Reliability index distribution method and device of power system

Abstract

The invention provides a reliability index distribution method and device for a power system, and is applied to the technical field of reliability design of the power system, and the method comprises the steps: obtaining the working condition characteristics and structure information of each part of the power system; inputting the working condition characteristics of each component into a trained severity quantification model based on a gradient boosting tree model to obtain a severity score of each component; inputting the component attribute and the connection relation of each component of the power system into a trained influence degree quantification model based on a graph neural network model to obtain an influence degree score of each component; and adopting a preset optimization algorithm to determine a reliability index distribution result of the power system based on the severity score of each component and the influence degree score of each component under a target constraint. According to the invention, the reliability design efficiency and accuracy of the system can be improved.

Description

Technical Field

[0001] This invention relates to the field of power system reliability design technology, and in particular to a method and apparatus for allocating reliability indicators for a power system. Background Technology

[0002] The reliability of a power system is crucial to its safety and performance. Traditional reliability design methods employ a top-down, experience-based decomposition model: first, the overall system reliability target is set, then engineers allocate the targets to each subsystem / component based on experience, followed by simulation verification and multiple rounds of manual iterative adjustments.

[0003] The above methods have the following significant drawbacks: the initial allocation lacks objective data support, the process is opaque, and the design quality is unstable. The "allocation-evaluation-adjustment" cycle is time-consuming and laborious, especially in complex systems, where finding the optimal solution is extremely inefficient. It is difficult to accurately quantify the severity of component operating conditions and their impact on the system, making it impossible to specifically strengthen key weak links, which can easily lead to design redundancy or inadequacy.

[0004] This shows that the allocation of reliability indicators for power systems in related technologies suffers from a technical problem of relying too heavily on subjective experience. Summary of the Invention

[0005] This invention provides a method and apparatus for allocating reliability indicators for a power system, which addresses the shortcomings of existing power system reliability indicator allocation methods that rely too heavily on subjective experience, thereby improving the efficiency and accuracy of system reliability design.

[0006] This invention provides a method for allocating reliability indicators for a power system, comprising the following steps: acquiring the operating characteristics and structural information of each component of the power system; inputting the operating characteristics of each component into a trained severity metric model based on a gradient boosting tree model to obtain a severity score for each component; inputting the component attributes and connectivity relationships of each component of the power system into a trained influence metric model based on a graph neural network model to obtain an influence score for each component; and using a preset optimization algorithm, under objective constraints, determining the reliability indicator allocation result of the power system based on the severity score and influence score of each component.

[0007] According to a method for allocating reliability indicators for a power system provided by the present invention, before acquiring the operating condition characteristics and structural information of each component of the power system, the method further includes: acquiring multi-source heterogeneous data, wherein the multi-source heterogeneous data includes historical operating data, simulation calculation data, and experimental test data of the power system; preprocessing and labeling the multi-source heterogeneous data to obtain a labeled dataset of the power system; and training a preset rigorous metric model based on a gradient boosting tree model and a preset impact metric model based on a graph neural network model based on the labeled dataset.

[0008] According to a method for allocating reliability indicators of a power system provided by the present invention, the method employs a preset optimization algorithm to determine the reliability indicator allocation result of the power system based on the severity score and the influence score of each component under objective constraints. The method includes: constructing a multi-objective constraint optimization model, wherein the objective function of the multi-objective constraint optimization model is to maximize the reliability of the power system, and the objective constraints of the multi-objective constraint optimization model include total cost constraints and total weight constraints; using the severity score and the influence score as optimization direction guidance parameters, employing an intelligent optimization algorithm to iteratively search under the objective constraints, dynamically adjusting the reliability indicator allocation values ​​of each component, and determining the optimal allocation combination of reliability indicators for each component as the reliability indicator allocation result of the power system; wherein a fast reliability evaluator is used to evaluate the reliability indicator allocation values ​​of each component obtained in each iterative search.

[0009] According to the present invention, a method for allocating reliability indicators of a power system is provided. The method employs a preset optimization algorithm, wherein the intelligent optimization algorithm is a deep reinforcement learning algorithm. The method uses the severity score and the impact score as optimization direction guidance parameters, and employs the deep reinforcement learning algorithm to iteratively search under the target constraint, dynamically adjusting the reliability indicator allocation values ​​of each component to determine the optimal allocation combination of reliability indicators for each component. This includes: calling an agent to obtain the current state, which includes the current reliability indicator allocation values ​​of each component, the current total cost, and the current total weight; the agent, based on a policy network, selects to adjust the reliability indicator allocation values ​​of the target component as an action; inputting the action to a fast reliability evaluator serving as the environment to obtain a new state and a reward value, wherein the new state includes the updated reliability indicator allocation values ​​of each component, the updated total cost, and the updated total weight, and the reward value is used to evaluate the contribution of the action to optimizing the objective function; updating the policy network based on the interaction experience formed by the current state, the action, the reward value, and the new state; repeating the above process until the optimal allocation combination of reliability indicators for each component that satisfies the target constraint and maximizes the reliability of the power system is obtained.

[0010] According to the present invention, a method for allocating reliability indicators for a power system is provided, wherein the fast reliability evaluator is a surrogate model or a Bayesian network.

[0011] According to a method for allocating reliability indicators for a power system provided by the present invention, after determining the reliability indicator allocation result of the power system based on the severity score and the influence score of each component under target constraints using a preset optimization algorithm, the method further includes: verifying the reliability indicator allocation result of the power system through a preset high-fidelity simulation model to obtain a verification result; and visually displaying the reliability indicator allocation result and the verification result.

[0012] The present invention also provides a reliability index allocation device for a power system, comprising the following modules: an acquisition module for acquiring the operating characteristics and structural information of each component of the power system; a severity scoring module for inputting the operating characteristics of each component into a trained severity quantification model based on a gradient boosting tree model to obtain a severity score for each component; an influence scoring module for inputting the component attributes and connection relationships of each component of the power system into a trained influence quantification model based on a graph neural network model to obtain an influence score for each component; and an index allocation module for using a preset optimization algorithm, under objective constraints, to determine the reliability index allocation result of the power system based on the severity score and influence score of each component.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the reliability index allocation method of any of the above-described power systems.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the reliability index allocation method for the power system as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the reliability index allocation method for any of the above-described power systems.

[0016] The reliability index allocation method and apparatus for a power system provided by this invention firstly acquires the operating characteristics and structural information of each component, laying a comprehensive and reliable data foundation for subsequent analysis; secondly, it uses a gradient boosting tree model to intelligently analyze the operating characteristics, generating objective and quantitative severity scores, effectively overcoming the subjectivity of traditional experience-based judgments; next, it employs a graph neural network model to perform deep learning on the system topology, accurately quantifying the system impact of each component, breaking through the limitation of traditional methods that ignore the correlation effects between components; finally, it uses an intelligent optimization algorithm to transform the severity and impact scores into the optimal allocation scheme under multi-objective constraints, achieving precise allocation of design resources while ensuring system reliability, significantly improving the scientific nature, efficiency, and accuracy of reliability design. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the reliability index allocation method for a power system provided by the present invention.

[0019] Figure 2 This is an overall flowchart of the reliability index allocation method for the power system provided by the present invention.

[0020] Figure 3 This is a flowchart illustrating the quantification of key factors provided by the present invention.

[0021] Figure 4 This is a flowchart of the intelligent optimization allocation process provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the module of the reliability index allocation device for the power system provided by the present invention.

[0023] Figure 6 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] This invention provides a method and apparatus for allocating reliability indicators for a power system. By constructing a multi-source dataset, using an artificial intelligence model to quantify key influencing factors, and employing an intelligent optimization algorithm to automatically search for the optimal reliability indicator allocation scheme, a paradigm shift from "experience-based design" to "scientific design" is achieved, thereby improving the efficiency and accuracy of system reliability design.

[0026] Figure 1 This is a flowchart illustrating the reliability index allocation method for a power system provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps.

[0027] Step 101: Obtain the operating characteristics and structural information of each component of the power system.

[0028] In this embodiment of the invention, real-time data of various components of the power system are acquired, including historical operating data, simulation calculation data, and experimental test data.

[0029] For each component, operating condition characteristics are calculated from time series data, including statistics (such as mean, variance, and peak value) and dynamic features (such as trend slope within a sliding window) to quantify the severity of the service environment.

[0030] Structural information is represented by topological features, which abstract the dynamic system into a graph structure (nodes represent components and edges represent connections). Topological attributes, such as the degree centrality and clustering coefficients of nodes, are parsed from the system block diagram or BOM table.

[0031] Step 102: Input the working condition features of each component into the trained severity quantification model based on the gradient boosting tree model to obtain the severity score of each component.

[0032] In this embodiment of the invention, the gradient boosting tree model is used as the core algorithm. The gradient boosting tree model achieves high-precision regression prediction by integrating multiple weak decision trees, which can effectively capture the nonlinear relationship between working condition characteristics and severity.

[0033] The input layer of the severity quantification model is designed to receive standardized operating condition features, such as statistical quantities (mean, variance, extreme values) and dynamic features (trend rate of change) of parameters like temperature, pressure, and vibration. The output layer is configured with continuous numerical nodes, corresponding to standardized severity scores (range 0-1).

[0034] In the inference application phase of the severity quantification model, the working condition features of each component are input into the trained severity quantification model based on the gradient boosting tree model for forward computation. The model combines decision paths through a tree structure and finally outputs the severity score of each component.

[0035] For example, for turbine components operating in high-temperature environments, their sustained high-temperature characteristics will trigger the corresponding decision rules within the model, resulting in a higher severity score.

[0036] The post-processing stage of the scoring standardizes and calibrates the severity scores, using Min-Max scaling to uniformly map the scores to the 0-1 range, and determining the severity level by percentile sorting.

[0037] Step 103: Input the component attributes and connection relationships of each component of the power system into the trained influence measurement model based on the graph neural network model to obtain the influence score of each component.

[0038] In this embodiment of the invention, the graph neural network model selects graph convolutional network (GCN) or graph attention network (GAT) as the basic architecture. These models can effectively aggregate neighborhood information through message passing mechanism and learn the importance representation of nodes in the global topology.

[0039] The component attributes and connectivity relationships of each part of the power system are input into a pre-trained influence metric model based on a graph neural network. Through multi-layer graph convolution operations, each node (component) obtains a high-dimensional embedding vector. This embedding vector contains comprehensive information about the node's own attributes and its topological position in the system.

[0040] By using fully connected layers and the Softmax activation function, the embedding vectors are mapped to standardized impact scores (ranging from 0 to 1). A higher score indicates a greater impact of the component on system reliability; for example, components located on the critical path or with high connectivity typically receive higher scores.

[0041] Step 104: Using a preset optimization algorithm, under the objective constraints, the reliability index allocation result of the power system is determined based on the severity score and influence score of each component.

[0042] In this embodiment of the invention, a mathematical model for a multi-objective constrained optimization model needs to be constructed. Maximizing the reliability of the power system is set as the primary objective function, which can be quantified as maximizing the mean time between failures (MTBF) of the power system. Constraints include total cost not exceeding the budget limit, total weight meeting design requirements, and other engineering constraints such as spatial layout limitations.

[0043] Severity score and impact score are used as key input parameters and are integrated into the weight coefficients of the optimization model. Components with higher severity scores or larger impact scores are assigned higher weight coefficients in the objective function to ensure that the optimization process prioritizes key weak points.

[0044] For example, components that have both high severity and high impact scores will receive higher optimization priority in reliability index allocation.

[0045] A variety of preset intelligent optimization algorithms can be used, including genetic algorithm (GA), Bayesian optimization (BO) or deep reinforcement learning (DRL).

[0046] Taking genetic algorithms as an example, the implementation process includes basic operations such as population initialization, selection, crossover, and mutation. Each individual in the population represents a possible reliability index allocation scheme, and the fitness function is calculated based on a weighted combination of the system reliability objective and the degree of constraint violation. Severity scores and impact scores are used to guide the direction of crossover and mutation operations, such as performing a local fine-grained search for index allocation of high-scoring components.

[0047] If a deep reinforcement learning algorithm is used, the allocation process is modeled as a Markov decision process. The agent's state space includes the current reliability index allocation values ​​of each component, the satisfaction status of cost and weight constraints, etc.; the action space consists of the adjustment operations on the reliability index of each component; the reward function design comprehensively considers the improvement of system reliability, cost control effect, and the utilization efficiency of severity and impact scores.

[0048] In the optimization process, surrogate models or Bayesian networks are used as fast evaluators to replace traditional time-consuming simulation calculations. These fast evaluation models can predict system-level reliability indicators and constraint satisfaction in real time based on the current allocation scheme, greatly improving optimization efficiency.

[0049] Optimization algorithms approach the optimal solution through iterative search. For genetic algorithms, the quality of the population is continuously improved through multiple generations of evolution; for reinforcement learning algorithms, the policy network is optimized through a large number of training rounds. Convergence conditions are set during the optimization process, such as the maximum number of iterations or an improvement threshold for the objective function, to ensure that the algorithm terminates within a reasonable time.

[0050] The final output of the reliability index allocation results includes the specific reliability index values ​​of each component, as well as the corresponding overall system reliability prediction value, cost weight and other constraints.

[0051] Through the embodiments of this invention, firstly, by acquiring the operating characteristics and structural information of each component, a comprehensive and reliable data foundation is laid for subsequent analysis; secondly, a gradient boosting tree model is used to intelligently analyze the operating characteristics, generating an objective and quantitative severity score, effectively overcoming the subjectivity of traditional experience-based judgment; nextly, a graph neural network model is used to perform deep learning on the system topology, accurately quantifying the system impact of each component, breaking through the limitation of traditional methods that ignore the correlation effects between components; finally, an intelligent optimization algorithm is used to transform the severity and impact scores into the optimal allocation scheme under multi-objective constraints, achieving precise allocation of design resources while ensuring system reliability, significantly improving the scientificity, efficiency, and accuracy of reliability design.

[0052] According to the reliability index allocation method for a power system provided by the present invention, before obtaining the operating condition characteristics and structural information of each component of the power system, the method further includes: Acquire multi-source heterogeneous data, which includes historical operating data of the power system, simulation calculation data, and experimental test data; A labeled dataset of dynamic systems is obtained by preprocessing and labeling multi-source heterogeneous data. Based on the labeled dataset, a pre-defined rigorous metric model based on a gradient boosting tree model and a pre-defined impact metric model based on a graph neural network model are trained.

[0053] In this embodiment of the invention, historical operating data, simulation calculation data, and experimental test data of the power system are collected and fused; the above-mentioned multi-source data are cleaned, aligned, and normalized preprocessed; based on mechanistic knowledge and data mining technology, feature engineering is performed to extract and construct feature vectors including operating condition features, degradation features, topological features, and historical features; the data are labeled to form a labeled dataset for model training.

[0054] Construct and train a service condition severity quantification model (such as a gradient boosting tree model), with the input being the extracted operating condition features of each component and the output being the standardized severity score of that component.

[0055] Construct and train a system influence quantification model (as shown in the figure neural network GNN model), which abstracts the dynamic system into a graph structure. The input is the component attributes and their connection relationships, and the output is the embedding vector of each node to quantify its system influence.

[0056] In this embodiment of the invention, a harsh metric model and an impact metric model are trained based on a prepared labeled dataset. The harsh metric model employs a gradient boosting tree (GBDT) architecture, and the training process focuses on optimizing hyperparameters such as tree depth, learning rate, and subsampling ratio. The training strategy adopts a phased approach: first, basic training is performed using the majority of the data; then, k-fold cross-validation is used to adjust the model parameters; and finally, the model performance is evaluated using the reserved test set. Model performance is primarily evaluated using mean squared error and coefficient of determination, while feature importance analysis is introduced to verify the physical rationality of the model.

[0057] The impact quantification model is built upon Graph Neural Networks (GNNs), and the training process requires special consideration of the characteristics of graph data. Graph structure construction abstracts the dynamical system as a network composed of nodes (components) and edges (connections). Node features contain component attribute information, and edge weights reflect connection strength. Training employs Graph Convolutional Networks (GCNs) or Graph Attention Networks (GATs) architectures, learning node representations through message passing mechanisms. Neighbor sampling techniques are used during training to handle large-scale graph data and avoid memory overflow issues. Model evaluation not only focuses on node classification accuracy but also verifies the model's understanding of the system's topology by visualizing node embedding results.

[0058] Through the embodiments of the present invention, multi-source heterogeneous data of the power system (covering historical operation, simulation calculation and experimental test data) are obtained, and a labeled dataset is formed after preprocessing and annotation. This dataset is then used to train a rigorous metric model based on a gradient boosting tree model and an impact metric model based on a graph neural network model, providing a more accurate and effective model foundation for subsequent acquisition of component operating condition characteristics and structural information and for carrying out reliability index allocation.

[0059] According to the present invention, a method for allocating reliability indicators for a power system employs a preset optimization algorithm to determine the allocation result of reliability indicators for the power system under objective constraints, based on the severity score and influence score of each component. The method includes: A multi-objective constrained optimization model is constructed, wherein the objective function of the multi-objective constrained optimization model is to maximize the reliability of the dynamic system, and the objective constraints of the multi-objective constrained optimization model include total cost constraints and total weight constraints. Severity score and impact score are used as optimization direction guiding parameters. Intelligent optimization algorithm is used to perform iterative search under target constraints, dynamically adjust the reliability index allocation values ​​of each component, and determine the optimal allocation combination of reliability indexes of each component as the reliability index allocation result of the power system. In this process, a fast reliability evaluator is used to evaluate the reliability index assignment values ​​of each component obtained in each iteration of the search.

[0060] In this embodiment of the invention, the reliability allocation problem is constructed as a multi-objective constrained optimization model (problem), with the objective function being the maximization of the (overall) reliability of the dynamic system. Constraints include total cost, total weight, etc. The output severity score and system impact are used to guide the optimization direction. Intelligent optimization algorithms (such as genetic algorithm GA, Bayesian optimization BO, or deep reinforcement learning DRL algorithm) are used to search for the optimal allocation combination of reliability indices (such as MTBF_i) of each component under the above constraints. In this process, a surrogate model or Bayesian network is used as a fast reliability estimator to approximate the time-consuming high-fidelity simulation and accelerate the optimization cycle.

[0061] In some embodiments, the objective function of the multi-objective constrained optimization model is set as maximizing the overall reliability of the power system, specifically quantified as maximizing the system-level mean time between failures (MTBF_system). The constraint system includes two levels: hard constraints and soft constraints. Hard constraints include that the total cost must not exceed the budget limit and the total weight must meet the design specifications. Soft constraints involve engineering feasibility conditions, such as the reasonable range of reliability indicators for each component. During model construction, severity scores and impact scores are introduced into the objective function as important weighting coefficients. Components with high severity or high impact are given higher priority weights during optimization, ensuring that the optimization direction can accurately focus on the key weaknesses in system reliability.

[0062] Intelligent optimization algorithms are used for automatic optimization, such as genetic algorithms, Bayesian optimization, or deep reinforcement learning.

[0063] Taking a genetic algorithm as an example, the implementation process first initializes the population, with each individual representing a complete reliability index allocation scheme. During the iteration process, the algorithm continuously evolves the population through selection, crossover, and mutation operations: selection, based on a fitness function (considering both system reliability and constraint satisfaction), retains superior individuals; crossover generates new schemes by exchanging component index values ​​among parent schemes; and mutation randomly perturbs the index values ​​of individual components to maintain population diversity. Severity and impact scores play a guiding role in this process; for example, a finer search step size is used on the index dimensions corresponding to high-scoring components, ensuring that the optimization process is both comprehensive and targeted.

[0064] To improve optimization efficiency, a fast reliability estimator is used to replace traditional time-consuming simulation calculations. This estimator is typically built based on a surrogate model (such as Gaussian process regression) or a Bayesian network, establishing a rapid mapping relationship from component reliability indicators to system-level reliability through offline learning of a large amount of historical simulation data. During the optimization iteration process, whenever the intelligent algorithm generates a new allocation scheme, the fast reliability estimator can complete an approximate calculation of system reliability within seconds and provide feedback on the satisfaction of constraints. This mechanism reduces the time for a single evaluation from hours in traditional simulations to seconds, enabling the optimization algorithm to complete a large number of iterative searches within a feasible timeframe, ultimately outputting the optimal allocation scheme that maximizes system reliability under given constraints.

[0065] Through the embodiments of the present invention, a multi-objective constrained optimization model is constructed with the goal of maximizing the reliability of the power system and including total cost and total weight constraints. The severity score and impact score are used as optimization direction guides. An intelligent optimization algorithm is used to iteratively search and dynamically adjust the allocation values ​​of reliability indicators of each component. At the same time, a fast reliability evaluator is used to evaluate the results of each iteration, and finally the optimal allocation combination of reliability indicators of each component is determined, which can effectively improve the scientificity and rationality of the allocation of reliability indicators of the power system.

[0066] According to the reliability index allocation method of the power system provided by the present invention, a preset optimization algorithm is adopted, and the intelligent optimization algorithm is a deep reinforcement learning algorithm. Using severity and impact scores as optimization guidance parameters, a deep reinforcement learning algorithm is employed to iteratively search under objective constraints, dynamically adjusting the reliability index allocation values ​​of each component to determine the optimal combination of reliability indices for each component, including: Call the intelligent agent to obtain the current state, which includes the current reliability index allocation values ​​of each component, the current total cost, and the current total weight; The agent selects and adjusts the reliability index allocation value of the target component as an action based on the policy network; The action is input into a fast reliability evaluator that serves as the environment, resulting in a new state and a reward value. The new state includes the updated reliability index assignments for each component, the updated total cost, and the updated total weight. The reward value is used to evaluate the contribution of the action to the optimization objective function. Update the policy network based on the current state, actions, reward values, and interactive experience formed by the new state; Repeat the above process until the optimal allocation combination of reliability indices of each component that satisfies the objective constraints and maximizes the reliability of the power system is obtained.

[0067] In this embodiment of the invention, the assignment process is modeled as a Markov decision process using a deep reinforcement learning (DRL) algorithm, and the optimal strategy is learned through the interaction between the agent and the environment.

[0068] The current state includes: the current reliability index allocation value of each component (e.g., MTBF_i), the current total cost (the sum of the costs of each component), and the current total weight (the sum of the weights of each component). The state representation adopts a normalized vector form, where the reliability index value is scaled to the [0,1] interval using Min-Max scaling, and the cost and quality values ​​are normalized relative to the upper bound of the constraints. The agent acts as the learning subject, and its policy network adopts a deep neural network architecture. The input layer dimension is consistent with the length of the state vector, the hidden layer is usually designed as a fully connected layer or an attention mechanism layer, and the output layer corresponds to the optional action space.

[0069] Based on the current state, the agent selects actions through a policy network. An action is defined as an adjustment operation to the assigned value of a reliability index for a specific target component, including the adjustment object (component identifier) ​​and the adjustment magnitude (continuous value or discrete level). The action selection strategy employs a greedy algorithm or probability-distribution-based random sampling to balance exploration and exploitation. Severity and impact scores play crucial guiding roles in this process: for high-scoring components, the action space is configured with finer adjustment granularity, and they are given higher weights in the reward function, ensuring that optimization resources are tilted towards critical components.

[0070] After the action is executed, the environment (i.e., the fast reliability estimator) receives adjustment instructions and generates new states and reward values. The fast reliability estimator, built on a surrogate model (such as Gaussian process regression or a neural network), can recalculate system-level reliability, total cost, and total weight within milliseconds. The new state updates the reliability metric assignments and derived parameters for each component. The reward function is designed as a multi-objective composite: the basic reward is positively correlated with the improvement in system reliability, the constraint penalty is negatively correlated with the degree of cost and weight exceeding limits, and severity and impact scores are introduced as weighting coefficients to ensure that optimization of highly important components yields higher returns. The reward value calculation formula integrates these factors, guiding the agent to learn in a direction that both improves reliability and satisfies constraints.

[0071] Each interaction generates an experience tuple (current state, action, reward value, new state), which is stored in the experience replay buffer. During training, batches of experience data are randomly sampled, and the policy gradient is calculated using temporal difference error to update the policy network parameters. Network optimization employs the Actor-Critic framework or the Proximal Policy Optimization (PPO) algorithm. The Critic network evaluates the state value function, and the Actor network optimizes the policy function. The learning rate is adaptively adjusted, initially using a larger learning rate to accelerate convergence, and later refining the search step size to improve accuracy. The experience replay mechanism breaks down data correlations, improving training stability.

[0072] The above process iterates repeatedly, with the agent gradually optimizing the strategy through numerous training rounds. Multiple requirements are set for the termination condition: a maximum number of training rounds to prevent infinite loops, a convergence threshold to check the stability of accumulated rewards, and constraint satisfaction to verify the feasibility of the scheme. The final output is the optimal allocation combination of component reliability indicators that maximizes system reliability, completing the intelligent allocation process from experience-driven to data-driven.

[0073] This invention applies deep reinforcement learning algorithms to the allocation of reliability indicators for power systems. Using severity and impact scores as optimization guidelines, the agent acquires the current state and selects actions based on a policy network. A fast reliability evaluator obtains new states and reward values, and the policy network is updated using interactive experience. This iterative process continues until the optimal allocation combination of reliability indicators for each component that satisfies the objective constraints and maximizes system reliability is obtained. This improves the accuracy and efficiency of indicator allocation and enhances the reliability of the power system.

[0074] According to the present invention, a method for allocating reliability indicators for a power system is provided, wherein the fast reliability evaluator is a surrogate model or a Bayesian network.

[0075] In this embodiment of the invention, the proxy model is a simplified mathematical model based on data-driven principles, capable of approximating time-consuming high-fidelity simulation calculations. In specific implementation, a training dataset is first established, containing a large amount of historical simulation results or experimental data. The input features are the reliability index assignment values ​​for each component, and the output labels are the corresponding system-level reliability indicators, as well as related parameters such as cost and weight.

[0076] The surrogate model can be implemented using various machine learning methods. Gaussian process regression models are suitable for small sample scenarios and can quantify the uncertainty of prediction results; neural network models excel at handling high-dimensional nonlinear relationships, capturing complex mapping relationships through deep architectures; support vector regression models demonstrate good generalization ability on medium-sized datasets. During model training, cross-validation is used to select optimal hyperparameters, and an early stopping strategy is employed to prevent overfitting. The trained surrogate model can complete system reliability assessment within milliseconds, achieving an efficiency improvement of several orders of magnitude compared to traditional simulation methods.

[0077] In the optimization loop, the surrogate model interacts with the intelligent algorithm as an environment model. Whenever the optimization algorithm generates a new reliability index allocation scheme, the surrogate model instantly predicts the system reliability performance and constraint satisfaction under that scheme, providing rapid feedback to the algorithm.

[0078] A fast reliability evaluator can also be implemented using Bayesian networks. A Bayesian network is a probabilistic graphical model that can effectively express causal relationships and uncertainty propagation between variables. In reliability assessment scenarios, the nodes of a Bayesian network represent system reliability-related variables, including component reliability indicators, environmental factors, and system-level reliability performance, while the edges represent dependencies between variables.

[0079] Bayesian network evaluators possess unique probabilistic reasoning capabilities. Given the reliability index assignment values ​​for each component, they calculate the probability distribution of system reliability using network propagation algorithms (such as confidence propagation), providing not only point estimation results but also confidence interval assessments.

[0080] In practical applications, surrogate models and Bayesian networks can be selected or combined depending on specific needs. For evaluation scenarios with high determinism, surrogate models provide efficient point estimation; for complex scenarios requiring uncertainty quantification, Bayesian networks provide probabilistic evaluation results. Both models can be deeply integrated with optimization algorithms, enabling data exchange and function calls through API interfaces.

[0081] According to the reliability index allocation method for a power system provided by the present invention, after determining the reliability index allocation result of the power system based on the severity score and influence score of each component under objective constraints using a preset optimization algorithm, the method further includes: The reliability index allocation results of the power system are verified by using a pre-set high-fidelity simulation model, and the verification results are obtained. The results of reliability index allocation and verification are visualized.

[0082] In this embodiment of the invention, a high-fidelity simulation model is used to verify the allocation results of the reliability indicators of the power system; the allocation scheme and verification results are presented to the designers through a human-computer interaction interface to support their decision-making.

[0083] In this embodiment of the invention, the high-fidelity simulation model is constructed based on the physical mechanism of the dynamic system. For example, it simulates the stress distribution of components through finite element analysis (FEA), calculates thermal management performance through fluid dynamics (CFD), or simulates the overall machine operation behavior through a system dynamics model. The parameters of the high-fidelity simulation model are strictly calibrated according to actual engineering data to ensure a high degree of consistency between the simulation accuracy and the real scene.

[0084] For example, firstly, the reliability indices (such as MTBF_i) of each component in the allocation results are used as input parameters to configure the boundary conditions of the simulation environment (such as workload and ambient temperature). Then, a high-fidelity simulation model is run to calculate the power system-level reliability indices (such as system MTBF), failure rate distribution, and key performance parameters. Finally, the simulation output is compared and analyzed with the allocation results to generate verification results. The verification results include consistency indices (such as error percentage) and feasibility indicators (such as constraint satisfaction status), used to quantify the accuracy and robustness of the allocation scheme. The entire process is automated and script-controlled, reducing manual intervention and improving verification efficiency.

[0085] The visualization phase aims to present the reliability index allocation results and verification results in a graphical form. In practice, the allocation and verification results are first integrated to extract key comparison dimensions, such as the reliability index values ​​of each component, the overall system reliability, and cost and weight constraint deviations. Subsequently, the information is displayed through a multi-view layout: the main view uses a parallel coordinate graph or radar chart to show the comparison between the allocated values ​​and simulation values ​​of each component index, highlighting key differences; the auxiliary view uses bar charts or line charts to show the historical optimization trajectory of system-level indicators, reflecting the convergence process; and the constraint check view uses a dashboard to mark the cost and weight compliance status in real time.

[0086] refer to Figure 2 , Figure 2 This is an overall flowchart of the reliability index allocation method for the power system provided by the present invention.

[0087] Step S100: Construct a multi-source heterogeneous reliability basic database.

[0088] The specific process includes: integrating historical, simulation, and experimental data; data cleaning and alignment; feature engineering and annotation.

[0089] Subsequently, the process enters the core AI processing stage (S200 and S300).

[0090] Step S200: Quantify key factors based on AI.

[0091] refer to Figure 3 , Figure 3 This is a flowchart illustrating the quantification of key factors provided by the present invention.

[0092] Historical operating conditions (such as temperature and pressure sequences) are input into a machine learning model (such as GBDT) for training and inference, and finally output the severity score S_i for each component.

[0093] The system's structural information (such as BOM and functional block diagram) is constructed into a graph network, which is then input into the GNN model for training and inference, and finally outputs the system influence embedding vector I_i for each component.

[0094] The steps mentioned above in step S200 can be executed in parallel, and S_i and I_i together constitute the quantitative basis for intelligent allocation.

[0095] The specific process includes: quantifying the severity of service conditions S_i and quantifying the system impact I_i.

[0096] Step S300: The intelligent optimization allocation reliability index takes S_i and I_i as input and uses an intelligent algorithm to search for the optimal MTBF scheme under constraints.

[0097] refer to Figure 4 , Figure 4 This is a flowchart of the intelligent optimization allocation process provided by the present invention.

[0098] This process is an iterative learning loop: The agent observes the current state, such as the current MTBF allocation scheme, cost, weight, etc.

[0099] The agent selects an action based on the policy network, namely, adjusting the MTBF value of a certain component.

[0100] This action operates on the environment, which is the reliability assessment proxy model. The environment quickly calculates the new system reliability, cost, etc., and provides a reward in return.

[0101] The interaction experience (state, action, reward, new state) is stored and used to update the agent's policy network.

[0102] This process repeats until the agent learns a strategy that maximizes cumulative rewards (i.e., finds the optimal solution). Finally, the process enters S400, which outputs and verifies the optimal solution.

[0103] Step S400: Generate and verify the optimal solution, output and verify the optimal allocation solution.

[0104] This invention overcomes the over-reliance on personal experience in related technologies, making the allocation process data-driven, and ensuring the results are reproducible and optimizable. By utilizing intelligent algorithms to automatically search for optimal solutions in high-dimensional space, it transforms lengthy manual iterations into efficient automated calculations, significantly shortening the design cycle. Through precise quantification of the "severity" and "impact" of components, it identifies truly critical weaknesses, enabling precise allocation of design resources and optimizing cost and weight while ensuring reliability. This invention not only proposes a method but also constructs a complete system framework, easily integrated into existing design processes to form an applicable industrial software tool.

[0105] The reliability index allocation device for the power system provided by the present invention is described below. The reliability index allocation device for the power system described below can be referred to in correspondence with the reliability index allocation method for the power system described above.

[0106] refer to Figure 5 , Figure 5 This is a schematic diagram of the module of the reliability index allocation device for the power system provided by the present invention.

[0107] The acquisition module 501 is used to acquire the operating characteristics and structural information of each component of the power system; The severity scoring module 502 is used to input the working condition characteristics of each component into the trained severity quantification model based on the gradient boosting tree model to obtain the severity score of each component. The influence rating module 503 is used to input the component attributes and connection relationships of each component of the power system into the trained influence measurement model based on the graph neural network model to obtain the influence rating of each component. The index allocation module 504 is used to determine the reliability index allocation result of the power system based on the severity score and influence score of each component under the target constraint by using a preset optimization algorithm.

[0108] Specifically, the reliability index allocation device for the power system provided by the present invention can realize all the method steps implemented in the above-mentioned power system reliability index allocation method embodiment, and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0109] Figure 6 This is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as... Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640. The processor 610, communication interface 620, and memory 630 communicate with each other via the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a reliability index allocation method for the power system. This method includes: acquiring the operating characteristics and structural information of each component of the power system; inputting the operating characteristics of each component into a trained severity quantification model based on a gradient boosting tree model to obtain a severity score for each component; inputting the component attributes and connection relationships of each component of the power system into a trained influence quantification model based on a graph neural network model to obtain an influence score for each component; and using a preset optimization algorithm, under objective constraints, determining the reliability index allocation result of the power system based on the severity score and influence score of each component.

[0110] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0111] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the reliability index allocation method for the power system provided by the above methods. The method includes: acquiring the operating characteristics and structural information of each component of the power system; inputting the operating characteristics of each component into a trained severity quantification model based on a gradient boosting tree model to obtain a severity score for each component; inputting the component attributes and connection relationships of each component of the power system into a trained influence quantification model based on a graph neural network model to obtain an influence score for each component; and using a preset optimization algorithm, under objective constraints, determining the reliability index allocation result of the power system based on the severity score and influence score of each component.

[0112] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program is implemented to perform the reliability index allocation method for a power system provided by the above methods. The method includes: acquiring the operating characteristics and structural information of each component of the power system; inputting the operating characteristics of each component into a trained severity quantification model based on a gradient boosting tree model to obtain a severity score for each component; inputting the component attributes and connection relationships of each component of the power system into a trained influence quantification model based on a graph neural network model to obtain an influence score for each component; and using a preset optimization algorithm, under objective constraints, determining the reliability index allocation result of the power system based on the severity score and influence score of each component.

[0113] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0114] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for allocating reliability indicators for a power system, characterized in that, include: Obtain the operating characteristics and structural information of each component of the power system; The operating condition characteristics of each component are input into a trained severity quantification model based on a gradient boosting tree model to obtain the severity score of each component. The component attributes and connection relationships of each component of the power system are input into a trained influence quantification model based on a graph neural network model to obtain the influence score of each component. Using a preset optimization algorithm, under target constraints, the reliability index allocation result of the power system is determined based on the severity score and the influence score of each component.

2. The method for allocating reliability indicators of a power system according to claim 1, characterized in that, Before acquiring the operating characteristics and structural information of each component of the power system, the method further includes: Acquire multi-source heterogeneous data, wherein the multi-source heterogeneous data includes historical operating data of the power system, simulation calculation data, and experimental test data; Preprocessing and labeling of the multi-source heterogeneous data yields the labeled dataset of the dynamic system. Based on the labeled dataset, a pre-defined rigorous metric model based on a gradient boosting tree model and a pre-defined influence metric model based on a graph neural network model are trained.

3. The method for allocating reliability indicators of a power system according to claim 1, characterized in that, The method employs a preset optimization algorithm to determine the reliability index allocation result of the power system under target constraints, based on the severity score and influence score of each component, including: A multi-objective constrained optimization model is constructed, wherein the objective function of the multi-objective constrained optimization model is to maximize the reliability of the power system, and the objective constraints of the multi-objective constrained optimization model include total cost constraints and total weight constraints; Using the severity score and the impact score as optimization direction guidance parameters, an intelligent optimization algorithm is used to iteratively search under the target constraints, dynamically adjust the reliability index allocation values ​​of each component, and determine the optimal allocation combination of the reliability indexes of each component as the reliability index allocation result of the power system. Specifically, a fast reliability evaluator is used to evaluate the reliability index assignment values ​​of each component obtained in each iteration search.

4. The method for allocating reliability indicators of a power system according to claim 3, characterized in that, The method employs a preset optimization algorithm, wherein the intelligent optimization algorithm is a deep reinforcement learning algorithm; The step of using the severity score and the impact score as optimization direction guidance parameters, employing a deep reinforcement learning algorithm to iteratively search under the target constraint, dynamically adjusting the reliability index allocation values ​​of each component, and determining the optimal allocation combination of the reliability indices of each component includes: The intelligent agent is invoked to obtain the current state, which includes the current reliability index allocation values ​​of each component, the current total cost, and the current total weight. The agent, based on a policy network, selects to adjust the reliability index allocation value of the target component as an action. The action is input into a fast reliability evaluator, which serves as the environment, to obtain a new state and a reward value. The new state includes the updated reliability index allocation values ​​of each component, the updated total cost, and the updated total weight. The reward value is used to evaluate the contribution of the action to optimizing the objective function. The policy network is updated based on the current state, the action, the reward value, and the interaction experience formed by the new state. Repeat the above process until the optimal allocation combination of reliability indices of each component is obtained that satisfies the target constraints and maximizes the reliability of the power system.

5. The method for allocating reliability indicators of a power system according to claim 3, characterized in that, The fast reliability evaluator is either a surrogate model or a Bayesian network.

6. The method for allocating reliability indicators of a power system according to claim 1, characterized in that, After determining the reliability index allocation result of the power system based on the severity score and influence score of each component under the target constraint using a preset optimization algorithm, the method further includes: The reliability index allocation results of the power system are verified by using a preset high-fidelity simulation model, and the verification results are obtained. The reliability index allocation results and the verification results are displayed visually.

7. A reliability index allocation device for a power system, characterized in that, include: The acquisition module is used to acquire the operating characteristics and structural information of various components of the power system; The severity scoring module is used to input the working condition characteristics of each component into a trained severity quantification model based on a gradient boosting tree model to obtain the severity score of each component. The influence rating module is used to input the component attributes and connection relationships of each component of the power system into a trained influence quantification model based on a graph neural network model to obtain the influence rating of each component. The index allocation module is used to determine the reliability index allocation result of the power system based on the severity score and the influence score of each component under the target constraint, using a preset optimization algorithm.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the reliability index allocation method for the power system as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the reliability index allocation method for the power system as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the reliability index allocation method for the power system as described in any one of claims 1 to 6.

Images