Machine learning algorithm-based medium and low voltage distribution network line loss abnormity intelligent diagnosis method

Abstract

The invention discloses a machine learning algorithm-based medium and low voltage power distribution network line loss abnormity intelligent diagnosis method, which belongs to the technical field of medium and low voltage power distribution networks, and comprises the following steps of collecting and preprocessing operation data of a medium and low voltage power distribution network; performing feature extraction and selection on the operating data of the medium-low voltage power distribution network; training a medium and low voltage distribution network line loss abnormity intelligent diagnosis model; and performing abnormality diagnosis on the line loss of the medium-low voltage power distribution network, positioning the abnormal position of the line loss, finding out the cause of the abnormal line loss, and performing intelligent management and control on the abnormal line loss of the medium-low voltage power distribution network. The problems that the existing detection precision is insufficient, the power distribution network line loss cannot be subjected to refined management, and the power grid operation efficiency is reduced are solved. According to the method, the identification precision of the line loss abnormity can be effectively improved, unnecessary energy waste is reduced, powerful support is provided for sustainable development of a power system, fine management of the line loss of the power distribution line can be realized, and the overall operation efficiency of a power grid is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medium and low voltage distribution networks, and specifically to an intelligent diagnosis method for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms. Background Art

[0002] Line loss refers to the electrical energy loss caused by various factors during the power transmission process. This loss is usually inevitable, and its main causes include physical phenomena such as resistance heating, and may also be affected by users' electricity consumption behaviors; in medium and low voltage distribution networks, abnormal line loss means that the actual line loss rate is much higher than the reasonable line loss level. Such abnormal situations will not only significantly increase the operating costs of the power system, but may also reflect problems existing in the technical level or management links of the power grid.

[0003] Among them, physical problems of the distribution network itself such as too long conductors and voltage quality problems; management problems such as technical failures of smart meters; human problems such as illegal electricity use, electricity theft and leakage; and extreme situations such as bad weather will all lead to abnormal line losses in the distribution network. Relying solely on manual statistics is not only time-consuming and laborious, but also the effect will be poor. Traditional manual statistics methods are not only inefficient, but also difficult to meet the complex abnormal diagnosis requirements. For example, manual analysis usually relies on historical experience and simple threshold judgment. This method often seems powerless when faced with a dynamically changing electricity consumption environment and diverse abnormal types. At present, most power companies use the method of setting assessment thresholds based on experience to detect abnormal line losses, but this method has obvious limitations: firstly, the design of the threshold is too dependent on manual experience and lacks scientific basis; secondly, the fixed threshold cannot adapt to the changes in electricity consumption characteristics in different regions and different time periods, resulting in insufficient detection accuracy, and cannot conduct refined management of the line losses in the distribution network, reducing the overall operating efficiency of the power grid. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent diagnosis method for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms, which can effectively improve the recognition accuracy of abnormal line losses, reduce unnecessary energy waste, provide strong support for the sustainable development of the power system, realize refined management of the line losses in the distribution network, and improve the overall operating efficiency of the power grid, and solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] An intelligent diagnosis method for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms, including:

[0007] Collect the operation data of the medium and low voltage distribution network and perform preprocessing;

[0008] Extract the characteristics of the operation data of the medium and low voltage distribution network, determine the influence degree of the current characteristics on the intelligent diagnosis and prediction results of the line loss anomaly of the medium and low voltage distribution network based on the target score value, and determine the operation characteristic data of the medium and low voltage distribution network;

[0009] Based on the machine learning algorithm, construct an intelligent diagnosis model for the line loss anomaly of the medium and low voltage distribution network;

[0010] According to the intelligent diagnosis model of the line loss anomaly of the medium and low voltage distribution network, analyze the operation characteristic data of the medium and low voltage distribution network and diagnose the line loss anomaly. Combine the line topology structure and equipment information to locate the line loss anomaly position and perform intelligent control on the line loss anomaly of the medium and low voltage distribution network.

[0011] Preferably, collect the operation data of the medium and low voltage distribution network and perform the following operations:

[0012] Apply the Internet of Things technology to install intelligent acquisition devices in the medium and low voltage distribution network;

[0013] Based on the intelligent acquisition devices, monitor and collect the voltage conditions in the operation process of the medium and low voltage distribution network in real time to obtain the voltage data of the medium and low voltage distribution network;

[0014] Based on the intelligent acquisition devices, monitor and collect the current conditions in the operation process of the medium and low voltage distribution network in real time to obtain the current data of the medium and low voltage distribution network;

[0015] Based on the intelligent acquisition devices, monitor and collect the power conditions in the operation process of the medium and low voltage distribution network in real time to obtain the power data of the medium and low voltage distribution network;

[0016] Based on the intelligent acquisition devices, monitor and collect the load conditions in the operation process of the medium and low voltage distribution network in real time to obtain the load data of the medium and low voltage distribution network;

[0017] Among them, based on the voltage data, current data, power data and load data of the medium and low voltage distribution network, determine the operation data of the medium and low voltage distribution network based on the Internet of Things.

[0018] Preferably, preprocess the operation data of the medium and low voltage distribution network, including:

[0019] Clean the operation data of the medium and low voltage distribution network based on the Internet of Things;

[0020] Remove the duplicate data, missing values and abnormal values that are useless for the intelligent diagnosis of the line loss anomaly of the medium and low voltage distribution network from the operation data of the medium and low voltage distribution network based on the Internet of Things;

[0021] Among them, for the missing values and abnormal values that are useful for the intelligent diagnosis of the line loss anomaly of the medium and low voltage distribution network, fill the missing values with the median and replace the abnormal values with the average value;

[0022] Normalize the operation data of medium and low voltage distribution networks based on the Internet of Things;

[0023] Convert the operation data of medium and low voltage distribution networks based on the Internet of Things into a unified format, remove the dimensional differences between the operation data of medium and low voltage distribution networks based on the Internet of Things, and determine the standardized operation data of medium and low voltage distribution networks.

[0024] Preferably, perform feature extraction and selection on the operation data of medium and low voltage distribution networks, including:

[0025] Extract static features and dynamic features related to line losses of medium and low voltage distribution networks from the operation data of medium and low voltage distribution networks based on the principal component analysis method;

[0026] Among them, the static features include line parameters, equipment parameters, and topological structure;

[0027] The line parameters include line length, conductor type, cross-sectional area, and resistivity; the equipment parameters include transformer capacity, type, and load rate; the topological structure includes network structure and node connection relationship;

[0028] Among them, the dynamic features include electrical measurement data, load characteristics, environmental factors, and time characteristics;

[0029] The electrical measurement data includes voltage, current, active power, reactive power, and power factor; the load characteristics include load rate, load fluctuation, peak-valley difference, and load curve; the environmental factors include temperature, humidity, and weather conditions; the time characteristics include hour, day, week, month, and season dimension features;

[0030] Select the extracted static features and dynamic features based on the random forest algorithm, evaluate the influence degree of each feature on the intelligent diagnosis and prediction results of abnormal line losses in medium and low voltage distribution networks, screen out the most valuable features for the intelligent diagnosis of abnormal line losses in medium and low voltage distribution networks, and determine the operation feature data of medium and low voltage distribution networks.

[0031] Preferably, the selection of the extracted static features and dynamic features based on the random forest algorithm and the evaluation of the influence degree of each feature on the intelligent diagnosis and prediction results of abnormal line losses in medium and low voltage distribution networks include:

[0032] Establish a feature set based on all dynamic features and static features, obtain the first change amplitude of all dynamic features relative to historical dynamic features, obtain the second change amplitude of all static features relative to historical static features, and use the ratio of the first change amplitude and the second change amplitude as the random sampling ratio for dynamic features and static features;

[0033] Randomly and with replacement extract features from the feature set according to the random sampling ratio to obtain multiple feature groups. Based on each feature group, construct a decision tree in combination with the abnormal wire loss information, and randomly select feature group features to determine the splitting point when splitting the nodes of the decision tree;

[0034] Input each feature in the feature set into each decision tree in turn, and determine the scoring set of the current feature in all decision trees according to the decision results;

[0035] Perform eigenvalue permutation on the current feature to obtain a permuted eigenvalue. Input the permuted eigenvalue into each decision tree in turn to obtain the set of performance degradation amplitudes of all decision trees. Determine the scoring weight of the current feature based on the difference between the average degradation value of the set of performance degradation amplitudes and the preset reference degradation value;

[0036] Perform weighted processing on the average score value determined by the scoring set based on the scoring weight to obtain the target score value of the current feature;

[0037] Determine the influence degree of the current feature on the intelligent diagnosis and prediction result of abnormal wire loss in the medium and low voltage distribution network based on the target score value.

[0038] Preferably, construct an intelligent diagnosis model for abnormal wire loss in the medium and low voltage distribution network and perform the following operations:

[0039] Collect the operation historical data of the medium and low voltage distribution network, and divide the collected operation historical data of the medium and low voltage distribution network to determine the training set and the test set;

[0040] Based on the machine learning algorithm, use the training set to train the machine learning model so that the machine learning model autonomously learns the intelligent diagnosis behavior of abnormal wire loss in the medium and low voltage distribution network, and determine the intelligent diagnosis model for abnormal wire loss in the medium and low voltage distribution network based on the machine learning algorithm;

[0041] Based on the test set, test the intelligent diagnosis model for abnormal wire loss in the medium and low voltage distribution network based on the machine learning algorithm, and evaluate whether the intelligent diagnosis model for abnormal wire loss in the medium and low voltage distribution network based on the machine learning algorithm can achieve the expected effect based on the accuracy rate, recall rate and F1 score;

[0042] According to the evaluation results, adjust the parameters and structure of the intelligent diagnosis model for abnormal wire loss in the medium and low voltage distribution network based on the machine learning algorithm. Through continuous iterative optimization, determine the optimal intelligent diagnosis model for abnormal wire loss in the medium and low voltage distribution network.

[0043] Preferably, analyze the operation characteristic data of the medium and low voltage distribution network, including:

[0044] Deploy the optimal intelligent diagnosis model for abnormal wire loss in the medium and low voltage distribution network to the actual intelligent diagnosis environment of abnormal wire loss in the medium and low voltage distribution network;

[0045] Input the operation characteristic data of the medium- and low-voltage distribution network into the intelligent diagnosis model for abnormal line losses of the medium- and low-voltage distribution network. Analyze the operation characteristic data of the medium- and low-voltage distribution network based on the intelligent diagnosis model for abnormal line losses of the medium- and low-voltage distribution network, and conduct intelligent diagnosis of abnormal line losses of the medium- and low-voltage distribution network to determine the predicted value of the line losses of the medium- and low-voltage distribution network.

[0046] Preferably, the abnormal diagnosis of the line losses of the medium- and low-voltage distribution network includes:

[0047] Compare the predicted value of the line losses with the actual value of the line losses, calculate the relative error, and judge whether there are abnormal line losses in the medium- and low-voltage distribution network based on the relative error;

[0048] Among them,

[0049] When the relative error is within 5%, it is determined that the line losses of the medium- and low-voltage distribution network are in a normal state;

[0050] When the relative error exceeds 5%, it is determined that the line losses of the medium- and low-voltage distribution network are in an abnormal state, trigger an alarm for abnormal line losses, and locate the position of the abnormal line losses;

[0051] Obtain the topological structure information and equipment information of the lines in the medium- and low-voltage distribution network. Through the topological structure information and equipment information, gradually narrow down the abnormal range and locate it to a specific line segment or equipment;

[0052] Among them, obtaining the topological structure information of the lines in the medium- and low-voltage distribution network includes the connection relationship of the lines, the length of the lines, the conductor type, the impedance parameters, and the position and status of the equipment;

[0053] Conduct zonal analysis on the medium- and low-voltage distribution network, divide the medium- and low-voltage distribution network into several regions, analyze the abnormal line losses in each region one by one, calculate the predicted value and the actual value of the line losses in each region, and find the region with the largest deviation;

[0054] Check the equipment status in the region with abnormal line losses, including: whether the transformer is overloaded or faulty, whether the switch is normally closed or open, whether the capacitor is working properly, and judge whether there are abnormalities in the equipment through the equipment monitoring data;

[0055] Analyze the line parameters, check the line parameters in the region with abnormal line losses, including: whether the line impedance is abnormal, whether there are problems of three-phase imbalance or harmonics, and analyze the operation status of the line through electrical measurement data;

[0056] If both the equipment and the line parameters are normal, there may be electricity theft behavior. Among them, compare the difference between the user's electricity consumption and the historical data, use the electricity theft detection algorithm, and combine on-site inspections to check whether the user's electricity meter has been tampered with;

[0057] By using topological structure information and device information, gradually narrow down the scope of anomalies, and locate the specific line segment or device. When the line loss of a certain feeder is abnormal, analyze each branch under this feeder. When the line loss of a certain transformer is abnormal, check the power consumption of users under this transformer.

[0058] Preferably, intelligent control of abnormal line losses in medium and low voltage distribution networks includes:

[0059] Analyze the intelligent diagnosis results of abnormal line losses in medium and low voltage distribution networks, and combine historical data and on-site inspections to find out the reasons for abnormal line losses, including equipment failures, line problems, power theft behaviors, operation problems, and management problems. According to the reasons for abnormal line losses, formulate intelligent control measures for reducing line losses in medium and low voltage distribution networks, and conduct intelligent control of abnormal line losses in medium and low voltage distribution networks;

[0060] Among them, for equipment failures, replace or repair faulty equipment, upgrade equipment, and conduct regular inspection and maintenance, including replacing overloaded transformers, repairing switches with poor contacts, using energy-efficient equipment, and establishing an equipment inspection system to detect and handle potential failures in a timely manner;

[0061] Among them, for line problems, replace aging lines, optimize line layouts, and strengthen joint maintenance, including using wires with high conductivity and low loss, reducing line lengths, reducing line impedance, and regularly checking line joints to ensure good contact of line joints;

[0062] Among them, for power theft behaviors, strengthen power consumption inspections, use power theft detection technologies and legal means, including real-time monitoring of users' power consumption behaviors through smart meters and power consumption information collection systems, using machine learning algorithms to detect abnormal power consumption behaviors, and severely cracking down on power theft behaviors to increase the cost of violations;

[0063] Among them, for operation problems, adjust three-phase balance, control harmonics, and manage loads, including reducing three-phase imbalance through load adjustment or installing three-phase balance devices, installing filters or reactive power compensation devices to reduce harmonic pollution, and optimizing load distribution to avoid local overload;

[0064] Among them, for management problems, improve the accuracy of data collection and transmission, reduce measurement errors, establish a line loss assessment mechanism to encourage operation and maintenance personnel to reduce losses and increase efficiency, and use technologies such as big data and artificial intelligence to achieve real-time monitoring and intelligent diagnosis of line losses.

[0065] Preferably, according to the evaluation results, adjust the parameters and structure of the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms, including:

[0066] Determine the types and distributions of abnormal line losses from the evaluation results, and determine the abnormal weights of the abnormal types based on the types and distributions of abnormal line losses;

[0067] Respectively obtain relevant historical operation data related to the abnormal type from the historical operation data of the power grid operation, and determine the data importance of the current relevant historical operation data based on the abnormal weight of the abnormal type;

[0068] Based on the data importance of each relevant historical operation data, perform weighted data extraction on the historical operation data of the power grid operation to obtain a new training set and a new validation set;

[0069] Determine the mean square error, error rate, and F1 score for the evaluation of abnormal line losses from the evaluation results, and determine the adjustment coefficient for the weights of the model structure based on the mean square error, error rate, and F1 score;

[0070] Determine the latest weights of the model structure based on the adjustment coefficient;

[0071] Based on the new training set, new validation set, and latest weights, retrain and adjust the intelligent diagnosis model for abnormal line losses in the low-voltage distribution network to obtain the latest intelligent diagnosis model for abnormal line losses.

[0072] Compared with the prior art, the beneficial effects of the present invention are:

[0073] The present invention collects voltage data, current data, power data, and load data during the operation of the medium- and low-voltage distribution network, determines the operation data of the medium- and low-voltage distribution network, preprocesses the operation data of the medium- and low-voltage distribution network, extracts features from the operation data of the medium- and low-voltage distribution network, extracts feature vectors related to the line losses of the medium- and low-voltage distribution network, selects the extracted feature vectors based on the random forest algorithm to determine the operation feature data of the medium- and low-voltage distribution network, trains a machine learning model using the historical operation data of the medium- and low-voltage distribution network based on the machine learning algorithm, optimizes the model parameters, determines the optimal intelligent diagnosis model for abnormal line losses in the medium- and low-voltage distribution network, analyzes the operation feature data of the medium- and low-voltage distribution network according to the intelligent diagnosis model for abnormal line losses in the medium- and low-voltage distribution network, diagnoses abnormal line losses in the medium- and low-voltage distribution network, combines the line topology structure and equipment information to locate the abnormal line loss position, find out the reasons for abnormal line losses, formulate intelligent control measures for abnormal line losses, and perform intelligent control on abnormal line losses in the medium- and low-voltage distribution network to reduce the line losses in the medium- and low-voltage distribution network, which can effectively improve the recognition accuracy of abnormal line losses, not only helps to optimize the operation efficiency of the power system, but also can reduce unnecessary energy waste, thus providing strong support for the sustainable development of the power system, and can achieve refined management of the line losses of the distribution network and improve the overall operation efficiency of the power grid. Description of the Drawings

[0074] Figure 1This is the flowchart of the intelligent diagnosis method for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms of the present invention. Specific embodiments

[0075] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0076] To solve the problems of insufficient existing detection accuracy, inability to perform refined management of distribution network line losses, and reduction of the overall operation efficiency of the power grid, please refer to Figure 1 , the following technical solutions are provided in this embodiment:

[0077] An intelligent diagnosis method for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms includes the following steps:

[0078] S1. Data collection and preprocessing: Collect voltage data, current data, power data, and load data during the operation of the medium and low voltage distribution network, determine the operation data of the medium and low voltage distribution network, and preprocess the operation data of the medium and low voltage distribution network;

[0079] In this embodiment, to determine the operation data of the medium and low voltage distribution network, the following operations are performed:

[0080] Apply Internet of Things technology to install intelligent collection devices in the medium and low voltage distribution network;

[0081] Based on the intelligent collection devices, monitor and collect the voltage conditions during the operation of the medium and low voltage distribution network in real time to obtain the voltage data of the medium and low voltage distribution network;

[0082] Based on the intelligent collection devices, monitor and collect the current conditions during the operation of the medium and low voltage distribution network in real time to obtain the current data of the medium and low voltage distribution network;

[0083] Based on the intelligent collection devices, monitor and collect the power conditions during the operation of the medium and low voltage distribution network in real time to obtain the power data of the medium and low voltage distribution network;

[0084] Based on the intelligent collection devices, monitor and collect the load conditions during the operation of the medium and low voltage distribution network in real time to obtain the load data of the medium and low voltage distribution network;

[0085] Among them, based on the voltage data of the medium and low voltage distribution network, the current data of the medium and low voltage distribution network, the power data of the medium and low voltage distribution network, and the load data of the medium and low voltage distribution network, determine the operation data of the medium and low voltage distribution network based on the Internet of Things.

[0086] In this embodiment, preprocessing the operation data of the medium and low voltage distribution network includes:

[0087] Cleaning the operation data of the medium and low voltage distribution network based on the Internet of Things;

[0088] Removing duplicate data, missing values, and outliers in the operation data of the medium and low voltage distribution network based on the Internet of Things that are useless for the intelligent diagnosis of abnormal line losses in the medium and low voltage distribution network;

[0089] Among them, for the missing values and outliers useful for the intelligent diagnosis of abnormal line losses in the medium and low voltage distribution network, the median is used to fill the missing values and the average value is used to replace the outliers;

[0090] Normalizing the operation data of the medium and low voltage distribution network based on the Internet of Things;

[0091] Converting the operation data of the medium and low voltage distribution network based on the Internet of Things into a unified format, removing the dimensional differences between the operation data of the medium and low voltage distribution network based on the Internet of Things, and determining the standardized operation data of the medium and low voltage distribution network.

[0092] S2. Feature extraction and selection: Extract features from the operation data of the medium and low voltage distribution network, extract feature vectors related to the line losses of the medium and low voltage distribution network, and select the extracted feature vectors based on the random forest algorithm to determine the operation feature data of the medium and low voltage distribution network;

[0093] In this embodiment, feature extraction and selection of the operation data of the medium and low voltage distribution network include:

[0094] Extracting static features and dynamic features related to the line losses of the medium and low voltage distribution network from the operation data of the medium and low voltage distribution network based on the principal component analysis method;

[0095] Among them, the static features include line parameters, equipment parameters, and topological structure;

[0096] The line parameters include line length, conductor type, cross-sectional area, and resistivity; the equipment parameters include transformer capacity, type, and load rate; the topological structure includes network structure and node connection relationship;

[0097] Among them, the dynamic features include electrical measurement data, load characteristics, environmental factors, and time features;

[0098] The electrical measurement data includes voltage, current, active power, reactive power, and power factor; the load characteristics include load rate, load fluctuation, peak-valley difference, and load curve; the environmental factors include temperature, humidity, and weather conditions; the time features include hour, day, week, month, and season dimension features;

[0099] Select the extracted static features and dynamic features based on the random forest algorithm, evaluate the influence degree of each feature on the intelligent diagnosis and prediction results of abnormal line losses in medium and low voltage distribution networks, screen out the features most valuable for the intelligent diagnosis of abnormal line losses in medium and low voltage distribution networks, and determine the operation characteristic data of medium and low voltage distribution networks.

[0100] In one embodiment, the selection of the extracted static features and dynamic features based on the random forest algorithm and the evaluation of the influence degree of each feature on the intelligent diagnosis and prediction results of abnormal line losses in medium and low voltage distribution networks include:

[0101] Establish a feature set based on all dynamic features and static features, obtain the first change range of all dynamic features relative to historical dynamic features, obtain the second change range of all static features relative to historical static features, and use the ratio of the first change range and the second change range as the random sampling ratio for dynamic features and static features;

[0102] Randomly draw features from the feature set with replacement according to the random sampling ratio to obtain multiple feature groups. Based on each feature group, combine the abnormal line loss information of the wire network to construct a decision tree, and randomly select the features of the feature group to determine the splitting point when splitting the nodes of the decision tree;

[0103] Input each feature in the feature set into each decision tree in turn, and determine the scoring set of the current feature in all decision trees according to the decision results;

[0104] Perform eigenvalue replacement on the current feature to obtain replacement eigenvalues, input the replacement eigenvalues into each decision tree in turn, obtain the set of performance degradation amplitudes of all decision trees, and determine the scoring weight of the current feature based on the difference between the average degradation value of the set of performance degradation amplitudes and the preset reference degradation value;

[0105] Perform weighted processing on the average score value determined by the scoring set based on the scoring weight to obtain the target score value of the current feature;

[0106] Determine the influence degree of the current feature on the intelligent diagnosis and prediction results of abnormal line losses in medium and low voltage distribution networks based on the target score value.

[0107] In this embodiment, the larger the target score value, the greater the influence degree.

[0108] In this embodiment, randomly selecting the features of the feature group to determine the splitting point when splitting the nodes of the decision tree can increase the independence of the decision tree.

[0109] In this embodiment, based on the ratio of the first change range and the second change range, as the random sampling ratio of dynamic features and static features, the rationality of the random extraction quantity allocation of dynamic features and static features is ensured, providing a basis for ensuring the accuracy of the decision tree.

[0110] In this embodiment, when the average decrease value is less than the preset reference decrease value, the scoring weight is less than 1; when the average decrease value is greater than the preset reference decrease value, the scoring weight is greater than 1, and the greater the difference, the greater the scoring weight.

[0111] The beneficial effects of the above design are as follows: By establishing a feature set based on all dynamic features and static features, obtaining the first change range of all dynamic features relative to historical dynamic features, obtaining the second change range of all static features relative to historical static features, using the ratio of the first change range and the second change range as the random sampling ratio of dynamic features and static features to ensure the rationality of the random extraction quantity allocation of features, providing a basis for ensuring the accuracy of the decision tree. Based on each feature group, a decision tree is constructed by combining the abnormal wire loss information, and when splitting the nodes of the decision tree, feature group features are randomly selected to determine the splitting point, increasing the independence of the decision tree and providing a basis for the multi-faceted scoring of features. Each feature in the feature set is sequentially input into each decision tree, and the scoring set of the current feature in all decision trees is determined according to the decision results; the feature value of the current feature is replaced to obtain a replacement feature value, and the replacement feature value is sequentially input into each decision tree to obtain a set of performance degradation ranges of all decision trees. Based on the difference between the average decrease value of the performance degradation range set and the preset reference decrease value, the scoring weight of the current feature is determined; based on the scoring weight, the average scoring value determined by the scoring set is weighted to obtain the target scoring value of the current feature. Through the analysis and fusion of the two factors before and after replacement, it is ensured that the obtained target scoring value of the corresponding feature can more accurately reflect the influence degree on the intelligent diagnosis and prediction results of the abnormal wire loss of the medium and low voltage distribution network.

[0112] S3. Model construction and training: Based on the machine learning algorithm, use the operation historical data of the medium and low voltage distribution network to train the machine learning model, optimize the model parameters, and determine the optimal intelligent diagnosis model for the abnormal wire loss of the medium and low voltage distribution network;

[0113] In this embodiment, to determine the optimal intelligent diagnosis model for the abnormal wire loss of the medium and low voltage distribution network, the following operations are performed:

[0114] Collect the operation historical data of the medium and low voltage distribution network, and divide the collected operation historical data of the medium and low voltage distribution network to determine the training set and the test set;

[0115] Based on the machine learning algorithm, the machine learning model is trained using the training set to enable the machine learning model to autonomously learn the intelligent diagnosis behavior of abnormal line losses in the medium and low voltage distribution network, and determine the intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network based on the machine learning algorithm;

[0116] Based on the test set, the intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network based on the machine learning algorithm is tested, and it is evaluated whether the intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network based on the machine learning algorithm can achieve the expected effect based on the accuracy rate, recall rate, and F1 score;

[0117] According to the evaluation results, the parameters and structure of the intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network based on the machine learning algorithm are adjusted. Through continuous iterative optimization, the optimal intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network is determined.

[0118] S4, Fault Prediction and Health Management: Analyze the operation characteristic data of the medium and low voltage distribution network according to the intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network, conduct abnormal diagnosis on the line losses of the medium and low voltage distribution network, combine the line topology structure and equipment information, locate the position of abnormal line losses, find out the reasons for abnormal line losses, formulate intelligent control measures for abnormal line losses, and conduct intelligent control on the abnormal line losses of the medium and low voltage distribution network to reduce the line losses of the medium and low voltage distribution network.

[0119] In this embodiment, analyzing the operation characteristic data of the medium and low voltage distribution network includes:

[0120] Deploy the optimal intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network to the actual intelligent diagnosis environment for abnormal line losses in the medium and low voltage distribution network;

[0121] Input the operation characteristic data of the medium and low voltage distribution network into the intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network, analyze the operation characteristic data of the medium and low voltage distribution network based on the intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network, and conduct abnormal intelligent diagnosis on the line losses of the medium and low voltage distribution network to determine the line loss prediction value of the medium and low voltage distribution network.

[0122] In this embodiment, conducting abnormal diagnosis on the line losses of the medium and low voltage distribution network includes:

[0123] Compare the line loss prediction value with the actual line loss value, calculate the relative error, and judge whether there is an abnormal line loss in the medium and low voltage distribution network based on the relative error;

[0124] Among them,

[0125] When the relative error is within 5%, it is determined that the line losses of the medium and low voltage distribution network are in a normal state;

[0126] When the relative error exceeds 5%, it is determined that the line loss of the medium and low voltage distribution network is in an abnormal state, triggering an alarm for abnormal line loss and locating the position of the abnormal line loss;

[0127] Obtain the topological structure information and equipment information of the medium and low voltage distribution network lines. Through the topological structure information and equipment information, gradually narrow down the abnormal range and locate it to a specific line segment or equipment;

[0128] Among them, obtaining the topological structure information of the medium and low voltage distribution network lines includes the connection relationship of the lines, the length of the lines, the conductor type, the impedance parameters, as well as the position and status of the equipment;

[0129] Conduct a zonal analysis of the medium and low voltage distribution network, divide the medium and low voltage distribution network into several regions, analyze the abnormal line loss in each region one by one, calculate the predicted line loss value and the actual line loss value of each region, and find the region with the largest deviation;

[0130] Check the equipment status in the abnormal line loss area, including: whether the transformer is overloaded or faulty, whether the switch is normally closed or open, whether the capacitor is working properly, and judge whether there is an abnormality in the equipment through the equipment monitoring data;

[0131] Analyze the line parameters, check the line parameters in the abnormal line loss area, including: whether the line impedance is abnormal, whether there is three-phase imbalance or harmonic problems, and analyze the line operation status through electrical measurement data;

[0132] If both the equipment and line parameters are normal, there may be electricity theft behavior. Among them, compare the difference between the user's electricity consumption and the historical data, use the electricity theft detection algorithm, and combine on-site inspections to check whether the user's electricity meter has been tampered with;

[0133] Through the topological structure information and equipment information, gradually narrow down the abnormal range and locate it to a specific line segment or equipment. When the line loss of a certain feeder is abnormal, analyze each branch under the feeder. When the line loss of a certain transformer is abnormal, check the electricity consumption of the users under the transformer.

[0134] In this embodiment, intelligent control of the abnormal line loss of the medium and low voltage distribution network includes:

[0135] Analyze the intelligent diagnosis results of the abnormal line loss of the medium and low voltage distribution network, combine historical data and on-site inspections to find out the reasons for the abnormal line loss, including equipment failures, line problems, electricity theft behavior, operation problems and management problems, formulate intelligent control measures for reducing the line loss of the medium and low voltage distribution network according to the reasons for the abnormal line loss, and conduct intelligent control of the abnormal line loss of the medium and low voltage distribution network;

[0136] Among them, for equipment failures, replace or repair faulty equipment, upgrade equipment, and conduct regular inspections and maintenance, including replacing overloaded transformers, repairing switches with poor contacts, using energy-efficient equipment, and establishing an equipment inspection system to promptly detect and handle potential failures;

[0137] Among them, for line problems, replace aging lines, optimize line layouts, and strengthen joint maintenance, including using wires with high conductivity and low losses, reducing line lengths, lowering line impedance, and regularly checking line joints to ensure good contact;

[0138] Among them, for electricity theft, strengthen electricity consumption inspections, use electricity theft detection technologies and legal means, including real-time monitoring of users' electricity consumption behaviors through smart meters and electricity information collection systems, using machine learning algorithms to detect abnormal electricity consumption behaviors, severely cracking down on electricity theft, and increasing the cost of violations;

[0139] Among them, for operation problems, adjust three-phase balance, control harmonics, and manage loads, including reducing three-phase imbalance through load adjustment or installing three-phase balance devices, installing filters or reactive power compensation devices to reduce harmonic pollution, and optimizing load distribution to avoid local overload;

[0140] Among them, for management problems, improve the accuracy of data collection and transmission, reduce measurement errors, establish a line loss assessment mechanism to motivate operation and maintenance personnel to reduce losses and increase efficiency, and use technologies such as big data and artificial intelligence to achieve real-time monitoring and intelligent diagnosis of line losses.

[0141] In one embodiment, adjusting the parameters and structure of the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms according to the evaluation results includes:

[0142] Determine the types and distributions of abnormal line losses from the evaluation results, and determine the abnormal weights of abnormal types based on the types and distributions of abnormal line losses;

[0143] Obtain relevant historical operation data related to abnormal types from the historical operation data of the power grid operation respectively, and determine the data importance degree of the current relevant historical operation data based on the abnormal weights of abnormal types;

[0144] The calculation formula for the data importance degree K of the current relevant historical operation data is as follows:

[0145]

[0146] Among them, C represents a constant with a value of 3, R represents the data volume of the current relevant historical operation data, R0 represents the reference data volume, n represents the number of abnormal types, δ i represents the abnormal weight of the i-th abnormal type, G iIndicates the data volume of all historical data related to the i-th abnormal type;

[0147] Based on the data importance of each relevant historical operation data, weighted data extraction is performed on the power grid operation historical data to obtain a new training set and a new validation set;

[0148] Determine the mean square error, error rate, and F1 score for the line loss anomaly assessment from the evaluation results, and determine the adjustment coefficient for the model structure weights based on the mean square error, error rate, and F1 score;

[0149] The calculation formula for the adjustment coefficient γ is as follows:

[0150]

[0151] where F1 represents the F1 score, T represents the error rate, σ 2 Mean square error, X represents the model comprehensive performance representation value, and e represents the natural constant, with a value of 2.72;

[0152] Determine the latest weights for the model structure based on the adjustment coefficient;

[0153] Retrain and adjust the intelligent diagnosis model for line loss anomalies in the low-voltage distribution network based on the new training set, new validation set, and latest weights to obtain the latest intelligent diagnosis model for line loss anomalies.

[0154] In this embodiment, the denser the distribution of the types of line loss anomalies, the greater the corresponding anomaly weight.

[0155] In this embodiment, the greater the data importance of the current relevant historical operation data, the more important the data is for diagnosis, and a larger extraction ratio is used when extracting the data set and the validation set.

[0156] In this embodiment, X is greater than If X is not greater than It indicates that there is no need to adjust the model structure weights.

[0157] In this embodiment, the greater the adjustment coefficient, the greater the adjustment amplitude of the structure weights.

[0158] The beneficial effects of the above design are as follows: By determining the types and distributions of abnormal line losses from the evaluation results, determining the abnormal weights of the abnormal types based on the types and distributions of abnormal line losses; respectively obtaining relevant historical operation data related to the abnormal types from the historical data of power grid operation, and determining the data importance of the current relevant historical operation data based on the abnormal weights of the abnormal types, and based on the data importance of each relevant historical operation data, performing weighted data extraction on the historical data of power grid operation to obtain a new training set and a new validation set, increasing the proportion of operation data related to frequent abnormal line losses, making the model trained have a more accurate recognition and judgment ability for the abnormal conditions of frequent line losses, then determining the mean square error, error rate, and F1 score for the evaluation of abnormal line losses from the evaluation results, determining the adjustment coefficient for the weights of the model structure based on the mean square error, error rate, and F1 score, and evaluating from the comprehensive performance of the model to adjust the model weights, improving the performance of the latest intelligent diagnosis model for abnormal line losses, and providing a highly accurate model for the intelligent diagnosis of abnormal line losses.

[0159] In summary, by analyzing the operation characteristic data of the medium and low voltage distribution network using the intelligent diagnosis model for abnormal line losses in the medium and low voltage distribution network, diagnosing the abnormal line losses in the medium and low voltage distribution network, combining the line topology structure and equipment information, locating the positions of abnormal line losses, finding out the causes of abnormal line losses, formulating intelligent control measures for abnormal line losses, and performing intelligent control on the abnormal line losses in the medium and low voltage distribution network to reduce the line losses in the medium and low voltage distribution network, it can effectively improve the recognition accuracy of abnormal line losses, not only help optimize the operation efficiency of the power system, but also reduce unnecessary energy waste, thus providing strong support for the sustainable development of the power system, and can achieve refined management of the line losses of the distribution network and improve the overall operation efficiency of the power grid.

[0160] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0161] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent diagnosis method for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms, characterized in that, Including: Collecting the operation data of the medium and low voltage distribution network and preprocessing it; Extracting features from the operation data of the medium and low voltage distribution network, determining the influence degree of the current feature on the intelligent diagnosis and prediction result of the line loss anomaly of the medium and low voltage distribution network based on the target score value, and determining the operation characteristic data of the medium and low voltage distribution network; Building an intelligent diagnosis model for line loss anomaly of the medium and low voltage distribution network based on machine learning algorithms; Analyzing the operation characteristic data of the medium and low voltage distribution network and diagnosing line loss anomaly according to the intelligent diagnosis model for line loss anomaly of the medium and low voltage distribution network, combining the line topology structure and equipment information, locating the line loss anomaly position, and performing intelligent control on the line loss anomaly of the medium and low voltage distribution network.

2. The intelligent diagnosis method for abnormal line loss of medium and low voltage distribution network based on machine learning algorithm according to claim 1, characterized in that Collecting the operation data of the medium and low voltage distribution network and performing the following operations: Using Internet of Things technology to install intelligent acquisition devices in the medium and low voltage distribution network; Based on the intelligent acquisition devices, real-time monitoring and collection of the voltage conditions during the operation of the medium and low voltage distribution network to obtain the voltage data of the medium and low voltage distribution network; Based on the intelligent acquisition devices, real-time monitoring and collection of the current conditions during the operation of the medium and low voltage distribution network to obtain the current data of the medium and low voltage distribution network; Based on the intelligent acquisition devices, real-time monitoring and collection of the power conditions during the operation of the medium and low voltage distribution network to obtain the power data of the medium and low voltage distribution network; Based on the intelligent acquisition devices, real-time monitoring and collection of the load conditions during the operation of the medium and low voltage distribution network to obtain the load data of the medium and low voltage distribution network; Among them, based on the voltage data, current data, power data and load data of the medium and low voltage distribution network, the operation data of the medium and low voltage distribution network based on the Internet of Things is determined.

3. The intelligent diagnosis method for abnormal line loss of medium and low voltage distribution network based on machine learning algorithm according to claim 2, wherein, Preprocessing the operation data of the medium and low voltage distribution network, including: Cleaning the operation data of the medium and low voltage distribution network based on the Internet of Things; Removing duplicate data, missing values and abnormal values that are useless for the intelligent diagnosis of line loss anomaly of the medium and low voltage distribution network from the operation data of the medium and low voltage distribution network based on the Internet of Things; Among them, for the missing values and abnormal values that are useful for the intelligent diagnosis of line loss anomaly of the medium and low voltage distribution network, the median is used to fill the missing values and the average value is used to replace the abnormal values; Normalizing the operation data of the medium and low voltage distribution network based on the Internet of Things; Converting the operation data of the medium and low voltage distribution network based on the Internet of Things into a unified format, removing the dimension differences between the operation data of the medium and low voltage distribution network based on the Internet of Things, and determining the standardized operation data of the medium and low voltage distribution network.

4. The intelligent diagnosis method for abnormal line loss of medium and low voltage distribution network based on machine learning algorithm according to claim 3, characterized in that Extracting features from the operation data of the medium and low voltage distribution network, including: Extracting static features and dynamic features related to the line loss of the medium and low voltage distribution network from the operation data of the medium and low voltage distribution network based on the principal component analysis method; Among them, the static features include line parameters, equipment parameters, and topology structure; The line parameters include line length, conductor type, cross-sectional area, and resistivity; the equipment parameters include transformer capacity, type, and load rate; the topology structure includes network structure and node connection relationship; Among them, the dynamic features include electrical measurement data, load characteristics, environmental factors, and time characteristics; Electrical measurement data includes voltage, current, active power, reactive power, and power factor; load characteristics include load rate, load fluctuation, peak-valley difference, and load curve; environmental factors include temperature, humidity, and weather conditions; time characteristics include hour, day, week, month, and seasonal dimension characteristics; Based on the random forest algorithm, select the extracted static features and dynamic features, evaluate the influence degree of each feature on the intelligent diagnosis and prediction results of abnormal line losses in medium and low voltage distribution networks, screen out the most valuable features for the intelligent diagnosis of abnormal line losses in medium and low voltage distribution networks, and determine the operation characteristic data of medium and low voltage distribution networks.

5. The intelligent diagnosis method for abnormal line loss in medium and low voltage distribution networks based on machine learning algorithms according to claim 4, characterized in that The selection of the extracted static features and dynamic features based on the random forest algorithm, and the evaluation of the influence degree of each feature on the intelligent diagnosis and prediction results of abnormal line losses in medium and low voltage distribution networks include: Establish a feature set based on all dynamic features and static features, obtain the first change range of all dynamic features relative to historical dynamic features, obtain the second change range of all static features relative to historical static features, and use the ratio of the first change range and the second change range as the random sampling ratio for dynamic features and static features; Randomly draw features with replacement from the feature set according to the random sampling ratio to obtain multiple feature groups. Based on each feature group, combine the abnormal information of line losses to construct a decision tree, and randomly select the features of the feature group to determine the splitting point when splitting the nodes of the decision tree; Input each feature in the feature set into each decision tree in turn, and determine the scoring set of the current feature in all decision trees according to the decision results; Perform eigenvalue replacement on the current feature to obtain replacement eigenvalues, input the replacement eigenvalues into each decision tree in turn, obtain the set of performance degradation amplitudes of all decision trees, and determine the scoring weight of the current feature based on the difference between the average degradation value of the set of performance degradation amplitudes and the preset reference degradation value; Perform weighted processing on the average score value determined by the scoring set based on the scoring weight to obtain the target score value of the current feature; Determine the influence degree of the current feature on the intelligent diagnosis and prediction results of abnormal line losses in medium and low voltage distribution networks based on the target score value.

6. The intelligent diagnosis method for abnormal line loss in medium and low voltage distribution networks based on machine learning algorithms according to claim 1, wherein Construct an intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks and perform the following operations: Collect the historical operation data of medium and low voltage distribution networks, and divide the collected historical operation data of medium and low voltage distribution networks to determine the training set and the test set; Based on the machine learning algorithm, use the training set to train the machine learning model, enable the machine learning model to autonomously learn the intelligent diagnosis behavior of abnormal line losses in medium and low voltage distribution networks, and determine the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks based on the machine learning algorithm; Based on the test set, test the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks based on the machine learning algorithm, and evaluate whether the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks based on the machine learning algorithm can achieve the expected effect based on accuracy, recall rate, and F1 score; According to the evaluation results, adjust the parameters and structure of the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks based on the machine learning algorithm. After continuous iterative optimization, determine the optimal intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks.

7. The intelligent diagnosis method for abnormal line loss in medium and low voltage distribution networks based on machine learning algorithms according to claim 6, characterized in that Analyze the operation characteristic data of medium and low voltage distribution networks, including: Deploy the optimal intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks in the actual intelligent diagnosis environment for abnormal line losses in medium and low voltage distribution networks; Input the operation characteristic data of medium and low voltage distribution networks into the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks, analyze the operation characteristic data of medium and low voltage distribution networks based on the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks, and conduct intelligent diagnosis of abnormal line losses in medium and low voltage distribution networks to determine the predicted value of line losses in medium and low voltage distribution networks.

8. The intelligent diagnosis method for abnormal line loss in medium and low voltage distribution network based on machine learning algorithm according to claim 7, characterized in that, Conduct abnormal diagnosis of line losses in medium and low voltage distribution networks, including: Compare the predicted value of line losses with the actual value of line losses, calculate the relative error, and determine whether there are abnormal line losses in medium and low voltage distribution networks based on the relative error; Among them, When the relative error is within 5%, it is determined that the line losses in medium and low voltage distribution networks are in a normal state; When the relative error exceeds 5%, it is determined that the line losses in medium and low voltage distribution networks are in an abnormal state, trigger an alarm for abnormal line losses, and locate the position of abnormal line losses; Obtain the topological structure information and equipment information of the lines in medium and low voltage distribution networks, and gradually narrow down the abnormal range through the topological structure information and equipment information to locate specific line segments or equipment; Among them, obtaining the topological structure information of the lines in medium and low voltage distribution networks includes the connection relationship of the lines, the length of the lines, the conductor type, impedance parameters, and the position and status of the equipment; Conduct zonal analysis on medium and low voltage distribution networks, divide medium and low voltage distribution networks into several regions, analyze abnormal line losses in each region one by one, calculate the predicted value and actual value of line losses in each region, and find the region with the largest deviation; Check the equipment status in the area with abnormal line losses, including: whether the transformer is overloaded or faulty, whether the switch is normally closed or open, whether the capacitor is working properly, and judge whether there are abnormalities in the equipment through equipment monitoring data; Analyze the line parameters, check the line parameters in the area with abnormal line losses, including: whether the line impedance is abnormal, whether there are problems of three-phase imbalance or harmonics, and analyze the operation status of the line through electrical measurement data; If both the equipment and line parameters are normal, there may be electricity theft behavior. Among them, compare the difference between the user's electricity consumption and historical data, use electricity theft detection algorithms, and combine on-site inspections to check whether the user's electricity meter has been tampered with; Gradually narrow down the abnormal range through the topological structure information and equipment information to locate specific line segments or equipment. When the line losses of a certain feeder are abnormal, analyze each branch under the feeder. When the line losses of a certain transformer are abnormal, check the electricity consumption of users under the transformer.

9. The intelligent diagnosis method for abnormal line loss in medium and low voltage distribution networks based on machine learning algorithms according to claim 6, characterized in that, According to the evaluation results, adjust the parameters and structure of the intelligent diagnosis model for abnormal line losses in medium and low voltage distribution networks based on machine learning algorithms, including: Determine the types and distributions of abnormal line losses from the evaluation results, and determine the abnormal weights of abnormal types based on the types and distributions of abnormal line losses; Respectively obtain the relevant historical operation data related to the abnormal type from the historical operation data of power grid operation, and determine the data importance of the current relevant historical operation data based on the abnormal weights of the abnormal type; Based on the data importance of each relevant historical operation data, conduct weighted data extraction on the historical operation data of power grid operation to obtain a new training set and a new validation set; Determine the mean square error, error rate, and F1 score for the line loss anomaly assessment from the evaluation results, and determine the adjustment coefficient for the model structure weights based on the mean square error, error rate, and F1 score; Determine the latest weights for the model structure based on the adjustment coefficient; Retrain and adjust the intelligent diagnosis model for line loss anomalies in the low-voltage distribution network based on the new training set, new validation set, and the latest weights to obtain the latest intelligent diagnosis model for line loss anomalies.

Images