A method and system for identifying defects in power equipment based on edge-cloud collaboration

Abstract

The present invention discloses a method and system for power equipment defect identification based on edge-cloud collaboration, which relates to the technical field of power equipment defect identification. The method includes: a data acquisition device collects monitoring data of power equipment in real time and sends the monitoring data to an edge computing gateway within the monitoring area; the edge computing gateway receives the monitoring data, preprocesses the monitoring data, obtains the preprocessed monitoring data and uploads it to the cloud data center; the cloud data center receives the preprocessed monitoring data, uses a multi-modal defect identification model to identify power equipment defects for the received data, obtains real-time defect identification results, uses a maintenance strategy generation model to process the real-time defect identification results, obtains real-time maintenance strategies, and sends the real-time maintenance strategies to the edge computing gateway. This method adopts an edge-cloud collaboration mechanism to timely and efficiently identify power equipment defects for the monitoring data of power equipment, and analyzes the defect identification results to obtain corresponding maintenance strategies.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment defect identification, and specifically relates to a method and system for power equipment defect identification based on edge-cloud collaboration. Background Art

[0002] Power equipment is an important part of the power system, and its health status is directly related to the safe and stable operation of the power system. Equipment defects may lead to accidents such as power outages, fires, or even explosions, which have a serious impact on people's lives and production. Therefore, timely identification of power equipment defects is crucial for ensuring the safety of the power system.

[0003] The existing power equipment defect identification technologies have the following defects:

[0004] Traditional power equipment defect identification usually performs data processing locally. However, this method is inefficient when dealing with a large amount of data and is easily limited by the computing power of the equipment, resulting in insufficient real-time monitoring and analysis capabilities of power equipment and an inability to detect potential defects and faults in a timely manner;

[0005] Existing power equipment defect identification often relies on a single type of monitoring data, such as only relying on image recognition or only relying on sensor data, lacking the fusion analysis of multi-source data. The limitations of a single data source may lead to low accuracy in defect identification, increasing the risk of misdiagnosis and missed diagnosis;

[0006] Existing power equipment defect identification lacks a response mechanism. After a power equipment defect is discovered, maintenance is carried out according to a preset maintenance strategy, and the preset maintenance strategy is often formulated based on experience, lacking data-driven intelligent decision support, which may lead to unreasonable resource allocation or incorrect processing decisions, affecting the stable operation and economic benefits of the power system. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and system for power equipment defect identification based on edge-cloud collaboration, which solves the problems of low data processing efficiency, insufficient real-time monitoring and analysis capabilities, and low accuracy in defect identification existing in the prior art.

[0008] The present invention is achieved through the following technical solutions:

[0009] In a first aspect, a method for power equipment defect identification based on edge-cloud collaboration provided by an embodiment of the present invention includes the following steps:

[0010] The data acquisition device collects the monitoring data of the power equipment in real time and sends the monitoring data of the power equipment to the edge computing gateway within the monitoring area;

[0011] The edge computing gateway receives the monitoring data of power equipment sent by the data acquisition device, preprocesses the monitoring data of the power equipment to obtain the preprocessed monitoring data, and uploads the preprocessed monitoring data to the cloud data center;

[0012] The cloud data center receives the preprocessed monitoring data sent by the edge computing gateway, uses the constructed multimodal defect recognition model to identify the defects of power equipment in the preprocessed monitoring data to obtain the real-time defect recognition result, uses the constructed maintenance strategy generation model to process the real-time defect recognition result to obtain the real-time maintenance strategy, and sends the real-time maintenance strategy to the edge computing gateway.

[0013] Further, before the step of the data acquisition device collecting the monitoring data of power equipment in real time, it also includes: taking the minimization of the edge-cloud collaborative deployment cost as the optimization goal, using an intelligent optimization algorithm to generate an edge-cloud collaborative deployment plan for the power system, and using an artificial intelligence algorithm to construct a multimodal defect recognition model and a maintenance strategy generation model in the cloud data center.

[0014] Further, the method of using the intelligent optimization algorithm to generate the edge-cloud collaborative deployment plan for the power system uses the ICPO optimization algorithm to generate the edge-cloud collaborative deployment plan for the power system, specifically including:

[0015] Taking the minimization of the edge-cloud collaborative deployment cost as the optimization goal, setting the objective function of the ICPO optimization algorithm, and encoding the ICPO individuals of the ICPO optimization algorithm to obtain the individual encoding format;

[0016] According to the objective function, setting the fitness function of the ICPO optimization algorithm, and setting the ICPO population parameters and the maximum number of iterations of the ICPO optimization algorithm;

[0017] According to the individual encoding format and the ICPO population parameters, initializing the ICPO population to obtain an initial ICPO population containing several initial ICPO individuals;

[0018] According to the fitness function, the ICPO population parameters, and the maximum number of iterations, performing iterative optimization on the initial ICPO population to obtain the optimal ICPO individual;

[0019] Decoding the individual encoding vector corresponding to the optimal ICPO individual to obtain the optimal edge-cloud collaborative deployment plan for the power system.

[0020] Further, the method of using an artificial intelligence algorithm to construct a multimodal defect recognition model and a maintenance strategy generation model in the cloud data center specifically includes:

[0021] Collecting a number of historical monitoring data, preprocessing the historical monitoring data to obtain the preprocessed historical monitoring data;

[0022] According to the preprocessed historical monitoring data, use the multi-modal fusion deep learning algorithm to construct a multi-modal defect recognition model and obtain several historical defect recognition results;

[0023] According to the historical defect recognition results, use the reinforcement learning algorithm to construct a maintenance strategy generation model and obtain several historical maintenance strategy generation experiences.

[0024] Furthermore, according to the preprocessed historical monitoring data, the specific method of using the multi-modal fusion deep learning algorithm to construct a multi-modal defect recognition model includes:

[0025] Divide several preprocessed historical monitoring data into a model training set and a model test set according to a ratio of 7:3;

[0026] Use the CNN-LSTM-Attention-Elman algorithm to construct an initial multi-modal defect recognition model. The multi-modal defect recognition model includes an image feature extraction module constructed based on the CNN algorithm, an operation feature extraction module constructed based on the LSTM algorithm, an attention weight module constructed based on the Attention mechanism, and a defect recognition module constructed based on the Elman algorithm. Use the model training set to optimize and train the initial multi-modal defect recognition model to obtain an optimized multi-modal defect recognition model;

[0027] Use the model test set to test the optimized multi-modal defect recognition model to obtain the corresponding model accuracy;

[0028] If the model accuracy is greater than the accuracy threshold, output the optimal multi-modal defect recognition model; otherwise, continue with the optimization training.

[0029] Furthermore, the maintenance strategy generation model is constructed based on the DQN algorithm. The maintenance strategy generation model includes an agent, an experience replay pool, and a deep Q network. According to the historical defect recognition results, the specific method of using the reinforcement learning algorithm to construct a maintenance strategy generation model includes:

[0030] According to the maintenance strategy generation problem, set the simulation environment of the DQN algorithm and construct an agent and an experience replay pool;

[0031] According to the states involved in the historical defect recognition results, define the state space of the DQN algorithm and construct the input layer of the deep Q network according to the data structure of the historical defect recognition results;

[0032] According to the actions involved in the maintenance strategy, define the action space of the DQN algorithm and construct the output layer of the deep Q network according to the data structure of the maintenance strategy;

[0033] Set several hidden layers according to the input layer and the output layer, construct the corresponding deep Q-network, connect the input layer to the state space, and connect the output layer to the action space;

[0034] Define the reward function of the DQN algorithm according to the influence of each action in the action space;

[0035] Based on the state space, the action space, and the reward function, use several historical defect recognition results to optimize and train the deep Q-network and the agent, construct a maintenance strategy generation model, and generate several historical maintenance strategy generation experiences;

[0036] Store several historical maintenance strategy generation experiences in the experience replay pool of the maintenance strategy generation model.

[0037] Further, the monitoring data of the power equipment includes real-time image monitoring data and real-time operation monitoring data. The specific method for the edge computing gateway to receive the monitoring data sent by the data acquisition device, preprocess the monitoring data of the power equipment, obtain the preprocessed monitoring data, and upload the preprocessed monitoring data to the cloud data center includes:

[0038] The edge computing gateway performs image preprocessing on the real-time image monitoring data to obtain preprocessed real-time image monitoring data;

[0039] The edge computing gateway performs data preprocessing on the real-time operation monitoring data to obtain preprocessed real-time operation monitoring data;

[0040] Integrate the preprocessed real-time image monitoring data and the preprocessed real-time operation monitoring data of the same power equipment to obtain the preprocessed monitoring data;

[0041] Compress and encrypt the preprocessed monitoring data to obtain an encrypted data compression package, and upload it to the cloud data center.

[0042] Further, the specific method for the cloud data center to receive the preprocessed monitoring data sent by the edge computing gateway and use the constructed multi-modal defect recognition model to identify power equipment defects in the preprocessed monitoring data to obtain real-time defect recognition results includes:

[0043] The cloud data center receives the preprocessed monitoring data, decrypts and decompresses the encrypted data compression package to obtain the decrypted preprocessed monitoring data;

[0044] The image feature extraction module extracts the real-time image features of the decrypted preprocessed real-time image monitoring data;

[0045] The operation feature extraction module extracts the real-time operation features of the decrypted preprocessed real-time operation monitoring data;

[0046] Using the attention weight values preset by the attention weight module, perform weighted splicing on the real-time image features and real-time operation features to obtain real-time weighted splicing features;

[0047] Use the defect recognition module to perform defect recognition on the real-time weighted splicing features to obtain real-time defect recognition results.

[0048] Further, the specific method for using the constructed maintenance strategy generation model to process the real-time defect recognition results to obtain real-time maintenance strategies includes:

[0049] The cloud data center updates the state space of the maintenance strategy generation model according to the real-time defect recognition results to obtain an updated state space;

[0050] Extract several historical maintenance strategy generation experiences from the experience replay pool, and update the action space of the maintenance strategy generation model according to the several historical maintenance strategy generation experiences to obtain an updated action space;

[0051] Input the updated state space into the input layer of the deep Q network of the maintenance strategy generation model, and connect the updated action space to the output layer of the deep Q network to obtain an updated deep Q network;

[0052] Based on the updated deep Q network, use an agent to control the updated deep Q network to output the Q values of possible actions in the updated action space;

[0053] Iteratively update the Q values according to a preset reward function to obtain updated Q values until the number of iterations reaches the number threshold;

[0054] According to the greedy strategy, take the possible action with the highest updated Q value in each iteration as the execution action corresponding to the corresponding state;

[0055] Integrate the execution actions corresponding to all states in the updated state space to obtain a real-time maintenance strategy and return it to the corresponding edge computing gateway.

[0056] In a second aspect, an edge-cloud collaborative power equipment defect recognition system provided by an embodiment of the present invention is used to implement the edge-cloud collaborative power equipment defect recognition method described in the above embodiment. The system includes a cloud data center, several edge computing gateways, and several data acquisition devices. The data acquisition devices are arranged at power equipment to collect monitoring data of the power equipment in real time and send the monitoring data of the power equipment to the edge computing gateway within the monitoring area;

[0057] The edge computing gateway is used to receive the monitoring data of power equipment sent by the data acquisition device, preprocess the monitoring data of the power equipment to obtain the preprocessed monitoring data, and upload the preprocessed monitoring data to the cloud data center;

[0058] The cloud data center is used to receive the preprocessed monitoring data sent by the edge computing gateway, identify power equipment defects in the preprocessed monitoring data by using the constructed multimodal defect recognition model to obtain real-time defect recognition results, process the real-time defect recognition results by using the maintenance strategy generation model to obtain real-time maintenance strategies, and send the real-time maintenance strategies to the edge computing gateway.

[0059] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0060] The embodiment of the present invention discloses a method for identifying power equipment defects based on edge-cloud collaboration, which introduces an edge-cloud collaboration mechanism, combines edge computing with cloud computing, effectively solves the deficiencies of traditional power equipment defects in data processing capabilities and real-time performance, strengthens data information interaction, and avoids the data island effect; the edge computing gateway in the monitoring area preprocesses the collected real-time monitoring data, reducing the computing burden on the cloud data center and improving the efficiency of data processing; the constructed multimodal defect recognition model can fuse different modalities of monitoring data for data analysis, timely discover potential defects and faults, improve efficiency, and at the same time improve the accuracy and reliability of defect identification; the constructed maintenance strategy generation model provides a response mechanism, which can intelligently generate maintenance strategies according to real-time defect recognition results, optimize resource allocation, provide guidance for processing decisions, and greatly improve the timeliness and reliability of defect maintenance.

[0061] This method can generate an adaptive edge-cloud collaboration deployment plan for the power system, complete the deployment of a distributed edge-cloud collaboration architecture, reduce the deployment cost, and improve the efficiency of data processing and transmission.

[0062] The embodiment of the present invention discloses a system for identifying power equipment defects based on edge-cloud collaboration and a method for identifying power equipment defects based on edge-cloud collaboration, which are based on the same inventive concept and have the same beneficial effects, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings. In the drawings:

[0064] Figure 1 Flow chart of a method for identifying defects in power equipment based on edge-cloud collaboration provided by an embodiment of the present invention;

[0065] Figure 2 Block diagram of the structure of a system for identifying defects in power equipment based on edge-cloud collaboration provided by another embodiment of the present invention. Detailed implementation manners

[0066] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the embodiments and the drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention. Embodiment

[0067] As Figure 1 shown, the first embodiment of the present invention provides a method for identifying defects in power equipment based on edge-cloud collaboration, including the following steps:

[0068] S1: Generate an edge-cloud collaborative deployment plan, and use artificial intelligence algorithms to build a multi-modal defect recognition model and a maintenance strategy generation model in the cloud data center, including the following steps:

[0069] S1-1: With the goal of minimizing the edge-cloud collaborative deployment cost, use the Improved Crested Porcupine Optimizer (ICPO) for optimization to generate an edge-cloud collaborative deployment plan for the power system, including the following steps:

[0070] S1-1-1: With the goal of minimizing the edge-cloud collaborative deployment cost, set the objective function of the ICPO optimization algorithm, and encode the ICPO individuals of the ICPO optimization algorithm to obtain the individual encoding format.

[0071] S1-1-2: According to the objective function, set the fitness function of the ICPO optimization algorithm, and set the ICPO population parameters and the maximum number of iterations of the ICPO optimization algorithm. The formula is:

[0072] (1)

[0073] In formula (1), is the fitness function; is the ICPO individual the th item of the edge-cloud collaborative deployment cost, including hardware cost, communication cost, monitoring range cost, etc.; is the th cost weight; is the cost indicator; The ICPO individual indication quantity, n is a natural number.

[0074] S1-1-3: According to the individual coding format and ICPO population parameters, use the Circle chaotic mapping sequence to initialize the ICPO population, obtaining an initial ICPO population containing several initial ICPO individuals. The formula is:

[0075] (2)

[0076] In formula (2), is the initial ICPO individual of the Circle chaotic mapping; is the randomly generated initial ICPO individual; is the ICPO individual indication quantity.

[0077] S1-1-4: According to the fitness function, ICPO population parameters, and the maximum number of iterations, perform iterative optimization on the initial ICPO population to obtain the optimal ICPO individual. Specifically, it includes the following steps:

[0078] S1-1-4-1: Introduce a cyclic population reduction mechanism to limit the number of individuals in the ICPO population parameters, obtaining the updated ICPO population parameters for the next iteration. The formula is:

[0079] (3)

[0080] In formula (3), is the number of individuals in the ICPO population parameters for the th iteration; is the number of individuals in the ICPO population parameters for the th iteration; is the minimum value of the number of individuals in the ICPO population parameters; is the function evaluation parameter; is the function evaluation loop parameter; is the maximum function evaluation loop parameter; t is the iteration number indication quantity.

[0081] S1-1-4-2: According to the fitness function, calculate the initial fitness values of the initial ICPO individuals in the initial ICPO population.

[0082] S1-1-4-3: According to the initial fitness values and the updated ICPO population parameters, use the first defense strategy, the second defense strategy, the third defense strategy, and the fourth defense strategy to update the initial ICPO population, obtaining the updated ICPO population.

[0083] Among them, the formula for the first defense strategy is:

[0084] (4)

[0085] In formula (4), is the updated ICPO individual within the first defense range; is the initial ICPO individual within the first defense range; is a random number based on the normal distribution; is a random value within the interval [0, 1]; is the optimal solution within the first defense range; is a vector generated between the true optimal solution and the randomly selected optimal solution from the ICPO population within the first defense range; is the ICPO individual indicator.

[0086] The formula for the second defense strategy is:

[0087] (5)

[0088] In formula (5), is the updated ICPO individual within the second defense range; is the initial ICPO individual within the second defense range; is the search upper limit vector of the second defense range; is a random value within the interval [0, 1]; are respectively the th initial ICPO individuals; are both two random integers between [1, S]; is a vector generated between the true optimal solution and the randomly selected optimal solution from the ICPO population within the second defense range.

[0089] The formula for the third defense strategy is:

[0090] (6)

[0091] In formula (6), is the updated ICPO individual within the third defense range; is the initial ICPO individual within the third defense range; is the search upper limit vector of the third defense range; are respectively the th initial ICPO individuals; is a random integer between [1, S]; is the odor diffusion factor defined by the fitness function; is the defense factor; is the search direction control parameter.

[0092] The formula for the fourth defense strategy is as follows:

[0093] (7)

[0094] In formula (7), is the updated ICPO individual within the fourth defense range; is the initial ICPO individual within the fourth defense range; is the optimal solution within the fourth defense range; are all random values within the interval [0, 1]; is the defense factor; is the search direction control parameter; is the average force affecting the search direction; is the convergence speed factor.

[0095] S1-1-4-4: Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the updated ICPO population to generate a dynamically reversed ICPO population. The formula is as follows:

[0096] (8)

[0097] In formula (8), is the dynamically reversed ICPO individual; is the decreasing inertia coefficient; are the maximum and minimum values of the vector space respectively; is the updated ICPO individual.

[0098] S1-1-4-5: According to the fitness function, calculate the fitness values of all ICPO individuals in the updated ICPO population and the dynamically reversed ICPO population. Take the ICPO individual with the minimum fitness value as the optimal ICPO individual and retain the optimal ICPO individual.

[0099] S1-1-4-6: If the number of iterations of iterative optimization reaches the maximum number of iterations or the fitness value of the optimal ICPO individual meets the requirements, output the optimal ICPO individual.

[0100] S1-1-5: Decode the individual coding vector corresponding to the optimal ICPO individual to obtain the optimal edge-cloud collaborative deployment plan for the power system. The optimal edge-cloud collaborative deployment plan includes the number of edge computing gateways, the locations of edge computing gateways, the number of data acquisition devices, the locations of data acquisition devices, the communication connection relationship between data acquisition devices and edge computing gateways, and so on.

[0101] S1-2: Collect a number of historical monitoring data and preprocess the number of historical monitoring data to obtain a number of preprocessed historical monitoring data.

[0102] The preprocessing includes image preprocessing of the historical image monitoring data in the historical monitoring data and data preprocessing of the historical operation monitoring data; the image preprocessing includes Gaussian denoising, translation, scaling, cropping, grayscale processing, and data augmentation, etc., and the data preprocessing includes data cleaning, data dimensionality reduction, and magnitude normalization processing, etc., to improve the data quality and provide data support for the subsequent model construction.

[0103] S1-3: According to a number of preprocessed historical monitoring data, use a multi-modal fusion deep learning algorithm to construct a multi-modal defect recognition model and generate a number of historical defect recognition results.

[0104] The multi-modal defect recognition model is constructed based on the CNN-LSTM-Attention-Elman algorithm, and the multi-modal defect recognition model includes an image feature extraction module constructed based on the Convolutional Neural Network (CNN) algorithm, an operation feature extraction module constructed based on the Long Short-Term Memory (LSTM) algorithm, an attention weight module constructed based on the Attention mechanism, and a defect recognition module constructed based on the Elman algorithm. Both the image feature extraction module and the operation feature extraction module are connected to the attention weight module, and the attention weight module is connected to the defect recognition module.

[0105] The CNN network highly conforms to the two-dimensional structure of the image, and the pixels of the image are arranged in a fixed grid form. The CNN can effectively extract local features through the convolutional layer. Through multiple convolutional and pooling operations, it can gradually abstract the high-level features of the image, so as to perform well in recognizing complex objects and be able to extract the deep features in the image; the LSTM network can learn to retain important information and ignore unimportant information in a long time series through the gate mechanism, making the LSTM network have significant advantages when processing sequence data. The LSTM network can capture the long-term dependence relationship in the data and extract the data features in the operation data sequence; the attention weight module automatically assigns different degrees of attention to different parts of the input feature sequence through the set attention weight values, significantly improving the model's ability to process multi-modal long sequence features, because it allows the model to focus on the most relevant parts of the input feature sequence when predicting each output; the Elman network enhances the traditional classifier by introducing context units (or called state units), and these context units can store past information, thereby affecting the current network output and achieving more accurate and efficient defect recognition prediction.

[0106] The specific method for constructing a multi-modal defect recognition model using a multi-modal fusion deep learning algorithm includes the following steps:

[0107] S1-3-1: Divide a number of preprocessed historical monitoring data into a model training set and a model test set according to a ratio of 7:3;

[0108] S1-3-2: Use the CNN-LSTM-Attention-Elman algorithm to construct an initial multi-modal defect recognition model, and input the model training set to optimize and train the initial multi-modal defect recognition model to obtain an optimized multi-modal defect recognition model;

[0109] S1-3-3: Input the model test set to test the optimized multi-modal defect recognition model to obtain the corresponding model accuracy;

[0110] S1-3-4: If the model accuracy is greater than the accuracy threshold, output the optimal multi-modal defect recognition model and generate a number of historical defect recognition results; otherwise, continue with the optimization training.

[0111] S1-4: Based on a number of historical defect recognition results, use the reinforcement learning algorithm to construct a maintenance strategy generation model and generate a number of historical maintenance strategy generation experiences. Among them, the maintenance strategy generation model is constructed based on the Deep Q Network (DQN) algorithm, and the maintenance strategy generation model is provided with an agent, an experience replay pool, and a deep Q network.

[0112] The specific method for constructing a maintenance strategy generation model using the reinforcement learning algorithm includes the following steps:

[0113] S1-4-1: According to the maintenance strategy generation problem, set the simulation environment of the DQN algorithm, and construct an agent and an experience replay pool;

[0114] S1-4-2: Define the state space of the DQN algorithm according to the states involved in the historical defect recognition results, and construct the input layer of the deep Q network according to the data structure of the historical defect recognition results;

[0115] S1-4-3: Define the action space of the DQN algorithm according to the actions involved in the maintenance strategy, and construct the output layer of the deep Q network according to the data structure of the maintenance strategy;

[0116] S1-4-4: Set a number of hidden layers according to the input layer and the output layer, construct the corresponding deep Q network, connect the input layer to the state space, and connect the output layer to the action space;

[0117] S1-4-5: Define the reward function of the DQN algorithm according to the influence of each action in the action space;

[0118] S1-4-6: Based on the state space, action space, and reward function, using several historical defect identification results, optimize and train the deep Q-network and the agent, construct an overhaul strategy generation model, and generate several historical overhaul strategy generation experiences;

[0119] S1-4-7: Store several historical overhaul strategy generation experiences in the experience replay pool of the overhaul strategy generation model.

[0120] S2: According to the edge-cloud collaborative deployment plan, set up edge computing gateways in each monitoring area of the power system, and set up data acquisition devices at each power equipment in the monitoring area.

[0121] S3: Use the edge computing gateway to preprocess the real-time monitoring data of the power equipment collected by the data acquisition devices in the monitoring area, obtain the preprocessed real-time monitoring data, and upload it to the cloud data center.

[0122] Among them, the real-time monitoring data includes real-time image monitoring data and real-time operation monitoring data; the real-time operation monitoring data includes the power data of the power equipment, current, voltage, power, load, operation time, etc., and the temperature, humidity, pressure, etc. of the environment where the power equipment is located; the real-time operation monitoring data is used to analyze the internal defects of the power equipment, such as operation defects at high temperatures, operation vibration defects, etc., and the real-time image monitoring data is used to analyze the external defects of the power equipment, such as insulation layer damage, insulator ash accumulation, chassis aging, etc.

[0123] The method of using the edge computing gateway to preprocess the real-time monitoring data of the power equipment collected by the data acquisition devices in the monitoring area, obtain the preprocessed real-time monitoring data, and upload it to the cloud data center includes the following steps:

[0124] S3-1: Use the data acquisition device to collect the real-time monitoring data of the power equipment, and send the real-time monitoring data to the edge computing gateway in the monitoring area;

[0125] S3-2: Use the edge computing gateway to perform image preprocessing on the real-time image monitoring data in the real-time monitoring data, and obtain the preprocessed real-time image monitoring data;

[0126] S3-3: Use the edge computing gateway to perform data preprocessing on the real-time operation monitoring data in the real-time monitoring data, and obtain the preprocessed real-time operation monitoring data;

[0127] S3-4: Integrate the preprocessed real-time image monitoring data and the preprocessed real-time operation monitoring data of the same power equipment to obtain the preprocessed real-time monitoring data;

[0128] S3-5: Compress and encrypt the real-time monitoring data after preprocessing to obtain an encrypted data compression package, and upload the encrypted data compression package to the cloud data center.

[0129] Among them, the source of the key used for encrypting the data compression package is: based on a trusted institution, using an asymmetric encryption algorithm to generate public and private keys for each edge computing gateway. The specific acquisition method includes the following steps:

[0130] A-1: Based on a trusted institution, perform key initialization to obtain public parameters, a master key, and an initial key. The formula is:

[0131] (9)

[0132] In formula (9), is the public parameter; is the master key; is the initial key; is the integer domain random number; are all target hash functions; are all cyclic groups random numbers of the generators; is the random number bilinear mapping.

[0133] A-2: Use an asymmetric encryption algorithm to generate a public-private key pair for the edge computing gateway;

[0134] (10)

[0135] In formula (10), is the private key of the edge computing gateway ; is the integer domain random number; is the second private key parameter of the edge computing gateway ; is the public parameter target hash function; is the edge computing gateway indicator; is the master key; is the initial key; is the edge computing gateway public key; is the cyclic group random number of the generator; is the edge computing gateway attribute information.

[0136] A-3: Store the private key in the public-private key pair in the corresponding edge computing gateway, and store the public key in the public-private key pair in the cloud data center. Encrypt the data compression package. The formula is:

[0137] (11)

[0138] In formula (11), is the encrypted data compression package; is the asymmetric encryption function; is the data compression package; is the edge computing gateway 's private key; is the edge computing gateway indicator.

[0139] S4: Based on the cloud data center, use the multi-modal defect recognition model to perform defect recognition on the preprocessed real-time monitoring data to obtain the real-time defect recognition result, including the following steps:

[0140] S4-1: Based on the cloud data center, receive the encrypted data compression package sent by the edge computing gateway, decrypt and decompress it to obtain the decrypted preprocessed real-time monitoring data. Decrypt the encrypted data compression package. The formula is:

[0141] (12)

[0142] In formula (12), is the decrypted data compression package; is the asymmetric decryption function; is the encrypted data compression package; is the edge computing gateway 's public key.

[0143] S4-2: Use the image feature extraction module of the multi-modal defect recognition model to extract the real-time image features of the real-time image monitoring data in the decrypted preprocessed real-time monitoring data.

[0144] S4-3: Use the operation feature extraction module to extract the real-time operation features of the real-time operation monitoring data in the decrypted preprocessed real-time monitoring data.

[0145] S4-4: According to the preset attention weight value of the attention weight module, perform weighted splicing on the real-time image features and the real-time operation features to obtain the real-time weighted splicing features.

[0146] S4-5: Use the defect recognition module to perform defect recognition according to the real-time weighted splicing features to obtain the real-time defect recognition result.

[0147] S5: Based on the cloud data center, according to the real-time defect recognition results, use the maintenance strategy generation model to generate a maintenance strategy, obtain the real-time maintenance strategy, and return it to the corresponding edge computing gateway, including the following steps:

[0148] S5-1: Based on the cloud data center, according to the real-time defect recognition results, update the state space of the maintenance strategy generation model to obtain the updated state space , where is the updated state value, is the state indicator, is the total number of state space dimensions.

[0149] S5-2: Extract several historical maintenance strategy generation experiences from the experience replay pool, and update the action space of the maintenance strategy generation model according to the several historical maintenance strategy generation experiences to obtain the updated action space , where is the updated action value, is the action indicator, is the total number of action space dimensions.

[0150] S5-3: Input the updated state space into the input layer of the deep Q-network of the maintenance strategy generation model, and connect the updated action space to the output layer of the deep Q-network to obtain the updated deep Q-network.

[0151] S5-4: Based on the updated deep Q-network, use the agent to control the updated deep Q-network and output the Q values of the possible actions in the updated action space.

[0152] S5-5: According to the preset reward function, iteratively update the Q values to obtain the updated Q values until the number of iterations reaches the threshold, and the formula is:

[0153] (13)

[0154] In formula (13), is the updated state value and the updated action value corresponding updated Q value; is the state value and the action value corresponding predicted Q value; is the learning rate; is the highest predicted Q value; is the comprehensive indicator; is the update parameter.

[0155] S5-6: According to the greedy strategy, use the possible action with the highest updated Q value in each iteration as the execution action for the corresponding state.

[0156] S5-7: Integrate the execution actions corresponding to all states in the updated state space to obtain a real-time maintenance strategy, and return it to the corresponding edge computing gateway.

[0157] A method for identifying power equipment defects based on edge-cloud collaboration provided by an embodiment of the present invention introduces an edge-cloud collaboration mechanism, combines edge computing with cloud computing, effectively solves the deficiencies of traditional power equipment defects in data processing capabilities and real-time performance, and strengthens the information interaction of data, avoiding the data island effect; generates an adaptive edge-cloud collaboration deployment plan for the power system, completes the deployment of a distributed edge-cloud collaboration architecture, reduces the deployment cost, and improves the efficiency of data processing and transmission; constructs a multi-modal defect identification model that can analyze data by fusing monitoring data of different modalities, discovers potential defects and faults in a timely manner, improves efficiency, and at the same time improves the accuracy and reliability of defect identification; preprocesses the collected real-time monitoring data at the edge computing gateway in the monitoring area, reduces the computing burden on the cloud data center, and improves the efficiency of data processing; constructs a maintenance strategy generation model, provides a response mechanism, can intelligently generate a maintenance strategy according to the real-time defect identification result, optimize resource allocation, and provide guidance for processing decisions, greatly improving the timeliness and reliability of defect maintenance. Embodiment

[0158] As Figure 2 shown, this embodiment provides a power equipment defect identification system based on edge-cloud collaboration for implementing the power equipment defect identification method based on edge-cloud collaboration. The system includes a cloud data center, several edge computing gateways, and several data collection devices. The cloud data center is respectively communicatively connected to the edge computing gateways in several monitoring areas in the power system. Each edge computing gateway is respectively communicatively connected to several data collection devices in the corresponding monitoring area. Each data collection device is arranged at a power equipment.

[0159] The data collection device is used to collect real-time monitoring data of the corresponding power equipment and send the real-time monitoring data to the edge computing gateway in the monitoring area.

[0160] The edge computing gateway is used to receive the monitoring data sent by the data collection device in the monitoring area, perform preprocessing to obtain the preprocessed monitoring data, and upload it to the cloud data center.

[0161] The cloud data center is used to receive the preprocessed monitoring data sent by the edge computing gateway, and uses the constructed multi-modal defect recognition model to identify power equipment defects in the preprocessed monitoring data, obtaining real-time defect recognition results. It uses the maintenance strategy generation model to process the real-time defect recognition results, obtaining real-time maintenance strategies, and sends the real-time maintenance strategies to the edge computing gateway.

[0162] The cloud data center is provided with an initialization unit, a defect recognition unit, and a maintenance strategy generation unit that are connected in sequence;

[0163] The initialization unit is used to generate an edge-cloud collaborative deployment plan, and uses artificial intelligence algorithms to construct a multi-modal defect recognition model and a maintenance strategy generation model in the cloud data center;

[0164] The defect recognition unit is used to perform defect recognition on the preprocessed monitoring data using the multi-modal defect recognition model to obtain real-time defect recognition results;

[0165] The maintenance strategy generation unit is used to generate a maintenance strategy according to the real-time defect recognition results using the maintenance strategy generation model, obtaining a real-time maintenance strategy, and returning it to the corresponding edge computing gateway.

[0166] A power equipment defect recognition system based on edge-cloud collaboration provided by an embodiment of the present invention introduces an edge-cloud collaboration mechanism, combines edge computing and cloud computing, effectively solves the deficiencies of traditional power equipment defects in terms of data processing capabilities and real-time performance, and strengthens data information interaction, avoiding the data island effect; generates an adaptive edge-cloud collaborative deployment plan for the power system, completes the deployment of a distributed edge-cloud collaborative architecture, reduces deployment costs, and improves the efficiency of data processing and transmission; the constructed multi-modal defect recognition model can fuse different modalities of monitoring data for data analysis, timely discover potential defects and faults, improve efficiency, and at the same time improve the accuracy and reliability of defect recognition; the edge computing gateway in the monitoring area preprocesses the collected real-time monitoring data, reducing the computing burden on the cloud data center and improving the efficiency of data processing; the constructed maintenance strategy generation model provides a response mechanism, can intelligently generate maintenance strategies according to real-time defect recognition results, optimize resource allocation, provide guidance for processing decisions, and greatly improve the timeliness and reliability of defect maintenance.

[0167] The above specific implementation manners further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above is only the specific implementation manner of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying defects in power equipment based on edge-cloud collaboration, characterized in that: The following steps are involved: The data acquisition device collects monitoring data of the power equipment in real time and sends the monitoring data to the edge computing gateway in the monitoring area; The edge computing gateway receives the monitoring data sent by the data acquisition device, preprocesses the monitoring data to obtain the preprocessed monitoring data, and uploads the preprocessed monitoring data to the cloud data center; The cloud data center receives the preprocessed monitoring data sent by the edge computing gateway, uses the constructed multimodal defect recognition model to identify the defects of the power equipment on the preprocessed monitoring data, obtains real-time defect recognition results, uses the constructed maintenance strategy generation model to process the real-time defect recognition results, obtains a real-time maintenance strategy, and sends the real-time maintenance strategy to the edge computing gateway; Before the step of real-time collection of monitoring data of power equipment by the data collection device, the method further includes: taking minimization of edge-cloud collaborative deployment cost as the optimization goal, using an intelligent optimization algorithm to generate an edge-cloud collaborative deployment plan for the power system, and using an artificial intelligence algorithm in a cloud data center to build a multimodal defect recognition model and a maintenance strategy generation model; the method of using an intelligent optimization algorithm to generate an edge-cloud collaborative deployment plan for the power system adopts an ICPO optimization algorithm to generate an edge-cloud collaborative deployment plan for the power system, specifically including: Taking minimizing the edge-cloud collaborative deployment cost as the optimization goal, the objective function of the ICPO optimization algorithm is set, and the ICPO individuals of the ICPO optimization algorithm are encoded to obtain the individual encoding format; According to the objective function, set the fitness function of the ICPO optimization algorithm, and set the ICPO population parameters and the maximum number of iterations of the ICPO optimization algorithm; According to the individual coding format and ICPO population parameters, the ICPO population is initialized to obtain an initial ICPO population including a number of initial ICPO individuals; According to the fitness function, ICPO population parameters and the maximum number of iterations, the initial ICPO population is iterated to find the best ICPO individual. The individual coding vector corresponding to the optimal ICPO individual is decoded to obtain the optimal edge-cloud collaborative deployment scheme for the power system. The optimal edge-cloud collaborative deployment scheme includes the number of edge computing gateways, the location of edge computing gateways, the number of data acquisition devices, the location of data acquisition devices, and the communication connection relationship between data acquisition devices and edge computing gateways.

2. The power equipment defect identification method based on edge-cloud collaboration according to claim 1 is characterized in that: The method of using an artificial intelligence algorithm to construct a multimodal defect recognition model and a maintenance strategy generation model in a cloud data center specifically includes: Collecting some historical monitoring data, preprocessing the historical monitoring data, and obtaining the preprocessed historical monitoring data; Based on the preprocessed historical monitoring data, a multimodal fusion deep learning algorithm is used to build a multimodal defect recognition model and obtain several historical defect recognition results; According to the historical defect recognition results, a maintenance strategy generation model is constructed using a reinforcement learning algorithm, and several historical maintenance strategy generation experiences are obtained.

3. The power equipment defect identification method based on edge-cloud collaboration according to claim 2 is characterized in that: The specific method of constructing a multimodal defect recognition model based on the preprocessed historical monitoring data using a multimodal fusion deep learning algorithm includes: The preprocessed historical monitoring data are divided into a model training set and a model test set in a ratio of 7:3; An initial multimodal defect recognition model is constructed using a CNN-LSTM-Attention-Elman algorithm, wherein the multimodal defect recognition model includes an image feature extraction module constructed based on a CNN algorithm, an operation feature extraction module constructed based on an LSTM algorithm, an attention weight module constructed based on an Attention mechanism, and a defect recognition module constructed based on an Elman algorithm. The initial multimodal defect recognition model is optimized and trained using a model training set to obtain an optimized multimodal defect recognition model. The optimized multimodal defect recognition model is tested using the model test set to obtain the corresponding model accuracy; If the model accuracy is greater than the accuracy threshold, the optimal multimodal defect recognition model is output, otherwise, the optimization training continues.

4. The power equipment defect identification method based on edge-cloud collaboration according to claim 3 is characterized in that: The maintenance strategy generation model is constructed based on the DQN algorithm. The maintenance strategy generation model includes an intelligent agent, an experience replay pool, and a deep Q network. According to the historical defect recognition results, the specific method of using the reinforcement learning algorithm to construct the maintenance strategy generation model includes: Generate questions based on the maintenance strategy, set up the simulation environment of the DQN algorithm, and build the intelligent agent and experience replay pool; According to the states involved in the historical defect recognition results, the state space of the DQN algorithm is defined, and the input layer of the deep Q network is constructed according to the data structure of the historical defect recognition results; According to the actions involved in the maintenance strategy, define the action space of the DQN algorithm, and construct the output layer of the deep Q network according to the data structure of the maintenance strategy; According to the input layer and the output layer, several hidden layers are set to construct the corresponding deep Q network, connecting the input layer to the state space and the output layer to the action space; Define the reward function of the DQN algorithm based on the impact of each action in the action space; Based on the state space, action space and reward function, several historical defect recognition results are used to optimize the training of the deep Q network and the intelligent agent, build a maintenance strategy generation model, and generate several historical maintenance strategy generation experiences; Several historical maintenance strategy generation experiences are stored in the experience replay pool of the maintenance strategy generation model.

5. The power equipment defect identification method based on edge-cloud collaboration according to claim 3 or 4 is characterized in that: The monitoring data includes real-time image monitoring data and real-time operation monitoring data. The edge computing gateway receives the monitoring data sent by the data acquisition device, and preprocesses the monitoring data to obtain preprocessed data. The specific method of uploading the preprocessed data to the cloud data center includes: The edge computing gateway performs image preprocessing on the real-time image monitoring data to obtain the preprocessed real-time image monitoring data; The edge computing gateway performs data preprocessing on the real-time operation monitoring data to obtain the preprocessed real-time operation monitoring data; Integrate the preprocessed real-time image monitoring data and the preprocessed real-time operation monitoring data of the same power equipment to obtain preprocessed data; The preprocessed data is compressed and encrypted to obtain an encrypted data compression package, which is then uploaded to the cloud data center.

6. The power equipment defect identification method based on edge-cloud collaboration according to claim 5 is characterized in that: The cloud data center receives the pre-processed monitoring data, and uses the constructed multi-modal defect recognition model to perform power equipment defect recognition on the pre-processed data, and the specific method for obtaining the real-time defect recognition result includes: The cloud data center receives the pre-processed monitoring data, decrypts and decompresses the encrypted data compression package, and obtains the decrypted pre-processed monitoring data; The image feature extraction module extracts the real-time image features of the decrypted pre-processed real-time image monitoring data; The operation feature extraction module extracts the real-time operation features of the decrypted pre-processed real-time operation monitoring data; Using the attention weight value preset by the attention weight module, the real-time image features and the real-time operation features are weightedly spliced ​​to obtain the real-time weighted splicing features; The defect recognition module is used to perform defect recognition on the real-time weighted splicing features to obtain real-time defect recognition results.

7. The power equipment defect identification method based on edge-cloud collaboration according to claim 6 is characterized in that: The specific method of using the maintenance strategy generation model to process the real-time defect identification result to obtain the real-time maintenance strategy includes: The cloud data center updates the state space of the maintenance strategy generation model based on the real-time defect identification results to obtain an updated state space; Extracting a number of historical maintenance strategy generation experiences from the experience replay pool, and updating the action space of the maintenance strategy generation model based on the number of historical maintenance strategy generation experiences to obtain an updated action space; Inputting the updated state space into the input layer of the deep Q network of the maintenance strategy generation model, and connecting the updated action space to the output layer of the deep Q network to obtain an updated deep Q network; Based on the updated deep Q network, use the agent to control the updated deep Q network and output the Q values ​​of possible actions in the updated action space; Iteratively update the Q value according to the preset reward function to obtain an updated Q value until the number of iterations reaches the threshold; According to the greedy strategy, the possible action with the highest updated Q value in each iteration is used as the execution action of the corresponding state; Integrate the execution actions corresponding to all states in the updated state space to obtain the real-time maintenance strategy and return it to the corresponding edge computing gateway.

8. A power equipment defect identification system based on edge-cloud collaboration, used to implement the power equipment defect identification method based on edge-cloud collaboration as described in any one of claims 1 to 7, characterized in that: The system includes a cloud data center, several edge computing gateways and several data acquisition devices, wherein the data acquisition devices are arranged at the power equipment and are used to collect monitoring data of the power equipment in real time and send the monitoring data to the edge computing gateways in the monitoring area; The edge computing gateway is used to receive the monitoring data sent by the data acquisition device, pre-process the monitoring data to obtain the pre-processed monitoring data, and upload the pre-processed monitoring data to the cloud data center; The cloud data center is used to receive the pre-processed monitoring data sent by the edge computing gateway, use the constructed multi-modal defect recognition model to identify the defects of the power equipment on the pre-processed monitoring data, obtain real-time defect recognition results, use the maintenance strategy generation model to process the real-time defect recognition results, obtain the real-time maintenance strategy, and send the real-time maintenance strategy to the edge computing gateway; Before the step of real-time collection of monitoring data of power equipment by the data collection device, the method further includes: taking minimization of edge-cloud collaborative deployment cost as the optimization goal, using an intelligent optimization algorithm to generate an edge-cloud collaborative deployment plan for the power system, and using an artificial intelligence algorithm in a cloud data center to build a multimodal defect recognition model and a maintenance strategy generation model; the method of using an intelligent optimization algorithm to generate an edge-cloud collaborative deployment plan for the power system adopts an ICPO optimization algorithm to generate an edge-cloud collaborative deployment plan for the power system, specifically including: Taking minimizing the edge-cloud collaborative deployment cost as the optimization goal, the objective function of the ICPO optimization algorithm is set, and the ICPO individuals of the ICPO optimization algorithm are encoded to obtain the individual encoding format; According to the objective function, set the fitness function of the ICPO optimization algorithm, and set the ICPO population parameters and the maximum number of iterations of the ICPO optimization algorithm; According to the individual coding format and ICPO population parameters, the ICPO population is initialized to obtain an initial ICPO population including a number of initial ICPO individuals; According to the fitness function, ICPO population parameters and the maximum number of iterations, the initial ICPO population is iterated to find the best ICPO individual. The individual coding vector corresponding to the optimal ICPO individual is decoded to obtain the optimal edge-cloud collaborative deployment scheme for the power system. The optimal edge-cloud collaborative deployment scheme includes the number of edge computing gateways, the location of edge computing gateways, the number of data acquisition devices, the location of data acquisition devices, and the communication connection relationship between data acquisition devices and edge computing gateways.

Images