Partial discharge detection model optimization method, device, equipment, medium and product

Abstract

The invention relates to an optimization method and device of a partial discharge detection model, equipment, a medium and a product. The method is applied to edge equipment, and comprises the following steps: firstly, constructing a first detection feature matrix at the current moment based on a local discharge signal feature vector of to-be-detected equipment at the current moment and a historical decision strategy of a local discharge detection model, and then inputting the first detection feature matrix into the local discharge detection model, and finally, determining a reward value of the partial discharge detection model based on the detection result, determining an optimization strategy of the partial discharge detection model based on the reward value, and carrying out optimization processing on the partial discharge detection model. By adopting the method, the efficiency and the accuracy of partial discharge detection can be improved.

Description

Technical Field

[0001] This application relates to the field of power system condition monitoring technology, and in particular to an optimization method, apparatus, equipment, medium and product for a partial discharge detection model. Background Technology

[0002] In high-voltage, long-distance power transmission scenarios, insulation defects can easily lead to distorted electric field distribution, resulting in partial discharge. This partial discharge gradually erodes the insulation material, eventually causing a breakdown accident, leading to line tripping or even large-scale power outages. Therefore, partial discharge detection is a key technology for ensuring the safe operation of transmission lines and equipment. It not only avoids economic losses caused by sudden power outages but also extends equipment lifespan and improves the reliability of power grid supply.

[0003] However, the detection of partial discharge currently relies on the experience of maintenance personnel, which results in low efficiency and accuracy. Summary of the Invention

[0004] Therefore, it is necessary to provide an optimization method, apparatus, equipment, medium, and product for a partial discharge detection model to address the aforementioned technical problems.

[0005] Firstly, this application provides an optimization method for a partial discharge detection model, applied to edge devices, the method comprising:

[0006] Based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model, the first detection feature matrix at the current moment is constructed.

[0007] The first detection feature matrix is ​​input into the partial discharge detection model to obtain the detection result of whether the device under test has a partial discharge phenomenon at the current time;

[0008] The reward value of the partial discharge detection model is determined based on the detection results. The optimization strategy of the partial discharge detection model is determined based on the reward value, and the partial discharge detection model is optimized.

[0009] In one embodiment, the optimization strategy for the partial discharge detection model is determined based on the reward value, including:

[0010] If the reward value is greater than the preset value, then the optimized sample of the partial discharge detection model is obtained, and the partial discharge detection model is optimized using the optimized sample.

[0011] If the reward value is less than the preset value, the parameter information of the partial discharge detection model is sent to the aggregation server. Based on the feedback parameters returned by the aggregation server, the partial discharge detection model is optimized. The aggregation server is used to comprehensively analyze the parameter information of partial discharge detection models corresponding to multiple different regions and determine the feedback parameters.

[0012] In one embodiment, an optimized sample of the partial discharge detection model is obtained, and the partial discharge detection model is optimized using the optimized sample, including:

[0013] Obtain the second detection feature matrix of the device under test at the next time step;

[0014] An optimized sample is generated based on the combination of the first detection feature matrix, the reward value, and the detection feature matrix at the next time step.

[0015] The optimized sample is input into the partial discharge detection model, and the parameters of the partial discharge detection model are updated to obtain the optimized partial discharge detection model.

[0016] In one embodiment, the parameter information of the partial discharge detection model is sent to a convergence server. Based on the feedback parameters returned by the convergence server, the partial discharge detection model is optimized, including:

[0017] Based on the parameter information of the partial discharge detection model and the initial parameters of the initial partial discharge detection model, the parameter changes of the partial discharge detection model are determined.

[0018] The parameter changes are sent to the aggregation server so that the aggregation server can optimize the stored global detection model and determine the feedback parameters based on the parameter information of the optimized global detection model.

[0019] The system receives feedback parameters from the aggregation server and optimizes the partial discharge detection model based on these parameters.

[0020] In one embodiment, the feedback parameters include model update parameters and model average score. Based on the feedback parameters returned by the aggregation server, the partial discharge detection model is optimized, including:

[0021] The target model score for edge devices is determined based on the reward values ​​corresponding to historical decision-making strategies.

[0022] If the target model score is greater than or equal to the model average score, the partial discharge detection model is optimized based on the updated parameters and the first preset learning rate.

[0023] If the target model score is less than the model average score, the partial discharge detection model is optimized based on the updated parameters and the second preset learning rate; the second preset learning rate is greater than the first preset learning rate.

[0024] In one embodiment, based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model, a first detection feature matrix for the current moment is constructed, including:

[0025] Based on historical decision-making strategies and the corresponding reward values, the model feature vector of the partial discharge detection model is obtained.

[0026] The state feature vector is determined based on the environmental and operating information of the device under test;

[0027] The first detection feature matrix is ​​constructed based on the partial discharge signal feature vector, the model feature vector, and the state feature vector.

[0028] Secondly, this application also provides an optimization device for a partial discharge detection model, applied to edge devices, the device comprising:

[0029] The construction module is used to construct the first detection feature matrix at the current moment based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model;

[0030] The input module is used to input the first detection feature matrix into the partial discharge detection model to obtain the detection result of whether the device under test has a partial discharge phenomenon at the current time;

[0031] The optimization module is used to determine the reward value of the partial discharge detection model based on the detection results, determine the optimization strategy of the partial discharge detection model based on the reward value, and optimize the partial discharge detection model.

[0032] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in any of the first aspects above.

[0033] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects above.

[0034] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in any of the first aspects above.

[0035] The aforementioned optimization method, apparatus, equipment, medium, and product for partial discharge detection models are applied to edge devices. First, based on the feature vector of the partial discharge signal of the device under test (DUT) at the current moment and the historical decision-making strategy of the partial discharge detection model, a first detection feature matrix for the current moment is constructed. Then, the first detection feature matrix is ​​input into the partial discharge detection model to obtain the detection result of whether partial discharge has occurred in the DUT at the current moment. Finally, based on the detection result, a reward value for the partial discharge detection model is determined, and based on the reward value, an optimization strategy for the partial discharge detection model is determined, and the partial discharge detection model is optimized. In this way, the partial discharge detection result of the DUT can be determined based on the first monitoring feature matrix of the DUT at the current moment according to the partial discharge detection model, eliminating the need to rely on the experience of maintenance personnel for partial discharge detection, thus improving efficiency. Furthermore, optimizing the partial discharge detection model based on the detection result improves the accuracy of partial discharge detection. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a diagram illustrating the application environment of an optimization method for a partial discharge detection model in one embodiment.

[0038] Figure 2 This is a flowchart illustrating the optimization method for a partial discharge detection model in one embodiment;

[0039] Figure 3 This is a flowchart illustrating the first monitoring feature matrix construction step in one embodiment;

[0040] Figure 4 This is a flowchart illustrating the local optimization steps of a partial discharge detection model in one embodiment;

[0041] Figure 5 This is a flowchart illustrating the federated learning optimization steps of a partial discharge detection model in one embodiment.

[0042] Figure 6 This is a flowchart illustrating the federated learning optimization steps of a partial discharge detection model in one embodiment.

[0043] Figure 7 This is a flowchart illustrating the optimization method for the partial discharge detection model in another embodiment;

[0044] Figure 8 This is a schematic diagram of the partial discharge detection system in another embodiment;

[0045] Figure 9 This is a structural block diagram of an optimization device for a partial discharge detection model in one embodiment;

[0046] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0049] The optimization method for the partial discharge detection model provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, edge device 102 communicates with aggregation server 104 via a network, and aggregation server 104 communicates with cloud server 106 via a network. Edge device 102 can be, but is not limited to, an edge gateway or a sensor. Aggregation server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services; aggregation server 104 can be deployed in a substation. Cloud server 106 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services; cloud server 106 can be deployed in a power company's cloud data center. Multiple edge devices 102 connect to one aggregation server 104, and multiple aggregation servers 104 connect to cloud server 106.

[0050] In one exemplary embodiment, such as Figure 2 As shown, an optimization method for a partial discharge detection model is provided, which is then applied to... Figure 1 Taking an edge device as an example, the explanation includes steps 201 to 203. Wherein:

[0051] Step 201: Based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model, construct the first detection feature matrix at the current moment.

[0052] The feature vector of the partial discharge signal of the device under test at the current moment can be obtained by extracting features from the partial discharge signal of the device under test at the current moment. The partial discharge signal can be acquired by a UHF electromagnetic wave sensor. The partial discharge signal is amplified and noise-reduced, and then the processed partial discharge signal is subjected to feature extraction by wavelet transform or S-transform to obtain the feature vector of the partial discharge signal at the current moment.

[0053] The partial discharge detection model can be trained by a cloud server and sent to the edge device via a aggregation server during the initialization phase, used for partial discharge detection on the edge device. Alternatively, it can be trained by the aggregation server and sent to the edge device if a cloud server is unavailable. The historical decision-making strategy of the partial discharge detection model represents the historical partial discharge detection results. The partial discharge detection results are discrete actions that the edge device can take. The set of discrete actions can be {A1: Ignore as noise, A2: Alarm as internal discharge, A3: Alarm as XX discharge, A4: Increase monitoring sensitivity to obtain more details, A5: Request collaborative analysis from the aggregation server, A6: Mark as an uncertain event and request manual annotation}, where XX represents the four types of partial discharge, such as tip discharge, particle discharge, suspended discharge, and air gap discharge. The partial discharge detection model can include a policy network and an output network. The policy network determines the probability of each discrete action, and the output network determines the partial discharge detection result based on the probabilities of each discrete action.

[0054] For example, construct the first detection feature matrix at the current time, such as Figure 3 As shown, it may include:

[0055] Step 301: Based on the historical decision-making strategies and the corresponding reward values, obtain the model feature vector of the partial discharge detection model.

[0056] The historical decision-making strategy of the partial discharge detection model can be represented by a_t. The reward value corresponding to the historical decision-making strategy is determined based on the consequences after the edge device executes the historical decision-making strategy, and can be represented by r_t. Optionally, the reward value can be determined based on the input information from the interface, or based on whether the edge device's operating status is normal after executing the historical decision-making strategy. For example, if the historical decision-making strategy correctly issues a warning: if a problem is confirmed after the alarm (e.g., through manual inspection or subsequent fault), a large positive reward of +R_success is given. If the historical decision-making strategy falsely issues a warning: if the alarm is confirmed as interference or a normal phenomenon, a negative reward of -R_false_alarm is given. If the historical decision-making strategy misses a warning: if it is ignored but a fault is later discovered, a large negative reward of -R_miss is given. If the historical decision-making strategy correctly ignores a warning: if it is ignored and the device continues to operate healthily, a small positive reward of +R_correct_ignore is given to encourage noise filtering. If the historical decision-making strategy is effective, and the manually labeled event is confirmed by experts as a valuable, atypical signal, then a moderate positive reward of +R_valuable_query is given. The values ​​of R_success, R_false_alarm, R_miss, R_correct_ignore, and R_valuable_query can be set according to the actual situation.

[0057] The historical decision measurement and the corresponding reward value are used as the model feature vector, which can be represented by H_t.

[0058] Step 302: Determine the state feature vector based on the environmental and operating information of the device under test.

[0059] The environmental information of the device under test (DUT) can be obtained from the sensors, including ambient temperature and humidity. The load rate, i.e. operating condition information, of the DUT can be determined based on the current and voltage of the DUT. The environmental information and operating condition information are used as state feature vectors, which can be represented by E_t.

[0060] Step 303: Construct the first detection feature matrix based on the partial discharge signal feature vector, the model feature vector, and the state feature vector.

[0061] The feature vector of the partial discharge signal can be represented by F_t. Based on the feature vector of the partial discharge signal F_t, the model feature vector H_t, and the state feature vector E_t, the first detection feature matrix s_t={F_t,E_t,H_t} is obtained.

[0062] Step 202: Input the first detection feature matrix into the partial discharge detection model to obtain the detection result of whether the device under test has a partial discharge phenomenon at the current time.

[0063] The first detection feature matrix s_t is input into the partial discharge detection model, which can be a policy network. The network outputs the probability of each possible action, thereby determining the detection result of whether the device under test has a partial discharge phenomenon at the current time based on the action with the highest probability.

[0064] Step 203: Determine the reward value of the partial discharge detection model based on the detection results, determine the optimization strategy of the partial discharge detection model based on the reward value, and optimize the partial discharge detection model.

[0065] Optionally, the reward value of the partial discharge detection model is determined based on the reward values ​​of the detection results within a preset time period. For example, the average reward value of the detection results within the preset time period can be used as the reward value of the partial discharge detection model. An optimization strategy for the partial discharge detection model is then determined based on the reward value. This optimization strategy can be local model self-optimization or optimization through federated learning with the aggregation server. The partial discharge detection model is then optimized based on the determined optimization strategy.

[0066] For example, the optimization strategy for the partial discharge detection model can also be determined based on time conditions. When the first preset period is reached, the optimization strategy for the partial discharge detection model can be local model self-optimization. When the second preset period is reached, the optimization strategy for the partial discharge detection model can be optimization through federated learning initiated with the aggregation server. The second preset period is greater than or equal to the first preset period.

[0067] In the above embodiments, applied to edge devices, firstly, based on the feature vector of the partial discharge signal of the device under test (DUT) at the current moment and the historical decision-making strategy of the partial discharge detection model, a first detection feature matrix for the current moment is constructed. Then, the first detection feature matrix is ​​input into the partial discharge detection model to obtain the detection result of whether the DUT has experienced partial discharge at the current moment. Finally, based on the detection result, the reward value of the partial discharge detection model is determined, and based on the reward value, an optimization strategy for the partial discharge detection model is determined, and the partial discharge detection model is optimized. In this way, the partial discharge detection result of the DUT can be determined based on the first monitoring feature matrix of the DUT at the current moment according to the partial discharge detection model, without relying on the experience of maintenance personnel for partial discharge detection, resulting in higher efficiency. Simultaneously, optimizing the partial discharge detection model based on the detection result leads to higher accuracy in partial discharge detection.

[0068] In the embodiments of this application, the optimization strategy for determining the partial discharge detection model based on the reward value includes the following two methods.

[0069] In the first approach, if the reward value is greater than the preset value, an optimized sample of the partial discharge detection model is obtained, and the partial discharge detection model is optimized using the optimized sample.

[0070] Optionally, the steps for obtaining optimized samples for the partial discharge detection model are as follows: Figure 4 As shown, it may include:

[0071] Step 401: Obtain the second detection feature matrix of the device under test at the next moment.

[0072] Based on the partial discharge signal feature vector, model feature vector, and state feature vector of the device under test at time t+1, the second detection feature matrix s_{t+1} at time t_1 is constructed.

[0073] Step 402: Generate optimized samples based on the combination of the first detection feature matrix, the reward value, and the detection feature matrix at the next time step.

[0074] The optimized sample is obtained based on the first detection feature matrix s_t, the detection result a_t corresponding to the first detection feature matrix, the reward value r_t corresponding to the detection result, and the detection feature matrix s_{t+1} at the next time step, and the optimized sample (s_t, a_t, r_t, s_{t+1}) is saved to the experience cache.

[0075] Step 403: Input the optimized sample into the partial discharge detection model, update the parameters of the partial discharge detection model, and obtain the optimized partial discharge detection model.

[0076] When the reward value of the partial discharge detection model is greater than the preset value, the optimized sample in the experience cache is obtained and input into the partial discharge detection model. The parameter θ_local of the partial discharge detection model is updated through reinforcement learning algorithms such as PPO, so that the future decisions of the partial discharge detection model can obtain high long-term cumulative rewards.

[0077] The second approach involves sending the parameter information of the partial discharge detection model to the aggregation server if the reward value is less than the preset value. Based on the feedback parameters returned by the aggregation server, the partial discharge detection model is then optimized.

[0078] The aggregation server is used to comprehensively analyze the parameter information of partial discharge detection models corresponding to multiple different regions and determine the feedback parameters. If the reward value of the partial discharge detection model is less than the preset value, it may be that the parameter optimization effect of the partial discharge detection model on the edge device is not good. In this case, the edge device can send the parameter information of its local partial discharge detection model to the aggregation server, which then initiates federated learning to all edge devices within its jurisdiction and determines the feedback parameters. The edge devices receive the feedback parameters returned by the aggregation server and optimize the partial discharge model according to the feedback parameters.

[0079] In the above embodiments, the federated learning mechanism of the aggregation server can break the data silo phenomenon of individual edge devices, enabling edge devices across the network to share learning results while protecting data privacy. At the same time, the experience of individual edge devices can also be quickly empowered to all edge devices, realizing the global collaborative evolution of the partial discharge detection capability of the entire network.

[0080] In one embodiment, to implement federated learning, the parameter information of the partial discharge detection model is sent to the aggregation server, such as... Figure 5 As shown, it includes:

[0081] Step 501: Based on the parameter information of the partial discharge detection model and the initial parameters of the initial partial discharge detection model, determine the parameter change of the partial discharge detection model.

[0082] Since the partial discharge detection model is continuously optimized, when the aggregation server to which the edge device belongs has not previously performed federated learning and has not sent feedback parameters to the edge device to optimize the partial discharge detection model, the initial parameters of the initial partial discharge detection model are the parameters of the partial discharge detection model sent by the cloud server to the edge device through the aggregation server during the initialization phase. When the aggregation server to which the edge device belongs has previously performed federated learning and sent feedback parameters to the edge device to optimize the partial discharge detection model, the initial parameters of the initial partial discharge detection model are the parameters of the previous global detection model on the aggregation server.

[0083] Based on the parameter information θ_local of the current partial discharge detection model and the initial parameters of the initial partial discharge detection model, the parameter change Δθlocal of the partial discharge detection model is determined.

[0084] Step 502: Send the parameter change to the aggregation server so that the aggregation server can optimize the stored global detection model and determine the feedback parameters based on the parameter information of the optimized global detection model.

[0085] Each edge device encrypts its own parameter changes and sends them to its respective aggregation server. After receiving a sufficient number of parameter changes, the aggregation server executes the federated aggregation algorithm to optimize the global detection model stored locally on the aggregation server to obtain the optimized global detection model.

[0086] θ global_new =θ global_old +η*Σ(w i *Δθ local_i )

[0087] Among them, w i It is the weight corresponding to edge device i within the jurisdiction of the aggregation server, Δθ local_i The parameter change θ sent by edge device i global_old To aggregate the parameters of the global detection model before server optimization, θ global_new Here are the parameters of the optimized global detection model, where η is the learning rate of the aggregation server, which can be set according to requirements. i It can be determined based on the amount of data from the edge device or its historical performance, i.e., its historical reward value.

[0088] Understandably, since the global detection model of the aggregation server is also continuously optimized, when the aggregation server performs federated learning for the first time, the global detection model stored locally on the aggregation server is sent by the cloud server during the initialization phase. When the aggregation server performs federated learning in subsequent sessions, the global detection model of the aggregation server is obtained after the previous federated learning optimization.

[0089] Step 503: Receive feedback parameters returned by the aggregation server, and optimize the partial discharge detection model based on the feedback parameters returned by the aggregation server.

[0090] The parameters of the optimized global detection model from the aggregation server are sent as feedback parameters to all edge devices. The edge devices then optimize their local partial discharge detection models based on these feedback parameters. Optionally, the edge devices can directly use the feedback parameters as new parameters for their local partial discharge detection models to obtain the optimized model. Alternatively, the edge devices can perform a weighted calculation based on the feedback parameters and the parameters of their local partial discharge detection models, using the weighted calculation result as new parameters for the optimized model.

[0091] In the above embodiments, the original, sensitive partial discharge waveform data is always kept at the edge device throughout the learning process, and only encrypted model parameters are transmitted in the network to avoid the security risks caused by data centralization.

[0092] For example, the feedback parameters include model update parameters and the model average score. The model update parameters are the parameters of the global detection model optimized by the aggregation server. The model average score can be the average score of the partial discharge detection models of all edge devices connected to the aggregation server. The score of the partial discharge detection model of each edge device can be determined based on the average of the historical reward values ​​of each partial discharge detection model. Based on the feedback parameters returned by the aggregation server, the partial discharge detection model is optimized, such as... Figure 6 As shown, it includes:

[0093] Step 601: Determine the target model score for the edge device based on the reward value corresponding to the historical decision-making strategy.

[0094] The average reward value corresponding to the historical decision-making strategies of the edge device is used as the target model score of the edge device.

[0095] Step 602: If the target model score is greater than or equal to the model average score, optimize the partial discharge detection model based on the updated parameters and the first preset learning rate.

[0096] If the target model score is greater than or equal to the model average score, meaning that the historical performance of this edge device is better than the historical average performance of other edge devices, then more individual capabilities can be retained when calculating the parameters of the partial discharge detection model.

[0097] θ local_new = θ local_old + η1 *θ global_new

[0098] Where, θ local_new θ represents the parameters of the optimized partial discharge detection model for the edge device, and θ represents the parameters of the unoptimized partial discharge detection model for the edge device. global_new For feedback parameters, η1 is the first preset learning rate.

[0099] Step 603: If the target model score is less than the model average score, optimize the partial discharge detection model based on the updated parameters and the second preset learning rate; the second preset learning rate is greater than the first preset learning rate.

[0100] If the target model score is lower than the model average score, meaning that the historical performance of this edge device is not as good as the historical average performance of other edge devices, then when calculating the parameters of the partial discharge detection model, more knowledge of the population can be drawn upon.

[0101] θ local_new = θ local_old + η2 *θ global_new

[0102] Where, θlocal_new θ represents the parameters of the optimized partial discharge detection model for the edge device, and θ represents the parameters of the unoptimized partial discharge detection model for the edge device. global_new η2 is the second preset learning rate, which is used as the feedback parameter.

[0103] In the above embodiments, based on the first preset learning rate or the second preset learning rate, the knowledge of the global detection model of the aggregation server is integrated into the local partial discharge detection model of the edge device, thereby achieving the goal of both drawing on the knowledge of the group (multiple edge devices) and retaining the individual adaptability.

[0104] In one embodiment, such as Figure 7 As shown, the optimization method for this partial discharge detection model may include:

[0105] Step 701: Based on the historical decision-making strategies and the corresponding reward values, obtain the model feature vector of the partial discharge detection model.

[0106] Step 702: Determine the state feature vector based on the environmental and operating information of the device under test.

[0107] Step 703: Construct the first detection feature matrix based on the partial discharge signal feature vector, the model feature vector, and the state feature vector.

[0108] Step 704: Input the first detection feature matrix into the partial discharge detection model to obtain the detection result of whether the device under test has a partial discharge phenomenon at the current time.

[0109] Step 705: Obtain the second detection feature matrix of the device under test at the next moment.

[0110] Step 706: Generate optimized samples based on the combination of the first detection feature matrix, the reward value, and the detection feature matrix at the next time step.

[0111] Step 707: Input the optimized sample into the partial discharge detection model, update the parameters of the partial discharge detection model, and obtain the optimized partial discharge detection model.

[0112] Step 708: Based on the parameter information of the partial discharge detection model and the initial parameters of the initial partial discharge detection model, determine the parameter change of the partial discharge detection model.

[0113] Step 709: Send the parameter change to the aggregation server so that the aggregation server can optimize the stored global detection model and determine the feedback parameters based on the parameter information of the optimized global detection model.

[0114] Step 710: Receive feedback parameters returned by the aggregation server, and optimize the partial discharge detection model based on the feedback parameters returned by the aggregation server.

[0115] In one embodiment, the optimization method for the partial discharge detection model can be applied to, for example... Figure 8 The partial discharge detection system shown adopts a three-layer intelligent agent architecture of "cloud-station-edge", including: 1) Edge Perception Agent (EPA): deployed on sensors or edge gateways, responsible for real-time decision-making and individual learning; 2) Site Coordination Agent (SCA): deployed on aggregation servers such as substations, responsible for federated learning and collaborative diagnosis within the region; 3) Cloud Evolution Agent (CEA) (optional): deployed in the cloud, responsible for macro model evolution and policy analysis across regions.

[0116] Edge-aware agents are typically integrated directly into the partial discharge detection sensor hardware or deployed on edge computing devices adjacent to the sensor. Core modules include a data preprocessing module for amplifying and denoising the raw partial discharge signal, and feature extraction; a local decision engine, comprising a partial discharge detection model that outputs the optimal action based on the current state of the device under test; and a local model update module that updates the partial discharge detection model parameters using reinforcement learning algorithms based on the received feedback parameters; and a secure communication module for encrypted communication with the site-coordinated agent, uploading only updated model parameters such as gradients or weights, rather than the raw partial discharge signal data.

[0117] The site coordination agent is typically deployed on the substation's main control computer or a regional data aggregation server. Core modules include a federated aggregation engine, which periodically collects parameter changes from all edge-aware agents under its jurisdiction for the partial discharge detection model and executes federated averaging or other aggregation algorithms to generate a higher-performance "site-level" global detection model. A collaborative diagnostic module, upon receiving a collaborative analysis request from an edge-aware agent, can combine information from multiple sensors of different locations and types (such as ultrasonic and vibration sensors) to make a comprehensive judgment. A human-computer interaction and feedback module provides a visual interface and generates feedback parameters based on expert annotations.

[0118] The cloud-evolved intelligent agent is typically deployed on the cloud data center servers of a power company. Its core modules include a cross-site model aggregation module, which aggregates global detection models from different aggregation servers to train and generate a more generalized "power grid-level" base model.

[0119] In the above embodiments, the edge agent can perceive environmental changes in real time and dynamically adjust its detection and alarm strategies, greatly reducing false alarms and missed alarms caused by changes in environmental factors, and the system has strong robustness. At the same time, through learning and interaction with the site coordination agent, it can achieve self-learning and self-improvement. As the running time increases and the number of events processed increases, the detection accuracy and decision-making level will continue to improve.

[0120] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0121] Based on the same inventive concept, this application also provides an optimization device for a partial discharge detection model to implement the optimization method for the partial discharge detection model described above. The solution provided by this device is similar to the implementation described in the above method. Therefore, the specific limitations in one or more embodiments of the optimization device for the partial discharge detection model provided below can be found in the limitations of the optimization method for the partial discharge detection model described above, and will not be repeated here.

[0122] In one exemplary embodiment, such as Figure 9 As shown, an optimization device for a partial discharge detection model is provided, comprising: a construction module, an input module, and an optimization module, wherein:

[0123] The construction module is used to construct the first detection feature matrix at the current moment based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model;

[0124] The input module is used to input the first detection feature matrix into the partial discharge detection model to obtain the detection result of whether the device under test has a partial discharge phenomenon at the current time;

[0125] The optimization module is used to determine the reward value of the partial discharge detection model based on the detection results, determine the optimization strategy of the partial discharge detection model based on the reward value, and optimize the partial discharge detection model.

[0126] In one embodiment, the optimization module is specifically used to: if the reward value is greater than a preset value, obtain an optimized sample of the partial discharge detection model and use the optimized sample to optimize the partial discharge detection model; if the reward value is less than the preset value, send the parameter information of the partial discharge detection model to the aggregation server and optimize the partial discharge detection model based on the feedback parameters returned by the aggregation server; wherein, the aggregation server is used to comprehensively analyze the parameter information of partial discharge detection models corresponding to multiple different regions and determine the feedback parameters.

[0127] In one embodiment, the optimization module is specifically used to obtain the second detection feature matrix of the device under test at the next moment; generate an optimized sample based on the combination of the first detection feature matrix, the reward value and the detection feature matrix at the next moment; input the optimized sample into the partial discharge detection model, update the parameters of the partial discharge detection model, and obtain the optimized partial discharge detection model.

[0128] In one embodiment, the optimization module is specifically used to determine the parameter change of the partial discharge detection model based on the parameter information of the partial discharge detection model and the initial parameters of the initial partial discharge detection model; send the parameter change to the aggregation server so that the aggregation server can optimize the stored global detection model and determine the feedback parameters based on the parameter information of the optimized global detection model; receive the feedback parameters returned by the aggregation server, and optimize the partial discharge detection model based on the feedback parameters returned by the aggregation server.

[0129] In one embodiment, the feedback parameters include model update parameters and model average score. Specifically, the optimization module is used to determine the target model score of the edge device based on the reward value corresponding to the historical decision-making strategy; if the target model score is greater than or equal to the model average score, the partial discharge detection model is optimized based on the update parameters and a first preset learning rate; if the target model score is less than the model average score, the partial discharge detection model is optimized based on the update parameters and a second preset learning rate; the second preset learning rate is greater than the first preset learning rate.

[0130] In one embodiment, the construction module is specifically used to obtain the model feature vector of the partial discharge detection model based on historical decision-making strategies and the reward values ​​corresponding to the historical decision-making strategies; determine the state feature vector based on the environmental information and operating condition information of the device under test; and construct a first detection feature matrix based on the partial discharge signal feature vector, the model feature vector, and the state feature vector.

[0131] Each module in the optimization device of the aforementioned partial discharge detection model can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0132] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements an optimization method for a partial discharge detection model. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0133] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0134] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: constructing a first detection feature matrix for the current moment based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model; inputting the first detection feature matrix into the partial discharge detection model to obtain a detection result of whether the device under test has experienced partial discharge at the current moment; determining a reward value for the partial discharge detection model based on the detection result; determining an optimization strategy for the partial discharge detection model based on the reward value; and optimizing the partial discharge detection model.

[0135] In one embodiment, when the processor executes the computer program, it further performs the following steps: if the reward value is greater than a preset value, it obtains an optimized sample of the partial discharge detection model and optimizes the partial discharge detection model using the optimized sample; if the reward value is less than the preset value, it sends the parameter information of the partial discharge detection model to the aggregation server and optimizes the partial discharge detection model based on the feedback parameters returned by the aggregation server; wherein, the aggregation server is used to comprehensively analyze the parameter information of partial discharge detection models corresponding to multiple different regions and determine the feedback parameters.

[0136] In one embodiment, when the processor executes the computer program, it further performs the following steps: obtaining the second detection feature matrix of the device under test at the next moment; generating an optimized sample based on the combination of the first detection feature matrix, the reward value, and the detection feature matrix at the next moment; inputting the optimized sample into the partial discharge detection model, updating the parameters of the partial discharge detection model, and obtaining the optimized partial discharge detection model.

[0137] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining the parameter change amount of the partial discharge detection model based on the parameter information of the partial discharge detection model and the initial parameters of the initial partial discharge detection model; sending the parameter change amount to the aggregation server so that the aggregation server can optimize the stored global detection model and determine the feedback parameters based on the parameter information of the optimized global detection model; receiving the feedback parameters returned by the aggregation server and optimizing the partial discharge detection model based on the feedback parameters returned by the aggregation server.

[0138] In one embodiment, the feedback parameters include model update parameters and model average score. When the processor executes the computer program, it further implements the following steps: determining the target model score of the edge device based on the reward value corresponding to the historical decision strategy; if the target model score is greater than or equal to the model average score, optimizing the partial discharge detection model based on the update parameters and a first preset learning rate; if the target model score is less than the model average score, optimizing the partial discharge detection model based on the update parameters and a second preset learning rate; the second preset learning rate is greater than the first preset learning rate.

[0139] In one embodiment, when the processor executes the computer program, it further performs the following steps: obtaining the model feature vector of the partial discharge detection model based on the historical decision-making strategy and the reward value corresponding to the historical decision-making strategy; determining the state feature vector based on the environmental information and operating condition information of the device under test; and constructing a first detection feature matrix based on the partial discharge signal feature vector, the model feature vector, and the state feature vector.

[0140] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it performs the following steps: constructing a first detection feature matrix for the current moment based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model; inputting the first detection feature matrix into the partial discharge detection model to obtain the detection result of whether the device under test has experienced partial discharge at the current moment; determining the reward value of the partial discharge detection model based on the detection result; determining the optimization strategy of the partial discharge detection model based on the reward value; and optimizing the partial discharge detection model.

[0141] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: if the reward value is greater than a preset value, then an optimized sample of the partial discharge detection model is obtained, and the partial discharge detection model is optimized using the optimized sample; if the reward value is less than the preset value, then the parameter information of the partial discharge detection model is sent to the aggregation server, and the partial discharge detection model is optimized based on the feedback parameters returned by the aggregation server; wherein, the aggregation server is used to comprehensively analyze the parameter information of the partial discharge detection models corresponding to multiple different regions and determine the feedback parameters.

[0142] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining the second detection feature matrix of the device under test at the next moment; generating an optimized sample based on the combination of the first detection feature matrix, the reward value, and the detection feature matrix at the next moment; inputting the optimized sample into the partial discharge detection model, updating the parameters of the partial discharge detection model, and obtaining the optimized partial discharge detection model.

[0143] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the parameter change of the partial discharge detection model based on the parameter information of the partial discharge detection model and the initial parameters of the initial partial discharge detection model; sending the parameter change to the aggregation server so that the aggregation server can optimize the stored global detection model and determine the feedback parameters based on the parameter information of the optimized global detection model; receiving the feedback parameters returned by the aggregation server and optimizing the partial discharge detection model based on the feedback parameters returned by the aggregation server.

[0144] In one embodiment, the feedback parameters include model update parameters and model average score. When the computer program is executed by the processor, it further implements the following steps: determining the target model score of the edge device based on the reward value corresponding to the historical decision strategy; if the target model score is greater than or equal to the model average score, optimizing the partial discharge detection model based on the update parameters and a first preset learning rate; if the target model score is less than the model average score, optimizing the partial discharge detection model based on the update parameters and a second preset learning rate; the second preset learning rate is greater than the first preset learning rate.

[0145] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining the model feature vector of the partial discharge detection model based on the historical decision-making strategy and the reward value corresponding to the historical decision-making strategy; determining the state feature vector based on the environmental information and operating condition information of the device under test; and constructing a first detection feature matrix based on the partial discharge signal feature vector, the model feature vector, and the state feature vector.

[0146] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps: constructing a first detection feature matrix for the current moment based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model; inputting the first detection feature matrix into the partial discharge detection model to obtain a detection result of whether the device under test has experienced partial discharge at the current moment; determining a reward value for the partial discharge detection model based on the detection result; determining an optimization strategy for the partial discharge detection model based on the reward value; and optimizing the partial discharge detection model.

[0147] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: if the reward value is greater than a preset value, then an optimized sample of the partial discharge detection model is obtained, and the partial discharge detection model is optimized using the optimized sample; if the reward value is less than the preset value, then the parameter information of the partial discharge detection model is sent to the aggregation server, and the partial discharge detection model is optimized based on the feedback parameters returned by the aggregation server; wherein, the aggregation server is used to comprehensively analyze the parameter information of the partial discharge detection models corresponding to multiple different regions and determine the feedback parameters.

[0148] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining the second detection feature matrix of the device under test at the next moment; generating an optimized sample based on the combination of the first detection feature matrix, the reward value, and the detection feature matrix at the next moment; inputting the optimized sample into the partial discharge detection model, updating the parameters of the partial discharge detection model, and obtaining the optimized partial discharge detection model.

[0149] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the parameter change of the partial discharge detection model based on the parameter information of the partial discharge detection model and the initial parameters of the initial partial discharge detection model; sending the parameter change to the aggregation server so that the aggregation server can optimize the stored global detection model and determine the feedback parameters based on the parameter information of the optimized global detection model; receiving the feedback parameters returned by the aggregation server and optimizing the partial discharge detection model based on the feedback parameters returned by the aggregation server.

[0150] In one embodiment, the feedback parameters include model update parameters and model average score. When the computer program is executed by the processor, it further implements the following steps: determining the target model score of the edge device based on the reward value corresponding to the historical decision strategy; if the target model score is greater than or equal to the model average score, optimizing the partial discharge detection model based on the update parameters and a first preset learning rate; if the target model score is less than the model average score, optimizing the partial discharge detection model based on the update parameters and a second preset learning rate; the second preset learning rate is greater than the first preset learning rate.

[0151] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: obtaining the model feature vector of the partial discharge detection model based on the historical decision-making strategy and the reward value corresponding to the historical decision-making strategy; determining the state feature vector based on the environmental information and operating condition information of the device under test; and constructing a first detection feature matrix based on the partial discharge signal feature vector, the model feature vector, and the state feature vector.

[0152] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0153] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0154] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0155] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An optimization method for a partial discharge detection model, characterized in that, Applied to edge devices, the method includes: Based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model, the first detection feature matrix at the current moment is constructed. The first detection feature matrix is ​​input into the partial discharge detection model to obtain the detection result of whether the device under test has a partial discharge phenomenon at the current time; Based on the detection results, a reward value for the partial discharge detection model is determined, an optimization strategy for the partial discharge detection model is determined based on the reward value, and the partial discharge detection model is optimized.

2. The method according to claim 1, characterized in that, The optimization strategy for determining the partial discharge detection model based on the reward value includes: If the reward value is greater than the preset value, then the optimized sample of the partial discharge detection model is obtained, and the partial discharge detection model is optimized using the optimized sample. If the reward value is less than the preset value, the parameter information of the partial discharge detection model is sent to the aggregation server. Based on the feedback parameters returned by the aggregation server, the partial discharge detection model is optimized. The aggregation server is used to comprehensively analyze the parameter information of partial discharge detection models corresponding to multiple different regions to determine the feedback parameters.

3. The method according to claim 2, characterized in that, The step of obtaining optimized samples of the partial discharge detection model and optimizing the partial discharge detection model using the optimized samples includes: Obtain the second detection feature matrix of the device under test at the next moment; The optimized sample is generated based on the combination of the first detection feature matrix, the reward value, and the detection feature matrix at the next time step. The optimized sample is input into the partial discharge detection model, and the parameters of the partial discharge detection model are updated to obtain the optimized partial discharge detection model.

4. The method according to claim 2, characterized in that, The step of sending the parameter information of the partial discharge detection model to the aggregation server, and optimizing the partial discharge detection model based on the feedback parameters returned by the aggregation server, includes: Based on the parameter information of the partial discharge detection model and the initial parameters of the initial partial discharge detection model, the parameter change of the partial discharge detection model is determined. The parameter change is sent to the aggregation server so that the aggregation server can optimize the stored global detection model and determine the feedback parameter based on the parameter information of the optimized global detection model. The system receives feedback parameters returned by the aggregation server and optimizes the partial discharge detection model based on these parameters.

5. The method according to claim 4, characterized in that, The feedback parameters include model update parameters and model average score. The optimization of the partial discharge detection model based on the feedback parameters returned by the aggregation server includes: The target model score of the edge device is determined based on the reward value corresponding to the historical decision-making strategy. If the target model score is greater than or equal to the model average score, the partial discharge detection model is optimized based on the update parameters and the first preset learning rate. If the target model score is less than the model average score, the partial discharge detection model is optimized based on the update parameters and the second preset learning rate; the second preset learning rate is greater than the first preset learning rate.

6. The method according to claim 1, characterized in that, The first detection feature matrix for the current moment is constructed based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model, including: Based on the historical decision-making strategy and the reward value corresponding to the historical decision-making strategy, the model feature vector of the partial discharge detection model is obtained; Based on the environmental and operating information of the device under test, determine the state feature vector; The first detection feature matrix is ​​constructed based on the partial discharge signal feature vector, the model feature vector, and the state feature vector.

7. An optimization device for a partial discharge detection model, characterized in that, Applied to edge devices, the device includes: The construction module is used to construct the first detection feature matrix at the current moment based on the feature vector of the partial discharge signal of the device under test at the current moment and the historical decision-making strategy of the partial discharge detection model; The input module is used to input the first detection feature matrix into the partial discharge detection model to obtain the detection result of whether the device under test has a partial discharge phenomenon at the current time; An optimization module is used to determine the reward value of the partial discharge detection model based on the detection results, determine the optimization strategy of the partial discharge detection model based on the reward value, and perform optimization processing on the partial discharge detection model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images