A machine learning-based abnormality identification system and identification method for substation equipment

Abstract

The present invention belongs to the technical field of abnormality identification, and discloses a substation abnormality identification system and identification method based on machine learning; the system comprises: collecting insulation parameters, temperature data and electrical parameters of transformers, combining machine learning models to evaluate the health status of equipment, and then using Actor-Critic algorithm to train a reinforcement learning model, which considers factors such as equipment status and resource status, and makes decisions in an action space including equipment selection, detection degree, personnel scheduling and frequency adjustment to balance detection effect, resource utilization and cost control.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormality identification, and more specifically, to a substation abnormality identification system and identification method based on machine learning. Background Art

[0002] Substation equipment is equipment used in power systems for voltage conversion, current conversion, DC to AC conversion, as well as protection, control, regulation, measurement and compensation.

[0003] The Chinese patent application publication number CN110889441A discloses a method for identifying abnormal data of substation equipment based on distance and point density. The method first uses cluster analysis, then uses a distance-based abnormal point identification method to identify the classified data, and at the same time integrates a density-based abnormal point identification method, and finally constructs an evaluation index for identification.

[0004] The Chinese patent with authorization announcement number CN104331521B discloses a method for automatic recognition of abnormalities of substation equipment based on image processing, which classifies the pixel points of the substation equipment image data to form a binary image; the fault type is determined based on the identified binary image information of the substation equipment combined with the infrared image database information.

[0005] The Chinese patent application publication number CN112418202A discloses a method and system for identifying and locating abnormalities of substation equipment based on MaskRCNN. This invention uses the iForest algorithm to identify outliers in preprocessed operating data, and uses the K-means clustering strategy to mark outliers. It combines the MaskRCNN target recognition network model for preliminary identification, and enhances the positioning and identification capabilities of abnormal equipment locations by adding the LSSVM positioning model.

[0006] However, monitoring and inspection of substation equipment may not be necessary every time. The above patents do not take into account reducing unnecessary inspections, balancing the frequency and depth of abnormality identification, optimizing resource utilization, and reducing resource waste in detection and labor costs.

[0007] In view of this, the present invention proposes a substation equipment abnormality identification system and identification method based on machine learning to solve the above problems. Summary of the invention

[0008] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a substation abnormality identification system and identification method based on machine learning.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A method for identifying substation equipment based on machine learning, comprising:

[0011] Collect transformer insulation parameters; transformer insulation parameters include: gas content in transformer oil and dielectric loss factor; collect transformer electrical parameters, transformer temperature data, equipment last maintenance time and equipment historical maintenance frequency; transformer temperature data Where: T is the transformer temperature, f is the transformer vibration frequency, a is the transformer vibration acceleration;

[0012] The transformer insulation parameters, transformer temperature data and transformer electrical parameters are used as equipment parameters; a machine learning model that outputs equipment health scores is trained based on equipment parameters;

[0013] Define state space and action space; the state space includes device state and resource state; device state includes device health score, device last maintenance time and device historical failure rate; resource state includes: available testing equipment, human resources and time budget; the action space includes: selecting the next device to be tested; calling the number of test personnel and their corresponding levels; determining the idle testing equipment to be used; determining the degree of testing; adjusting the testing frequency;

[0014] Set the reward function and cumulative reward, reward function r_t = detection completion reward + health score improvement reward + anomaly detection reward + fault prevention reward - resource usage penalty + time management reward + detection frequency adjustment reward - false alarm penalty + equipment utilization reward; time step t means that in reinforcement learning, the number of consecutive steps from the initial state is t; cumulative reward R_t = r_t + γ*r_{t+1} + γ^2*r_{t+2} + ... + γ^(Tt)*r_T; where: R_t is the discounted sum of all instant rewards from the current time step t to the end time step T, r_{t+1} is the instant reward value at time step t+1; r_{t+2} is the instant reward value at time step t+2;

[0015] The reinforcement learning model is trained through the Actor-Critic algorithm to complete the anomaly recognition strategy, and the trained anomaly recognition strategy is applied to the actual anomaly recognition and detection of substation equipment.

[0016] Preferably, the transformer insulation parameters are synchronously updated and uploaded to the cloud after each measurement by the staff. When the transformer insulation parameters are not updated, the most recent measurement result is selected as the transformer insulation parameter;

[0017] The transformer electrical parameters include the current and voltage of the transformer;

[0018] The historical maintenance frequency of the equipment is the number of maintenance times within 1000 hours before the current time; the last maintenance time of the equipment is the specific time of the most recent maintenance of the equipment from the current time, accurate to the minute;

[0019] Available testing equipment is presented in a list, including the number of each testing equipment and the idle / occupied status; the testing equipment is the testing equipment used to inspect the transformer; human resources include the number of idle testing personnel classified by skill level; the skill level classification includes junior testing engineer, intermediate testing engineer and senior testing engineer;

[0020] The time budget is the number of working hours remaining in the current working day.

[0021] Preferably, the training process of the machine learning model includes:

[0022] A set of device parameters is converted into a feature vector, and the device health score is used as the label corresponding to the feature vector, and the device health score is a positive integer between 0 and 100; each set of feature vectors and the label corresponding to each set of feature vectors are constructed as a sample, and multiple samples are collected to construct a data set; the data set is divided into a training set, a validation set and a test set, wherein the training set accounts for 70% of the data set, the validation set accounts for 20% of the data set, and the test set accounts for 10% of the data set;

[0023] The training set is used as the input of the machine learning model, and the machine learning model uses the real-time device health score as the output; the device health score corresponding to a set of real-time feature vectors is used as the prediction target, and minimizing the loss function value of the machine learning model is used as the training target; when the loss function value of the machine learning model is less than or equal to the preset target loss value, the training is stopped;

[0024] The machine learning model loss function is the mean square error; the mean square error is calculated by Minimize to train the model; in the loss function, MSE is the loss function value, i is the feature vector group number; u is the number of feature vector groups; y i is the label corresponding to the i-th group of feature vectors, The device health score predicted for the i-th group of feature vectors;

[0025] The machine learning model is a random forest model;

[0026] Preferably, the degree of inspection includes quick inspection, routine inspection and in-depth inspection; quick inspection is a visual inspection or quick measurement using simple tools; routine inspection uses conventional equipment to perform standard inspection procedures, and in-depth inspection uses professional equipment for comprehensive inspection; the inspection frequency adjustment range is [-30%, +30%], which represents the adjustment ratio relative to the historical maintenance frequency of the current equipment.

[0027] Preferably, the reward function design process includes:

[0028] Reward for completing the test: +5 points for each completed equipment test;

[0029] Health score improvement reward: For every 1-point increase in the device health score, an additional reward of +2 points;

[0030] Abnormal detection reward: For successfully detecting an abnormal situation, a reward of +10 points;

[0031] Fault prevention reward: If a problem that may lead to a major fault is detected and resolved, a reward of +50 points;

[0032] Resource usage penalty: For using a quick check, a penalty of -1 point; for using a regular check, a penalty of -3 points; for using a deep check, a penalty of -5 points; at the same time, the penalty coefficient for junior detection engineers is 1; the penalty coefficient for intermediate detection engineers is 1.5; the penalty coefficient for senior detection engineers is 3; Multiply the degree of check used by the penalty coefficient corresponding to the level of the engineer called as the resource usage penalty. When there are multiple engineers called, the resource usage penalty is stacked for each engineer in turn;

[0033] Time management reward: If all planned detections are completed before the end of the working day, a reward of +7 points;

[0034] Detection frequency adjustment reward, set the detection frequency adjustment evaluation value E = δH / δF; where, δH is the change in the device health score this time compared to the previous time, and δF is the difference in the detection frequency adjustment this time compared to the previous time. When E > E1, a reward of +10 points; when E2 < E ≤ E1, a reward of +5 points; E1 and E2 are preset detection frequency adjustment evaluation thresholds;

[0035] False alarm penalty: For reporting a non-existent problem, a penalty of -15 points, and for failing to detect an actually existing problem, a penalty of -30 points;

[0036] Device utilization reward: If the average non-idle rate of all available detection devices and engineers on a working day exceeds 70%, a reward of +5 points; the time discount factor γ is set to 0.98.

[0037] Preferably, the Actor-Critic network model is trained through cumulative rewards, and the process includes:

[0038] Step 1: Initialize the network parameters of the Actor network and the Critic network; the parameters include but are not limited to the dimension of the state input layer of the Actor network, the number and size of the hidden layers, the dimension of the action output layer, the dimension of the state input layer of the Critic network, the number and size of the hidden layers, and the dimension of the reward value function output layer, the learning rate, the activation function, and the network optimization algorithm; at the beginning of each training loop, reset the transformer to the initial state;

[0039] Step 2: Make the Actor network output the action space according to the current state;

[0040] Step 3: Execute the action space, obtain immediate rewards, and update the state space;

[0041] Step 4: Use the Critic network to estimate the state value function of the current state and calculate the time difference error; the time difference error reflects the accuracy of the current state value function estimation; use the time difference error to update the Critic network to minimize the estimation error of the state value function;

[0042] Step 5: Use the policy gradient theorem to update the Actor network to maximize the expected cumulative reward.

[0043] Repeat steps 2 to 5 until the termination condition is met to complete the anomaly identification strategy.

[0044] A system for identifying abnormalities of substation equipment based on machine learning, the system comprising:

[0045] Data collection module, collects transformer insulation parameters; transformer insulation parameters include: gas content in transformer oil and dielectric loss factor; collects transformer electrical parameters, transformer temperature data, equipment last maintenance time and equipment historical maintenance frequency; transformer temperature data Where: T is the transformer temperature, f is the transformer vibration frequency, a is the transformer vibration acceleration;

[0046] The model training module uses transformer insulation parameters, transformer temperature data and transformer electrical parameters as equipment parameters; a machine learning model that outputs equipment health scores based on equipment parameter training;

[0047] The reinforcement learning module defines the state space and action space; the state space includes the equipment state and resource state; the equipment state includes the equipment health score, the time of the last equipment maintenance and the equipment historical failure rate; the resource state includes: available testing equipment, human resources and time budget; the action space includes: selecting the next equipment to be tested; the number of testing personnel to be called and the corresponding level; determining the idle testing equipment to be used; determining the degree of testing; adjusting the testing frequency;

[0048] Reinforcement learning reward module; set reward function and cumulative reward, reward function r_t = detection completion reward + health score improvement reward + anomaly detection reward + fault prevention reward - resource usage penalty + time management reward + detection frequency adjustment reward - false alarm penalty + equipment utilization reward; time step t means that in reinforcement learning, the number of consecutive steps from the initial state is t; cumulative reward R_t = r_t + γ*r_{t+1} + γ^2*r_{t+2} + ... + γ^(Tt)*r_T; where: R_t is the discounted sum of all instant rewards from the current time step t to the end time step T, r_{t+1} is the instant reward value at time step t+1; r_{t+2} is the instant reward value at time step t+2;

[0049] The reinforcement learning training module trains the reinforcement learning model through the Actor-Critic algorithm to complete the anomaly recognition strategy, and applies the trained anomaly recognition strategy to the actual abnormality recognition and detection of substation equipment.

[0050] An electronic device comprises: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the above-mentioned method for identifying abnormalities of substation equipment based on machine learning by calling the computer program stored in the memory.

[0051] A computer-readable storage medium stores instructions. When the instructions are executed on a computer, the computer executes the above-mentioned method for identifying abnormalities of substation equipment based on machine learning.

[0052] The technical effects and advantages of the present invention's abnormal identification system and identification method for substation equipment based on machine learning are as follows: by collecting insulation parameters, temperature data and electrical parameters of the transformer, combining machine learning models to evaluate the health status of the equipment, and using the Actor-Critic algorithm to train a reinforcement learning model, the model considers factors such as equipment status and resource status, and makes decisions in an action space including equipment selection, detection level, personnel scheduling and frequency adjustment to balance detection effect, resource utilization and cost control. It can timely detect equipment abnormalities, prevent possible failures, reduce unnecessary inspections, optimize resource use, balance the frequency and depth of inspections, neither over-inspect nor neglect, effectively allocate resources, and contribute to social energy conservation and emission reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of a method for identifying abnormalities of substation equipment based on machine learning according to the present invention;

[0054] Figure 2 A schematic diagram of a system for identifying abnormalities of substation equipment based on machine learning according to the present invention;

[0055] Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application;

[0056] Figure 4 is a schematic diagram of the structure of a computer-readable storage medium provided by an embodiment of the present application;

[0057] Figure 5 This is a schematic diagram of reinforcement learning logic provided by an embodiment of the present application. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] Example 1

[0060] See also Figure 1 As shown, the method for identifying abnormality of substation equipment based on machine learning in this embodiment includes:

[0061] Collect transformer insulation parameters;

[0062] Transformer insulation parameters include: gas content and dielectric loss factor in transformer oil;

[0063] The gas content in transformer oil is obtained by the staff through ultrasonic detection or infrared spectroscopy;

[0064] Infrared spectroscopy uses the infrared absorption characteristics of gas molecules to detect the gas components in oil samples through an infrared spectrometer. Ultrasonic detection uses the fact that the propagation speed of ultrasound in oil is related to the dissolved gas content in the oil, and estimates the gas content by measuring the propagation speed of ultrasound.

[0065] Transformer is a kind of substation equipment, which plays a very important role in the power system. The main function of transformer is voltage conversion, that is, converting AC voltage from one level to another through the principle of electromagnetic induction;

[0066] Transformer oil, also known as insulating oil or power oil, is an oil specially used for insulation and cooling in power systems; transformer oil can be used as a medium to dissolve and transport gases and moisture generated inside the transformer. These dissolved gases and moisture can be used to monitor the status of the transformer through oil chromatography analysis.

[0067] The dielectric loss factor is obtained by the reverse connection method of the AC bridge;

[0068] It should be noted that the transformer insulation parameters are not available online in real time. They are updated synchronously and uploaded to the cloud after each measurement by the staff. When the transformer insulation parameters are not updated, the most recent measurement result is selected as the transformer insulation parameter.

[0069] Changes in the type and content of dissolved gas in the oil may indicate partial discharge, overheating and other faults inside the transformer; while an increase in dielectric loss factor may indicate aging or contamination of the insulation material. By regularly testing and analyzing the insulation parameters of the transformer, potential problems of the transformer can be discovered in time, so that corresponding maintenance measures can be taken to ensure the safe operation of the power system.

[0070] Collect transformer temperature data; Transformer temperature data Where: T is the transformer temperature, f is the transformer vibration frequency, and a is the transformer vibration acceleration; the transformer temperature data reflects the temperature and vibration level of the transformer. The higher the transformer temperature data, the higher the transformer temperature, the stronger the vibration state, and the worse the safety condition of the transformer. The transformer temperature is obtained in real time by the temperature sensor, and the transformer vibration frequency and transformer vibration acceleration are obtained in real time by the acceleration sensor;

[0071] Collect transformer electrical parameters, which include the current and voltage of the transformer; specifically, they are obtained from a current transformer and a voltage transformer respectively.

[0072] The transformer insulation parameters, transformer temperature data and transformer electrical parameters are used as equipment parameters; a machine learning model that outputs equipment health scores is trained based on equipment parameters;

[0073] The training process of the machine learning model includes:

[0074] A set of equipment parameters is converted into a feature vector, and the equipment health score is used as the label corresponding to the feature vector. The equipment health score is obtained by expert evaluation of the substation equipment, and the equipment health score is a positive integer of 0-100; the higher the equipment health score, the higher the safety level of the equipment and the lower the probability of abnormality; each set of feature vectors and the label corresponding to each set of feature vectors are constructed as a sample, and multiple samples are collected to construct a data set; the data set is divided into a training set, a validation set and a test set, wherein the training set accounts for 70% of the data set, the validation set accounts for 20% of the data set, and the test set accounts for 10% of the data set.

[0075] The training set is used as the input of the machine learning model, and the machine learning model uses the real-time device health score as the output; the device health score corresponding to a set of real-time feature vectors is used as the prediction target, and minimizing the loss function value of the machine learning model is used as the training target; when the loss function value of the machine learning model is less than or equal to the preset target loss value, the training is stopped;

[0076] The machine learning model loss function is the mean square error; the mean square error is calculated by Minimize to train the model; in the loss function, MSE is the loss function value, i is the feature vector group number; u is the number of feature vector groups; y i is the label corresponding to the i-th group of feature vectors, The device health score predicted for the i-th group of feature vectors;

[0077] The machine learning model is a random forest model;

[0078] Collect the last maintenance time of the equipment; the last maintenance time of the equipment is the specific time of the equipment's most recent maintenance to the current time, accurate to the minute; the last maintenance time of the equipment is uploaded to the cloud by the engineer during maintenance.

[0079] Collect the historical maintenance frequency of the equipment; the historical maintenance frequency of the equipment is the number of maintenance times within 1000 hours before the current time; obtained from the historical maintenance database of the equipment;

[0080] Define the state space; the state space includes the device state and resource state;

[0081] In reinforcement learning, state space refers to the set of all possible states, that is, multiple groups of state vectors constitute the state space. The logical diagram of reinforcement learning is as follows Figure 5 .

[0082] Equipment status includes equipment health score, the last maintenance time of the collected equipment and the equipment historical failure rate;

[0083] Resource status includes:

[0084] Available testing equipment, which is presented in a list format, including the quantity and status of each testing equipment (idle / occupied, idle is represented by 0, occupied is represented by 1); the testing equipment is the testing equipment used to inspect the transformer, including but not limited to partial discharge detectors, etc.

[0085] Human resources, the number of idle testing personnel classified by skill level; for example, there is 1 junior testing engineer, none of whom are idle; there are 2 intermediate testing engineers, 1 of whom is idle; there are 2 senior testing engineers, none of whom are idle;

[0086] Time budget, which is the number of working hours remaining in the current working day;

[0087] Resource status data is obtained in real time by the enterprise resource planning system and the human resource management system.

[0088] Define the action space; the action space defines all possible actions that the agent can take in any given state in reinforcement learning. The action space includes:

[0089] Select the next device to be tested, the number of testers to be called, and the corresponding level;

[0090] Decide on the idle detection device to use;

[0091] Determine the level of testing; testing levels include rapid inspection, routine inspection and in-depth inspection;

[0092] A quick check is a visual inspection or quick measurement using simple tools. It takes a short time but may miss some potential problems.

[0093] Routine inspections follow standard testing procedures and use conventional equipment to detect most common problems.

[0094] In-depth inspection uses advanced equipment to conduct a comprehensive inspection, including oil sample analysis, partial discharge testing, etc., which can detect potential serious problems, but it takes a long time and is costly.

[0095] Adjust the inspection frequency; the range is [-30%, +30%], which indicates the adjustment ratio relative to the historical inspection frequency of the current equipment.

[0096] The reward function defines the goal of the agent in reinforcement learning;

[0097] The reward function design process includes:

[0098] Detection completion reward: Each time a device detection is completed, the reward is +5 points. This encourages the agent to actively perform device detection.

[0099] Health score improvement reward, for every 1 point improvement in the equipment health score, an additional reward of +2 points. This encourages the agent to choose detection methods that can effectively improve the equipment condition.

[0100] Anomaly detection reward: +10 points for successfully detecting anomalies. This encourages the agent to check carefully and not miss potential problems.

[0101] Failure Prevention Reward: +50 points if a problem that could lead to a major failure is detected and resolved. This incentivizes the agent to focus on important potential risks.

[0102] Resource usage penalty: For using quick inspection, the penalty is -1 point; for using regular inspection, the penalty is -3 points; for using in-depth inspection, the penalty is -5 points. Meanwhile, the penalty coefficient for junior inspection engineers is 1; for intermediate inspection engineers is 1.5; for senior inspection engineers is 3. Multiply the inspection level used by the penalty coefficient corresponding to the level of the engineer called as the resource usage penalty. When there are multiple engineers called, the resource usage penalty is accumulated for each engineer in turn.

[0103] This encourages the agent to use more in-depth detection methods only when necessary and avoid overusing resources.

[0104] Time management reward: If all planned inspections are completed before the end of the working day, the reward is +7 points. This encourages the agent to manage time effectively.

[0105] Detection frequency adjustment reward: Set the detection frequency adjustment evaluation value E = δH / δF; where δH is the change in the device health score this time compared to the last time, and δF is the difference in the detection frequency adjustment this time compared to the last time. The smaller δF and the larger δH, the greater the detection frequency adjustment evaluation value, the more meaningful the detection frequency adjustment, and the more obvious the effect, and vice versa. When E > E1, the reward is +10 points; when E2 < E ≤ E1, the reward is +5 points. This encourages the agent to find the optimal detection frequency. E1 and E2 are preset detection frequency adjustment evaluation thresholds.

[0106] False alarm penalty: For reporting non-existent problems, the penalty is -15 points; for failing to detect actual existing problems, the penalty is -30 points. This encourages the agent to improve the accuracy of detection.

[0107] Device utilization reward: If the average non-idle rate of all available detection devices and engineers per working day exceeds 70%, the reward is +5 points. This encourages the agent to efficiently utilize all available resources.

[0108] Set the time discount factor γ to 0.98 to balance immediate rewards and long-term benefits.

[0109] Reward function r_t = detection completion reward + health score improvement reward + anomaly detection reward + fault prevention reward - resource usage penalty + time management reward + detection frequency adjustment reward - false alarm penalty + device utilization reward; The reward function is a function for calculating the immediate reward value, which maps the environmental state and the agent's actions to a scalar reward value. The time step t refers to the number of consecutive steps passed from the initial state in reinforcement learning as t.

[0110] Cumulative reward R_t = r_t + γ * r_{t + 1} + γ^2 * r_{t + 2} +... + γ^(T - t) * r_T; In the formula:

[0111] R_t is the discounted sum of all immediate rewards from the current time step t to the end time step T, r_{t+1} is the immediate reward value at time step t+1; r_{t+2} and so on.

[0112] The reinforcement learning model is trained through the Actor-Critic algorithm to complete the anomaly identification strategy.

[0113] Use the Actor network in the Actor-Critic network model to output the recommended action space and train the Actor-Critic network model through cumulative rewards. The process includes:

[0114] Step 1: Initialize the network parameters of the Actor network and the Critic network; the parameters include but are not limited to the dimensions of the state input layer of the Actor network, the number and size of hidden layers, the dimension of the action output layer, the dimensions of the state input layer of the Critic network, the number and size of hidden layers, and the dimension of the reward value function output layer, the learning rate, the activation function, and the network optimization algorithm (gradient descent method or Adam optimization algorithm, etc.); at the beginning of each training cycle, reset the transformer to the initial state;

[0115] Step 2: Make the Actor network output the action space according to the current state;

[0116] Step 3: Execute the action, obtain the immediate reward, and update the state space;

[0117] Step 4: Use the Critic network to estimate the state value function of the current state and calculate the time difference error; the time difference error reflects the accuracy of the current state value function estimation. Use the time difference error to update the Critic network to minimize the estimation error of the state value function.

[0118] Step 5: Use the policy gradient theorem to update the Actor network to maximize the expected cumulative reward.

[0119] Repeat steps 2 to 5 until the termination condition is met to complete the anomaly identification strategy (such as reaching the maximum cumulative reward).

[0120] The trained anomaly recognition strategy is applied to the actual anomaly recognition and detection of substation equipment.

[0121] Example 1 collects insulation parameters, temperature data and electrical parameters of the transformer and combines them with a machine learning model to assess the health of the equipment. Then the Actor-Critic algorithm is used to train a reinforcement learning model, which considers factors such as equipment status and resource status, and makes decisions in an action space that includes equipment selection, detection level, personnel scheduling and frequency adjustment to balance detection effect, resource utilization and cost control.

[0122] Example 1 realizes timely detection of equipment abnormalities, prevention of possible failures, reduction of unnecessary inspections, optimization of resource usage, balance of inspection frequency and depth, neither excessive inspection nor negligence, effective allocation of resources, and contributes to energy conservation and emission reduction for society.

[0123] Example 2

[0124] See also Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description of Embodiment 1. A system and method for identifying abnormalities of substation equipment based on machine learning are provided, and the system includes: each module is connected by a wired and / or wireless network.

[0125] Example 3

[0126] See also Figure 3 As shown, according to another aspect of the present application, an electronic device 500 for identifying abnormality of substation equipment based on machine learning and an identification method is also provided. The electronic device 500 may include one or more processors and one or more memories. Among them, the memory stores a computer readable code, and when the computer readable code is executed by one or more processors, it can execute the above-mentioned abnormality identification system and identification method of substation equipment based on machine learning.

[0127] The method or system according to the embodiment of the present application can also be used by Figure 3 The electronic device architecture shown in FIG. Figure 3 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, and the like. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store a substation abnormality identification system and identification method based on machine learning provided in the present application. The path planning method may, for example, include: obtaining environmental information of the current position, constructing a local map corresponding to the current position based on the environmental information, the local map including multiple candidate nodes; obtaining a reference path from the starting position to the target position; determining the sub-target position in the local map based on the reference path; screening the candidate nodes based on the reference path and the sub-target position to obtain the path node; and planning the target path pointing to the target position based on the path node. Furthermore, the electronic device 500 may also include a user interface 508. Of course, Figure 3 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 3 One or more components of an electronic device are shown.

[0128] Example 4

[0129] See also Figure 4 As shown, a computer-readable storage medium 600 according to an embodiment of the present application is shown. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are executed by the processor, a substation abnormality identification system and identification method based on machine learning according to an embodiment of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0130] In addition, according to the implementation of the present application, the process described in the above reference flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application, for example: obtaining environmental information of the current position, constructing a local map corresponding to the current position according to the environmental information, and the local map includes multiple candidate nodes; obtaining a reference path from the starting position to the target position; determining the sub-target position in the local map according to the reference path; screening the candidate nodes based on the reference path and the sub-target position to obtain the path node; planning the target path pointing to the target position according to the path node. When the computer program is executed by the central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0131] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through a wired network or a wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0132] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0133] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0134] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a substation abnormality identification system and identification method based on machine learning. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0135] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0136] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0137] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0138] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying abnormalities in substation equipment based on machine learning, characterized in that: include: Collect transformer insulation parameters; transformer insulation parameters include: gas content in transformer oil and dielectric loss factor; collect transformer electrical parameters, transformer temperature data, equipment last maintenance time and equipment historical maintenance frequency; transformer temperature data Where: T is the transformer temperature, f is the transformer vibration frequency, a is the transformer vibration acceleration; The transformer insulation parameters, transformer temperature data and transformer electrical parameters are used as equipment parameters; a machine learning model that outputs equipment health scores is trained based on equipment parameters; Define state space and action space; the state space includes device state and resource state; device state includes device health score, device last maintenance time and device historical failure rate; resource state includes available testing equipment, human resources and time budget; the action space includes: selecting the next device to be tested, selecting the number of testing personnel to be called and the corresponding skill level classification, determining the idle testing equipment to be used, determining the degree of testing and adjusting the testing frequency; Set the reward function and cumulative reward, reward function r_t = detection completion reward + health score improvement reward + anomaly detection reward + fault prevention reward - resource usage penalty + time management reward + detection frequency adjustment reward - false alarm penalty + equipment utilization reward; time step t refers to the number of consecutive steps from the initial state in reinforcement learning; cumulative reward R_t = r_t + γ*r_{t+1} + γ^2*r_{t+2} + ... + γ^(Tt)*r_T; where: R_t is the discounted sum of all instant rewards from the current time step t to the end time step T, r_{t+1} is the instant reward value at time step t+1; r_{t+2} is the instant reward value at time step t+2; the time discount factor γ is set to 0.98; The reinforcement learning model is trained through the Actor-Critic algorithm to complete the anomaly recognition strategy, and the trained anomaly recognition strategy is applied to the actual anomaly recognition and detection of substation equipment.

2. The method for identifying abnormality of substation equipment based on machine learning according to claim 1, characterized in that: The transformer insulation parameters are updated and uploaded to the cloud synchronously after each measurement by the staff. When the transformer insulation parameters are not updated, the most recent measurement result is selected as the transformer insulation parameter; The transformer electrical parameters include the current and voltage of the transformer; The historical maintenance frequency of the equipment is the number of maintenance times within 1000 hours before the current time; the last maintenance time of the equipment is the specific time of the most recent maintenance of the equipment from the current time, accurate to the minute; Available testing equipment is presented in a list, including the number of each testing equipment and the idle / occupied status; the testing equipment is the testing equipment used to inspect the transformer; human resources include the idle number of testing personnel classified by skill level; Skill level classification includes junior testing engineer, intermediate testing engineer and senior testing engineer; The time budget is the number of working hours remaining in the current working day.

3. The method for identifying abnormality of substation equipment based on machine learning according to claim 2 is characterized in that: The training process of the machine learning model includes: A set of device parameters is converted into a feature vector, and the device health score is used as the label corresponding to the feature vector, and the device health score is a positive integer between 0 and 100; each set of feature vectors and the label corresponding to each set of feature vectors are constructed as a sample, and multiple samples are collected to construct a data set; the data set is divided into a training set, a validation set and a test set, wherein the training set accounts for 70% of the data set, the validation set accounts for 20% of the data set, and the test set accounts for 10% of the data set; Use the training set as the input of the machine learning model, and the machine learning model outputs the real-time device health score; use the device health score corresponding to a set of real-time feature vectors as the prediction target, and minimize the loss function value of the machine learning model as the training target; stop training when the loss function value of the machine learning model is less than or equal to the preset target loss value; The machine learning model loss function is the mean square error; the mean square error is calculated by Minimize to train the model; in the loss function, MSE is the loss function value, i is the feature vector group number; u is the number of feature vector groups; y i is the label corresponding to the i-th group of feature vectors, The device health score predicted for the i-th group of feature vectors; The machine learning model is a random forest model.

4. The method for identifying abnormality of substation equipment based on machine learning according to claim 3 is characterized in that: The detection levels include quick inspection, regular inspection and in-depth inspection; quick inspection is visual inspection or quick measurement using simple tools; regular inspection uses regular equipment for standard detection processes, and in-depth inspection uses professional equipment for comprehensive inspection; the detection frequency adjustment range is [-30%, +30%], indicating the adjustment ratio relative to the historical maintenance frequency of the current device.

5. The method for identifying abnormality of substation equipment based on machine learning according to claim 4 is characterized in that: The design process of the reward function includes: Reward for detection completion, +5 points are rewarded for each device detection completed; Reward for health score improvement, an additional +2 points are rewarded for every 1-point increase in the device health score; Reward for anomaly detection, +10 points are rewarded for successfully detecting an abnormal situation; Reward for fault prevention, if a problem that may lead to a major fault is detected and resolved, +50 points are rewarded; Resource usage penalty, -1 point is penalized for using quick inspection; -3 points are penalized for using regular inspection; -5 points are penalized for using in-depth inspection; at the same time, the penalty coefficient for junior inspection engineers is 1; the penalty coefficient for intermediate inspection engineers is 1.5; the penalty coefficient for senior inspection engineers is 3; multiply the inspection level used by the penalty coefficient corresponding to the level of the engineer called as the resource usage penalty. When there are multiple engineers called, the resource usage penalty is stacked for each engineer in turn; Reward for time management, +7 points are rewarded if all planned detections are completed before the end of the working day; Reward for detection frequency adjustment, set the detection frequency adjustment evaluation value E = δH / δF; where δH is the change in the device health score this time compared to the previous time, and δF is the difference in the detection frequency adjustment this time compared to the previous time. When E > E1, +10 points are rewarded; when E2 < E ≤ E1, +5 points are rewarded; E1 and E2 are preset detection frequency adjustment evaluation thresholds; False alarm penalty, -15 points are penalized for reporting non-existent problems, and -30 points are penalized for failing to detect actual existing problems; Reward for device utilization rate, +5 points are rewarded if the average non-idle rate of all available detection devices and engineers on weekdays exceeds 70%; 6. The method for identifying abnormality of substation equipment based on machine learning according to claim 5, characterized in that: Train the Actor-Critic network model through cumulative rewards. The process includes: Step 1: Initialize the network parameters of the Actor network and the Critic network; the parameters include the dimension of the state input layer of the Actor network, the number and size of the hidden layers, the dimension of the action output layer, the dimension of the state input layer of the Critic network, the number and size of the hidden layers, and the dimension of the reward value function output layer, the learning rate, the activation function, and the network optimization algorithm; at the beginning of each training loop, reset the transformer to the initial state; Step 2: Make the Actor network output the action space according to the current state; Step 3: Execute the action space, obtain immediate rewards, and update the state space; Step 4: Use the Critic network to estimate the state value function of the current state and calculate the time difference error; the time difference error reflects the accuracy of the current state value function estimation; use the time difference error to update the Critic network to minimize the estimation error of the state value function; Step 5: Use the policy gradient theorem to update the Actor network to maximize the expected cumulative reward; Repeat steps 2 to 5 until the termination condition is met to complete the anomaly identification strategy.

7. A system for identifying abnormalities in substation equipment based on machine learning, characterized in that: The system comprises: Data collection module, collects transformer insulation parameters; transformer insulation parameters include: gas content in transformer oil and dielectric loss factor; collects transformer electrical parameters, transformer temperature data, equipment last maintenance time and equipment historical maintenance frequency; transformer temperature data Where: T is the transformer temperature, f is the transformer vibration frequency, a is the transformer vibration acceleration; The model training module uses transformer insulation parameters, transformer temperature data and transformer electrical parameters as equipment parameters; a machine learning model that outputs equipment health scores based on equipment parameter training; The reinforcement learning module defines the state space and action space; the state space includes the equipment state and resource state; the equipment state includes the equipment health score, the time of the last equipment maintenance and the equipment historical failure rate; the resource state includes: available testing equipment, human resources and time budget; the action space includes: selecting the next equipment to be tested, selecting the number of testing personnel to be called and the corresponding skill level classification, determining the idle testing equipment to be used, determining the degree of testing and adjusting the testing frequency; Reinforcement learning reward module, set reward function and cumulative reward, reward function r_t = detection completion reward + health score improvement reward + anomaly detection reward + fault prevention reward - resource usage penalty + time management reward + detection frequency adjustment reward - false alarm penalty + equipment utilization reward; time step t means that in reinforcement learning, the number of consecutive steps from the initial state is t; cumulative reward R_t = r_t + γ*r_{t+1} + γ^2*r_{t+2} + ... + γ^(Tt)*r_T; where: R_t is the discounted sum of all instant rewards from the current time step t to the end time step T, r_{t+1} is the instant reward value when the time step is t+1; r_{t+2} is the instant reward value when the time step is t+2; the time discount factor γ is set to 0.98; The reinforcement training module trains the reinforcement learning model through the Actor-Critic algorithm to complete the anomaly recognition strategy, and applies the trained anomaly recognition strategy to the actual abnormality recognition and detection of substation equipment.

8. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor executes the method for identifying abnormalities of substation equipment based on machine learning as described in any one of claims 1 to 6 by calling the computer program stored in the memory.

9. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer executes a method for identifying abnormalities of substation equipment based on machine learning as described in any one of claims 1 to 6.

Images