Relay protection fault diagnosis method and device, electronic equipment and storage medium

Abstract

The invention provides a relay protection fault diagnosis method and device, electronic equipment and a storage medium, and relates to the technical field of relay protection, and the method comprises the steps: obtaining operation data of a power system, carrying out the preprocessing of the operation data to generate standardized time sequence data, and carrying out the spatial feature extraction and time sequence dependence analysis of the data to obtain fault feature representation; and subsequently, an early warning parameter and a protection action strategy are dynamically optimized by adopting a reinforcement learning algorithm, and finally a graded early warning signal is output and a control instruction is generated. The problems that in the prior art, due to the fact that fixed protection parameters lack self-adaptive capacity, multi-source data cannot be efficiently fused, time sequence data long-term dependency is difficult to capture, real-time performance is insufficient, and manual intervention is depended can be solved. The technical effects of improving the relay protection fault diagnosis precision of the power system, enhancing the adaptive capability, guaranteeing real-time and efficient early warning, reducing manual intervention and optimizing protection action decision are achieved.

Description

Technical Field

[0001] This disclosure relates to the field of relay protection technology, and in particular to a relay protection fault diagnosis method and device, electronic equipment and storage medium. Background Technology

[0002] Power system relay protection, as a core technology for ensuring the safe and stable operation of the power grid, is widely used in fault detection and condition assessment. With the development of smart grids and the integration of distributed energy resources, traditional relay protection methods face the dual challenges of dynamic operating environments and complex fault modes. Among related technologies, physical model-based diagnostic methods rely on power grid parameter calculations, making it difficult to cope with load fluctuations and system complexity; experience-based rule-driven schemes are limited by manually set thresholds, resulting in insufficient scenario adaptability; data-driven machine learning methods, while improving recognition accuracy through feature engineering, are limited by the reliance on manual intervention in feature selection and insufficient high-dimensional time-series modeling capabilities. Specifically, the existing technological system covers the entire process from electrical parameter acquisition to fault classification, including key aspects such as threshold-based static protection mechanisms, multi-source data acquisition architectures, and traditional classification algorithms. Its technological evolution path shows a trend from single physical modeling to the integration of multiple technologies.

[0003] However, existing relay protection methods directly use fixed parameter thresholds for fault diagnosis without adaptive adjustments based on the dynamic characteristics of the power grid. This can lead to malfunctions or missed trips. The substantial economic losses caused by relay protection malfunctions each year highlight the urgent need for improvements in the dynamic environmental adaptability and intelligent decision-making capabilities of existing technologies. Summary of the Invention

[0004] This disclosure provides a method and apparatus for fault diagnosis of relay protection, as well as electronic equipment and storage medium. Its main purpose is to at least partially solve one of the technical problems in related technologies.

[0005] According to a first aspect of this disclosure, a method for diagnosing relay protection faults is provided, comprising: Acquire power system operation data and preprocess the operation data to generate standardized time-series data; Spatial feature extraction and temporal dependency analysis are performed on the standardized time-series data to obtain fault feature representations; Based on the fault feature representation, a reinforcement learning algorithm is used to dynamically optimize the early warning parameters and protection action strategies; Based on the optimized warning parameters and protection action strategies, graded warning signals are output and control commands are generated.

[0006] Optionally, acquiring power system operation data and preprocessing the operation data includes: The running data is segmented using a sliding window to form time-series data segments; The time-series data segment is subjected to data augmentation processing, including superimposing noise perturbations in the time domain and adjusting the amplitude and phase in the frequency domain.

[0007] Optionally, the step of performing spatial feature extraction and temporal dependency analysis on the standardized time-series data includes: Based on a pre-trained convolutional neural network, local spatial features of the standardized time-series data are extracted. Long-term temporal dependencies are captured by using a pre-trained long short-term memory network to standardize the time-series data.

[0008] Optionally, the reinforcement learning algorithm is based on a preset learning mechanism, wherein the state space is defined as a joint set of system operating states and fault feature representations, the action space includes adjusting alarm thresholds, setting warning levels, and selecting protection actions, and the effect of the actions is evaluated through a reward function to drive strategy optimization.

[0009] Optionally, the graded early warning signal is divided into multiple levels according to severity, and the control command includes at least one of switching lines, adjusting load, isolating faulty equipment, tripping or closing.

[0010] Optional, also includes: Feature fusion is performed on multi-source operational data. An attention mechanism is used to calculate the weight coefficients of each data source. The fused features are then input into a fully connected layer for nonlinear transformation to output the fault probability distribution.

[0011] According to a second aspect of this disclosure, a relay protection fault diagnosis device is provided, comprising: An acquisition unit is used to acquire power system operation data and preprocess the operation data to generate standardized time-series data; The analysis unit is used to perform spatial feature extraction and temporal dependency analysis on the standardized time series data to obtain fault feature representations; The optimization unit is used to dynamically optimize the early warning parameters and protection action strategies based on the fault feature representation using a reinforcement learning algorithm; The output unit is used to output graded early warning signals and generate control commands based on the optimized early warning parameters and protection action strategies.

[0012] Optionally, the acquisition unit is also used for: The running data is segmented using a sliding window to form time-series data segments; The time-series data segment is subjected to data augmentation processing, including superimposing noise perturbations in the time domain and adjusting the amplitude and phase in the frequency domain.

[0013] Optionally, the analysis unit is also used for: Based on a pre-trained convolutional neural network, local spatial features of the standardized time-series data are extracted. Long-term temporal dependencies are captured by using a pre-trained long short-term memory network to standardize the time-series data.

[0014] Optionally, the reinforcement learning algorithm is based on a preset learning mechanism, wherein the state space is defined as a joint set of system operating states and fault feature representations, the action space includes adjusting alarm thresholds, setting warning levels, and selecting protection actions, and the effect of the actions is evaluated through a reward function to drive strategy optimization.

[0015] Optionally, the graded early warning signal is divided into multiple levels according to severity, and the control command includes at least one of switching lines, adjusting load, isolating faulty equipment, tripping or closing.

[0016] Optional, also includes: The fusion unit is used to perform feature fusion on multi-source operational data. It uses an attention mechanism to calculate the weight coefficients of each data source and inputs the fused features into a fully connected layer for nonlinear transformation to output a fault probability distribution.

[0017] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0018] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0019] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0020] The relay protection fault diagnosis method, device, electronic equipment, and storage medium disclosed herein acquire power system operating data and preprocess it to generate standardized time-series data. Then, spatial feature extraction and time-series dependency analysis are performed to obtain fault feature representations. Subsequently, reinforcement learning algorithms are used to dynamically optimize early warning parameters and protection action strategies. Finally, hierarchical early warning signals are output and control commands are generated. Therefore, it can solve the problems in the prior art, such as the lack of adaptive capability of fixed protection parameters, inability to efficiently integrate multi-source data, difficulty in capturing long-term dependencies of time-series data, insufficient real-time performance, and reliance on manual intervention. It achieves the technical effects of improving the accuracy of power system relay protection fault diagnosis, enhancing adaptive capability, ensuring real-time and efficient early warning, reducing manual intervention, and optimizing protection action decisions.

[0021] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0022] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a flowchart illustrating a relay protection fault diagnosis method provided in an embodiment of the present disclosure. Figure 2 This is a schematic diagram of the structure of a relay protection fault diagnosis device provided in an embodiment of the present disclosure; Figure 3 A schematic block diagram of an example electronic device provided for embodiments of this disclosure. Detailed Implementation

[0023] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0024] The following description, with reference to the accompanying drawings, outlines a relay protection fault diagnosis method and apparatus, electronic device, and storage medium according to embodiments of the present disclosure.

[0025] Figure 1 This is a flowchart illustrating a relay protection fault diagnosis method provided in an embodiment of the present disclosure.

[0026] like Figure 1 As shown, the method includes the following steps: Step 101: Obtain power system operation data and preprocess the operation data to generate standardized time-series data.

[0027] In the embodiments of this disclosure, the first step in the technical process of fault diagnosis and early warning for power system relay protection is to acquire power system operating data and preprocess it to generate standardized time-series data. The power system operating data encompasses various parameter information reflecting the operating status of the power grid and relay protection equipment. This data can be obtained from various components in the power system with data acquisition capabilities. The preprocessing operation is used to eliminate dimensional differences, numerical range deviations, and redundant interference between different parameter data, so that the processed data forms standardized time-series data with a unified format and temporal consistency, providing data support that meets technical requirements for subsequent fault characteristic analysis. As one implementation method, the operating data may include parameters such as voltage, current, frequency, and power. Data acquisition can be completed through relay protection devices, sensors, and distributed control systems (DCS). Preprocessing can employ standardized algorithms for data normalization, sliding window segmentation with specific window lengths and step sizes, and data enhancement through time-domain and frequency-domain perturbations.

[0028] By comprehensively acquiring power system operation data and performing standardized preprocessing, the integrity and standardization of the data used in subsequent fault analysis are effectively ensured, avoiding subsequent processing deviations caused by inconsistent data formats or interference. This lays a reliable data foundation for improving the accuracy of fault diagnosis and early warning in power system relay protection.

[0029] Step 102: Spatial feature extraction and temporal dependency analysis are performed on the standardized time series data to obtain fault feature representation.

[0030] In the embodiments of this disclosure, in the technical process of fault diagnosis for power system relay protection, step 102 aims to perform deep feature mining on the generated standardized time-series data. Through the synergistic effect of spatial feature extraction and temporal dependency analysis, a fault feature representation that can accurately reflect the fault state of the power system is formed. Spatial feature extraction is used to mine local correlation features and key differences between different parameter dimensions in the standardized time-series data, while temporal dependency analysis is used to capture the dynamic evolution patterns and long-term correlations of the data in the time dimension. The combination of the two can comprehensively mine fault-related feature information from both spatial and temporal dimensions, ensuring that the final fault feature representation can fully cover the key feature attributes at the time of fault occurrence, providing a core basis for subsequent fault diagnosis and decision-making. As one implementation method, a network model with local feature extraction capabilities can be used to implement spatial feature extraction, and a network model with time-series data processing capabilities can be used to implement temporal dependency analysis. For example, a convolutional neural network (CNN) can be used to extract spatial features from the standardized time-series data, and a long short-term memory network (LSTM) can be used to analyze the temporal dependencies of the data. The two types of features are then fused to obtain the fault feature representation.

[0031] By performing feature mining on standardized time-series data from both spatial and temporal dimensions, the problem of traditional methods being unable to fully capture fault characteristics is effectively solved. This allows for more complete and accurate extraction of key information related to faults, providing core feature support for the accuracy of subsequent fault diagnosis. It also lays a feature foundation for adapting to the complex and ever-changing operating environment of the power system.

[0032] Step 103: Based on the fault feature representation, a reinforcement learning algorithm is used to dynamically optimize the early warning parameters and protection action strategy.

[0033] In the embodiments of this disclosure, in the fault diagnosis and early warning process of power system relay protection, step 103 uses the fault feature representation obtained in step 102 as the core basis. Leveraging the dynamic interactive learning capability of reinforcement learning algorithms, a decision optimization mechanism adapted to the operating characteristics of the power system is constructed to achieve adaptive adjustment of early warning parameters (such as alarm thresholds, early warning level definition parameters, etc.) and protection action strategies. During this process, the reinforcement learning algorithm establishes a continuous perception-decision-feedback closed loop with the power system operating environment. Based on the fault state reflected by the fault feature representation and the system operating state, it dynamically evaluates the implementation effect of different early warning parameter configurations and protection action strategies, and iteratively optimizes the decision logic accordingly. This ensures that the early warning parameters and protection action strategies can be adjusted in real time according to changes in system operating conditions and differences in fault types, avoiding the limitations of fixed parameters or strategies in complex dynamic environments. As one implementation method, the Q-learning algorithm can be used to construct a decision optimization model. The joint set of system operating status (including electrical parameters and equipment operating information) and fault feature representation is used as the state space, and adjusting alarm thresholds, setting early warning levels, and selecting protection actions are used as the action space. The reward function is set according to the system operating effect after the action is implemented, and the optimization strategy is updated through algorithm iteration.

[0034] By using reinforcement learning algorithms to dynamically optimize early warning parameters and protection action strategies, the problem of insufficient adaptive capability caused by the reliance on fixed parameters in traditional methods is effectively solved. This significantly reduces the false alarm rate and the missed alarm rate, while improving the matching degree between protection action strategies and actual fault scenarios. This provides more targeted decision support for power system relay protection and enhances the safety and stability of system operation.

[0035] Step 104: Based on the optimized warning parameters and protection action strategy, output graded warning signals and generate control commands.

[0036] In the embodiments of this disclosure, in the complete process of fault diagnosis and early warning for power system relay protection, step 104 uses the early warning parameters and protection action strategies optimized in step 103 as the core basis to construct the execution link of early warning output and command generation. Specifically, the output of graded early warning signals needs to be combined with the optimized early warning parameters to differentiate and present the severity of potential fault risks, ensuring that different levels of fault risk correspond to clearly distinguishable early warning information. The generation of control commands needs to be closely matched with the optimized protection action strategies, forming executable commands that meet the system's safety and stability requirements based on the current fault state and actual operational needs of the power system. This allows the early warning information and protection actions to form a coordinated response, meeting the power system's timeliness and adaptability requirements for fault handling. As one implementation method, graded early warning signals can be divided into three levels: general, severe, and emergency. Control commands can include operational suggestions for maintenance personnel (such as switching lines and adjusting loads) and control commands that the system can automatically execute (such as tripping and closing).

[0037] By outputting tiered early warning signals and generating matching control commands, the problem of traditional early warning signals being too simple and protection action commands lacking specificity is effectively solved. This not only improves the risk differentiation and intuitiveness of fault early warning, but also ensures the accuracy and timeliness of protection actions, providing direct execution support for the power system to quickly handle faults and ensure operational safety and stability.

[0038] The relay protection fault diagnosis method disclosed herein acquires power system operating data and preprocesses it to generate standardized time-series data. Then, it extracts spatial features and performs time-series dependency analysis to obtain fault feature representations. Subsequently, it uses reinforcement learning algorithms to dynamically optimize early warning parameters and protection action strategies, and finally outputs hierarchical early warning signals and generates control commands. Therefore, it can solve the problems in the prior art, such as the lack of adaptive capability of fixed protection parameters, inability to efficiently integrate multi-source data, difficulty in capturing long-term dependencies of time-series data, insufficient real-time performance, and reliance on manual intervention. It achieves the technical effects of improving the accuracy of power system relay protection fault diagnosis, enhancing adaptive capability, ensuring real-time and efficient early warning, reducing manual intervention, and optimizing protection action decisions.

[0039] As a specific implementation of this disclosure, based on the basic scheme, the acquisition of power system operation data and the preprocessing of the operation data are further defined as follows: segmenting the operation data through a sliding window to form time-series data segments; performing data enhancement processing on the time-series data segments, including superimposing noise disturbances in the time domain and adjusting the amplitude and phase in the frequency domain.

[0040] Specifically, in this particular embodiment of the disclosure, during the preprocessing of power system operation data, the sliding window segmentation operation is specifically set with a window length of 200ms and a step size of 100ms. The sliding window configured with these parameters slides and captures continuous operation data in time sequence, dividing the originally continuous operation data into multiple time-series data segments with temporal continuity. Each segment contains complete operation parameter change information within 200ms, thereby ensuring accurate capture of transient data characteristics when short-term faults occur. In the data augmentation stage, time-domain noise perturbation is achieved by superimposing Gaussian noise. The mean of this Gaussian noise is set to 0, and the variance is dynamically adjusted according to the actual noise level of the operating data to be processed. It is usually taken as 0.1-0.3 times the standard deviation of the original data. By superimposing this Gaussian noise with the original time-series data segment, the external electromagnetic interference that may occur in the operation of the power system is simulated. The frequency domain adjustment first converts the time-series data segment in the time domain into a frequency domain signal through Fast Fourier Transform (FFT). Then, in the frequency domain, the amplitude and phase of the frequency domain signal are randomly perturbed according to the frequency characteristics of the actual power system. The range of perturbation amplitude and phase change strictly matches the frequency fluctuation range during normal operation of the power grid. After the perturbation is completed, the frequency domain signal is converted back to the time domain through Inverse Fast Fourier Transform (IFFT) to form the enhanced time-series data segment.

[0041] By using a sliding window segmentation with specific parameters, transient fault information can be effectively preserved, avoiding the loss of fault features due to improper data extraction methods. At the same time, data augmentation processing combining the time and frequency domains can simulate various interference scenarios in the actual operation of power systems, significantly increasing the diversity of training data, improving the adaptability of subsequent fault diagnosis models to noise and disturbances, and providing high-quality data support for model training that is closer to actual operating conditions.

[0042] As a specific implementation of this disclosure, based on the basic scheme, the spatial feature extraction and temporal dependency analysis of the standardized time series data are further defined as follows: extracting local spatial features of the standardized time series data based on a pre-trained convolutional neural network; and capturing long-term temporal dependencies of the standardized time series data using a pre-trained long short-term memory network.

[0043] Specifically, in this particular embodiment, the spatial feature extraction and temporal dependency analysis of standardized time-series data are both based on a pre-trained model. The pre-trained convolutional neural network (CNN) used for extracting local spatial features employs normal and various fault time-series data from the historical operation of the power system as training samples. Through multiple rounds of iterative optimization, the kernel size, number, and stride of the convolutional layers are determined. Simultaneously, the pooling layers are configured to use either max pooling or average pooling strategies. In the actual extraction of local spatial features, the pre-trained CNN takes the standardized time-series data as input and performs sliding operations on the data using the convolutional kernels of the convolutional layers. This deeply mines the local correlation features of parameters such as voltage and current in different dimensions. The pooling layers then downsample the feature map output by the convolution, compressing the data dimensionality while preserving key local features, thus obtaining the local spatial features of the standardized time-series data. The pre-trained Long Short-Term Memory (LSTM) network used to capture long-term temporal dependencies is also based on historical time-series data of the power system during its pre-training phase, optimizing the weight parameters of the input gate, forget gate, and output gate. In actual processing, the pre-trained LSTM receives standardized time-series data in time step order, controls the input of new time-series information through the input gate, filters and discards useless historical time-series information through the forget gate, and outputs the effective time-series features that are relevant to the past at the current moment through the output gate. This accurately captures the dynamic correlation of standardized time-series data over a long period of time, thus completing the extraction of long-term temporal dependencies.

[0044] By using pre-trained CNN and LSTM models, on the one hand, the number of training iterations in practical applications is reduced by the pre-trained parameters, thus improving the efficiency of feature extraction; on the other hand, pre-trained CNN can more accurately mine the local spatial correlation features of standardized time series data, and pre-trained LSTM can effectively capture long-term temporal dependencies. The combination of the two makes the extracted fault-related features more comprehensive and accurate, providing reliable support for obtaining high-quality fault feature representations in the future.

[0045] As a specific implementation of this disclosure, based on the basic scheme, the reinforcement learning algorithm is further defined as being based on a preset learning mechanism, wherein the state space is defined as a joint set of system operating states and fault feature representations, the action space includes adjusting alarm thresholds, setting warning levels and selecting protection actions, and the effect of the actions is evaluated through a reward function to drive strategy optimization.

[0046] Specifically, in this particular embodiment, the reinforcement learning algorithm uses Q-learning as the preset learning mechanism, and its core parameters and space definitions are designed around the scenario of fault diagnosis and early warning for power system relay protection. The state space is explicitly defined to include two parts: first, the system operating state, specifically covering electrical parameters such as voltage, current, frequency, and power during real-time operation of the power system, as well as the operating status information of relay protection equipment (such as equipment temperature and operating time); second, the fault feature representation obtained in the aforementioned steps. These two together constitute a joint set, ensuring that the algorithm can comprehensively perceive the current operating conditions of the system and the associated fault characteristics. The action space is strictly defined as three types of executable operations: adjusting alarm thresholds (such as dynamically correcting the critical value of alarm triggering parameters based on fault probability), setting early warning levels (such as dividing different early warning levels according to the severity of the fault), and selecting protection actions (such as determining whether to perform line switching, tripping, or closing operations). The reward function is designed based on the actual operating effect of the system after the action is implemented. Specifically, it is set as follows: when the system correctly alarms based on the action and the protection action effectively improves the operating state, the reward value r = 1; when a false alarm occurs (an alarm is triggered even though there is no fault), the reward value r = -0.5; when a missed alarm occurs (a fault exists but no alarm is triggered), the reward value r = -1. During the algorithm's operation, the action strategy is continuously iterated and optimized based on the Q-learning update formula, combined with the real-time feedback of the state space and the evaluation results of the reward function, to achieve dynamic adjustment of the warning parameters and protection actions.

[0047] By clearly defining the scope of the state space and action space, and quantifying the reward function, reinforcement learning algorithms can accurately perceive the system state, flexibly execute decision-making operations, and objectively evaluate the effects of actions, effectively avoiding blind decision-making. At the same time, based on the Q-learning learning mechanism, the algorithm can significantly reduce the false alarm rate and false negative rate through continuous iterative optimization of strategies, improve the adaptability of early warning parameters and protection action strategies to actual fault scenarios, and further ensure the safety and stability of power system operation.

[0048] As a specific implementation of this disclosure, based on the basic scheme, the graded early warning signal is further defined as being divided into multiple levels according to severity, and the control command includes at least one of switching lines, adjusting load, isolating faulty equipment, tripping or closing.

[0049] Specifically, in this particular embodiment, the graded early warning signals are clearly divided into three levels according to the severity of the impact of the fault on the operation of the power system: general early warning, severe early warning, and emergency early warning. A general early warning corresponds to a slight deviation of the power system operating parameters from the normal range, with a low fault risk that does not currently affect the core functions of the system. A severe early warning corresponds to the detection of a clear potential fault, such as persistently abnormal local parameters with a worsening trend, which, if not addressed promptly, could lead to an expansion of the fault range. An emergency early warning corresponds to a fault that directly threatens the stable operation of the system, such as a short circuit or grounding, requiring immediate action to prevent equipment damage or system shutdown. The control commands specifically include line switching, load adjustment, faulty equipment isolation, tripping, and closing operations. The line switching command is used to transfer the load on a line to a backup line when an anomaly occurs. The load adjustment command is used to alleviate the operational pressure in the faulty area by adjusting the load distribution across different areas of the power grid. The faulty equipment isolation command is used to disconnect confirmed faulty equipment from the main power grid circuit to prevent the fault from spreading to other equipment. The tripping command is used to quickly disconnect the faulty circuit in the event of a serious fault (such as a short circuit). The closing command is used to restore normal power supply to equipment or lines after the fault has been cleared. In practical applications, at least one control command can be selected to respond to the fault based on the warning level and fault type.

[0050] By classifying early warning signals according to severity, maintenance personnel can quickly identify the level of fault risk and prioritize the allocation of resources to handle high-priority faults, thus avoiding delays in handling the situation. Meanwhile, diverse control commands can be accurately matched to different fault scenarios, ensuring that fault handling measures are both targeted and flexible, effectively improving the efficiency of power system fault response and reducing the impact of faults on system operation.

[0051] As a specific implementation of this disclosure, based on the basic scheme, the embodiments of this disclosure further include: performing feature fusion on multi-source operating data, using an attention mechanism to calculate the weight coefficients of each data source, and inputting the fused features into a fully connected layer for nonlinear transformation to output a fault probability distribution.

[0052] Specifically, in this particular embodiment, the feature fusion of multi-source operating data and the subsequent output of fault probability distribution are clearly designed. The multi-source operating data specifically includes current and voltage data collected by relay protection devices, equipment temperature and vibration data collected by sensors, and frequency and power data collected by distributed control systems (DCS). An attention mechanism is introduced in the feature fusion stage. An attention weight calculation module is constructed, using the local spatial features and time-series dependency features obtained from each data source after preprocessing as input. First, the correlation score between each data source and the fault diagnosis target is calculated. Then, the correlation score is normalized to a weight coefficient within the 0-1 range using the Softmax function, so that data sources with higher correlation to the fault (such as current and voltage data at the time of the fault) receive a larger weight coefficient. Finally, the features of each data source are weighted and summed with their corresponding weight coefficients to obtain the comprehensive feature fused from the multi-source information. The fused features are further input into a fully connected layer. The fully connected layer has at least two hidden layers. Each hidden layer uses the ReLU activation function to perform a nonlinear transformation on the fused features in order to uncover the deep correlation between features. Finally, the processed features are mapped to the probability values ​​of various common power system faults such as short circuits, grounding, and overloads through the Softmax activation function of the output layer, forming a fault probability distribution. The fault type with the highest probability value is the most likely fault in the current system.

[0053] By dynamically allocating weights for each data source through an attention mechanism, the contribution of key data to fault diagnosis can be effectively highlighted, interference from irrelevant or low-correlation data can be avoided, and the effectiveness of feature fusion can be improved. Meanwhile, the nonlinear transformation of the fully connected layer can deeply mine the deep information of the fused features, and the final output fault probability distribution can intuitively quantify the probability of various faults, providing accurate quantitative basis for fault diagnosis and further improving the reliability of fault identification.

[0054] It should be noted that the embodiments of this disclosure may include multiple steps. For ease of description, these steps are numbered, but these numbers are not a limitation on the execution time slots or execution order between the steps; these steps can be implemented in any order, and the embodiments of this disclosure do not limit this.

[0055] Corresponding to the aforementioned relay protection fault diagnosis method, this disclosure also proposes a relay protection fault diagnosis device. Since the device embodiments of this disclosure correspond to the aforementioned method embodiments, details not disclosed in the device embodiments can be referred to the aforementioned method embodiments, and will not be repeated here.

[0056] Figure 2 This is a schematic diagram of the structure of a relay protection fault diagnosis device provided in an embodiment of this disclosure, as shown below. Figure 2 As shown, it includes: The acquisition unit 21 is used to acquire power system operation data and preprocess the operation data to generate standardized time series data; Analysis unit 22 is used to perform spatial feature extraction and temporal dependency analysis on the standardized time series data to obtain fault feature representation; Optimization unit 23 is used to dynamically optimize early warning parameters and protection action strategies based on the fault feature representation using a reinforcement learning algorithm; Output unit 24 is used to output graded early warning signals and generate control commands based on the optimized early warning parameters and protection action strategies.

[0057] The relay protection fault diagnosis device provided in this disclosure acquires power system operating data and generates standardized time-series data through preprocessing. Then, it extracts spatial features and performs time-series dependency analysis to obtain fault feature representations. Subsequently, it uses reinforcement learning algorithms to dynamically optimize early warning parameters and protection action strategies, and finally outputs hierarchical early warning signals and generates control commands. Therefore, it can solve the problems in the prior art, such as the lack of adaptive capability of fixed protection parameters, inability to efficiently integrate multi-source data, difficulty in capturing long-term dependencies of time-series data, insufficient real-time performance, and reliance on manual intervention. It achieves the technical effects of improving the accuracy of power system relay protection fault diagnosis, enhancing adaptive capability, ensuring real-time and efficient early warning, reducing manual intervention, and optimizing protection action decisions.

[0058] Furthermore, in one possible implementation of this embodiment, the acquisition unit 21 is also used for: The running data is segmented using a sliding window to form time-series data segments; The time-series data segment is subjected to data augmentation processing, including superimposing noise perturbations in the time domain and adjusting the amplitude and phase in the frequency domain.

[0059] Furthermore, in one possible implementation of this embodiment, the analysis unit 22 is also used for: Based on a pre-trained convolutional neural network, local spatial features of the standardized time-series data are extracted. Long-term temporal dependencies are captured by using a pre-trained long short-term memory network to standardize the time-series data.

[0060] Furthermore, in one possible implementation of this embodiment, the reinforcement learning algorithm is based on a preset learning mechanism, wherein the state space is defined as a joint set of system operating states and fault feature representations, the action space includes adjusting alarm thresholds, setting warning levels, and selecting protection actions, and the effect of the actions is evaluated through a reward function to drive strategy optimization.

[0061] Furthermore, in one possible implementation of this embodiment, the graded early warning signal is divided into multiple levels according to severity, and the control command includes at least one of switching lines, adjusting load, isolating faulty equipment, tripping or closing.

[0062] Furthermore, in one possible implementation of this embodiment, such as Figure 2 As shown, it also includes: The fusion unit 25 is used to perform feature fusion on multi-source operational data. It uses an attention mechanism to calculate the weight coefficients of each data source and inputs the fused features into a fully connected layer for nonlinear transformation to output the fault probability distribution.

[0063] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of this embodiment, and the principle is the same, so it is not limited in this embodiment.

[0064] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0065] Figure 3 A schematic block diagram of an example electronic device 300 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0066] like Figure 3 As shown, the electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 302 or a computer program loaded from storage unit 308 into RAM (Random Access Memory) 303. The RAM 303 may also store various programs and data required for the operation of the electronic device 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An I / O (Input / Output) interface 305 is also connected to the bus 304.

[0067] Multiple components in electronic device 300 are connected to I / O interface 305, including: input unit 306, such as keyboard, mouse, etc.; output unit 307, such as various types of displays, speakers, etc.; storage unit 308, such as disk, optical disk, etc.; and communication unit 309, such as network card, modem, wireless transceiver, etc. Communication unit 309 allows electronic device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0068] The computing unit 301 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as relay protection fault diagnosis methods. For example, in some embodiments, the relay protection fault diagnosis method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by the computing unit 301, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 301 may be configured to perform the aforementioned relay protection fault diagnosis method by any other suitable means (e.g., by means of firmware).

[0069] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0070] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0071] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0072] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0073] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0074] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0075] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0076] The various numerical designations such as "first," "second," etc., used in this disclosure are merely for ease of description and are not intended to limit the scope of the embodiments of this disclosure, nor do they indicate a sequential order.

[0077] At least one of the features described in this disclosure can also be described as one or more, and multiple features can be two, three, four or more, and this disclosure does not impose any limitations. In the embodiments of this disclosure, for a technical feature, the technical features in that technical feature are distinguished by "first", "second", "third", "A", "B", "C" and "D", etc., and there is no sequential order or size order among the technical features described by "first", "second", "third", "A", "B", "C" and "D".

[0078] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0079] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for fault diagnosis of relay protection, characterized in that, include: Acquire power system operation data and preprocess the operation data to generate standardized time-series data; Spatial feature extraction and temporal dependency analysis are performed on the standardized time-series data to obtain fault feature representations; Based on the fault feature representation, a reinforcement learning algorithm is used to dynamically optimize the early warning parameters and protection action strategies; Based on the optimized warning parameters and protection action strategies, graded warning signals are output and control commands are generated.

2. The method according to claim 1, characterized in that, The acquisition of power system operation data and the preprocessing of the operation data include: The running data is segmented using a sliding window to form time-series data segments; The time-series data segment is subjected to data augmentation processing, including superimposing noise perturbations in the time domain and adjusting the amplitude and phase in the frequency domain.

3. The method according to claim 1, characterized in that, The spatial feature extraction and temporal dependency analysis of the standardized time-series data include: Based on a pre-trained convolutional neural network, local spatial features of the standardized time-series data are extracted. Long-term temporal dependencies are captured by using a pre-trained long short-term memory network to standardize the time-series data.

4. The method according to claim 1, characterized in that, The reinforcement learning algorithm is based on a preset learning mechanism, where the state space is defined as a joint set of system operating states and fault feature representations, the action space includes adjusting alarm thresholds, setting warning levels, and selecting protection actions, and the effect of the actions is evaluated through a reward function to drive strategy optimization.

5. The method according to claim 1, characterized in that, The graded early warning signals are divided into multiple levels according to their severity, and the control commands include at least one of switching lines, adjusting load, isolating faulty equipment, tripping or closing.

6. The method according to claim 1, characterized in that, Also includes: Feature fusion is performed on multi-source operational data. An attention mechanism is used to calculate the weight coefficients of each data source. The fused features are then input into a fully connected layer for nonlinear transformation to output the fault probability distribution.

7. A relay protection fault diagnosis device, characterized in that, include: An acquisition unit is used to acquire power system operation data and preprocess the operation data to generate standardized time-series data; The analysis unit is used to perform spatial feature extraction and temporal dependency analysis on the standardized time series data to obtain fault feature representations; The optimization unit is used to dynamically optimize the early warning parameters and protection action strategies based on the fault feature representation using a reinforcement learning algorithm; The output unit is used to output graded early warning signals and generate control commands based on the optimized early warning parameters and protection action strategies.

8. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-6.

Images