguration submodule is used to describe the circuit dynamic fault reconfiguration problem as an optimal control problem with the goal of minimizing the weighted power loss and switching actions of the system, and to construct a grid uniformity index as an auxiliary decision-making tool. The agent submodule is used to construct a DQN-based agent for dynamically adjusting the population splitting ratio. The input submodule is used to input load information and fault information for preliminary iteration work; The evolutionary submodule is used to construct a two-population evolutionary strategy based on sex determination. The parameter submodule is used to adjust and iterate the scaling parameters based on the results of the dual-population evolution strategy. The verification submodule is used to perform corresponding tests in system scenarios to verify the effectiveness of the algorithm.

10. A computer memory, characterized in that: It contains a computer program that can be executed by a computer processor, which performs the dynamic reconfiguration method for fault networks of multi-objective ship power systems based on DRL as described in any one of claims 1 to 8.

Images