System operation and maintenance method and device based on deep learning, equipment and storage medium

Abstract

The invention relates to the technical field of power grid operation and maintenance, in particular to a system operation and maintenance method and device based on deep learning, equipment and a storage medium. A primary deep learning model is firstly constructed, network system characteristics and a deep learning algorithm are fused, and training is optimized by means of historical data and a regularization method, so that the model masters a power grid operation rule, over-fitting is prevented, and generalization and prediction capabilities are improved; state data are widely and completely obtained from multiple devices and multiple scenes through a network management protocol, the data quality is improved through data preprocessing cleaning, standardization and other operations, a deep power grid operation and maintenance prediction model predicts and evaluates the result confidence coefficient according to standard data, when the confidence coefficient is low, an adjustment instruction and a maintenance task are generated, and potential problems are early warned and processed in advance. And efficient, safe and stable operation of a power grid is powerfully ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid operation and maintenance, and particularly to a system operation and maintenance method, device, equipment and storage medium based on deep learning. Background Art

[0002] In the technical field of power grid operation and maintenance, traditional methods have many limitations in power grid state prediction and fault handling. In the past, the monitoring and analysis of power grid operation states mostly relied on simple statistical models and manual experience, making it difficult to handle the increasingly complex operation data and system structures of the power grid. As the scale of the power grid continues to expand, the number of devices increases, and the complexity of the network system rises sharply. Traditional means cannot effectively learn the complex operation rules of the power grid, and both the prediction accuracy and timeliness are insufficient. At the same time, the collection and processing of data lack systematicness, often being limited to single sources and simple forms, resulting in incomplete information and poor data quality, making it difficult to support refined analysis and decision-making. Moreover, in the face of possible power grid operation anomalies, it is impossible to accurately judge potential problems, lacking an effective early warning mechanism and automated response solutions, and it is difficult to adjust controller parameters or arrange maintenance tasks in advance, bringing great challenges to the stable operation of the power grid. In view of this, this solution emerges as the times require, aiming to improve the overall efficiency of power grid operation and maintenance by introducing deep learning technology, various network management protocols, and advanced data preprocessing and optimization means. Summary of the Invention

[0003] To solve the above-mentioned drawbacks in the prior art, the present invention proposes a system operation and maintenance method, system, equipment and storage medium based on deep learning.

[0004] To solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0005] A system operation and maintenance method based on deep learning, comprising the steps of:

[0006] Construct a primary deep learning model based on the characteristics of the preset network system and deep learning algorithms; obtain historical power grid operation and maintenance data in the preset database; perform iterative optimization training on the primary deep learning model according to the historical power grid operation and maintenance data and regularization methods to obtain a deep power grid operation and maintenance prediction model; collect the status data of network devices according to the preset network management protocol; preprocess the status data according to the data preprocessing method to obtain standard status data; perform system status analysis on the standard status data based on the deep power grid operation and maintenance prediction model to obtain a power grid status prediction result; obtain the confidence level in the power grid status prediction result and determine whether the confidence level is lower than the preset confidence threshold; when the confidence level is lower than the confidence threshold, generate an adjustment instruction for adjusting the controller parameters and a power grid maintenance task according to the power grid status prediction result. By constructing a primary deep learning model, the characteristics of the network system and deep learning algorithms are combined. On this basis, using historical data and regularization methods for optimization training not only enables the model to learn the complex operation rules of the power grid, but also avoids overfitting of the model, enhancing the generalization ability and prediction accuracy of the model; the application of the network management protocol ensures that the source of status data is extensive and complete, obtaining accurate information from multiple devices and multiple scenarios; the data preprocessing method's cleaning, standardization, transformation, and encoding operations on the status data improve the data quality and prepare for subsequent analysis; the deep power grid operation and maintenance prediction model can obtain a prediction result based on the standard status data, and at the same time, the confidence level evaluation of the result can accurately judge its reliability. Importantly, when the confidence level is low, the corresponding adjustment instruction and maintenance task generation mechanism can early warn and handle potential problems of the power grid, effectively avoiding failures or reducing the impact of failures, enhancing the stability and reliability of the power grid, and providing a strong guarantee for the efficient and safe operation of the power grid system.

[0007] Furthermore, iteratively optimizing and training the primary deep learning model based on historical power grid operation and maintenance data and the regularization method to obtain a deep power grid operation and maintenance prediction model includes: iteratively optimizing and training the primary deep learning model based on historical power grid operation and maintenance data and the regularization method to obtain an improved deep learning model; calculating the loss of the secondary deep learning model according to a preset cross-entropy loss function to obtain the cross-entropy loss; using the preset Adam optimization algorithm and the cross-entropy loss to adjust the model parameters of the secondary deep learning model to obtain a deep power grid operation and maintenance prediction model. Using historical power grid operation and maintenance data to iteratively optimize and train the primary deep learning model, combined with the regularization method, helps prevent the model from overfitting, enables the model to learn the complex patterns and potential laws of power grid operation from a large amount of historical information, and improves the model performance to obtain a secondary deep learning model; then, using the cross-entropy loss function to calculate the loss can effectively measure the difference between the model prediction result and the actual situation, especially suitable for classification tasks, and promotes the model to have a more accurate judgment on multi-class problems; finally, using the Adam optimization algorithm to adjust the parameters of the secondary deep learning model can adaptively optimize and adjust the model parameters according to the training status, make the model more reasonable, further improve the model performance, and finally obtain a deep power grid operation and maintenance prediction model, which can achieve accurate prediction of power grid operation and maintenance, better assist power grid management personnel to discover potential problems in advance, formulate reasonable maintenance strategies, improve the stability and reliability of power grid operation, optimize the power grid resource allocation at the same time, and provide strong technical support for the safe, efficient and stable operation of the power grid system.

[0008] Furthermore, iteratively optimizing and training the primary deep learning model according to historical power grid operation and maintenance data and the regularization method to obtain an improved deep learning model includes: allocating weights to the parameters of the primary deep learning model according to the regularization method to obtain a secondary deep learning model; obtaining task requirements from the database, and configuring the network structure of the secondary deep learning model according to the network system characteristics and task requirements to obtain a tertiary deep learning model; dividing the historical power grid operation and maintenance data into data sets according to a preset first ratio to obtain a sample set; using the chi-square test statistical method to perform feature selection on the sample set to obtain a training set; iteratively training the tertiary deep learning model according to the training set and the regularization method to obtain a quaternary deep learning model; dividing the historical power grid operation and maintenance data into data sets according to a preset second ratio to obtain a test set; evaluating the quaternary deep learning model based on the cross-validation method and the test set to obtain an evaluation result; obtaining the regularization strength parameter of the quaternary deep learning model; and iteratively optimizing the quaternary deep learning model according to the evaluation result and the regularization strength parameter to obtain an improved deep learning model. Using the regularization method to allocate weights to the parameters of the primary deep learning model to obtain a secondary deep learning model is crucial for processing complex power grid operation and maintenance data, effectively avoiding the overfitting problem, greatly enhancing the generalization ability of the model, and enabling it to perform well in different power grid operation and maintenance scenarios; then, configuring the network structure of the secondary model according to the task requirements and network system characteristics, the obtained tertiary deep learning model has strong task pertinence, can fully fit the actual application scenario, improves the adaptability to specific tasks, gives full play to the characteristics of the network system, and lays a solid foundation for subsequent work; by dividing the historical power grid operation and maintenance data to obtain a sample set and using the chi-square test for feature selection to form a training set, redundant information is removed, significantly improving the model training efficiency and the attention to key information, enabling the tertiary deep learning model to better capture data patterns during the training process, and obtaining a quaternary deep learning model through iterative training, ensuring its reliability in tasks such as power grid operation and maintenance prediction; providing an effective test set for evaluation, combined with cross-validation evaluation, making the performance evaluation of the quaternary deep learning model more comprehensive and objective, and avoiding the bias brought by a single test set; finally, by obtaining the regularization strength parameter and combining the evaluation result to iteratively optimize the quaternary model, fine-tuning is achieved, and an improved deep learning model is obtained. This improved model has greatly improved in terms of prediction accuracy and stability, provides high-quality decision support for power grid operation and maintenance, can effectively improve the operation and management efficiency of the power grid system, enhance the reliability and security of the power grid, can detect potential problems in advance, promote the development of power grid operation and maintenance towards a more intelligent and efficient direction, ensure the stable and reliable operation of the power grid system, and provide strong technical support for the intelligent upgrade of power grid operation and maintenance.

[0009] Further, the system state analysis of the standard state data based on the deep grid operation and maintenance prediction model to obtain the grid state prediction result includes: calculating the order of a preset autoregressive model using a preset information criterion to obtain a set of model orders; obtaining the minimum value from the set of model orders and setting the minimum value as the model order of the autoregressive model; converting the autoregressive model into a matrix regression model according to the model order and the least squares method; performing short-term fluctuation smoothing processing on the standard state data using a smoothing method to obtain smoothed state data; converting the smoothed state data into a feature matrix based on the matrix regression model; and performing system state analysis based on the deep grid operation and maintenance prediction model and the feature matrix to obtain the grid state prediction result. Determining the model order of the autoregressive model through the information criterion balances the goodness of fit and complexity of the autoregressive model, enhancing the performance of the autoregressive model; then, using the smoothing method to process the standard state data can effectively eliminate the short-term fluctuations of the data, highlight the long-term trend of the data, and improve the data quality; furthermore, converting the smoothed state data into a feature matrix through the matrix regression model formed by converting the autoregressive model and using the feature matrix as the input of the deep grid operation and maintenance prediction model can provide a more structured and regular data form for the deep grid operation and maintenance prediction model to mine the information in the data, thereby better supporting the deep grid operation and maintenance prediction model to perform complex non-linear fitting, and further improving the prediction performance of the entire system for the grid state and ensuring the intelligent decision-making and management of grid operation and maintenance.

[0010] Further, the preprocessing of the state data according to the data preprocessing method to obtain the standard state data includes: performing denoising processing on the state data using a statistical method and a preset deviation from the normal range to obtain denoised state data; obtaining the time series data in the denoised state data and filling in its missing values according to the linear interpolation method to obtain the standard time series data, and the expression is as follows:

[0011] where x t is the missing value to be filled, t is the timestamp where the missing value is located, x t-1 and x t+1 are adjacent time series data, t -1 and t +1are the corresponding timestamps of two adjacent time series data; the operation and maintenance status data and status feature data are selected from the denoised status data by using a feature selection method and a preset system operation and maintenance task; the standard time series data, operation and maintenance status data, and status feature data are unified in magnitude according to a preset normalization formula to obtain standard status data. Denoising the status data by using a statistical method can effectively remove noise, ensure data quality, and reduce interference caused by measurement errors, etc. Then, linear interpolation is used to fill in the missing values of the time series data, and complete standard time series data can be obtained, ensuring the continuity of the data and facilitating subsequent analysis; then, a feature selection method is used to screen the operation and maintenance status and status feature data, which can reduce the dimension and improve the model training efficiency. Finally, normalization is used to unify the data magnitude, making the data on the same scale, enabling the model to process features equally, enhancing the training and prediction effects, providing high-quality, continuous, concise, and normalized data for the operation and maintenance and status prediction of the power grid system, and ensuring the stable operation and efficient decision-making of the system.

[0012] Further, when the confidence level is lower than the confidence threshold, generating an adjustment instruction for adjusting the controller parameters and a power grid maintenance task according to the power grid status prediction result includes: when the confidence level is lower than the confidence threshold, obtaining the power grid topology structure and equipment connection relationship from the database; analyzing the power grid status prediction result according to the power grid topology structure, equipment connection relationship, and standard status data to obtain the actual operating status of the power grid; obtaining power system analysis knowledge and operation and maintenance experience from the database; generating the working parameters that need to be adjusted by the controller according to the power system analysis knowledge, operation and maintenance experience, and actual operating status of the power grid; obtaining the communication protocol and instruction format of the controller, and generating an adjustment instruction for the controller parameters according to the working parameters, communication protocol, and instruction format; generating a power grid maintenance task according to a preset fault diagnosis model and the adjustment instruction for the controller parameters. Obtaining information such as the power grid topology structure from the database and accurately analyzing the actual operating status of the power grid in combination with the standard status data provides a basis for subsequent operations; then, using power system analysis knowledge and operation and maintenance experience to scientifically generate the working parameters of the controller, optimizing the power grid performance, and enhancing stability and security; at the same time, generating a highly targeted and compatible adjustment instruction for the controller parameters according to the communication protocol and instruction format to ensure the accurate and reliable operation of the controller; finally, generating a maintenance task based on the fault diagnosis model and adjustment instruction can discover problems in advance, reasonably arrange maintenance work, improve the maintenance efficiency and effect, reduce fault losses, provide comprehensive guarantee for the stable operation of the power grid, and promote the intelligent operation and maintenance management of the power grid.

[0013] Further, generating a power grid maintenance task according to a preset fault diagnosis model and a controller parameter adjustment instruction includes: obtaining a fault log from a database; establishing a mapping relationship between a power grid state prediction result and the fault log through the fault diagnosis model; performing a simulated power grid test according to the mapping relationship, the controller parameter adjustment instruction, and the fault diagnosis model to obtain a simulated power grid performance result; analyzing the result of the simulated power grid performance result; when the simulated power grid performance result is that the power grid dynamic performance value is greater than or equal to the third-level value, generating a power grid stable operation guarantee task according to the power grid state prediction result and the actual operation state of the power grid; when the simulated power grid performance result is that the power grid dynamic performance value is less than the third-level value, generating a power grid dynamic performance optimization task according to the power grid state prediction result and the actual operation state of the power grid; generating a power grid maintenance task according to the power grid stable operation guarantee task and the power grid dynamic performance optimization task. By obtaining the fault log from the database and establishing the mapping relationship, it helps to insight into the power grid fault mode and provide data support for subsequent operations; using the mapping relationship, the controller parameter adjustment instruction, and the fault diagnosis model to perform the simulated power grid test can evaluate the performance in advance and avoid the risks of actual tests; the analysis of the simulation results can accurately evaluate the power grid performance, generate targeted stable operation guarantee or dynamic performance optimization tasks according to different performance performances, reflect the differential task generation mechanism, ensure the stability of the power grid and the optimization of performance; finally, combining the two tasks to generate the power grid maintenance task makes the maintenance work more systematic and scientific, can respond to various situations in advance, reduce the possibility of faults, improve the maintenance efficiency and effect, ensure the stable, efficient, and reliable operation of the power grid system, and ensure the continuous safety of power supply.

[0014] Furthermore, a system operation and maintenance device based on deep learning includes: the model construction module, which is used to construct a primary deep learning model according to the preset characteristics of the network system and deep learning algorithms; the operation and maintenance data acquisition module, which is used to acquire historical power grid operation and maintenance data in a preset database; the model training module, which is used to perform iterative optimization training on the primary deep learning model according to the historical power grid operation and maintenance data and regularization methods to obtain a deep power grid operation and maintenance prediction model; the status data collection module, which is used to collect the status data of network devices according to a preset network management protocol; the status data processing module, which is used to preprocess the status data according to a data preprocessing method to obtain standard status data; the power grid status prediction module, which is used to perform system status analysis on the standard status data based on the deep power grid operation and maintenance prediction model to obtain a power grid status prediction result; the confidence threshold judgment module, which is used to obtain the confidence in the power grid status prediction result and judge whether the confidence is lower than a preset confidence threshold; the instruction and task generation module, which is used to generate an adjustment instruction for adjusting the controller parameters and a power grid maintenance task according to the power grid status prediction result when the confidence is lower than the confidence threshold. By using the characteristics of the network system and deep learning algorithms through the model construction module to construct a primary deep learning model, and through the model training module to combine historical power grid operation and maintenance data and regularization training to obtain a deep power grid operation and maintenance prediction model, the model can learn the operation rules of the power grid, avoid overfitting, and enhance the generalization and prediction accuracy of the model; the network management protocol ensures the wide and complete status data, and data preprocessing is used to improve the data quality. The deep power grid operation and maintenance prediction model can analyze, predict and evaluate the confidence of the standard status data. When the confidence is low, the instruction and task generation module will generate an adjustment instruction and a maintenance task to pre-warn and handle potential problems in advance, avoid or reduce the impact of faults, enhance the stability and reliability of the power grid. At the same time, the status data collection module ensures the data source, providing a strong guarantee for the efficient and safe operation of the power grid system.

[0015] Furthermore, a system operation and maintenance device based on deep learning, the system operation and maintenance device based on deep learning includes: a memory and at least one processor, instructions are stored in the memory; at least one of the processors calls the instructions in the memory so that the system operation and maintenance device based on deep learning executes each step of the system operation and maintenance method based on deep learning as described above.

[0016] Furthermore, a computer-readable storage medium, instructions are stored on the computer-readable storage medium, and when the instructions are executed by a processor, each step of the system operation and maintenance method based on deep learning as described above is implemented.

[0017] The beneficial effects of the system operation and maintenance method based on deep learning of the present invention are:

[0018] By constructing a primary deep learning model, the characteristics of the network system are combined with deep learning algorithms. On this basis, historical data and regularization methods are used to optimize the training, which not only enables the model to learn the complex operation laws of the power grid, but also avoids overfitting of the model, enhances the generalization ability and prediction accuracy of the model; the application of network management protocols ensures that the source of status data is extensive and complete, and accurate information is obtained from multiple devices and multiple scenarios; data preprocessing methods clean, standardize, transform and encode the status data, improving the data quality and preparing for subsequent analysis; the deep power grid operation and maintenance prediction model can obtain prediction results based on standard status data, and at the same time evaluate the confidence of the results, which can accurately judge its reliability. Importantly, when the confidence is low, the corresponding adjustment instructions and maintenance task generation mechanism can give early warnings and handle potential problems of the power grid, effectively avoiding failures or reducing the impact of failures, enhancing the stability and reliability of the power grid, and providing strong guarantee for the efficient and safe operation of the power grid system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of the embodiments in conjunction with the accompanying drawings, wherein:

[0020] Figure 1 FIG. 9 is the first flow chart of the system operation and maintenance method based on deep learning provided by the embodiment of the present invention;

[0021] Figure 2 FIG. 13 is the second flow chart of the system operation and maintenance method based on deep learning provided by the embodiment of the present invention;

[0022] Figure 3 FIG. 17 is the third flow chart of the system operation and maintenance method based on deep learning provided by the embodiment of the present invention;

[0023] Figure 4 FIG. 21 is the fourth flow chart of the system operation and maintenance method based on deep learning provided by the embodiment of the present invention;

[0024] Figure 5 FIG. 25 is the fifth flow chart of the system operation and maintenance method based on deep learning provided by the embodiment of the present invention;

[0025] Figure 6 FIG. 29 is the sixth flow chart of the system operation and maintenance method based on deep learning provided by the embodiment of the present invention;

[0026] Figure 7 FIG. 33 is the seventh flow chart of the system operation and maintenance method based on deep learning provided by the embodiment of the present invention;

[0027] Figure 8 FIG. 37 is the structural schematic diagram of the system operation and maintenance device based on deep learning provided by the embodiment of the present invention;

[0028] Figure 9 The structural schematic diagram of the system operation and maintenance device based on deep learning provided by the embodiment of the present invention. Specific implementation manners

[0029] The technical solutions of the present invention will be described clearly and completely below with reference to the accompanying drawings. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0030] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order different from that illustrated or described herein. In addition, the term "comprising" or "having" and any variation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0031] For ease of understanding, the specific process of the embodiment of the present invention will be described below. Please refer to Figure 1 One embodiment of the system operation and maintenance method based on deep learning in the embodiment of the present invention includes:

[0032] 101. Construct a primary deep learning model according to the preset network system characteristics and deep learning algorithms;

[0033] 102. Obtain the historical power grid operation and maintenance data in the preset database;

[0034] 103. Iteratively optimize and train the primary deep learning model according to the historical power grid operation and maintenance data and the regularization method to obtain a deep power grid operation and maintenance prediction model;

[0035] In this embodiment, the regularization methods adopted include Lasso regularization. In the power grid operation and maintenance prediction model, using Lasso regularization can make some parameters of the model become 0, thus achieving the effect of feature selection, removing some features that contribute little to the power grid operation and maintenance prediction, simplifying the model structure, and preventing the model from overfitting. The regularization method adopted can also be ridge regression regularization. Ridge regression regularization will make the values of the model parameters smaller, but will not make the parameters become 0. It can reduce the weights of the parameters, reduce the dependence of the model on certain features, make the model smoother, and improve the generalization ability of the model. In the power grid operation and maintenance prediction, it can make the model more balanced in its dependence on different power grid operation characteristics and avoid the model from over-relying on certain specific features.

[0036] Based on a large amount of historical power grid operation and maintenance data, the model is trained to enable the model to learn the complex patterns and rules of power grid operation. The regularization method prevents the model from overfitting, ensures the stable performance of the model on different data, and thus improves the accuracy of power grid state prediction.

[0037] 104. Collect the status data of network devices according to the preset network management protocol;

[0038] In this embodiment, the network management protocols adopted include: Simple Network Management Protocol (SNMP), Common Management Information Protocol (CMIP), Remote Monitoring Protocol (RMON), Network Configuration Protocol (NETCONF), and Link Layer Discovery Protocol (LLDP). Collecting the status data of network devices according to the preset network management protocol ensures that comprehensive and accurate information reflecting the power grid operation status can be obtained. Different network management protocols are applicable to different types of devices and data collection scenarios, ensuring the diversity and integrity of data sources.

[0039] 105. Preprocess the status data according to the data preprocessing method to obtain the standard status data;

[0040] In this embodiment, the data preprocessing methods adopted include: data cleaning method, data standardization and normalization (Z-score standardization, min-max normalization, etc.), data transformation methods (logarithmic transformation, Box-Cox transformation), and feature encoding.

[0041] Processing the status data according to the data preprocessing method to obtain the standard status data unifies the data format, eliminates data noise and outliers, makes the data more suitable for model analysis, improves the data quality, and provides reliable data support for subsequent status analysis and prediction.

[0042] 106. Perform system status analysis on the standard status data based on the deep power grid operation and maintenance prediction model to obtain the power grid state prediction result;

[0043] 107. Obtain the confidence level in the power grid status prediction result, and determine whether the confidence level is lower than a preset confidence threshold;

[0044] 108. When the confidence level is lower than the confidence threshold, generate an adjustment instruction for adjusting the controller parameters and a power grid maintenance task according to the power grid status prediction result.

[0045] In this embodiment, when the confidence level is lower than the preset threshold, it means that the reliability of the prediction result needs attention. At this time, generating an instruction for adjusting the controller parameters and a power grid maintenance task according to the prediction result can discover potential problems in the power grid operation in advance, and take corresponding measures for prevention and handling, avoid the occurrence of faults or reduce the impact of faults, and enhance the stability and reliability of the power grid operation;

[0046] In this embodiment, by constructing a primary deep learning model, the characteristics of the network system and the deep learning algorithm are combined. On this basis, using historical data and regularization methods for optimization training not only enables the model to learn the complex operation rules of the power grid, but also avoids overfitting of the model, enhancing the generalization ability and prediction accuracy of the model; the application of the network management protocol ensures that the source of status data is extensive and complete, and accurate information is obtained from multiple devices and multiple scenarios; the data preprocessing method performs cleaning, standardization, transformation, and encoding operations on the status data, improving the data quality and preparing for subsequent analysis; the deep power grid operation and maintenance prediction model can obtain prediction results based on standard status data, and at the same time evaluate the confidence level of the results, accurately judging its reliability. Importantly, when the confidence level is low, the corresponding adjustment instruction and maintenance task generation mechanism can give early warnings and handle potential problems in the power grid, effectively avoiding faults or reducing the impact of faults, enhancing the stability and reliability of the power grid, and providing a strong guarantee for the efficient and safe operation of the power grid system.

[0047] Please refer to Figure 2 , the second embodiment of the system operation and maintenance method based on deep learning in the embodiment of the present invention includes:

[0048] 201. Iteratively optimize and train the primary deep learning model according to historical power grid operation and maintenance data and regularization methods to obtain an improved deep learning model;

[0049] In this embodiment, the historical power grid operation and maintenance data includes: power grid topology data, equipment maintenance data, equipment failure data, equipment status data, power grid operation scheduling data, etc.;

[0050] 202. Calculate the loss of the secondary deep learning model according to a preset cross-entropy loss function to obtain the cross-entropy loss;

[0051] 203. Adjust the model parameters of the secondary deep learning model by using a preset Adam optimization algorithm and cross-entropy loss to obtain a deep power grid operation and maintenance prediction model.

[0052] In this embodiment, the model parameter adjustment includes adjusting the learning rate, data batch size and other hyperparameters using the Adam optimization algorithm; for example, if the model converges too slowly, the learning rate can be appropriately increased; if the model oscillates, the learning rate can be reduced; for another example, generally speaking, a smaller data batch size may lead to unstable training, but a larger data batch size may require more memory. Common batch sizes are 32, 64, 128, etc., and the data batch size can be adjusted according to hardware resources and data set size; the adjustment of hyperparameters can also be achieved through grid search, random search or more advanced hyperparameter optimization algorithms (such as Bayesian optimization); in addition to adjusting hyperparameters, the model parameter adjustment also includes adding the regularization term in the iterative optimization training process and the cross entropy loss calculated by the cross entropy loss function to the cross entropy loss function to modify the cross entropy loss function;

[0053] In this embodiment, historical power grid operation and maintenance data is used to iteratively optimize and train the primary deep learning model. Combined with the regularization method, it helps to prevent model overfitting, so that the model can learn the complex patterns and potential laws of power grid operation from a large amount of historical information, and improve the model performance to obtain a secondary deep learning model; then, the cross entropy loss function is used for loss calculation, which can help effectively measure the difference between the model prediction results and the actual situation, which is particularly suitable for classification tasks, and enables the model to have more accurate judgments on multi-category problems; finally, the Adam optimization algorithm is used to adjust the parameters of the secondary deep learning model, and the model parameters can be adaptively optimized and adjusted according to the training state to make the model more reasonable and further improve the model performance. The deep power grid operation and maintenance prediction model finally obtained can realize accurate prediction of power grid operation and maintenance, and can better assist power grid managers to discover potential problems in advance, formulate reasonable maintenance strategies, improve the stability and reliability of power grid operation, and optimize the allocation of power grid resources at the same time, providing strong technical support for the safe, efficient and stable operation of the power grid system.

[0054] See also Figure 3 , a third embodiment of the system operation and maintenance method based on deep learning in the embodiment of the present invention includes:

[0055] 301. Adjust the weights of the parameters of the primary deep learning model according to the regularization method to obtain a secondary deep learning model;

[0056] In this embodiment, the parameters of the primary deep learning model are weighted by regularization method to obtain the secondary deep learning model, which helps to prevent the model from overfitting, so that when the model processes complex power grid operation and maintenance data, it can avoid overfitting of the training data while learning data features, thereby enhancing the generalization ability of the model;

[0057] 302. Obtain task requirements from the database, and configure the network structure of the secondary deep learning model according to the characteristics of the network system and the task requirements to obtain a tertiary deep learning model;

[0058] In this embodiment, configuring the network structure of the model according to the characteristics of the network system and the task requirements to obtain a tertiary deep learning model can make the model structure more suitable for the actual application scenario, improve the adaptability of the model to specific tasks, ensure that the model can make full use of the characteristics of the network system, and specifically complete the task;

[0059] 303. Divide the historical power grid operation and maintenance data according to a preset first ratio to obtain a sample set;

[0060] 304. Use the chi-square test statistical method to perform feature selection on the sample set to obtain a training set;

[0061] In this embodiment, using the chi-square test statistical method to perform feature selection on the sample set to obtain a training set can screen out the features most relevant to the target model, reduce redundant information, improve the model training efficiency, and at the same time enhance the model's attention to key information, further improving the model performance and prediction accuracy;

[0062] 305. Iteratively train the tertiary deep learning model according to the training set and the regularization method to obtain a quaternary deep learning model;

[0063] In this embodiment, using the training set and the regularization method to iteratively train the tertiary deep learning model to obtain a quaternary deep learning model enables the model to improve its performance in continuous learning and optimization, better capture the potential laws and patterns in the data, and ensure the reliability of the model in tasks such as power grid operation and maintenance prediction;

[0064] 306. Divide the historical power grid operation and maintenance data according to a preset second ratio to obtain a test set;

[0065] In this embodiment, dividing the historical power grid operation and maintenance data according to a preset ratio can reasonably allocate data resources, ensure the effectiveness and representativeness of the training set and the test set, and provide a high-quality data basis for model training and evaluation;

[0066] 307. Evaluate the quaternary deep learning model based on the cross-validation method and the test set to obtain an evaluation result;

[0067] In this embodiment, evaluating the quaternary deep learning model based on the cross-validation method and the test set can more comprehensively and objectively evaluate the model performance, and the obtained evaluation result can accurately reflect the performance of the model on different data subsets, avoiding the evaluation bias that may be brought by a single test set;

[0068] 308. Obtain the regularization intensity parameter of the four - level deep learning model;

[0069] 309. Iteratively optimize the four - level deep learning model according to the evaluation result and the regularization intensity parameter to obtain an improved deep learning model.

[0070] In this embodiment, obtaining the regularization intensity parameter and combining the evaluation result to iteratively optimize the four - level deep learning model to obtain an improved deep learning model realizes the fine - tuning of the model, optimizes the model performance according to the evaluation feedback, ensures that the finally obtained improved deep learning model has higher prediction accuracy and stability, provides better decision - making support for tasks such as power grid operation and maintenance, improves the operation and management efficiency of the power grid system, enhances the reliability and security of the power grid, and at the same time helps to detect potential problems in advance, realizing more intelligent and effective power grid operation and management;

[0071] Using the regularization method to allocate the weights of the parameters of the primary deep learning model to obtain the secondary deep learning model is crucial for processing complex power grid operation and maintenance data, effectively avoiding the over - fitting problem, greatly enhancing the generalization ability of the model, and enabling it to perform well in different power grid operation and maintenance scenarios; then, configuring the network structure of the secondary model according to the task requirements and the characteristics of the network system, the obtained three - level deep learning model has strong task pertinence, can fully fit the actual application scenario, improves the adaptability to specific tasks, and gives full play to the characteristics of the network system, laying a solid foundation for the subsequent work; by dividing the historical power grid operation and maintenance data to obtain a sample set and using the chi - square test for feature selection to form a training set, redundant information is removed, significantly improving the model training efficiency and the attention to key information, enabling the three - level deep learning model to better capture the data law during the training process, and obtaining the four - level deep learning model through iterative training, ensuring its reliability in tasks such as power grid operation and maintenance prediction; providing an effective test set for evaluation, combined with cross - validation evaluation, making the performance evaluation of the four - level deep learning model more comprehensive and objective, and avoiding the deviation brought by a single test set; finally, by obtaining the regularization intensity parameter and combining the evaluation result to iteratively optimize the four - level model, fine - tuning is realized, and an improved deep learning model is obtained. This improved model has been greatly improved in terms of prediction accuracy and stability, provides high - quality decision - making support for power grid operation and maintenance, can effectively improve the operation and management efficiency of the power grid system, enhance the reliability and security of the power grid, can detect potential problems in advance, promote the development of power grid operation and maintenance towards a more intelligent and efficient direction, ensure the stable and reliable operation of the power grid system, and provide strong technical support for the intelligent upgrade of power grid operation and maintenance.

[0072] Please refer to Figure 4 , the fourth embodiment of the system operation and maintenance method based on deep learning in the embodiment of the present invention includes:

[0073] 401. Calculate the order of a preset autoregressive model using a preset information criterion to obtain a set of model orders;

[0074] 402. Obtain the minimum value from the set of model orders and set the minimum value as the model order of the autoregressive model;

[0075] In this embodiment, the autoregressive model achieves a good balance between goodness of fit and complexity at the model order of the minimum value; for example, if the minimum value in the calculation results of information criterion AIC or information criterion BIC is obtained at order three, then the model order of the final autoregressive model is set to three;

[0076] 403. Convert the autoregressive model into a matrix regression model according to the model order and the least squares method;

[0077] 404. Use a smoothing method to smooth the short-term fluctuations of the standard state data to obtain smoothed state data;

[0078] In this embodiment, common smoothing methods include the moving average method, the exponential smoothing method, etc.; the calculation formula involved when using the exponential smoothing method to calculate the smoothed state data is as follows:

[0079] S t =αy t +(1 - α)S t-1 , where represents the smoothed state value of the standard state data at time t, α represents the smoothing coefficient, and its value range is 0 < α < 1, S t-1 represents the smoothed value of the standard state data at the previous moment, and y t represents the smoothed state data;

[0080] 405. Convert the smoothed state data into a feature matrix based on the matrix regression model;

[0081] In this embodiment, the smoothed data is constructed into a corresponding feature matrix according to the requirements of the matrix regression model; for example, for a p-order matrix regression model, the smoothed state data is constructed into a feature matrix X in the form of y t for subsequent prediction analysis;

[0082] 406. Perform system state analysis based on the deep grid operation and maintenance prediction model and the feature matrix to obtain the grid state prediction result.

[0083] In this embodiment, the feature matrix is input into the deep grid operation and maintenance prediction model, and the powerful non-linear fitting ability of the deep grid operation and maintenance prediction model is used to analyze and predict the system state of the grid, and the final grid state prediction result is obtained through the forward propagation of the model.

[0084] In this embodiment, the model order of the autoregressive model is determined by the information criterion, so that the autoregressive model achieves a balance between the goodness of fit and complexity, enhancing the performance of the autoregressive model; then, the standard state data is processed by a smoothing method, which can effectively eliminate the short-term fluctuations of the data, highlight the long-term trend of the data, and improve the data quality; furthermore, the smoothed state data is transformed into a feature matrix through the matrix regression model formed by the transformation of the autoregressive model, and the feature matrix is used as the input of the deep grid operation and maintenance prediction model, which can provide a more structured and regular data form for the deep grid operation and maintenance prediction model to mine the information in the data, thereby better supporting the deep grid operation and maintenance prediction model to perform complex non-linear fitting, and further improving the prediction performance of the entire system for the grid state, and ensuring the intelligent decision-making and management of grid operation and maintenance.

[0085] Please refer to Figure 5 , the fifth embodiment of the system operation and maintenance method based on deep learning in the embodiments of the present invention includes:

[0086] 501. Use a statistical method and a preset deviation from the normal range to denoise the state data to obtain denoised state data;

[0087] In this embodiment, the statistical method can be mean filtering, median filtering, Z-score method, etc. These methods use the statistical characteristics of the data to remove noise; the deviation from the normal range usually determines a normal range according to prior knowledge, and data points outside this range may be considered noise; for example, for the voltage data measured by a sensor, according to long-term statistical experience, the normal voltage range is between [V min , V max . If the measured value at a certain moment exceeds this range, it may be noise caused by measurement error or abnormal interference and will be identified and processed;

[0088] 502. Obtain the time series data in the denoised state data and fill in its missing values according to the linear interpolation method to obtain standard time series data, and the expression is as follows:

[0089] In the formula, x t is the missing value to be filled, t is the time stamp where the missing value is located, x t-1 and x t+1 are adjacent time series data, and t -1 and t +1 are the corresponding time stamps corresponding to two adjacent time series data;

[0090] 503. Use a feature selection method and a preset system operation and maintenance task to select operation and maintenance state data and state feature data from the denoised state data;

[0091] In this embodiment, a correlation-based feature selection method or a tree-based feature importance evaluation method (such as the feature importance of a random forest) can be used to select the data that is most relevant and important to the system operation and maintenance tasks, so as to obtain the operation and maintenance status data and status feature data, reduce the data dimension, and improve the model training efficiency;

[0092] 504. Unify the magnitudes of the standard time series data, operation and maintenance status data, and status feature data according to a preset normalization formula to obtain standard status data;

[0093] In this embodiment, a common minimum-maximum normalization formula can be used to unify the magnitudes of the data, ensure that the data is processed on the same scale, enable the model to treat each feature equally, and improve the training and prediction effects of the model;

[0094] Using statistical methods to denoise the status data can effectively remove noise, ensure data quality, and reduce interference caused by measurement errors, etc. Then, linear interpolation is used to fill in the missing values of the time series data, and complete standard time series data can be obtained, ensuring the continuity of the data and facilitating subsequent analysis. Then, feature selection methods are used to screen the operation and maintenance status and status feature data, which can reduce the dimension and improve the model training efficiency. Finally, normalization is used to unify the data magnitudes, making the data on the same scale, enabling the model to process features equally, enhancing the training and prediction effects, providing high-quality, continuous, concise, and normalized data for the operation and maintenance and status prediction of the power grid system, and ensuring the stable operation and efficient decision-making of the system.

[0095] Please refer to Figure 6 , the sixth embodiment of the system operation and maintenance method based on deep learning in the embodiments of the present invention includes:

[0096] 601. When the confidence level is lower than the confidence threshold, obtain the power grid topology structure and device connection relationship from the database;

[0097] 602. Analyze the power grid status prediction result according to the power grid topology structure, device connection relationship, and standard status data to obtain the actual operation status of the power grid;

[0098] In this embodiment, analyzing the power grid status prediction result in combination with the power grid topology structure, device connection relationship, and standard status data helps to accurately grasp the actual operation status of the power grid, which can provide a reliable basis for subsequent operations;

[0099] 603. Obtain the power system analysis knowledge and operation and maintenance experience from the database;

[0100] In this embodiment, the power system analysis knowledge includes power flow analysis, short-circuit analysis, stability analysis, and power system planning, and the operation and maintenance experience includes equipment inspection, fault troubleshooting and handling, equipment maintenance and upkeep, safety management, and operation data monitoring and analysis;

[0101] 604. Generate the working parameters that the controller needs to adjust according to the power system analysis knowledge, operation and maintenance experience, and the actual operation status of the power grid;

[0102] In this embodiment, by introducing the power system analysis knowledge and operation and maintenance experience and combining with the actual operation status of the power grid, the working parameters required by the controller can be generated scientifically and reasonably. This knowledge- and experience-based method ensures the scientific nature of parameter adjustment, makes the working parameters of the controller more in line with the actual needs of the power grid, avoids blind adjustment, thereby optimizing the operation performance of the power grid and improving the stability and security of the power grid;

[0103] 605. Obtain the communication protocol and instruction format of the controller, and generate a controller parameter adjustment instruction according to the working parameters, communication protocol, and instruction format;

[0104] In this embodiment, the communication protocols involved include Modbus protocol, Profibus protocol, CAN protocol, etc. The instruction format taking the Modbus protocol as an example includes: function code, slave address, data address, data length, data content, check code. The instruction format taking the Profibus protocol as an example includes start byte, address field, control field, data field, check field, end byte. Generating the controller parameter adjustment instruction according to the working parameters, communication protocol, and instruction format ensures the pertinence and compatibility of the instruction, enables the generated instruction to be accurately transmitted to the controller, avoids control failure caused by communication protocol mismatch or instruction format error, and ensures that the controller can correctly receive and execute the adjustment operation, improving the accuracy and reliability of control.

[0105] 606. Generate a power grid maintenance task according to the preset fault diagnosis model and the controller parameter adjustment instruction.

[0106] In this embodiment, generating a power grid maintenance task based on the fault diagnosis model and the controller parameter adjustment instruction can comprehensively consider the fault diagnosis and parameter adjustment requirements, formulate a more targeted and reasonable maintenance task, which helps to discover potential problems in advance, reasonably arrange maintenance work, improve the efficiency and effect of maintenance work, ensure the long-term stable operation of the power grid, and at the same time reduce the losses caused by failures not being discovered and handled in time, providing comprehensive guarantee for the stable operation of the power grid system;

[0107] In this embodiment, information such as the power grid topology is obtained from the database, and the actual operating state of the power grid is accurately analyzed in combination with the standard state data, providing a basis for subsequent operations. Then, the working parameters of the controller are scientifically generated using power system analysis knowledge and operation and maintenance experience to optimize the power grid performance and enhance stability and security. At the same time, highly targeted and compatible controller parameter adjustment instructions are generated according to the communication protocol and instruction format to ensure the accurate and reliable operation of the controller. Finally, maintenance tasks are generated based on the fault diagnosis model and adjustment instructions, which can detect problems in advance, reasonably arrange maintenance work, improve the efficiency and effect of maintenance, reduce fault losses, provide comprehensive protection for the stable operation of the power grid, and promote the intelligent operation and maintenance management of the power grid.

[0108] Please refer to Figure 7 , the seventh embodiment of the system operation and maintenance method based on deep learning in the embodiment of the present invention includes:

[0109] 701. Obtain the fault log from the database;

[0110] 702. Establish a mapping relationship between the power grid state prediction result and the fault log through the fault diagnosis model;

[0111] 703. Conduct a simulated power grid test according to the mapping relationship, the controller parameter adjustment instruction, and the fault diagnosis model to obtain the simulated power grid performance result;

[0112] In this embodiment, obtaining the fault log from the database and establishing a mapping relationship between the power grid state prediction result and the fault log based on the fault diagnosis model helps to deeply understand the potential patterns and rules of power grid faults, providing strong data support for subsequent testing and optimization. By conducting a simulated power grid test based on this mapping relationship, the controller parameter adjustment instruction, and the fault diagnosis model, the performance of the power grid in different situations can be evaluated in advance without directly testing in the actual power grid, avoiding the risks and losses that may be brought by actual operations;

[0113] 704. Analyze the result of the simulated power grid performance result;

[0114] 705. When the simulated power grid performance result is that the power grid dynamic performance value is greater than or equal to the third-level value, generate a power grid stable operation guarantee task according to the power grid state prediction result and the actual operating state of the power grid;

[0115] 706. When the simulated power grid performance result is that the power grid dynamic performance value is less than the third-level value, generate a power grid dynamic performance optimization task according to the power grid state prediction result and the actual operating state of the power grid;

[0116] In this embodiment, by analyzing the simulation results of the power grid, the performance of the power grid can be accurately evaluated. According to different dynamic performance values, corresponding power grid stable operation guarantee tasks or power grid dynamic performance optimization tasks are generated, making the task generation targeted and meeting different performance requirements of the power grid. This differentiated task generation mechanism can better cope with the actual state of the power grid, ensuring the stable operation and performance optimization of the power grid.

[0117] 707. Generate a power grid maintenance task according to the power grid stable operation guarantee task and the power grid dynamic performance optimization task.

[0118] In this embodiment, a power grid maintenance task is generated according to the power grid stable operation guarantee task and the power grid dynamic performance optimization task. By comprehensively considering the maintenance requirements in different aspects, a comprehensive and effective power grid maintenance task is generated. This helps improve the systematicness and scientific nature of power grid maintenance work, make preparations in advance for various situations, ensure that the power grid system is always in a good operating state, reduce the possibility of failures, and at the same time improve the maintenance efficiency and effect under different performance states of the power grid, promoting the stable, efficient and reliable operation of the power grid system and ensuring the continuity and security of power supply.

[0119] In this embodiment, by obtaining the fault log from the database and establishing a mapping relationship, it helps to understand the power grid fault mode and provide data support for subsequent operations; using the mapping relationship, controller parameter adjustment instructions and the fault diagnosis model to conduct simulation tests on the power grid can evaluate the performance in advance and avoid the risks of actual tests; the analysis of the simulation results can accurately evaluate the power grid performance, generate targeted stable operation guarantee or dynamic performance optimization tasks according to different performance manifestations, reflecting the differentiated task generation mechanism, ensuring the power grid stability and performance optimization; finally, combining the two tasks to generate a power grid maintenance task makes the maintenance work more systematic and scientific, can prepare for various situations in advance, reduce the possibility of failures, improve the maintenance efficiency and effect, ensure the stable, efficient and reliable operation of the power grid system, and ensure the continuous and safe power supply.

[0120] The above describes the system operation and maintenance method based on deep learning in the embodiments of the present invention. Next, the device for the system operation and maintenance method based on deep learning in the embodiments of the present invention will be described. Please refer to Figure 8 , an embodiment of the device for the system operation and maintenance method based on deep learning in the embodiments of the present invention includes:

[0121] The model construction module 1 is used to construct a primary deep learning model according to the preset characteristics of the network system and deep learning algorithms.

[0122] The operation and maintenance data acquisition module 2 is used to acquire historical power grid operation and maintenance data in the preset database.

[0123] The model training module 3 is used to iteratively optimize and train the primary deep learning model according to historical power grid operation and maintenance data and regularization methods to obtain a deep power grid operation and maintenance prediction model;

[0124] The status data collection module 4 is used to collect the status data of network devices according to a preset network management protocol;

[0125] The status data processing module 5 is used to preprocess the status data according to a data preprocessing method to obtain standard status data;

[0126] The power grid status prediction module 6 is used to perform system status analysis on the standard status data based on the deep power grid operation and maintenance prediction model to obtain a power grid status prediction result;

[0127] The confidence threshold judgment module 7 is used to obtain the confidence in the power grid status prediction result and judge whether the confidence is lower than a preset confidence threshold;

[0128] The instruction and task generation module 8 is used to generate an adjustment instruction for adjusting the controller parameters and a power grid maintenance task according to the power grid status prediction result when the confidence is lower than the confidence threshold.

[0129] In this embodiment, the model construction module 1 uses the characteristics of the network system and deep learning algorithms to construct a primary deep learning model. The model training module 3 combines historical power grid operation and maintenance data and regularization training to obtain a deep power grid operation and maintenance prediction model, enabling the model to learn the power grid operation rules, avoid overfitting, and enhance the generalization and prediction accuracy of the model. The network management protocol ensures the wide and complete status data. Data preprocessing is used to improve the data quality. The deep power grid operation and maintenance prediction model can analyze, predict, and evaluate the confidence of the standard status data. When the confidence is low, the instruction and task generation module 8 will generate adjustment instructions and maintenance tasks to pre-warn and handle potential problems in advance, avoid or reduce the impact of faults, and enhance the stability and reliability of the power grid. At the same time, the status data collection module 4 ensures the data source, providing a strong guarantee for the efficient and safe operation of the power grid system.

[0130] Figure 9FIG. 0 is a schematic structural diagram of a system operation and maintenance device based on deep learning provided by an embodiment of the present invention. The system operation and maintenance device 900 based on deep learning may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPUs) 910 (for example, one or more processors) and a memory 920, and one or more storage media 930 for storing application programs 933 or data 932 (for example, one or more mass storage devices). Among them, the memory 920 and the storage media 930 may be transient storage or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the system operation and maintenance device 900 based on deep learning. Further, the processor 910 may be configured to communicate with the storage media 930 and execute a series of instruction operations in the storage media 930 on the system operation and maintenance device 900 based on deep learning to implement the steps of the system operation and maintenance method based on deep learning provided in the above method embodiments.

[0131] The system operation and maintenance device 900 based on deep learning may further include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that Figure 9 The shown structural diagram of the system operation and maintenance device based on deep learning does not limit the system operation and maintenance device based on deep learning, and may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements.

[0132] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium, and when the instructions run on a computer, the computer is caused to execute the steps of the system operation and maintenance method based on deep learning.

[0133] The above describes the present invention and its embodiments. This description is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual content is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural ways and embodiments without creative work without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A system operation and maintenance method based on deep learning, characterized in that: Includes steps: A primary deep learning model is constructed based on the preset network system characteristics and deep learning algorithm; Obtain historical power grid operation and maintenance data from a preset database; The primary deep learning model is iteratively optimized and trained based on historical power grid operation and maintenance data and regularization methods to obtain a deep power grid operation and maintenance prediction model; Collect status data of network devices according to preset network management protocols; Preprocessing the state data according to the data preprocessing method to obtain standard state data; Based on the deep power grid operation and maintenance prediction model, the standard status data is analyzed to obtain the power grid status prediction results; Obtaining the confidence level in the power grid state prediction result, and determining whether the confidence level is lower than a preset confidence threshold; When the confidence level is lower than the confidence threshold, controller parameter adjustment instructions and power grid maintenance tasks are generated according to the power grid state prediction results.

2. The system operation and maintenance method based on deep learning according to claim 1, characterized in that: The iterative optimization training of the primary deep learning model according to the historical power grid operation and maintenance data and the regularization method to obtain the deep power grid operation and maintenance prediction model includes: Iteratively optimize and train the primary deep learning model based on historical power grid operation and maintenance data and regularization methods to obtain an improved deep learning model; The loss of the secondary deep learning model is calculated according to the preset cross entropy loss function to obtain the cross entropy loss; The preset Adam optimization algorithm and cross entropy loss are used to adjust the model parameters of the secondary deep learning model to obtain a deep power grid operation and maintenance prediction model.

3. The system operation and maintenance method based on deep learning according to claim 2, characterized in that: The iterative optimization training of the primary deep learning model according to the historical power grid operation and maintenance data and the regularization method to obtain the improved deep learning model includes: The parameters of the primary deep learning model are weighted according to the regularization method to obtain a secondary deep learning model; Obtain task requirements from the database, and configure the network structure of the secondary deep learning model according to the characteristics of the network system and task requirements to obtain a third-level deep learning model; Dividing the historical power grid operation and maintenance data into a data set according to a preset first ratio to obtain a sample set; The chi-square test statistical method was used to perform feature selection on the sample set to obtain the training set; Iteratively train the three-level deep learning model according to the training set and regularization method to obtain a four-level deep learning model; Dividing the historical power grid operation and maintenance data into a data set according to a preset second ratio to obtain a test set; The four-level deep learning model is evaluated based on the cross-validation method and the test set to obtain the evaluation results; Get the regularization strength parameter of the four-level deep learning model; The four-level deep learning model is iteratively optimized according to the evaluation results and the regularization strength parameters to obtain an improved deep learning model.

4. The system operation and maintenance method based on deep learning according to claim 1, characterized in that: The system status analysis of the standard status data based on the deep power grid operation and maintenance prediction model to obtain the power grid status prediction result includes: The order of the preset autoregressive model is calculated using a preset information criterion to obtain a model order set; Get the minimum value from the model order set and set the minimum value as the model order of the autoregressive model; The autoregressive model is converted into a matrix regression model according to the model order and the least squares method; The smoothing method is used to smooth the short-term fluctuations of the standard state data to obtain smooth state data; The smooth state data is converted into a feature matrix based on a matrix regression model; Based on the deep power grid operation and maintenance prediction model and feature matrix, system status analysis is performed to obtain the power grid status prediction results.

5. The system operation and maintenance method based on deep learning according to claim 1, characterized in that: The preprocessing of the state data according to the data preprocessing method to obtain standard state data includes: De-noising the state data using a statistical method and a preset deviation from a normal range to obtain de-noised state data; Get the time series data in the denoised state data and fill its missing values ​​according to the linear interpolation method to obtain the standard time series data. The expression is as follows: In the formula, x t is the missing value to be filled, t is the timestamp of the missing value, x t-1 and x t+1 is the adjacent time series data, t -1 and t +1 It is the corresponding timestamp of two adjacent time series data; Using feature selection methods and preset system operation and maintenance tasks to select operation and maintenance status data and status feature data from the denoised status data; The standard time series data, operation and maintenance status data, and status feature data are unified in magnitude according to the preset normalization formula to obtain standard status data.

6. The system operation and maintenance method based on deep learning according to claim 1, characterized in that: When the confidence level is lower than the confidence threshold, a controller parameter adjustment instruction and a power grid maintenance task are generated according to the power grid state prediction result, including: When the confidence level is lower than the confidence threshold, the power grid topology and equipment connection relationship are obtained from the database; Analyze the grid state prediction results according to the grid topology, equipment connection relationship and standard state data to obtain the actual operation status of the grid; Acquire power system analysis knowledge and operation and maintenance experience from the database; Generate the working parameters that the controller needs to adjust based on power system analysis knowledge, operation and maintenance experience and the actual operation status of the power grid; Acquire the communication protocol and instruction format of the controller, and generate controller parameter adjustment instructions according to the working parameters, communication protocol and instruction format; Generate power grid maintenance tasks based on preset fault diagnosis models and controller parameter adjustment instructions.

7. The system operation and maintenance method based on deep learning according to claim 6, characterized in that: The generating of the power grid maintenance task according to the preset fault diagnosis model and the controller parameter adjustment instruction includes: Get the fault log from the database; A mapping relationship between the power grid state prediction results and the fault log is established through the fault diagnosis model; Perform simulated power grid testing according to the mapping relationship, controller parameter adjustment instructions and fault diagnosis model to obtain simulated power grid performance results; Analyze the simulated power grid performance results; When the simulated power grid performance result shows that the power grid dynamic performance value is greater than or equal to the third-level value, a power grid stable operation guarantee task is generated according to the power grid state prediction result and the actual power grid operation state; When the simulated power grid performance result shows that the power grid dynamic performance value is less than the third-level value, a power grid dynamic performance optimization task is generated according to the power grid state prediction result and the actual operation state of the power grid; The grid maintenance tasks are generated according to the grid stable operation guarantee tasks and grid dynamic performance optimization tasks.

8. A system operation and maintenance device based on deep learning, characterized in that: include: The model building module is used to build a primary deep learning model according to the preset network system characteristics and deep learning algorithm; The operation and maintenance data acquisition module is used to acquire historical power grid operation and maintenance data from a preset database; The model training module is used to iteratively optimize and train the primary deep learning model according to historical power grid operation and maintenance data and a regularization method to obtain a deep power grid operation and maintenance prediction model; The status data collection module is used to collect status data of network devices according to a preset network management protocol; The state data processing module is used to preprocess the state data according to the data preprocessing method to obtain standard state data; The power grid state prediction module is used to perform system state analysis on the standard state data based on the deep power grid operation and maintenance prediction model to obtain a power grid state prediction result; The confidence threshold judgment module is used to obtain the confidence in the power grid state prediction result and judge whether the confidence is lower than a preset confidence threshold; The instruction and task generation module is used to generate controller parameter adjustment instructions and power grid maintenance tasks according to the power grid state prediction results when the confidence level is lower than the confidence threshold.

9. System operation and maintenance equipment based on deep learning, characterized in that: The system operation and maintenance device based on deep learning includes: a memory and at least one processor, wherein instructions are stored in the memory; At least one of the processors calls the instructions in the memory so that the deep learning-based system operation and maintenance device performs each step of the deep learning-based system operation and maintenance method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the deep learning-based system operation and maintenance method as described in any one of claims 1 to 7 are implemented.

Images