Transformer area intelligent operation and maintenance device and method based on machine learning expert model

Abstract

The invention relates to the technical field of electric power operation and maintenance management, in particular to a transformer area intelligent operation and maintenance device and method based on a machine learning expert model, and the device comprises a logistic regression machine learning module and a robot process automation module. Real-time perception and data analysis of a business scene are realized by means of artificial intelligence, total data information of a master station measurement domain and an operation and maintenance work task is dynamically collected, incremental data is continuously utilized for data analysis and rule mining, a machine learning expert model is introduced, and the operation and maintenance efficiency is improved. A transformer area intelligent operation and maintenance system integrating human-computer interaction such as operation and maintenance index intelligent monitoring and analysis, intelligent abnormity diagnosis, intelligent operation list generation, intelligent dispatching, intelligent path planning and field operation guidance is constructed, service data is analyzed in real time, meanwhile, a robot process automation technology is integrated, automatic operation of a part of processes is achieved, and operation efficiency is improved. And support is provided for on-site acquisition operation and maintenance work.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power operation and maintenance management, and in particular to a substation intelligent operation and maintenance device and method based on a machine learning expert model. Background Art

[0002] As one of the key pillars of the power industry's digital transformation, the collection and operation of electricity information is being implemented in depth along with the comprehensive deployment of a new generation of electricity information systems. This process not only relies on the advancement of data collection, efficient processing and in-depth analysis technology, but also successfully builds a standardized collection and operation closed-loop management system, and implements standardized management of each on-site disposal link, thereby ensuring the full controllability of the collection and operation work and a substantial improvement in the level of online refined management of on-site operations.

[0003] However, in the face of increasingly complex business scenarios and stringent requirements for data timeliness, although existing technologies have achieved a certain degree of automation in the collection of on-site operation and maintenance management, there are still many shortcomings. Grassroots operation and maintenance personnel have a large workload and need to frequently switch system interfaces to find data, involving multiple diagnostic topics and methods, which is arduous and costly; monitoring planning is not intuitive, and as the business volume grows, there is a lack of overall evaluation and intelligent tool support, which affects operation and maintenance efficiency; there is redundancy in work orders, multiple sources of receiving tasks, inaccurate diagnostic analysis, and inability to filter invalid business data well; there is no order in the work, no reasonable work arrangement, and the current task priority and dispatch rules lack flexibility, and cannot be dynamically adjusted according to the unit's phased work priorities and the specific changes in employees' abilities; the work instructions are not accurate, and the work instructions are not flexible for operation and maintenance personnel with different abilities, and the root cause analysis and intelligent diagnosis are relatively vague, lacking accurate guidance on treatment measures. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the existing technology and propose an intelligent operation and maintenance device and method for the substation based on machine learning expert model, which can improve the operation and maintenance efficiency and accuracy, realize the automation of some processes, and increase the speed of fault repair.

[0005] To achieve the above object, the present invention adopts the following specific technical solutions: The intelligent operation and maintenance device for a substation area based on a machine learning expert model provided by the present invention includes a logistic regression machine learning module and a robotic process automation (RPA) module; The logistic regression machine learning module includes data preprocessing and feature engineering module, model training and updating module, and fault prediction and diagnosis module; the robotic process automation module includes data collection automation module, system interaction and integration module, and work order automatic generation and processing module; The data preprocessing and feature engineering module processes the outliers in the electricity meter reading data, normalizes the voltage and current data in different ranges, and extracts features such as the change rate of electricity meter readings and the voltage fluctuation frequency; The model training and updating module uses historical data including normal operation data and fault data of the equipment to train a logistic regression model to adapt to the changes in the substation area equipment and electricity consumption situation, and improve the accuracy and generalization ability of the logistic regression model; The fault prediction and diagnosis module inputs the data obtained in real time by the data preprocessing and feature engineering module into the trained logistic regression model to obtain the probability of equipment failure, determines whether the equipment is likely to fail according to the set threshold, and diagnoses the fault type; The data acquisition automation module regularly collects power consumption, voltage, and current data from all smart electricity meters in the substation area according to the set time interval and rules; The system interaction and integration module automatically transfers the data collected by the data acquisition automation module to the power marketing system and the equipment management system through robotic process automation software, and obtains user information and equipment file information to achieve data interconnection; When the work order automatic generation and processing module detects equipment failure or data abnormality, it automatically generates an operation and maintenance work order including faulty equipment information, fault type, discovery time, and recommended handling measures, and tracks the processing progress of the work order through robotic process automation software.

[0006] The present invention also provides a substation area intelligent operation and maintenance method based on a machine learning expert model, which applies the above operation and maintenance device, including: constructing an integrated learning algorithm model based on Logistic Regression, learning the user's behavior habits and the data of indicators concerned in daily work, adjusting the algorithm rules of data mining, and extracting key concerned indicators; Power consumption anomaly monitoring, by analyzing the historical data of user equipment including normal operation data and fault data, training an integrated learning algorithm model, and setting a threshold or probability to judge whether there is an anomaly in power consumption; Equipment fault warning, by analyzing the real-time operation data of the equipment, when the operation parameters of the equipment change and this change is similar to the fault mode, the integrated learning algorithm model issues a warning signal; Intelligent work order generation and dispatching, the integrated learning algorithm model generates an operation and maintenance work order according to the set work order generation probability threshold, and intelligently dispatches according to the equipment type, fault type, skills and location of the operation and maintenance personnel; Intelligent path planning and operation guidance, the integrated learning algorithm model combines with the Geographic Information System (GIS) to plan the optimal operation and maintenance path for the operation and maintenance personnel, calculates the probability of the fault type, and improves the response speed of fault elimination.

[0007] Furthermore, in power consumption anomaly monitoring and equipment fault warning, the device operation data for integrated learning algorithm model training and analysis is collected at regular intervals and according to rules by the robotic process automation (RPA) robots of smart meters and sensors deployed in the distribution area. The collected data is transmitted to the power marketing system and the equipment management system, and user information and equipment file information are obtained. The operation data of the distribution area equipment, user power consumption data, and marketing data are integrated to generate reports on the operation status of distribution area equipment, line loss analysis reports, and user power consumption behavior analysis reports.

[0008] Furthermore, the integrated learning algorithm model based on logistic regression models the output based on the logistic function, also known as the Sigmoid function. The Sigmoid function can map the linearly combined features to probability values between 0 and 1. The integrated learning algorithm model can be expressed as: h(x)=sigmoid(w^Tx); where x is the input feature vector, which usually represents the features of the sample. In the model for predicting whether an electricity meter will fail, x includes features such as the service life of the electricity meter and the time interval since the last abnormal fault recovery; w is the weight vector, whose dimension is the same as that of x. Each element in the weight vector w corresponds to the weight of a feature, which determines the influence degree of this feature on the final result; w^Tx represents the dot product of the feature vector and the weight vector, and h(x) represents the probability value of the output result between 0 and 1. The Sigmoid function is a common S-shaped function in biology. In deep learning, the Sigmoid function is used as the activation function of the neural network to map the variable to between [0, 1]. The formula is as follows: The integrated learning algorithm model based on logistic regression learns the model parameters by maximizing the likelihood function or minimizing the loss function.

[0009] Furthermore, the integrated learning algorithm model based on logistic regression includes: an input layer that receives feature data, and each feature corresponds to a node in the input layer; weights, where each input feature has a corresponding weight to measure the influence degree of this feature on the final output; a linear combination, where the input features and the corresponding weights are linearly combined to obtain a weighted sum; an activation function, and the result of the linear combination passes through the activation function, usually the Sigmoid function, to map the result to between [0, 1], representing the probability that the sample belongs to the positive class; an output layer, and the output layer outputs the classification result, usually classifying with a threshold of 0.5, greater than 0.5 is the positive class, and less than 0.5 is the negative class.

[0010] Furthermore, the integrated learning algorithm model extracts the key attention indicators as follows: Through robotic process automation, the meter data including readings, voltage, current, and power factor are collected; the equipment operation data including the temperature of the terminal and the status of the communication module; the user information including the type of electricity usage and the time period; the operation and maintenance records including the fault type, maintenance time, and maintenance personnel; the integrated learning algorithm model sets the attention weight ratio according to business needs and phased work goals, and uses the weight vector of the integrated learning algorithm model to evaluate the importance of each feature to the target event.

[0011] The present invention can achieve the following technical effects: The present invention uses artificial intelligence to achieve real-time perception and data analysis of business scenarios, dynamically collects and acquires full data information on the measurement domain and operation and maintenance tasks of the main station, and continuously uses incremental data for data analysis and rule mining. It introduces machine learning expert models to build an intelligent operation and maintenance system for the area that integrates human-computer interaction such as intelligent monitoring and analysis of collection and operation and maintenance indicators, intelligent abnormality diagnosis, intelligent job order generation, intelligent dispatching, intelligent path planning, and on-site operation guidance. It analyzes business data in real time and integrates RPA technology to realize partial process automation, provide support for on-site collection and operation and maintenance work, and improve operation and maintenance efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a structural diagram of a substation intelligent operation and maintenance device based on a machine learning expert model provided according to an embodiment of the present invention.

[0013] Figure 2 It is a schematic diagram of an application process in Robotic Process Automation (RPA) operation and maintenance provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0014] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same modules are represented by the same reference numerals. In the case of the same reference numerals, their names and functions are also the same. Therefore, the detailed description thereof will not be repeated.

[0015] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0016] The embodiment of the present invention provides a substation intelligent operation and maintenance device based on a machine learning expert model, and its structure is as follows: Figure 1 As shown, it includes a logistic regression machine learning module and a robotic process automation (RPA) module; The logistic regression machine learning module includes a data preprocessing and feature engineering module, a model training and updating module, and a fault prediction and diagnosis module; the robotic process automation module includes a data collection automation module, a system interaction and integration module, and a work order automatic generation and processing module.

[0017] 1. Data preprocessing and feature engineering module: Perform preprocessing operations such as cleaning, normalization, and feature extraction on the large amount of collected data. For example, handle outliers in the electricity meter reading data, normalize voltage and current data in different ranges, and extract valuable features such as the change rate of electricity meter readings and the voltage fluctuation frequency to provide high-quality data input for the logistic regression model.

[0018] 2. Model training and updating module: Use historical data (including normal operation data and fault data of equipment) to train the logistic regression model and determine the model parameters (such as weight vectors and bias terms). At the same time, as new data accumulates, update the model regularly to adapt to changes in the equipment and electricity consumption situation in the substation area and improve the accuracy and generalization ability of the model.

[0019] 3. Fault prediction and diagnosis module: Input the real-time collected data into the trained logistic regression model, and the model outputs the probability of equipment failure. Judge whether the equipment is likely to fail according to the set threshold and diagnose the type of failure.

[0020] 4. Data collection automation module: Automatically log in to each power consumption collection system and sensor data management system by simulating manual operations, and collect data according to the set time interval and rules. For example, regularly collect data such as electricity consumption, voltage, and current from all smart electricity meters in the substation area without manual intervention, improving the efficiency and accuracy of data collection.

[0021] 5. System interaction and integration module: Realize the interaction and data sharing between this device and other relevant systems (such as power marketing systems, equipment management systems). The RPA software can automatically transfer the collected data to other systems and obtain necessary information (such as user information, equipment files, etc.) from other systems at the same time, breaking information silos and achieving data interconnection and interoperability.

[0022] 6. Work order automatic generation and processing module: Automatically generate operation and maintenance work orders when equipment failures or data anomalies are detected. The work order content includes faulty equipment information, fault type, discovery time, recommended treatment measures, etc. At the same time, the RPA software can track the processing progress of the work order, automatically remind the operation and maintenance personnel to process the work order in a timely manner, and file and record it after the work order is processed.

[0023] The embodiment of the present invention also provides a substation intelligent operation and maintenance method based on a machine learning expert model. Based on the above-mentioned substation intelligent operation and maintenance device, it uses artificial intelligence to realize real-time perception and data analysis of business scenarios, dynamically collects and collects all data information of the measurement domain and operation and maintenance tasks of the main station, and continuously uses incremental data for data analysis and rule mining. It introduces a machine learning expert model to build an intelligent operation and maintenance system for substations that integrates human-computer interaction such as intelligent monitoring and analysis of collection and maintenance indicators, intelligent abnormal diagnosis, intelligent job order generation, intelligent dispatching, intelligent path planning, and on-site operation guidance. It analyzes business data in real time and integrates RPA technology to realize partial process automation and provide support for on-site collection and operation and maintenance work. The operation and maintenance method is as follows: 1. Build an integrated learning algorithm model based on logistic regression.

[0024] Learn users' behavioral habits and indicator data they pay attention to in their daily work, gradually adjust the algorithm rules of data mining, form personal preferences based on the topics that users pay attention to at different stages, automatically form key indicators for collection and operation, analyze and predict key indicators of collection and operation in real time, realize instant warning of abnormal situations, provide proactive data insights for operation and maintenance monitoring personnel, and provide visual display.

[0025] 1) Definition and purpose of the ensemble learning algorithm model based on logistic regression: Logistic regression is a classification algorithm used to deal with binary classification problems. It determines which category a sample belongs to by mapping input features to the probability of the output category. The output of logistic regression is a probability value between 0 and 1, indicating the probability that a sample belongs to a certain category.

[0026] 2) Principle of ensemble learning algorithm model: Logistic regression models output based on the logistic function (also known as the Sigmoid function). The Sigmoid function can map the features of the linear combination to a probability value between 0 and 1. The ensemble learning algorithm model based on logistic regression can be expressed as: h(x)=sigmoid(w^Tx); (1) Where \(x\) is the input feature vector, which usually represents the features of the sample. For example, in a model for predicting whether an electricity meter will malfunction, \(x\) may include features such as the service life of the electricity meter and the time interval since the last abnormal fault recovery. \(w\) is the weight vector, whose dimension is the same as that of \(x\). Each element in the weight vector corresponds to the weight of a feature, which determines the degree of influence of that feature on the final result. During the training process, these weights are adjusted so that the model can better fit the data. For example, if the feature of service life is very important for judging whether the electricity meter malfunctions, the corresponding weight value will be relatively large. \(w^Tx\) represents the dot product of the feature vector and the weight vector, and \(h(x)\) represents the probability value between 0 and 1 of the output result.

[0027] The Sigmoid function is an S-shaped function commonly found in biology, also known as the S-shaped growth curve. In deep learning, due to its properties such as being monotonically increasing and having a monotonically increasing inverse function, the Sigmoid function is often used as the activation function of neural networks, mapping variables to the interval \([0, 1]\). The formula is as follows: (2) During the training process, logistic regression learns the model parameters by maximizing the likelihood function or minimizing the loss function (such as the cross-entropy loss function). Commonly used optimization algorithms include gradient descent, etc.

[0028] 3)Algorithm model structure: The structure of the logistic regression algorithm is relatively simple and intuitive, including the following parts: the input layer, which receives the feature data, and each feature corresponds to a node in the input layer; weights, each input feature has a corresponding weight, which is used to measure the degree of influence of that feature on the final output; linear combination, the input features and the corresponding weights are linearly combined to obtain the weighted sum; activation function, the result of the linear combination passes through the activation function, usually the Sigmoid function, which maps the result to the interval \([0, 1]\), representing the probability that the sample belongs to the positive class; output layer, the output layer outputs the classification result, usually classifying with a threshold of 0.5, greater than 0.5 is the positive class, and less than 0.5 is the negative class.

[0029] 4) Focus on index extraction, collect comprehensive data from the power consumption acquisition system, including electricity meter data (such as readings, voltage, current, power factor, etc.), equipment operation data (such as the temperature of the acquisition terminal, the status of the communication module, etc.), user information (such as electricity consumption type, electricity consumption period, etc.), and operation and maintenance records (such as fault type, repair time, maintenance personnel, etc.). Extract representative feature data, set reasonable attention weight ratios according to business requirements and phased work objectives, and use the coefficients (i.e., weight vectors) of the logistic regression model to evaluate the importance of each feature to the target event. The absolute value of the coefficient reflects the degree of influence of the corresponding feature on the probability of the target event. For each feature, determine whether it is a key indicator based on its importance in the logistic regression model and its relevance to the operation and maintenance objectives.

[0030] 2. Abnormal power consumption monitoring.

[0031] By analyzing the historical power consumption data of users, establish a power consumption behavior model, and set reasonable thresholds or probabilities to determine whether there is abnormal power consumption. For example, when the power consumption of a user suddenly increases or decreases significantly, and this change exceeds the range predicted by the model, the algorithm can determine that the power consumption is abnormal, and then issue an alarm in a timely manner to remind the operation and maintenance personnel to conduct inspections and repairs.

[0032] 3. Equipment fault warning.

[0033] By analyzing the operation data of equipment, establish an equipment fault prediction model. When the operation parameters of the equipment change, and this change is similar to the fault mode, the algorithm can issue a warning signal to remind the operation and maintenance personnel to conduct equipment maintenance or replacement in advance, so as to avoid the impact of equipment faults on power consumption.

[0034] 4. Intelligent work order generation and dispatching.

[0035] When new power consumption acquisition data is input into the trained logistic regression model, the model will output a work order generation probability. For example, if the output probability is greater than the set threshold (such as 0.5), it can be determined that a maintenance work order needs to be generated. This threshold can be adjusted according to the actual situation. For example, for maintenance scenarios with higher requirements, the threshold can be lowered to detect problems more timely. According to the evaluation results of the logistic regression machine learning model, if the model performance is not good, the model can be optimized. For example, readjust features, add new features, adjust the hyperparameters of the model (such as learning rate, number of iterations, etc.), or adopt more complex model fusion strategies to improve the accuracy and timeliness of work order generation.

[0036] The dispatch of intelligent operation and maintenance work orders for transformer substations often needs to consider multiple factors, such as equipment type, fault type, skills and location of operation and maintenance personnel, etc. Logistic regression can easily model these factors as input features, comprehensively consider the impact of multiple factors on work order dispatch, so as to achieve more accurate dispatch decisions.

[0037] 5. Intelligent path planning and operation guidance.

[0038] Combined with the Geographic Information System (GIS), it plans the optimal operation and maintenance path for operation and maintenance personnel. The model can calculate the most time-saving and efficient operation and maintenance route according to factors such as the distance between transformer substations, traffic conditions, and fault probability, reduce the time spent by operation and maintenance personnel on the road, and improve the response speed of fault elimination.

[0039] When there are multiple possible fault types (such as electricity meter metering faults, communication faults, power line overload faults, etc.), multi-class logistic regression can be used. Through the model, calculate the probability of each fault type. Operation and maintenance personnel can infer the most likely fault type according to the probabilities of various fault types output by the model, and carry out fault elimination work according to the fault elimination steps recommended by the operation and maintenance knowledge base.

[0040] The transformer substation intelligent operation and maintenance method based on the machine learning expert model provided by the embodiments of the present invention integrates the Robotic Process Automation (RPA) technology to realize partial process automation operations and provide support for on-site collection operation and maintenance work. The RPA technical solution is applied in collection operation and maintenance as Figure 2 shown.

[0041] Robotic Process Automation (RPA) is a digital workflow solution that can automatically execute a series of daily, time-consuming, and error-prone tasks such as data entry, data processing, file transfer, and form filling through pre-set rules and logic.

[0042] The technical solution of RPA technology includes the following key parts:

[0043] 1) Process analysis: Conduct a detailed analysis of the business process to determine which links are suitable for automated processing. Such as the verification of transformer substation meter reading data, remote automated analysis of equipment faults, export of report data, etc., evaluate factors such as the complexity, frequency, and rule clarity of the process to ensure the effective application of RPA technology.

[0044] 2) Automated data collection: RPA robots are deployed to automatically collect data from smart meters and various sensors within the substation area. By simulating manual login to the data collection system, information such as meter readings, voltage, and current is collected at preset times (such as once per hour). RPA robots can handle data collection tasks for multiple substations simultaneously, greatly improving the efficiency and timeliness of data collection.

[0045] 3) Automated remote troubleshooting and repair of collection faults: Using RPA technology and backend logic processing services to replace in-house staff to automatically complete remote device fault analysis and defect elimination. For example, if a power meter has no data for multiple consecutive days, the RPA robot screens for the anomaly of "no data for multiple consecutive days on the power meter", extracts detailed data, performs data transcribing, clock comparison, comparison of measurement point archives, comparison of collection tasks, formulates diagnostic opinions, automatically dispatches work orders, and remotely resumes monitoring, etc., to carry out defect elimination work on equipment faults.

[0046] 4) Automated report generation and analysis: RPA technology is used to connect and integrate data from multiple systems such as the power consumption collection system, equipment management system, and marketing system. Through automated interface calls and data transmission, the operation data of substation area equipment, user power consumption data, and marketing data are integrated onto a unified platform. For example, data is extracted from each system at a fixed time every day, matched and correlated to form a complete view of substation area operation and maintenance data. Based on the integrated data, RPA automatically generates various operation and maintenance reports, such as substation area equipment operation status reports, line loss analysis reports, user power consumption behavior analysis reports, etc. It provides intuitive data display and deeply mines the report data through built-in data analysis algorithms. For example, by analyzing the line loss report, it is found that the line loss of certain lines in the substation area is abnormal, providing a basis for fault troubleshooting and energy-saving measures.

[0047] 5) System integration and expansion: Through methods such as API calls, the integration of RPA technology with business systems is achieved.

[0048] By adopting the idea of ensemble learning, the model of the present invention can be dynamically adjusted according to real-time data and new environmental conditions. Compared with existing static models, this advantage significantly improves the accuracy of fault prediction and enhances the robustness and reliability of fault diagnosis. It enables the system to respond quickly when there are abnormal changes in the equipment, solving the problem of relatively fixed and procedural model diagnosis in traditional technologies. It uniformly standardizes data reception, receiving data such as intelligent diagnosis processing, data pushed by each system, and details of failed remote execution of collection. It uses feature detection and extraction, deep learning algorithms, pattern matching and semantic analysis, as well as large-scale data processing and parallel computing to identify key information, convert it into a unified format, summarize task data, and comprehensively consider aspects such as filtering invalid data, influencing key indicators, and merging business data to construct a work order generation model. It integrates RPA technology to achieve seamless connection with business applications. By simulating human operations on a computer, it can automatically execute tasks according to preset rules and processes, thus reducing human errors and omissions, being easy to expand, and easily adapting to changing business needs. Only a small modification to the RPA script is required to implement new functions.

[0049] In the description of this specification, the description referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0050] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0051] The above specific implementation manners of the present invention do not constitute a limitation to the protection scope of the present invention. Any other corresponding changes and deformations made according to the technical concept of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A substation intelligent operation and maintenance device based on a machine learning expert model, characterized in that: Includes logistic regression machine learning module and Robotic Process Automation (RPA) module; The logistic regression machine learning module includes a data preprocessing and feature engineering module, a model training and updating module, and a fault prediction and diagnosis module; The robotic process automation module includes a data acquisition automation module, a system interaction and integration module, and a work order automatic generation and processing module; The data preprocessing and feature engineering module processes abnormal values ​​in the meter reading data, normalizes the voltage and current data in different ranges, and extracts the characteristics of the meter reading change rate and voltage fluctuation frequency; The model training and updating module uses historical data including normal operation data and fault data of the equipment to train the logistic regression model to adapt to changes in equipment and power consumption in the substation area and improve the accuracy and generalization ability of the logistic regression model; The fault prediction and diagnosis module inputs the data obtained in real time by the data preprocessing and feature engineering module into the trained logistic regression model to obtain the probability of equipment failure, determines whether the equipment is likely to fail according to the set threshold, and diagnoses the type of failure; The data collection automation module collects electricity, voltage and current data from all smart meters in the area at regular intervals and according to set rules; The system interaction and integration module automatically transmits the data collected by the data collection automation module to the power marketing system and the equipment management system through the robotic process automation software, and obtains user information and equipment file information to achieve data interconnection and interoperability; When a device failure or data anomaly is detected, the work order automatic generation and processing module automatically generates an operation and maintenance work order including the faulty device information, fault type, discovery time, and recommended processing measures, and tracks the processing progress of the work order through the robotic process automation software.

2. A substation intelligent operation and maintenance method based on a machine learning expert model, using the substation intelligent operation and maintenance device according to claim 1, characterized in that: include: Build an integrated learning algorithm model based on logistic regression to learn users' behavioral habits and indicator data of daily work, adjust the algorithm rules of data mining, and extract key indicators; Abnormal power consumption monitoring: by analyzing the historical data of user equipment, including normal operation data and fault data, training the integrated learning algorithm model, and setting thresholds or probabilities to determine whether there is abnormal power consumption; Equipment failure early warning: by analyzing the real-time operation data of the equipment, when the operating parameters of the equipment change and this change is similar to the failure mode, the integrated learning algorithm model will issue an early warning signal; Intelligent work order generation and dispatching: The integrated learning algorithm model generates operation and maintenance work orders according to the set work order generation probability threshold, and intelligently dispatches work orders based on equipment type, fault type, and the skills and location of the operation and maintenance personnel; Intelligent path planning and operation guidance, integrated learning algorithm model combined with geographic information system (GIS), plan the optimal operation and maintenance path for operation and maintenance personnel, calculate the probability of fault types, and improve the response speed of fault elimination.

3. The intelligent operation and maintenance method for substations based on a machine learning expert model according to claim 2, characterized in that: In the monitoring of abnormal electricity consumption and equipment failure warning, the equipment operation data trained and analyzed by the integrated learning algorithm model is collected regularly according to the set time intervals and rules through the Robotic Process Automation (RPA) robots of the smart meters and sensors deployed in the substation, and the collected data is transmitted to the power marketing system and equipment management system, and the user information and equipment file information are obtained. The operation data of the substation equipment, the user's electricity consumption data and the marketing data are integrated to generate the substation equipment operation status report, line loss analysis report and user electricity consumption behavior analysis report.

4. The intelligent operation and maintenance method for substations based on a machine learning expert model according to claim 3 is characterized in that: The ensemble learning algorithm model based on logistic regression is based on the logistic function, also known as the Sigmoid function, to model the output. The Sigmoid function can map the features of the linear combination to a probability value between 0 and 1. The ensemble learning algorithm model can be expressed as: h(x)=sigmoid(w^Tx); Among them, x is the input feature vector, which usually represents the characteristics of the sample. In the model for predicting whether the electric energy meter will fail, x contains the characteristics of the service life of the electric energy meter and the time interval of the most recent abnormal fault recovery; w is the weight vector, whose dimension is the same as x. Each element in the weight vector w corresponds to the weight of a feature, which determines the degree of influence of the feature on the final result; w^Tx represents the dot product of the feature vector and the weight vector, and h(x) represents the probability value of the output result between 0 and 1; The Sigmoid function is a common S-shaped function in biology. In deep learning, the Sigmoid function is used as the activation function of the neural network to map the variables to [0, 1]. The formula is as follows: The ensemble learning algorithm model based on logistic regression learns model parameters by maximizing the likelihood function or minimizing the loss function.

5. The intelligent operation and maintenance method for substations based on a machine learning expert model according to claim 4 is characterized in that: The ensemble learning algorithm model based on logistic regression includes: input layer, which receives feature data, and each feature corresponds to a node in the input layer; weight, each input feature has a corresponding weight, which is used to measure the influence of the feature on the final output; linear combination, the input feature and the corresponding weight are linearly combined to obtain the weighted sum; activation function, the result of the linear combination is passed through the activation function, usually the Sigmoid function, and the result is mapped to [0, 1], indicating the probability that the sample belongs to the positive class; output layer, the output layer outputs the classification result, usually with 0.5 as the threshold for classification, greater than 0.5 is positive class, and less than 0.5 is negative class.

6. The intelligent operation and maintenance method for substations based on machine learning expert models according to claim 5 is characterized in that: The key focus indicators of the ensemble learning algorithm model extraction are as follows: Through robotic process automation, collect meter data including reading, voltage, current, and power factor; equipment operation data including terminal temperature and communication module status; user information including electricity usage type and electricity usage time; Including fault type, maintenance time, and maintenance records of maintenance personnel; the integrated learning algorithm model sets the attention weight ratio according to business needs and phased work goals, and uses the weight vector of the integrated learning algorithm model to evaluate the importance of each feature to the target event.

Images