Real-time monitoring and optimization method and system for source, grid, load and storage equipment on a virtual power plant platform

Abstract

The present invention provides a real-time monitoring and optimization method and system for source-grid-load-storage equipment on a virtual power plant platform, which relates to the technical field of intelligent optimization and scheduling of virtual power plants. The method comprises collecting equipment operation data and performing noise reduction and feature compression to obtain feature data, identifying equipment fault types based on the feature data, decoupling and reconstructing the feature data in time and space dimensions to obtain an equipment association matrix, constructing an equipment topology map based on the equipment association matrix and credible fault types, performing bidirectional tracking on the equipment topology map according to the fault diffusion probability, and generating a fault propagation path; hierarchically sorting the equipment on the fault propagation path, and outputting a fault impact prediction result and an equipment priority list; and using a hierarchical reinforcement learning algorithm to perform collaborative scheduling optimization on the equipment in the virtual power plant, thereby realizing collaborative scheduling optimization of the virtual power plant equipment.

Description

Technical Field

[0001] The present invention relates to intelligent optimization and scheduling technology for virtual power plants, and in particular to a real-time monitoring and optimization method and system for source, grid, load and storage equipment on a virtual power plant platform. Background Art

[0002] Existing power dispatch systems primarily rely on traditional monitoring methods to collect and analyze the operating status of power generation, grid, load, and storage equipment. However, in complex operating environments, equipment data is subject to significant noise interference, resulting in low data quality and affecting the accuracy of fault identification. Furthermore, traditional methods typically rely on fixed rules or empirical judgments for fault diagnosis, making it difficult to fully utilize historical data for accurate matching. This leads to significant errors in the identification of equipment fault types and an inability to effectively assess the risk level and credibility of faults.

[0003] Furthermore, current equipment monitoring methods often focus on the independent analysis of individual devices, ignoring the dynamic coupling relationships between devices. This makes it difficult to accurately predict the propagation path and impact range of equipment failures, and effectively assess the spread of failures, hindering the efficiency of fault warning and emergency dispatch. Furthermore, existing scheduling optimization methods are mostly based on static scheduling strategies, failing to dynamically adjust equipment combinations and control strategies based on real-time fault analysis results, making it difficult to address the optimal scheduling needs in the event of sudden failures.

[0004] Therefore, there is an urgent need for a method that can improve the data processing capability of equipment, accurately identify faults and dynamically evaluate their propagation impact, while combining optimized scheduling strategies to improve system stability, so as to improve the operating reliability and scheduling efficiency of source-grid-load-storage equipment. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for real-time monitoring and optimization of source, grid, load and storage equipment on a virtual power plant platform, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] A method for real-time monitoring and optimization of source, grid, load and storage equipment on a virtual power plant platform is provided, comprising:

[0008] Collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Identify equipment fault types based on the feature data, perform similarity matching between the equipment fault type and historical fault data, generate a fault risk score, and verify the credibility of the equipment fault type based on the fault risk score to obtain a credible fault type.

[0009] The feature data is decoupled and reconstructed in time and space dimensions to obtain a device association matrix. A device topology map is constructed based on the device association matrix and trusted fault types. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Bidirectional tracking is performed on the device topology map based on the fault diffusion probability to generate a fault propagation path. The devices on the fault propagation path are hierarchically sorted and the fault impact prediction results and device priority list are output.

[0010] A hierarchical reinforcement learning algorithm is used to coordinate and optimize the equipment scheduling in a virtual power plant. The upper-level intelligent agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list, and the lower-level intelligent agent generates equipment control instructions based on the equipment combination plan and the fault risk score, thereby realizing the coordinated scheduling optimization of the virtual power plant equipment.

[0011] In an optional embodiment,

[0012] Collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Based on the feature data, identify the equipment fault type. Match the equipment fault type with historical fault data for similarity, generate a fault risk score, and verify the credibility of the equipment fault type based on the fault risk score. The credible fault types obtained include:

[0013] Performing a wavelet transform on the equipment operation data to obtain multi-scale coefficients, calculating the ratios of adjacent scale coefficients to construct a feature matrix; performing a singular value decomposition on the feature matrix to obtain a signal subspace and a noise subspace, calculating the orthogonal vector product of the signal subspace and the noise subspace to obtain an adaptive threshold, projecting the equipment operation data into the signal subspace, and reconstructing based on the adaptive threshold to obtain denoised data;

[0014] Construct an autoencoder model, in which the encoding part maps the denoised data into intermediate features, and obtains latent variables from the intermediate features through a nonlinear activation function; construct a discriminator network, input the latent variables and preset Gaussian distribution samples into the discriminator network to obtain discrimination probabilities, and calculate the cross entropy based on the discrimination probabilities and the true labels to obtain adversarial loss; perform gradient descent optimization on the latent variables based on the adversarial loss to obtain fault features;

[0015] The fault features are subjected to sliding convolution through a one-dimensional convolution kernel to obtain a time series feature sequence, and the gated recurrent unit is used to calculate the state vector of the time series feature sequence time step by time step; the conditional entropy of the state vector at different time steps is calculated to obtain a time series weight, and the time series weight is weighted and summed with the time series feature sequence to obtain a global feature, which is input into a softmax classifier to obtain an initial fault type;

[0016] Based on conditional entropy, historical fault samples are stratified according to operating condition parameters. The fault type transition frequency is calculated in each layer of samples to obtain a state transition matrix. The global features are spliced ​​with the current operating condition parameters and input into the state transition matrix to obtain the predicted probability. The mutual information between the initial fault type and the predicted probability is calculated to obtain a credibility score. When the credibility score exceeds a preset threshold, it is confirmed as a credible fault type.

[0017] In an optional embodiment,

[0018] Based on conditional entropy, historical fault samples are stratified according to operating condition parameters. The fault type transition frequency is calculated in each layer of samples to obtain a state transition matrix. The global features and current operating condition parameters are concatenated and input into the state transition matrix to obtain a predicted probability. The mutual information between the initial fault type and the predicted probability is calculated to obtain a credibility score.

[0019] Divide the numerical interval of the operating condition parameter into multiple subintervals, count the occurrence frequency of the fault type in each subinterval, and calculate the conditional entropy of the operating condition parameter on the fault type; calculate the mean and standard deviation of the conditional entropy, and obtain an adaptive threshold based on the mean minus the standard deviation; construct a hierarchical parameter set for the operating condition parameters whose conditional entropy is less than the adaptive threshold;

[0020] Calculate the local density and minimum distance of historical fault samples in the space corresponding to the hierarchical parameter set, determine the density peak point as the cluster center based on the local density, and assign the sample points to the cluster center based on the minimum distance to obtain the working condition layer;

[0021] In each operating condition layer, the total number of samples of the fault type and the number of transition samples between the fault types are calculated; when the total number of samples is greater than a preset sample number threshold, the ratio of the number of transition samples to the total number of samples is calculated to obtain the fault type transition frequency; otherwise, the fault type transition frequency is set to zero; and a state transition matrix is ​​constructed based on the fault type transition frequency.

[0022] Obtain the global features and current operating parameters of the sample to be verified, and determine the target operating layer based on the current operating parameters; combine the global features and the current operating parameters to form a feature vector, and input the state transition matrix corresponding to the target operating layer to obtain the predicted probability distribution;

[0023] Obtain the initial fault type probability distribution of the sample to be verified, divide the probability of each fault type in the initial fault type probability distribution by the probability of the corresponding fault type in the predicted probability distribution to obtain a probability ratio; and obtain a credibility score based on the sum of the logarithm of the probability ratio and the product of the initial fault type probability.

[0024] In an optional embodiment,

[0025] The feature data is decoupled and reconstructed in time and space dimensions to obtain a device association matrix. A device topology map is constructed based on the device association matrix and trusted fault types. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Bidirectional tracking is performed on the device topology map based on the fault diffusion probability to generate a fault propagation path. The devices on the fault propagation path are hierarchically sorted and the fault impact prediction results and device priority list are output, including:

[0026] Decomposing the feature data in a time dimension into hourly, daily, and weekly subsequences, extracting periodic patterns and state trends from the hourly, daily, and weekly subsequences, and constructing a time dimension feature vector;

[0027] Extracting physical connection distances, electrical circuit connections, and device functional relationships from the feature data, calculating correlation coefficients between devices, and constructing spatial dimension feature vectors;

[0028] Performing a tensor fusion operation on the time dimension feature vector and the space dimension feature vector to obtain a device association matrix;

[0029] Based on the device association matrix and the trusted fault types, a device topology graph including device nodes and connection edges is constructed. A spectral clustering algorithm with adaptive similarity is used to dynamically partition the device topology graph to obtain multiple device partitions. The device density distribution within each device partition is calculated to obtain the intra-regional eigenvalue. The connection edge weights between the device partitions are calculated to obtain the inter-regional eigenvalue. A transfer function including a delay factor and an attenuation coefficient is constructed. The intra-regional eigenvalue and the inter-regional eigenvalue are input into the transfer function to calculate the fault diffusion probability between the device partitions.

[0030] Marking the location of the faulty device as the starting node in the device topology map, performing forward propagation calculation based on the fault diffusion probability to obtain an impact propagation path; marking the device nodes on the impact propagation path whose fault diffusion probability exceeds a preset probability threshold as the affected device locations, calculating the transfer importance of each node in the device topology map, and obtaining the source device location through reverse tracing;

[0031] The impact propagation path and the source device location are combined and optimized to obtain the fault propagation path, the risk level of each device on the fault propagation path is calculated and ranked, and the fault impact prediction result and device priority list are output.

[0032] In an optional embodiment,

[0033] Construct a transfer function including a delay factor and an attenuation coefficient, input the characteristic values ​​within the region and the characteristic values ​​between regions into the transfer function, and calculate the fault diffusion probability between device partitions, including:

[0034] Obtaining physical propagation delay and response delay between devices, classifying them according to transmission type to obtain delay classification results, performing segmented mapping on the delay classification results to obtain delay mapping values, and constructing a delay factor based on the delay mapping values;

[0035] Collecting the physical distance, electrical impedance, and functional correlation between devices, calculating a spatial attenuation value based on the physical distance, calculating a signal attenuation value based on the electrical impedance, and calculating a coupling attenuation value based on the functional correlation, hierarchically combining the spatial attenuation value, signal attenuation value, and coupling attenuation value to obtain an attenuation combination value, and constructing an attenuation coefficient based on the attenuation combination value; and constructing a nonlinear transfer function based on the delay factor and the attenuation coefficient.

[0036] Calculate the spatial distribution density value of devices in the area and the distance distribution value between devices, calculate the regional density characteristics based on the spatial distribution density value and the distance distribution value, and fuse the regional density characteristics with the device operating parameters to obtain the regional characteristic value; calculate the number of physical connections and energy transfer power values ​​of devices at the regional boundary, calculate the connection transmission characteristics based on the physical connection number and the energy transfer power value, and fuse the connection transmission characteristics with the device operating status to obtain the inter-regional characteristic value; establish device connection constraints and energy transfer constraints based on the intra-regional characteristic value and the inter-regional characteristic value;

[0037] Historical fault samples are classified according to fault type and propagation path to obtain training samples. The training samples are input into a nonlinear transfer function, and iterative training is performed using device connection constraints and energy transfer constraints to obtain corrected values ​​for the delay factor and attenuation coefficient. The intra-regional and inter-regional eigenvalues ​​are input into the nonlinear transfer function, and the propagation calculation results are dynamically corrected based on the corrected values ​​to output the fault diffusion probability between device partitions.

[0038] In an optional embodiment,

[0039] A hierarchical reinforcement learning algorithm is used to coordinate and optimize the dispatch of equipment within a virtual power plant. The upper-layer agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list. The lower-layer agent generates equipment control instructions based on the equipment combination plan and fault risk score. The coordinated dispatch optimization of virtual power plant equipment includes:

[0040] Obtain the predicted value of equipment failure impact, equipment operating status value, and equipment priority score to construct the upper state space, and construct the upper action space based on the equipment combination plan; construct the upper reward function based on the weighted combination of system reliability index, operating efficiency index, and scheduling cost;

[0041] The device combination scheme, device failure risk score, and real-time operating parameters are used to construct a lower-level state space, and the device control instructions are used to construct a lower-level action space; a lower-level reward function is constructed based on a dynamic weight combination of a system stability index, a response time index, and the failure risk score;

[0042] The device collaborative impact is calculated based on the fault impact prediction results and the priority list, and the device collaborative impact is used as the weight coefficient to construct a weighted state feature. Historical scheduling data is obtained to calculate the scheduling association strength between devices. Based on the scheduling association strength, the weighted state feature is adaptively adjusted. The adjusted feature is input into the upper-level state space, and the upper-level state action value is calculated according to the upper-level reward function to generate a device combination plan.

[0043] The device combination scheme and the fault risk score are deeply fused to obtain the device association features, which are then input into the lower-level state space. The lower-level state action value is calculated based on the lower-level reward function to generate device control instructions.

[0044] Execute device control instructions to obtain control results, calculate reward values ​​based on the control results, determine the sampling weights of the experience replay pool based on the historical reward distribution of different scheduling scenarios, and obtain training samples from historical data; based on the training samples and reward values, perform online updates on the collaborative influence and scheduling association strength;

[0045] The updated collaborative influence and scheduling association strength are input into the upper-level intelligent agent and the lower-level intelligent agent respectively to generate optimized equipment combination plans and optimized equipment control instructions, thereby realizing the collaborative scheduling optimization of virtual power plant equipment.

[0046] In an optional embodiment,

[0047] Execute device control instructions to obtain control results, calculate reward values ​​based on the control results, determine the sampling weights of the experience replay pool based on the historical reward distribution of different scheduling scenarios, and obtain training samples from historical data. Online updates of the collaborative influence and scheduling association strength based on the training samples and reward values ​​include:

[0048] Executing device control instructions to obtain control results consisting of power balance, response time, energy loss, and system stability; calculating a system reliability score based on the power balance and system stability, and calculating a scheduling efficiency score based on the response time and energy loss, and using a weighted combination of the system reliability score and the scheduling efficiency score as a reward value;

[0049] According to the scheduling scenarios of different device combinations, state transition samples are classified and stored to build an experience replay pool. The state transition samples include device combination plans, device control instructions, reward values, and post-control states. Sampling weights are determined based on the historical reward distribution of scheduling scenarios, and training samples are selected from the experience replay pool based on the sampling weights.

[0050] Extracting response timing characteristics and power coupling characteristics between devices based on the device combination scheme in the training sample, and constructing a device collaborative influence matrix; constructing a collaborative influence loss function based on the temporal correlation between the device collaborative influence matrix and the reward value, and optimizing the collaborative influence loss function to perform online update of the collaborative influence;

[0051] Based on the device control instructions in the training samples, the scheduling dependency characteristics and control coupling characteristics between devices are extracted to construct a device scheduling association matrix; according to the dynamic correlation between the device scheduling association matrix and the reward value, a scheduling association strength loss function is constructed, and the scheduling association strength is updated online by optimizing the scheduling association strength loss function.

[0052] According to a second aspect of the embodiments of the present invention,

[0053] Provides a real-time monitoring and optimization system for source, grid, load and storage equipment on a virtual power plant platform, including:

[0054] The first unit is used to collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Based on the feature data, the equipment fault type is identified. The equipment fault type is matched with historical fault data for similarity to generate a fault risk score. Based on the fault risk score, the credibility of the equipment fault type is verified to obtain a credible fault type.

[0055] The second unit is used to decouple feature data in time and space dimensions to reconstruct a device association matrix. Based on the device association matrix and trusted fault types, a device topology map is constructed. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Based on the fault diffusion probability, bidirectional tracking is performed on the device topology map to generate a fault propagation path. Devices along the fault propagation path are hierarchically ranked and output as fault impact prediction results and a device priority list.

[0056] The third unit is used to use a hierarchical reinforcement learning algorithm to coordinate and optimize the scheduling of equipment within the virtual power plant. The upper-level intelligent agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list, and the lower-level intelligent agent generates equipment control instructions based on the equipment combination plan and fault risk score, thereby realizing the coordinated scheduling optimization of virtual power plant equipment.

[0057] According to a third aspect of the embodiments of the present invention,

[0058] An electronic device is provided, comprising:

[0059] processor;

[0060] a memory for storing processor-executable instructions;

[0061] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0062] According to a fourth aspect of the embodiments of the present invention,

[0063] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0064] In this embodiment, noise reduction and feature compression are performed on device operation data to improve data quality, reduce noise interference, and ensure accurate fault identification. Furthermore, by matching historical fault data with a risk scoring mechanism, the type of device fault can be accurately identified, increasing the confidence of fault diagnosis and avoiding misdiagnosis and missed diagnosis, thereby improving the accuracy and reliability of device operation and maintenance. This solution constructs an association matrix based on device feature data to dynamically analyze the coupling relationships between devices. It also predicts fault propagation paths through topology analysis and fault diffusion probability calculation. This not only provides early warning of potentially affected devices but also prioritizes devices based on their impact, facilitating precise location of critical fault sources, improving the targeted and efficient nature of fault resolution, and mitigating the impact of faults on the entire system. Regarding scheduling optimization, a hierarchical reinforcement learning algorithm is employed to achieve coordinated optimal scheduling of source, grid, load, and storage devices. The upper-level agent formulates a reasonable device combination plan based on the fault impact prediction results and the device priority list. The lower-level agent generates specific control instructions based on the device combination plan and risk scores. This enables dynamic optimization and control of devices, improving the intelligent level of scheduling and ensuring the safety, economy, and stability of the power system even under complex operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 Schematic diagram of the flow of a method for real-time monitoring and optimization of source, grid, load and storage equipment on a virtual power plant platform according to an embodiment of the present invention;

[0066] Figure 2 This is a simulation diagram of the effect of the real-time monitoring and optimization method for source, grid, load and storage equipment on the virtual power plant platform according to an embodiment of the present invention;

[0067] Figure 3 This is a diagram analyzing the fault propagation delay and attenuation characteristics of an embodiment of the present invention;

[0068] Figure 4This is a structural diagram of the real-time monitoring and optimization system for source, grid, load and storage equipment of a virtual power plant platform according to an embodiment of the present invention. DETAILED DESCRIPTION

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0070] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0071] Figure 1 Schematic diagram of the flow of the real-time monitoring and optimization method of the source, grid, load and storage equipment of the virtual power plant platform according to the embodiment of the present invention. Figure 1 As shown, the method includes:

[0072] Collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Identify equipment fault types based on the feature data, perform similarity matching between the equipment fault type and historical fault data, generate a fault risk score, and verify the credibility of the equipment fault type based on the fault risk score to obtain a credible fault type.

[0073] The feature data is decoupled and reconstructed in time and space dimensions to obtain a device association matrix. A device topology map is constructed based on the device association matrix and trusted fault types. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Bidirectional tracking is performed on the device topology map based on the fault diffusion probability to generate a fault propagation path. The devices on the fault propagation path are hierarchically sorted and the fault impact prediction results and device priority list are output.

[0074] A hierarchical reinforcement learning algorithm is used to coordinate and optimize the equipment scheduling in a virtual power plant. The upper-level intelligent agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list, and the lower-level intelligent agent generates equipment control instructions based on the equipment combination plan and the fault risk score, thereby realizing the coordinated scheduling optimization of the virtual power plant equipment.

[0075] Equipment operation data refers to the time series data generated by various types of equipment in a virtual power plant during operation, including but not limited to parameters such as voltage, current, power, temperature, frequency, and equipment operating status. This data may contain noise and redundant information, so it needs to be processed through noise reduction and feature compression before analysis to remove irrelevant information and reduce computational complexity.

[0076] Feature data refers to key information extracted from equipment operating data that characterizes the equipment's operating status. After noise reduction and feature compression, the raw data is converted into a smaller-dimensional dataset that still retains the essential characteristics of the equipment's operation, facilitating subsequent analysis and modeling.

[0077] Equipment fault types are a classification of possible abnormal conditions or operational failures that may occur in equipment, typically including short circuits, overloads, overtemperatures, and electrical faults. Fault type identification is based on feature data and is combined with historical data for classification to determine the current fault category of the equipment.

[0078] Historical fault data refers to fault information recorded during past equipment operation, including fault type, occurrence time, impact scope, and repair measures. This data is used to perform similarity matching with currently detected fault types to assess the potential risk of the current fault.

[0079] The Fault Risk Score is a metric calculated based on the similarity between the current device fault type and historical fault data. This score quantifies the severity of a device fault and its potential impact on the system. A higher score indicates a greater fault risk.

[0080] Trusted fault types refer to fault categories verified by fault risk scoring. Because misjudgments can occur during fault identification, the credibility of the initially identified fault types needs to be verified based on historical data to ensure the reliability of the final results.

[0081] The device topology diagram is a network structure diagram constructed based on the device association matrix and trusted fault types. Nodes represent devices, and edges represent the relationships between devices. This topology diagram is used to analyze the impact propagation paths between devices, particularly when a fault occurs, to help identify other potentially affected devices.

[0082] Dynamic partitioning involves using an adaptive algorithm to divide the entire virtual power plant's network into local areas based on the structure of the device topology and the device's operating status. This method reduces computational complexity and ensures that fault analysis focuses on the affected areas.

[0083] Local features refer to key characteristic information extracted within a specific partition of the device topology. These features describe the operating status of devices within that area, the distribution of faults, and the potential impact of faults. Inter-region feature transfer refers to the process of exchanging information between different dynamic partitions of the device topology. Due to the mutual influence of devices, analysis of a single region may not fully reflect the status of the entire system. Therefore, it is necessary to transfer fault-related features between regions to calculate the probability of fault spread.

[0084] Device ranking is the process of prioritizing devices based on the fault propagation path. Devices that are most affected by the fault are assigned higher priority, allowing them to be prioritized for adjustment or protection during scheduling optimization.

[0085] The fault impact prediction result is a predictive analysis result of the possible scope and extent of the fault impact. It combines the fault propagation path, equipment priority and historical fault impact pattern to help the scheduling system make preventive decisions.

[0086] The device priority list is generated based on the device hierarchical sorting results and contains the scheduling priority information for each device. During the scheduling optimization process, high-priority devices are typically prioritized in the scheduling plan to reduce system risk.

[0087] In an optional embodiment,

[0088] Collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Based on the feature data, identify the equipment fault type. Match the equipment fault type with historical fault data for similarity, generate a fault risk score, and verify the credibility of the equipment fault type based on the fault risk score. The credible fault types obtained include:

[0089] Performing a wavelet transform on the equipment operation data to obtain multi-scale coefficients, calculating the ratios of adjacent scale coefficients to construct a feature matrix; performing a singular value decomposition on the feature matrix to obtain a signal subspace and a noise subspace, calculating the orthogonal vector product of the signal subspace and the noise subspace to obtain an adaptive threshold, projecting the equipment operation data into the signal subspace, and reconstructing based on the adaptive threshold to obtain denoised data;

[0090] Construct an autoencoder model, in which the encoding part maps the denoised data into intermediate features, and obtains latent variables from the intermediate features through a nonlinear activation function; construct a discriminator network, input the latent variables and preset Gaussian distribution samples into the discriminator network to obtain discrimination probabilities, and calculate the cross entropy based on the discrimination probabilities and the true labels to obtain adversarial loss; perform gradient descent optimization on the latent variables based on the adversarial loss to obtain fault features;

[0091] The fault features are subjected to sliding convolution through a one-dimensional convolution kernel to obtain a time series feature sequence, and the gated recurrent unit is used to calculate the state vector of the time series feature sequence time step by time step; the conditional entropy of the state vector at different time steps is calculated to obtain a time series weight, and the time series weight is weighted and summed with the time series feature sequence to obtain a global feature, which is input into a softmax classifier to obtain an initial fault type;

[0092] Based on conditional entropy, historical fault samples are stratified according to operating condition parameters. The fault type transition frequency is calculated in each layer of samples to obtain a state transition matrix. The global features are spliced ​​with the current operating condition parameters and input into the state transition matrix to obtain the predicted probability. The mutual information between the initial fault type and the predicted probability is calculated to obtain a credibility score. When the credibility score exceeds a preset threshold, it is confirmed as a credible fault type.

[0093] For example, a wavelet transform is first performed on the device operation data, decomposing the original signal into coefficients of different scales to extract information from each frequency band. The ratios of coefficients at adjacent scales are calculated, and a feature matrix is ​​constructed based on these ratios, preserving the multi-scale nature of the data. Furthermore, a singular value decomposition is performed on the feature matrix to partition the data into a signal subspace and a noise subspace. An adaptive threshold is obtained by calculating the orthogonal vector product of the signal and noise subspaces. This threshold is used to project the original device operation data into the signal subspace to remove noise and reconstruct the de-noised data.

[0094] An autoencoder model is constructed to further extract features from the denoised data. The encoding component maps the denoised data into intermediate features and uses a nonlinear activation function to obtain latent variables. To enhance the model's generalization capabilities, a discriminator network is introduced. This network inputs latent variables and samples from a preset Gaussian distribution and outputs a discriminant probability. A cross-entropy loss is calculated based on the discriminant probability and the true label. Adversarial training is then used to optimize the latent variables to obtain more discriminative fault features.

[0095] Fault signatures are modeled using a time series model, employing a sliding convolution operation with a one-dimensional convolution kernel to capture the local temporal characteristics of the data. A gated recurrent unit is used to calculate the temporal characteristics on a time-step basis to obtain a state vector. The importance of the temporal characteristics is determined by calculating the conditional entropy at different time steps. The state vectors for each time step are weighted and summed according to their importance to obtain a global feature. This global feature is then input into a softmax classifier to identify the initial fault type of the device.

[0096] Historical fault data is stratified based on conditional entropy, and sample sets are divided according to operating condition parameters. For each stratified set, the transition frequencies between different fault types are counted to construct a state transition matrix. Global features are concatenated with current operating condition parameters and input into the state transition matrix to calculate the predicted probability of each fault type. A credibility score is calculated using the mutual information between the initial fault type and the predicted probability. When the credibility score exceeds a set threshold, the fault type is considered credible.

[0097] In this embodiment, a multi-scale feature analysis method is used to improve the accuracy of signal processing and avoid the defect of traditional methods that may lose key information. By constructing an autoencoder and discriminator network, the extraction of fault features is made more accurate, overcoming the problem of insufficient learning ability for high-dimensional data in existing methods. Combined with the time series analysis method, the problem of time information loss caused by single-frame data analysis is effectively solved, so that the fault evolution process can be accurately captured. In addition, the introduction of statistical analysis of historical fault data makes fault identification under different working conditions more reasonable and improves the stability and reliability of the prediction. Overall, this solution not only improves the accuracy of equipment fault detection, but also reduces the false alarm rate, making equipment management more intelligent.

[0098] In an optional embodiment,

[0099] Based on conditional entropy, historical fault samples are stratified according to operating condition parameters. The fault type transition frequency is calculated in each layer of samples to obtain a state transition matrix. The global features and current operating condition parameters are concatenated and input into the state transition matrix to obtain a predicted probability. The mutual information between the initial fault type and the predicted probability is calculated to obtain a credibility score.

[0100] Divide the numerical interval of the operating condition parameter into multiple subintervals, count the occurrence frequency of the fault type in each subinterval, and calculate the conditional entropy of the operating condition parameter on the fault type; calculate the mean and standard deviation of the conditional entropy, and obtain an adaptive threshold based on the mean minus the standard deviation; construct a hierarchical parameter set for the operating condition parameters whose conditional entropy is less than the adaptive threshold;

[0101] Calculate the local density and minimum distance of historical fault samples in the space corresponding to the hierarchical parameter set, determine the density peak point as the cluster center based on the local density, and assign the sample points to the cluster center based on the minimum distance to obtain the working condition layer;

[0102] In each operating condition layer, the total number of samples of the fault type and the number of transition samples between the fault types are calculated; when the total number of samples is greater than a preset sample number threshold, the ratio of the number of transition samples to the total number of samples is calculated to obtain the fault type transition frequency; otherwise, the fault type transition frequency is set to zero; and a state transition matrix is ​​constructed based on the fault type transition frequency.

[0103] Obtain the global features and current operating parameters of the sample to be verified, and determine the target operating layer based on the current operating parameters; combine the global features and the current operating parameters to form a feature vector, and input the state transition matrix corresponding to the target operating layer to obtain the predicted probability distribution;

[0104] Obtain the initial fault type probability distribution of the sample to be verified, divide the probability of each fault type in the initial fault type probability distribution by the probability of the corresponding fault type in the predicted probability distribution to obtain a probability ratio; and obtain a credibility score based on the sum of the logarithm of the probability ratio and the product of the initial fault type probability.

[0105] For example, we first need to conduct an in-depth analysis of historical fault data to determine the relationship between operating parameters and fault types. Operating parameters refer to data describing the system state under different operating conditions, such as temperature, pressure, and speed. At each moment, the system has a specific set of operating parameter values, and there is a certain correlation between these values ​​and the fault type.

[0106] In order to effectively analyze these operating parameters, it is first necessary to divide the numerical range of the operating parameters. The numerical range of the operating parameters is divided into multiple sub-intervals, and the operating parameter values ​​in each sub-interval correspond to different fault types. Specifically, first, by statistically analyzing historical fault data, the minimum and maximum values ​​that can appear in the historical samples of each operating parameter are determined, and based on these values, the intervals are divided into equal intervals or other methods. Within each interval, the frequency of the corresponding fault type is counted. For example, if a certain operating parameter value is in the first interval, the frequency of the fault type occurring in this interval is counted.

[0107] Calculate the conditional entropy for each operating parameter value. Conditional entropy measures the uncertainty in the distribution of fault types under given operating parameters. Conditional entropy is calculated based on the probability distribution of fault types within each interval and indicates the uncertainty of the fault type within a given operating parameter interval. A low conditional entropy for a given operating parameter indicates strong predictive power for the fault type, meaning the fault type is relatively clear. A high conditional entropy indicates weak predictive power for the fault type. To further identify important operating parameters, first calculate the mean and standard deviation of the conditional entropy. These two values ​​reflect the distribution of the entire dataset. Then, determine an adaptive threshold by subtracting the standard deviation from the mean. The adaptive threshold is dynamically adjusted by subtracting one standard deviation from the mean, adjusting the tolerance for conditional entropy based on different operating data conditions. If the conditional entropy for a given operating parameter is below the threshold, it is considered to have good predictive power for fault type prediction and warrants further attention. If the conditional entropy is above the adaptive threshold, the predictive power of the operating parameter is poor and may need to be eliminated.

[0108] A hierarchical parameter set is constructed based on the selected important operating parameters. This hierarchical parameter set is a space containing multiple dimensions, each representing a single operating parameter, and the data in each dimension is influenced by conditional entropy. By filtering out operating parameters with low conditional entropy, a space containing important operating parameters is obtained.

[0109] Further analysis is performed on historical fault samples. In the hierarchical parameter set, the local density and minimum distance of each sample are first calculated. Local density reflects the density of samples around a certain location. Local density can be achieved by calculating the number of samples within a certain range around each sample point. The minimum distance refers to the closest distance between a sample point and its surrounding sample points. By calculating the local density and minimum distance, the density peak points in the sample space can be found. These peak points represent the cluster centers. Based on this, the remaining samples are assigned to the closest cluster center based on the minimum distance method, thereby dividing the samples into different operating condition layers.

[0110] In each operating condition layer, the total number of samples of the fault type and the number of transition samples between fault types need to be calculated. The number of transition samples refers to the number of times a sample transitions from one fault type to another in a certain operating condition layer. The transition frequency of a fault type is expressed as the ratio of the number of transition samples to the total number of samples. If the total number of samples in a certain operating condition layer exceeds the preset threshold, the transition frequency is calculated based on the ratio of transition samples. If the total number of samples is small, the transition frequency is set to zero. This transition frequency reflects the mutual conversion law of different fault types under specific operating conditions. Based on these transition frequencies, a state transition matrix can be constructed, which can show the transition probability of a fault type under different operating conditions.

[0111] When the system needs to predict faults for a new sample to be verified, it first needs to extract the global features and current operating parameters of the sample to be verified. Global features may include the sample's historical fault records, system configuration, etc., while operating parameters describe the system's current operating status. Based on the current operating parameters, the target operating layer to which the sample belongs can be determined. The global features and current operating parameters are then combined into a feature vector that contains all the key information about the sample. This feature vector is then input into the state transition matrix corresponding to the target operating layer, and the system outputs the probability distribution of the sample belonging to different fault types.

[0112] The system compares the initial fault type probability distribution of the sample to be verified with the predicted probability distribution of the target operating condition layer. A confidence score is calculated by taking the logarithm of the ratio of the initial probability to the predicted probability and multiplying it by the initial fault type probability. This confidence score indicates the reliability of the current prediction. A high confidence score indicates that the system's prediction of the fault type for the sample is reliable; a low confidence score indicates that further data review or model adjustment is required.

[0113] Existing fault prediction technologies typically rely on fixed thresholds or traditional machine learning methods, failing to fully analyze the relationship between operating parameters and fault types in a hierarchical manner. This results in low accuracy and reliability in prediction results. Furthermore, some methods lack in-depth analysis of historical fault data and fail to effectively utilize state transition patterns to model fault evolution trends, making it difficult to dynamically adapt to changing fault modes under varying operating conditions.

[0114] This application performs a hierarchical analysis of historical fault samples based on conditional entropy, screens out the operating parameters that have a greater impact on the fault type, and constructs a hierarchical parameter set, thereby achieving more refined fault prediction. Furthermore, based on the local density and minimum distance method, the density peak point is determined, and the samples are divided into different operating condition layers to fully utilize the spatial distribution characteristics of the data and improve the rationality of the operating condition modeling. On this basis, the state transfer matrix is ​​constructed by calculating the fault type transition frequency, so that the prediction model can learn the fault evolution law under different working conditions. For the samples to be verified, this application combines the global features with the operating condition parameters, and calculates the prediction probability in combination with the state transfer matrix, and then calculates the credibility score through mutual information to quantify the reliability of the prediction results.

[0115] Compared to existing technologies, this application improves upon existing methods by introducing conditional entropy to screen operating parameters, stratifying operating conditions based on density peaks, and combining predictions with state transition matrices. This overcomes the limitations of traditional methods in adapting to complex operating conditions and making it difficult to measure prediction credibility. Through these optimizations, this application improves the accuracy of fault predictions under different operating conditions and provides a credibility assessment of prediction results, thereby enhancing the system's intelligent diagnostic capabilities and reliability.

[0116] Figure 2 This is a simulation diagram of the effect of the real-time monitoring and optimization method of the source, grid, load and storage equipment of the virtual power plant platform according to the embodiment of the present invention. Figure 2 The figure shows the conditional entropy distribution and feature importance assessment results for different operating parameters. The horizontal axis represents six key operating parameters: temperature, pressure, speed, vibration, current, and voltage; the vertical axis represents the conditional entropy value, ranging from 0 to 0.8. The comparison between this technical solution (circles) and the traditional entropy method (squares) clearly demonstrates the advantages of this solution.

[0117] For the speed parameter, this technical solution achieved the lowest conditional entropy value of 0.64, indicating that this parameter has the strongest predictive ability for the fault type. In comparison, the corresponding conditional entropy of the traditional entropy method is 0.44, which is significantly weaker in predictive ability. The pressure parameter is second, with the conditional entropy of this technical solution being 0.56, while the traditional method is 0.38. The conditional entropy values ​​for temperature parameters are 0.44 and 0.30, respectively, for vibration parameters are 0.50 and 0.34, for current parameters are 0.38 and 0.24, and for voltage parameters are 0.30 and 0.18.

[0118] Overall, this technical solution achieved higher conditional entropy values ​​for all operating parameters, with an average improvement of approximately 47%. This demonstrates that it more fully utilizes operating parameters and better captures the correlation between parameters and fault types. Performance improvements were particularly significant for the two key parameters of speed and pressure, reaching 45.5% and 47.4%, respectively, fully demonstrating the superiority of this solution in feature importance assessment.

[0119] In an optional embodiment,

[0120] The feature data is decoupled and reconstructed in time and space dimensions to obtain a device association matrix. A device topology map is constructed based on the device association matrix and trusted fault types. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Bidirectional tracking is performed on the device topology map based on the fault diffusion probability to generate a fault propagation path. The devices on the fault propagation path are hierarchically sorted and the fault impact prediction results and device priority list are output, including:

[0121] Decomposing the feature data in a time dimension into hourly, daily, and weekly subsequences, extracting periodic patterns and state trends from the hourly, daily, and weekly subsequences, and constructing a time dimension feature vector;

[0122] Extracting physical connection distances, electrical circuit connections, and device functional relationships from the feature data, calculating correlation coefficients between devices, and constructing spatial dimension feature vectors;

[0123] Performing a tensor fusion operation on the time dimension feature vector and the space dimension feature vector to obtain a device association matrix;

[0124] Based on the device association matrix and the trusted fault types, a device topology graph including device nodes and connection edges is constructed. A spectral clustering algorithm with adaptive similarity is used to dynamically partition the device topology graph to obtain multiple device partitions. The device density distribution within each device partition is calculated to obtain the intra-regional eigenvalue. The connection edge weights between the device partitions are calculated to obtain the inter-regional eigenvalue. A transfer function including a delay factor and an attenuation coefficient is constructed. The intra-regional eigenvalue and the inter-regional eigenvalue are input into the transfer function to calculate the fault diffusion probability between the device partitions.

[0125] Marking the location of the faulty device as the starting node in the device topology map, performing forward propagation calculation based on the fault diffusion probability to obtain an impact propagation path; marking the device nodes on the impact propagation path whose fault diffusion probability exceeds a preset probability threshold as the affected device locations, calculating the transfer importance of each node in the device topology map, and obtaining the source device location through reverse tracing;

[0126] The impact propagation path and the source device location are combined and optimized to obtain the fault propagation path, the risk level of each device on the fault propagation path is calculated and ranked, and the fault impact prediction result and device priority list are output.

[0127] For example, first, the feature data is decomposed in the time dimension. Feature data usually includes the operating status, performance indicators, fault records, etc. of various types of equipment, and these data are presented in the form of time series. Through time dimension decomposition, the data is decomposed into hourly, daily, and weekly subsequences. Each subsequence represents the data characteristics at a different time granularity. During the decomposition process, the hourly subsequence can capture the dynamic changes of the equipment in a short period of time, the daily subsequence reveals the stability and fluctuation of the equipment in a longer period, and the weekly subsequence helps identify the trend changes of the equipment in a longer time range. By analyzing these subsequences, the periodic patterns and status trends therein are extracted. For example, the equipment may exhibit periodic failures at certain time points, or there may be obvious trend changes in certain time periods. This information can help identify the key characteristics of the equipment operation.

[0128] Next, the physical connection distance, electrical circuit connection, and device functional relationship are extracted from the feature data from a spatial perspective. Physical connection distance refers to the physical distance between devices, for example, whether the devices are located in the same room or adjacent to each other. Electrical circuit connection refers to the connection between devices through circuits, reflecting the interdependence of devices in terms of power supply or data transmission. Device functional relationship refers to whether there is a functional dependency or collaboration between devices. For example, the failure of one device may cause the shutdown of another. By extracting these features, the correlation coefficient between devices can be calculated, indicating the degree of association between them. The higher the correlation coefficient, the stronger the degree of association between the devices and the greater the probability of fault propagation between them. Based on the calculated correlation coefficient, spatial dimension feature vectors are constructed, each of which reflects the spatial relationship between the devices.

[0129] The time- and space-dimensional feature vectors are combined through tensor fusion to produce a device association matrix. Tensor fusion combines multidimensional data, integrating information from both the time and space dimensions to construct a complete device association matrix. This matrix accurately describes the relationships between devices under varying temporal and spatial conditions, reflecting the potential risk of fault propagation.

[0130] Based on the device association matrix and trusted fault types, a device topology graph consisting of device nodes and edges is constructed. Device nodes represent individual devices, while edges represent the relationships between devices. By analyzing the connectivity between devices, the device topology graph is dynamically partitioned using a spectral clustering algorithm based on adaptive similarity. This algorithm automatically divides devices into different partitions based on their similarities, with devices within each partition exhibiting strong correlations. After obtaining the device partitions, the device density distribution within each partition is calculated to represent the distribution of devices within each partition and to obtain the regional eigenvalues ​​for that partition. Furthermore, the edge weights between device partitions are calculated. These weights represent the strength of the relationships between partitions, and therefore the likelihood of fault propagation between different zones. Based on these intra- and inter-zone eigenvalues, a transfer function is constructed. This transfer function integrates the delay factor and attenuation coefficient to simulate the propagation of device faults between device partitions.

[0131] By inputting the characteristic values ​​within and between regions into the transfer function, the probability of fault propagation between device partitions is calculated. This probability reflects the likelihood that a fault occurring within a device partition will spread to other regions. This process helps identify critical areas for fault propagation and predict the devices that may be affected.

[0132] In the device topology diagram, the location of the faulty device is marked as the starting node. The starting node represents the device where the fault occurred. Starting from this device, the fault will propagate to surrounding devices. Based on the fault diffusion probability between devices, a forward propagation calculation is performed to determine the impact propagation path. The fault diffusion probability of each device node along the propagation path is calculated to determine which devices will be affected by the fault. When the fault diffusion probability of a device on the propagation path exceeds a preset probability threshold, it is marked as an affected device location. This threshold is set based on the device's importance and historical fault data to screen out devices that are likely to cause serious failures.

[0133] By calculating the transfer importance of each node in the device topology, we can trace back to the source device. Transfer importance refers to the degree of influence a device has on fault propagation, reflecting its impact on the spread of the fault. By tracing back, we can find the source device of the fault and identify the key devices that caused the fault to propagate.

[0134] Finally, the impact propagation path and the location of the source device are combined and optimized to obtain a complete fault propagation path. The fault propagation path shows the propagation process of the fault from the source device to other devices. By ranking the risk level of each device along this path, a priority list of devices is generated, helping operations and maintenance personnel prioritize high-risk devices, taking proactive countermeasures, and reducing the impact of the fault on the overall system.

[0135] In this embodiment, the device association matrix constructed by combining time and space features through tensor fusion can fully reflect the relationship between devices and predict the possible fault propagation path based on the fault type. The construction of the device topology map and the dynamic partitioning algorithm further optimize the partition management of the device, making the identification of the fault propagation path more accurate and avoiding the limitations of the influence of a single factor in the traditional method. The spectral clustering algorithm based on adaptive similarity can dynamically adjust the device partitions and calculate the density distribution and connection weight of each partition to accurately identify the key areas of fault propagation. During the fault propagation process, the transmission importance of the device can guide the tracing of the source of the fault, thereby making the fault response more timely and reducing potential losses. This solution not only improves the accuracy of equipment fault prediction, but also optimizes the equipment maintenance strategy, helping operation and maintenance personnel to accurately identify high-risk equipment before the fault occurs and implement priority processing, thereby improving the reliability and stability of the system and reducing the impact of the fault on the overall system.

[0136] In an optional embodiment,

[0137] Construct a transfer function including a delay factor and an attenuation coefficient, input the characteristic values ​​within the region and the characteristic values ​​between regions into the transfer function, and calculate the fault diffusion probability between device partitions, including:

[0138] Obtaining physical propagation delay and response delay between devices, classifying them according to transmission type to obtain delay classification results, performing segmented mapping on the delay classification results to obtain delay mapping values, and constructing a delay factor based on the delay mapping values;

[0139] Collecting the physical distance, electrical impedance, and functional correlation between devices, calculating a spatial attenuation value based on the physical distance, calculating a signal attenuation value based on the electrical impedance, and calculating a coupling attenuation value based on the functional correlation, hierarchically combining the spatial attenuation value, signal attenuation value, and coupling attenuation value to obtain an attenuation combination value, and constructing an attenuation coefficient based on the attenuation combination value; and constructing a nonlinear transfer function based on the delay factor and the attenuation coefficient.

[0140] Calculate the spatial distribution density value of devices in the area and the distance distribution value between devices, calculate the regional density characteristics based on the spatial distribution density value and the distance distribution value, and fuse the regional density characteristics with the device operating parameters to obtain the regional characteristic value; calculate the number of physical connections and energy transfer power values ​​of devices at the regional boundary, calculate the connection transmission characteristics based on the physical connection number and the energy transfer power value, and fuse the connection transmission characteristics with the device operating status to obtain the inter-regional characteristic value; establish device connection constraints and energy transfer constraints based on the intra-regional characteristic value and the inter-regional characteristic value;

[0141] Historical fault samples are classified according to fault type and propagation path to obtain training samples. The training samples are input into a nonlinear transfer function, and iterative training is performed using device connection constraints and energy transfer constraints to obtain corrected values ​​for the delay factor and attenuation coefficient. The intra-regional and inter-regional eigenvalues ​​are input into the nonlinear transfer function, and the propagation calculation results are dynamically corrected based on the corrected values ​​to output the fault diffusion probability between device partitions.

[0142] For example, first, the physical propagation delay and response delay between devices are obtained. Physical propagation delay refers to the time it takes for a signal to propagate through a transmission medium (such as a cable or optical fiber), and is typically related to the propagation distance and the physical properties of the medium. Response delay refers to the time it takes for a device to respond after receiving a signal, including both device processing delay and signal transmission time. These two delay values ​​can be calculated by monitoring inter-device communication delay and device response speed in real time. To classify this delay data, it is first necessary to categorize it based on the propagation path and transmission medium type, for example, distinguishing between wired and wireless signal propagation, or by network topology. These classification results are then segmented and mapped, with the delay values ​​mapped to different delay intervals to generate delay mapping values. These delay mapping values ​​represent the degree of signal propagation delay between devices. These mapped values ​​can then be used to construct a delay factor, which reflects the impact of signal propagation delay between devices.

[0143] Next, the physical distance, electrical impedance, and functional correlation between devices are collected. Physical distance refers to the straight-line distance between devices and can be obtained through a sensor network or a positioning system between devices. Electrical impedance refers to the impedance of the electrical lines between devices, typically determined by factors such as cable material and length, and signal frequency. It can be measured using a network analyzer. Functional correlation measures the degree of functional interdependence between devices and is typically assessed based on how well they collaborate during operation. For example, if two devices in a power system jointly control a critical load, their functional correlation is high. When calculating attenuation based on these parameters, the spatial attenuation value is first calculated using the physical distance. Spatial attenuation is generally inversely proportional to the distance between devices. Electrical impedance affects signal attenuation, so the signal attenuation value is calculated based on the resistance value. Functional correlation reflects the coupling effect when devices work together, and a coupling attenuation value is calculated. These three attenuation values ​​are then weighted and combined to form a combined attenuation value. This is then used to construct the attenuation coefficient, which reflects the comprehensive attenuation effect between devices during signal propagation.

[0144] Next, a nonlinear transfer function is constructed based on the delay factor and attenuation coefficient. This transfer function describes how signals or fault information propagate between devices. Since the delay factor and attenuation coefficient affect the speed and strength of signal propagation, they are used as inputs in the transfer function for nonlinear modeling, reflecting the actual propagation characteristics between devices. Machine learning algorithms can be used for model training, enabling the transfer function to adaptively adjust based on historical data and optimize propagation results.

[0145] When calculating the spatial distribution density and inter-device distance distribution of devices within a region, the distribution of devices within the region must be considered. The spatial distribution density indicates the number density of devices within a specific region and is typically calculated using the spatial layout of the devices. The inter-device distance distribution indicates the distribution of distances between devices within the region and requires calculation based on the device's location information. These two values ​​are used to calculate a regional density feature, which reflects the potential for fault propagation within the region. A high density of devices within a region can lead to rapid propagation of fault information between devices. The regional density feature is combined with device operating parameters (such as load, temperature, and operating status) to generate a regional feature value that comprehensively reflects the risk of fault propagation within the region.

[0146] To calculate the number of physical connections and energy transfer power for devices at the region boundary, we first need to know the number of physical connections between the device and other external devices. This number can be calculated using device interface data or communication protocols. The energy transfer power indicates the efficiency of energy transfer between devices and is typically correlated with the device's operating status and energy transfer efficiency. These connection transfer characteristics are combined with the device's operating status (such as workload and fault type) through feature fusion to generate an inter-region feature value. This value reflects the device's inter-region energy transfer capability and its importance in fault propagation.

[0147] Based on the eigenvalues ​​within and between regions, device connectivity constraints and energy transfer constraints can be established. Connectivity constraints reflect the strength of the physical connection between devices, while energy transfer constraints represent the resistance to energy transfer between devices. These constraints are used to restrict the propagation paths of signals between devices, thereby more accurately simulating fault propagation.

[0148] By classifying historical fault samples, we can label fault types and propagation paths, generating training samples. These training samples are then fed into a nonlinear transfer function. Through iterative training, the delay factor and attenuation coefficient are adjusted to ensure that the transfer function more accurately reflects the actual fault propagation process. During the training process, device connection constraints and energy transfer constraints are used to optimize the propagation results, ensuring that the propagation process is more realistic.

[0149] Finally, based on the corrected delay factor and attenuation coefficient, the intra-regional and inter-regional eigenvalues ​​are input into a nonlinear transfer function for dynamic correction, outputting the probability of fault spread between device partitions. These probabilities can help operations and maintenance personnel accurately predict the extent of fault spread and provide a basis for maintenance decisions.

[0150] Existing technologies for fault propagation prediction typically rely on simple linear models or static rules, making it difficult to accurately characterize fault propagation characteristics in complex device environments. For example, some methods evaluate signal attenuation between devices based solely on a single physical parameter, such as distance or electrical impedance, while ignoring the impact of functional relevance on fault propagation. Furthermore, traditional methods lack detailed modeling of inter-device delays, resulting in inaccurate descriptions of the temporal characteristics of signal propagation, which in turn affects prediction accuracy.

[0151] This application improves the calculation method of the fault diffusion probability by constructing a nonlinear transfer function that includes a delay factor and an attenuation coefficient. The delay factor is introduced to accurately characterize the time characteristics of signal transmission between devices by classifying and segmenting the physical propagation delay and the response delay. Taking into account the physical distance, electrical impedance and functional correlation, an attenuation coefficient is constructed to comprehensively evaluate the attenuation of signal propagation between devices. By calculating the regional density characteristics and the connection transmission characteristics, the fusion of intra-regional and inter-regional characteristics is achieved, so that the fault propagation prediction not only considers the operating status of the equipment, but also combines its spatial distribution characteristics to improve the comprehensiveness and accuracy of the prediction. In addition, by iteratively optimizing the delay factor and attenuation coefficient through training samples, the model can be adaptively adjusted to improve the robustness and applicability of the prediction.

[0152] The improvements proposed in this application aim to enhance the accuracy and adaptability of fault propagation prediction, overcoming the limited adaptability of traditional methods to complex equipment environments. The improved model will more accurately reflect the dynamics of fault propagation, providing a more precise prediction basis for equipment operation and maintenance. This will optimize maintenance strategies, improve system reliability, reduce maintenance costs, and ensure long-term stable equipment operation.

[0153] Figure 3 This is a diagram analyzing the fault propagation delay and attenuation characteristics of an embodiment of the present invention. Figure 3 This figure shows the changing characteristics of fault signal propagation strength at different physical distances. The horizontal axis represents the physical distance between devices (0-50 meters), and the vertical axis represents the fault signal propagation strength (0.0-0.8). The figure compares three different propagation models: this technical solution (circles), the traditional exponential decay model (squares), and the linear decay model (triangles).

[0154] The data shows that at close distances (0-10 meters), all three models exhibit high propagation intensities, with initial values ​​of 0.75. However, as distance increases, the attenuation characteristics of the different models show significant differences. This technical solution maintains a propagation intensities of 0.68 at 10 meters, while the exponential and linear models drop to 0.55 and 0.50, respectively. At medium distances (20-30 meters), this technical solution exhibits a more gradual attenuation, maintaining a propagation intensities of 0.52 at 20 meters, while the other two models drop to 0.35 and 0.30, respectively.

[0155] Especially at long distances (40-50 meters), this technical solution exhibits better propagation characteristics, maintaining a propagation strength of 0.12 at 50 meters, while the exponential model and linear model drop to 0.08 and 0.05, respectively. This demonstrates that this technical solution significantly improves the propagation characteristics of long-distance fault signals by introducing dynamic adjustment of the delay factor and attenuation coefficient, increasing the effective distance of fault detection and making fault warnings for remote equipment more reliable. Overall, the nonlinear transfer function of this technical solution exhibits superior propagation characteristics across all distance ranges, which is of great significance for fault warnings for large equipment clusters.

[0156] In an optional embodiment,

[0157] A hierarchical reinforcement learning algorithm is used to coordinate and optimize the dispatch of equipment within a virtual power plant. The upper-layer agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list. The lower-layer agent generates equipment control instructions based on the equipment combination plan and fault risk score. The coordinated dispatch optimization of virtual power plant equipment includes:

[0158] Obtain the predicted value of equipment failure impact, equipment operating status value, and equipment priority score to construct the upper state space, and construct the upper action space based on the equipment combination plan; construct the upper reward function based on the weighted combination of system reliability index, operating efficiency index, and scheduling cost;

[0159] The device combination scheme, device failure risk score, and real-time operating parameters are used to construct a lower-level state space, and the device control instructions are used to construct a lower-level action space; a lower-level reward function is constructed based on a dynamic weight combination of a system stability index, a response time index, and the failure risk score;

[0160] The device collaborative impact is calculated based on the fault impact prediction results and the priority list, and the device collaborative impact is used as the weight coefficient to construct a weighted state feature. Historical scheduling data is obtained to calculate the scheduling association strength between devices. Based on the scheduling association strength, the weighted state feature is adaptively adjusted. The adjusted feature is input into the upper-level state space, and the upper-level state action value is calculated according to the upper-level reward function to generate a device combination plan.

[0161] The device combination scheme and the fault risk score are deeply fused to obtain the device association features, which are then input into the lower-level state space. The lower-level state action value is calculated based on the lower-level reward function to generate device control instructions.

[0162] Execute device control instructions to obtain control results, calculate reward values ​​based on the control results, determine the sampling weights of the experience replay pool based on the historical reward distribution of different scheduling scenarios, and obtain training samples from historical data; based on the training samples and reward values, perform online updates on the collaborative influence and scheduling association strength;

[0163] The updated collaborative influence and scheduling association strength are input into the upper-level intelligent agent and the lower-level intelligent agent respectively to generate optimized equipment combination plans and optimized equipment control instructions, thereby realizing the collaborative scheduling optimization of virtual power plant equipment.

[0164] For example, first, the monitoring system and data acquisition module obtain the predicted fault impact value, real-time operating status information, and equipment priority score for each device. The predicted fault impact value can be obtained through a deep learning prediction model based on the device's historical fault records and maintenance logs. Its numerical value describes the impact of a possible future device failure on the overall system. Operating status information includes multiple indicators such as the device's current load, temperature, and vibration. The equipment priority score is determined based on the device's criticality and historical reliability in the virtual power plant. For example, if the failure prediction of a key device shows an impact of approximately 5%, all status indicators are at a good level, and the priority score is rated the highest, the system will assign a higher weight to the device in subsequent scheduling.

[0165] Next, the above-mentioned collected data is used to construct an upper-level state space, in which each device is presented as a description vector containing fault prediction, operating status and priority score. Historical scheduling data and expert experience are used to construct the action space of the equipment combination plan, which contains various combinations of switches and operating modes between different devices. For example, in a certain scheduling scenario, the equipment combination plan may be that device A and device C run at the same time while device B is on standby. At the same time, by analyzing the equipment failure prediction results and priority list, the collaborative impact between devices is calculated, that is, the degree of mutual influence between devices when a fault occurs is described. The description is expressed in the form of text and data cases. For example, the collaborative impact between device A and device B is rated as medium, indicating that the two have a certain degree of dependence in scheduling.

[0166] Furthermore, based on the overall system reliability, operational efficiency, and dispatch costs, a higher-level reward function is constructed using specific rules (such as empirical rules or expert-defined weighting strategies). Equipment combinations that reduce dispatch costs while ensuring system reliability and efficient operation are rewarded higher. The reward function is constructed not by relying on mathematical formulas but by comprehensively describing the contributions of various indicators. For example, if an equipment combination improves system efficiency by nearly 10% while significantly reducing dispatch costs, its reward value will be increased accordingly.

[0167] In terms of lower-level scheduling, the upper-level generated device combination plan, each device's failure risk score, and real-time operating parameters are integrated into the lower-level state space. The lower-level action space is defined as a collection of device control instructions, covering operations such as startup, shutdown, load adjustment, and fault self-detection. The lower-level reward function depends on the dynamic performance of system stability, response time, and device failure risk. It is determined through textual descriptions and empirical data. For example, positive incentives are given when device control instructions can reduce response time to a predetermined level while simultaneously improving system stability.

[0168] During the state feature construction phase, the collaborative impact of the devices is calculated through fault impact prediction and priority list calculations. This impact is then combined with the original state features as a weight coefficient to generate a weighted state feature. For example, if device A has a high failure risk and a high priority, a larger weight is added to its state description. Simultaneously, the scheduling association strength between devices is extracted from historical scheduling data to describe the frequency and dependencies of collaborative scheduling between devices. If devices A and B frequently collaborated in past scheduling, the information about these two devices is appropriately enhanced in the weighted state feature. The adaptively adjusted weighted state features are then input into the upper-level state space, allowing the upper-level agent to evaluate the state action value of various device combination schemes, ultimately generating an optimal or near-optimal device combination scheme. This scheme can be described as a combination configuration in which devices A, B, and C are in operation, standby, or backup states, respectively.

[0169] Subsequently, the device combination plan output by the upper layer and the real-time fault risk score undergo deep feature fusion to generate device association features. This process comprehensively considers the operating status, fault risk, and synergistic impact of each device. The generated device association features reflect the correlation between devices in a detailed manner, such as information such as device A's risk score being medium to high and device B having a high scheduling correlation with device A. This device association feature is passed as input to the lower-level state space. Based on a pre-set reward function, the lower-level intelligent agent generates highly targeted and responsive device control instructions by evaluating the value of state actions. These instructions may include operations such as adjusting the load of the equipment, switching operating modes, and early warning detection.

[0170] After the device control instruction is executed, the system collects the control results in real time, such as device response time, load changes, and scheduling effects, as feedback data. The feedback data is used to calculate the actual reward value and determine the sampling weight of each training sample in the experience replay pool based on the historical reward distribution under different scheduling scenarios. For example, in a certain scheduling, if the device scheduling success rate reaches more than 90% and the response time is significantly shortened, the data samples under this scheduling scenario are given a higher sampling weight so that they can be used first in subsequent training. The training samples extracted from the experience replay pool are combined with the actual reward value to update the parameters of the collaborative influence between devices and the scheduling association strength online. The update process adopts a descriptive feedback mechanism, that is, the previously set weights are fine-tuned according to the new sampled data to ensure that the description of the synergy between devices and the scheduling association can timely reflect the current system operation status.

[0171] The updated synergy influence and scheduling association strength are fed back to the upper-level agent and lower-level agent, respectively, prompting the upper-level agent to further optimize the equipment combination plan, while the lower-level agent generates more precise equipment control instructions. The entire scheduling process forms a closed-loop feedback system, adjusting and continuously improving the overall performance of the equipment scheduling plan in real time, ensuring that the virtual power plant system can respond to various emergencies and maintain stable and efficient operation.

[0172] In this embodiment, hierarchical reinforcement learning is used to achieve intelligent collaborative scheduling and adaptive optimization of virtual power plant equipment, making full use of fault prediction, equipment status, and priority information to build a closed-loop scheduling system. The system can generate and adjust equipment combination plans and control instructions in real time, improving overall operating efficiency and system reliability, while effectively reducing scheduling costs and failure risks. The online update of the collaborative influence between equipment and the scheduling association strength makes scheduling decisions more accurate, with the ability to quickly respond to sudden failures, optimize load distribution, and provide dynamic early warning. The overall effect is reflected in the continuous improvement of intelligent scheduling management, stable and efficient operation, and autonomous learning capabilities, thereby enhancing the safety and economic benefits of virtual power plants.

[0173] In an optional embodiment,

[0174] Execute device control instructions to obtain control results, calculate reward values ​​based on the control results, determine the sampling weights of the experience replay pool based on the historical reward distribution of different scheduling scenarios, and obtain training samples from historical data. Online updates of the collaborative influence and scheduling association strength based on the training samples and reward values ​​include:

[0175] Executing device control instructions to obtain control results consisting of power balance, response time, energy loss, and system stability; calculating a system reliability score based on the power balance and system stability, and calculating a scheduling efficiency score based on the response time and energy loss, and using a weighted combination of the system reliability score and the scheduling efficiency score as a reward value;

[0176] According to the scheduling scenarios of different device combinations, state transition samples are classified and stored to build an experience replay pool. The state transition samples include device combination plans, device control instructions, reward values, and post-control states. Sampling weights are determined based on the historical reward distribution of scheduling scenarios, and training samples are selected from the experience replay pool based on the sampling weights.

[0177] Extracting response timing characteristics and power coupling characteristics between devices based on the device combination scheme in the training sample, and constructing a device collaborative influence matrix; constructing a collaborative influence loss function based on the temporal correlation between the device collaborative influence matrix and the reward value, and optimizing the collaborative influence loss function to perform online update of the collaborative influence;

[0178] Based on the device control instructions in the training samples, the scheduling dependency characteristics and control coupling characteristics between devices are extracted to construct a device scheduling association matrix; according to the dynamic correlation between the device scheduling association matrix and the reward value, a scheduling association strength loss function is constructed, and the scheduling association strength is updated online by optimizing the scheduling association strength loss function.

[0179] For example, the effectiveness of the equipment control instructions executed in the virtual power plant is evaluated to obtain four key indicators: power balance reflects the degree of matching between the output power of the source, grid, load and storage equipment and the load demand; response time represents the time interval from the equipment receiving the control instruction to reaching the target state; energy loss measures the loss level during the equipment adjustment process; system stability describes the fluctuation of key parameters such as voltage and frequency. The system reliability score is calculated based on the power balance and system stability. When the power balance is greater than 95% and the fluctuation range of key parameters is less than 3% of the rated value, the system reliability score is high; the dispatch efficiency score is calculated based on the response time and energy loss. When the average response time is less than ten seconds and the total energy loss rate is less than 5%, the dispatch efficiency score is high. The two types of scores are weighted and combined to obtain a comprehensive reward value.

[0180] Establish a scenario-specific experience replay pool storage mechanism, and construct independent experience replay pools for different scheduling scenarios, such as peak-valley regulation, accident recovery, and economic scheduling. Each state transition sample contains four elements: the equipment combination plan records the equipment combination involved in the scheduling; the equipment control instruction contains the target state and control parameters of each device; the reward value reflects the effectiveness of the control; and the post-control state describes the final operating state of the system. Based on historical scheduling data, the reward distribution characteristics of each scenario are statistically analyzed, and the scenario sampling weight is calculated. The higher the reward value, the greater the sampling probability of the scenario. Training samples are selected from the experience replay pool of different scenarios based on the sampling weight.

[0181] Analyze the response characteristics of each device in the device combination scheme to extract timing characteristics such as device start-up and shutdown sequence and power regulation rate. Analyze the power feedback relationship between devices to extract power coupling characteristics such as device capacity matching and transmission loss. Based on the extracted characteristics, construct a device synergistic influence matrix, where the matrix elements represent the degree of synergistic influence between devices. By analyzing the correlation between the synergistic influence matrix and the historical reward sequence, construct a synergistic influence loss function, which reflects the deviation between the current synergistic influence and the influence corresponding to the optimal reward. The synergistic influence is updated in real time using optimization algorithms such as gradient descent.

[0182] The execution process of device control commands is analyzed to extract scheduling dependency features, such as command response sequence and control authority. The control relationships between devices are analyzed to extract control coupling features, such as control parameter coupling and state constraints. A device scheduling association matrix is ​​constructed based on these extracted features, with the matrix elements representing the strength of scheduling associations between devices. By analyzing the correlation between the scheduling association matrix and dynamic reward changes, a scheduling association strength loss function is constructed, which characterizes the difference between the current association strength and the strength corresponding to the optimal scheduling effect. The scheduling association strength is dynamically updated using online learning methods.

[0183] Existing technologies for equipment scheduling optimization typically employ rule-driven or simple static models, making them incapable of accurately adapting to the dynamic changes of complex scheduling scenarios. For example, some methods adjust equipment control strategies based solely on fixed rules, failing to fully leverage historical data for learning. This results in difficulties optimizing scheduling efficiency and system reliability. Furthermore, existing methods lack precise modeling of the impact of equipment collaboration and the strength of scheduling associations, ignoring the dynamic interactions between equipment and thus hindering scheduling optimization effectiveness.

[0184] This application improves the accuracy and adaptability of equipment scheduling decisions by constructing an optimization method based on experience replay and online update. An experience replay pool is constructed for different scheduling scenarios, and the sampling weights are dynamically adjusted based on the historical reward distribution to make the training samples more representative, thereby improving the generalization ability of the model. An online update mechanism for collaborative influence and scheduling association strength is proposed. By extracting the response timing characteristics, power coupling characteristics, scheduling dependency characteristics and control coupling characteristics between devices, a collaborative influence matrix and a scheduling association matrix are constructed respectively, and a loss function is constructed based on the correlation of the reward values ​​to achieve adaptive optimization of key parameters. Through methods such as gradient optimization, the model can dynamically adjust the collaborative influence and scheduling association strength between devices according to the actual scheduling effect, thereby improving the intelligence level of the scheduling strategy. Taking into account indicators such as power balance, system stability, response time and energy loss, the reward calculation method is optimized so that the scheduling strategy can take into account both system reliability and scheduling efficiency.

[0185] The improvements in this application aim to enhance the accuracy and intelligence of equipment scheduling optimization, overcoming the lack of adaptability of traditional methods to dynamic scheduling scenarios. The improved model can fully leverage historical data to dynamically optimize scheduling decisions, making equipment scheduling strategies more flexible and adaptable. This improves system stability, reduces energy loss, and increases overall scheduling efficiency, achieving better resource allocation.

[0186] The core innovation of this invention lies in the organic integration of equipment fault diagnosis, propagation prediction, and scheduling optimization. By deeply mining equipment operating data, a comprehensive fault warning and handling framework is established. This method not only considers the operating status of individual devices but also fully accounts for the interactions between devices, making fault warning and handling more accurate and effective. Furthermore, the use of a hierarchical reinforcement learning algorithm for equipment scheduling gives the system greater environmental adaptability and optimization effects.

[0187] Improved fault warning accuracy significantly reduces operation and maintenance costs; optimized fault propagation prediction improves fault handling efficiency; and improved equipment scheduling enhances overall system efficiency. These innovations not only address key issues in existing technologies but also provide new technical insights for the intelligent operation of virtual power plants.

[0188] Figure 4 Schematic diagram of the structure of the real-time monitoring and optimization system for source, grid, load and storage equipment of the virtual power plant platform according to the embodiment of the present invention. Figure 4 As shown, the system includes:

[0189] The first unit is used to collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Based on the feature data, the equipment fault type is identified. The equipment fault type is matched with historical fault data for similarity to generate a fault risk score. Based on the fault risk score, the credibility of the equipment fault type is verified to obtain a credible fault type.

[0190] The second unit is used to decouple feature data in time and space dimensions to reconstruct a device association matrix. Based on the device association matrix and trusted fault types, a device topology map is constructed. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Based on the fault diffusion probability, bidirectional tracking is performed on the device topology map to generate a fault propagation path. Devices along the fault propagation path are hierarchically ranked and output as fault impact prediction results and a device priority list.

[0191] The third unit is used to use a hierarchical reinforcement learning algorithm to coordinate and optimize the scheduling of equipment within the virtual power plant. The upper-level intelligent agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list, and the lower-level intelligent agent generates equipment control instructions based on the equipment combination plan and fault risk score, thereby realizing the coordinated scheduling optimization of virtual power plant equipment.

[0192] According to a third aspect of the embodiments of the present invention,

[0193] An electronic device is provided, comprising:

[0194] processor;

[0195] a memory for storing processor-executable instructions;

[0196] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0197] According to a fourth aspect of the embodiments of the present invention,

[0198] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0199] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time monitoring and optimization method for source, grid, load and storage equipment on a virtual power plant platform, characterized in that: include: Collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Based on the feature data, identify the equipment fault type. Match the equipment fault type with historical fault data for similarity to generate a fault risk score. Based on the fault risk score, verify the credibility of the equipment fault type to obtain a credible fault type. Equipment operation data refers to the time series data generated by various types of equipment in the virtual power plant during operation, including but not limited to voltage, current, power, temperature, frequency, and equipment operating status. The feature data is decoupled and reconstructed in time and space dimensions to obtain a device association matrix. A device topology map is constructed based on the device association matrix and trusted fault types. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Bidirectional tracking is performed on the device topology map based on the fault diffusion probability to generate a fault propagation path. The devices on the fault propagation path are hierarchically sorted and the fault impact prediction results and device priority list are output. A hierarchical reinforcement learning algorithm is used to coordinate and optimize the equipment scheduling in a virtual power plant. The upper-level intelligent agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list, and the lower-level intelligent agent generates equipment control instructions based on the equipment combination plan and the fault risk score, thereby realizing the coordinated scheduling optimization of the virtual power plant equipment.

2. The method according to claim 1, characterized in that Collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Based on the feature data, identify the equipment fault type. Match the equipment fault type with historical fault data for similarity, generate a fault risk score, and verify the credibility of the equipment fault type based on the fault risk score. The credible fault types obtained include: Performing a wavelet transform on the equipment operation data to obtain multi-scale coefficients, calculating the ratios of adjacent scale coefficients to construct a feature matrix; performing a singular value decomposition on the feature matrix to obtain a signal subspace and a noise subspace, calculating the orthogonal vector product of the signal subspace and the noise subspace to obtain an adaptive threshold, projecting the equipment operation data into the signal subspace, and reconstructing based on the adaptive threshold to obtain denoised data; Construct an autoencoder model, in which the encoding part maps the denoised data into intermediate features, and obtains latent variables from the intermediate features through a nonlinear activation function; construct a discriminator network, input the latent variables and preset Gaussian distribution samples into the discriminator network to obtain discrimination probabilities, and calculate the cross entropy based on the discrimination probabilities and the true labels to obtain adversarial loss; perform gradient descent optimization on the latent variables based on the adversarial loss to obtain fault features; The fault features are subjected to sliding convolution through a one-dimensional convolution kernel to obtain a time series feature sequence, and the gated recurrent unit is used to calculate the state vector of the time series feature sequence time step by time step; the conditional entropy of the state vector at different time steps is calculated to obtain a time series weight, and the time series weight is weighted and summed with the time series feature sequence to obtain a global feature, which is input into a softmax classifier to obtain an initial fault type; Based on conditional entropy, historical fault samples are stratified according to operating condition parameters. The fault type transition frequency is calculated in each layer of samples to obtain a state transition matrix. The global features are spliced ​​with the current operating condition parameters and input into the state transition matrix to obtain the predicted probability. The mutual information between the initial fault type and the predicted probability is calculated to obtain a credibility score. When the credibility score exceeds a preset threshold, it is confirmed as a credible fault type.

3. The method according to claim 2, characterized in that Based on conditional entropy, historical fault samples are stratified according to operating condition parameters. The fault type transition frequency is calculated in each layer of samples to obtain a state transition matrix. The global features and current operating condition parameters are concatenated and input into the state transition matrix to obtain a predicted probability. The mutual information between the initial fault type and the predicted probability is calculated to obtain a credibility score. Divide the numerical interval of the operating condition parameter into multiple subintervals, count the occurrence frequency of the fault type in each subinterval, and calculate the conditional entropy of the operating condition parameter on the fault type; calculate the mean and standard deviation of the conditional entropy, and obtain an adaptive threshold based on the mean minus the standard deviation; construct a hierarchical parameter set for the operating condition parameters whose conditional entropy is less than the adaptive threshold; Calculate the local density and minimum distance of historical fault samples in the space corresponding to the hierarchical parameter set, determine the density peak point as the cluster center based on the local density, and assign the sample points to the cluster center based on the minimum distance to obtain the working condition layer; In each operating condition layer, the total number of samples of the fault type and the number of transition samples between the fault types are calculated; when the total number of samples is greater than a preset sample number threshold, the ratio of the number of transition samples to the total number of samples is calculated to obtain the fault type transition frequency; otherwise, the fault type transition frequency is set to zero; and a state transition matrix is ​​constructed based on the fault type transition frequency. Obtain the global features and current operating parameters of the sample to be verified, and determine the target operating layer based on the current operating parameters; combine the global features and the current operating parameters to form a feature vector, and input the state transition matrix corresponding to the target operating layer to obtain the predicted probability distribution; Obtain the initial fault type probability distribution of the sample to be verified, divide the probability of each fault type in the initial fault type probability distribution by the probability of the corresponding fault type in the predicted probability distribution to obtain a probability ratio; and obtain a credibility score based on the sum of the logarithm of the probability ratio and the product of the initial fault type probability.

4. The method according to claim 1, wherein The feature data is decoupled and reconstructed in time and space dimensions to obtain a device association matrix. A device topology map is constructed based on the device association matrix and trusted fault types. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Perform bidirectional tracing on the device topology map based on the fault diffusion probability to generate the fault propagation path; The devices on the fault propagation path are ranked and the fault impact prediction results and device priority list are output, including: Decomposing the feature data in a time dimension into hourly, daily, and weekly subsequences, extracting periodic patterns and state trends from the hourly, daily, and weekly subsequences, and constructing a time dimension feature vector; Extracting physical connection distances, electrical circuit connections, and device functional relationships from the feature data, calculating correlation coefficients between devices, and constructing spatial dimension feature vectors; Performing a tensor fusion operation on the time dimension feature vector and the space dimension feature vector to obtain a device association matrix; Based on the device association matrix and the trusted fault types, a device topology graph including device nodes and connection edges is constructed. A spectral clustering algorithm with adaptive similarity is used to dynamically partition the device topology graph to obtain multiple device partitions. The device density distribution within each device partition is calculated to obtain the intra-regional eigenvalue. The connection edge weights between the device partitions are calculated to obtain the inter-regional eigenvalue. A transfer function including a delay factor and an attenuation coefficient is constructed. The intra-regional eigenvalue and the inter-regional eigenvalue are input into the transfer function to calculate the fault diffusion probability between the device partitions. Marking the location of the faulty device as the starting node in the device topology map, performing forward propagation calculation based on the fault diffusion probability to obtain an impact propagation path; marking the device nodes on the impact propagation path whose fault diffusion probability exceeds a preset probability threshold as the affected device locations, calculating the transfer importance of each node in the device topology map, and obtaining the source device location through reverse tracing; The impact propagation path and the source device location are combined and optimized to obtain the fault propagation path, the risk level of each device on the fault propagation path is calculated and ranked, and the fault impact prediction result and device priority list are output.

5. The method according to claim 4, characterized in that Construct a transfer function including a delay factor and an attenuation coefficient, input the characteristic values ​​within the region and the characteristic values ​​between regions into the transfer function, and calculate the fault diffusion probability between device partitions, including: Obtaining physical propagation delay and response delay between devices, classifying them according to transmission type to obtain delay classification results, performing segmented mapping on the delay classification results to obtain delay mapping values, and constructing a delay factor based on the delay mapping values; Collecting the physical distance, electrical impedance, and functional correlation between devices, calculating a spatial attenuation value based on the physical distance, calculating a signal attenuation value based on the electrical impedance, and calculating a coupling attenuation value based on the functional correlation, hierarchically combining the spatial attenuation value, signal attenuation value, and coupling attenuation value to obtain an attenuation combination value, and constructing an attenuation coefficient based on the attenuation combination value; and constructing a nonlinear transfer function based on the delay factor and the attenuation coefficient. Calculate the spatial distribution density value of devices in the area and the distance distribution value between devices, calculate the regional density characteristics based on the spatial distribution density value and the distance distribution value, and fuse the regional density characteristics with the device operating parameters to obtain the regional characteristic value; calculate the number of physical connections and energy transfer power values ​​of devices at the regional boundary, calculate the connection transmission characteristics based on the physical connection number and the energy transfer power value, and fuse the connection transmission characteristics with the device operating status to obtain the inter-regional characteristic value; establish device connection constraints and energy transfer constraints based on the intra-regional characteristic value and the inter-regional characteristic value; Historical fault samples are classified according to fault type and propagation path to obtain training samples. The training samples are input into a nonlinear transfer function, and iterative training is performed using device connection constraints and energy transfer constraints to obtain corrected values ​​for the delay factor and attenuation coefficient. The intra-regional and inter-regional eigenvalues ​​are input into the nonlinear transfer function, and the propagation calculation results are dynamically corrected based on the corrected values ​​to output the fault diffusion probability between device partitions.

6. The method according to claim 1, characterized in that A hierarchical reinforcement learning algorithm is used to coordinate and optimize the dispatch of equipment within a virtual power plant. The upper-layer agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list. The lower-layer agent generates equipment control instructions based on the equipment combination plan and fault risk score. The coordinated dispatch optimization of virtual power plant equipment includes: Obtain the predicted value of equipment failure impact, equipment operating status value, and equipment priority score to construct the upper state space, and construct the upper action space based on the equipment combination plan; construct the upper reward function based on the weighted combination of system reliability index, operating efficiency index, and scheduling cost; The device combination scheme, device failure risk score, and real-time operating parameters are used to construct a lower-level state space, and the device control instructions are used to construct a lower-level action space; a lower-level reward function is constructed based on a dynamic weight combination of a system stability index, a response time index, and the failure risk score; The device collaborative impact is calculated based on the fault impact prediction results and the priority list, and the device collaborative impact is used as the weight coefficient to construct a weighted state feature. Historical scheduling data is obtained to calculate the scheduling association strength between devices. Based on the scheduling association strength, the weighted state feature is adaptively adjusted. The adjusted feature is input into the upper-level state space, and the upper-level state action value is calculated according to the upper-level reward function to generate a device combination plan. The device combination scheme and the fault risk score are deeply fused to obtain the device association features, which are then input into the lower-level state space. The lower-level state action value is calculated based on the lower-level reward function to generate device control instructions. Execute device control instructions to obtain control results, calculate reward values ​​based on the control results, determine the sampling weights of the experience replay pool based on the historical reward distribution of different scheduling scenarios, and obtain training samples from historical data; based on the training samples and reward values, perform online updates on the collaborative influence and scheduling association strength; The updated collaborative influence and scheduling association strength are input into the upper-level intelligent agent and the lower-level intelligent agent respectively to generate optimized equipment combination plans and optimized equipment control instructions, thereby realizing the collaborative scheduling optimization of virtual power plant equipment.

7. The method according to claim 6, characterized in that Execute device control instructions to obtain control results, calculate reward values ​​based on the control results, determine the sampling weights of the experience replay pool based on the historical reward distribution of different scheduling scenarios, and obtain training samples from historical data; Online updates of collaborative influence and scheduling association strength based on training samples and reward values ​​include: Executing device control instructions to obtain control results consisting of power balance, response time, energy loss, and system stability; calculating a system reliability score based on the power balance and system stability, and calculating a scheduling efficiency score based on the response time and energy loss, and using a weighted combination of the system reliability score and the scheduling efficiency score as a reward value; According to the scheduling scenarios of different device combinations, state transition samples are classified and stored to build an experience replay pool. The state transition samples include device combination plans, device control instructions, reward values, and post-control states. Sampling weights are determined based on the historical reward distribution of scheduling scenarios, and training samples are selected from the experience replay pool based on the sampling weights. Extracting response timing characteristics and power coupling characteristics between devices based on the device combination scheme in the training sample, and constructing a device collaborative influence matrix; constructing a collaborative influence loss function based on the temporal correlation between the device collaborative influence matrix and the reward value, and optimizing the collaborative influence loss function to perform online update of the collaborative influence; Based on the device control instructions in the training samples, the scheduling dependency characteristics and control coupling characteristics between devices are extracted to construct a device scheduling association matrix; according to the dynamic correlation between the device scheduling association matrix and the reward value, a scheduling association strength loss function is constructed, and the scheduling association strength is updated online by optimizing the scheduling association strength loss function.

8. A real-time monitoring and optimization system for source, grid, load and storage equipment on a virtual power plant platform, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to collect equipment operation data and perform noise reduction and feature compression to obtain feature data. Based on the feature data, the equipment fault type is identified, the equipment fault type is matched with historical fault data for similarity, a fault risk score is generated, and the credibility of the equipment fault type is verified based on the fault risk score to obtain a credible fault type. The equipment operation data refers to the time series data generated by various types of equipment in the virtual power plant during operation, including but not limited to voltage, current, power, temperature, frequency, and equipment operating status. The second unit is used to decouple feature data in time and space dimensions to reconstruct a device association matrix. Based on the device association matrix and trusted fault types, a device topology map is constructed. The device topology map is dynamically partitioned and local features are extracted. The fault diffusion probability is calculated by transferring features between regions. Based on the fault diffusion probability, bidirectional tracking is performed on the device topology map to generate a fault propagation path. Devices along the fault propagation path are hierarchically ranked and output as fault impact prediction results and a device priority list. The third unit is used to use a hierarchical reinforcement learning algorithm to coordinate and optimize the scheduling of equipment within the virtual power plant. The upper-level intelligent agent generates an equipment combination plan based on the fault impact prediction results and the equipment priority list, and the lower-level intelligent agent generates equipment control instructions based on the equipment combination plan and fault risk score, thereby realizing the coordinated scheduling optimization of virtual power plant equipment.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images