CN111738462A - Fault first-aid repair active service early warning method for electric power metering device - Google Patents
Fault first-aid repair active service early warning method for electric power metering device Download PDFInfo
- Publication number
- CN111738462A CN111738462A CN202010511520.4A CN202010511520A CN111738462A CN 111738462 A CN111738462 A CN 111738462A CN 202010511520 A CN202010511520 A CN 202010511520A CN 111738462 A CN111738462 A CN 111738462A
- Authority
- CN
- China
- Prior art keywords
- metering device
- fault
- data
- power
- risk
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
The invention provides a fault first-aid repair active service early warning method for an electric power metering device, which comprises the following steps: the method comprises the steps that firstly, portrait analysis is carried out on power consumers on the basis of big data of a power metering device, and the power consumers are divided into three categories, namely key consumers, important consumers and ordinary consumers according to importance from high to low; constructing a fault recognition model of the electric power metering device, performing high-risk, medium-risk or low-risk fault risk prediction on the electric power metering device by using the trained fault recognition model, and outputting the electric energy metering device with the fault risk prediction result of high risk and medium risk as a suspected fault electric energy metering device; and thirdly, correspondingly determining the priority level of the fault first-aid repair active service of the electric power metering device of each electric power user according to the importance category of the electric power user determined in the first step and the risk prediction level output in the second step: and the early warning is divided into first to sixth levels from high to low. The invention can improve the predictability, pertinence and working efficiency of the emergency repair of the power metering device of the power supply department.
Description
Technical Field
The invention relates to the technical field of power supply department metering device fault emergency repair, in particular to a power metering device fault emergency repair active service early warning method.
Background
With the acceleration of economic development and the improvement of life quality of people, the demand for electric power is continuously increased, and the fault first-aid repair of the electric power metering device is an important business of relevant departments of power supply and an important aspect of the quality of service of power supply experienced by users. However, various problems of the metering device are easy to occur due to the influence of various factors such as electricity stealing behavior, system interference, external environment and the like. To electric power metering device trouble, the power supply department traditional way adopts the passive mode of salvageing to handle, receives the trouble feedback promptly after, just can arrange relevant fortune dimension personnel to look over the trouble condition and salvage, and concrete flow is: the method comprises the steps of business acceptance, investigation and dispatching, on-site investigation, fault treatment of the metering device and information filing. The mode is salvageed to traditional electric power metering device trouble, and the maintenance personal is not known the trouble condition in advance, and the field is looked over and is implemented the maintenance again and prolonged the maintenance duration greatly, also can cause the power consumer to have a power failure time overlength when the strength of salvageing is not enough moreover. Therefore, it is necessary to research a more efficient method for repairing and handling a fault of an electric power metering device.
Disclosure of Invention
The purpose of the invention is: the method starts from the perspective of intelligently predicting the fault risk of the metering device, combines a user portrait technology, constructs a power supply department to construct a power metering device fault first-aid repair active service early warning model, realizes early warning and accurate first-aid repair of the power metering device fault, and improves the power metering device fault first-aid repair efficiency.
The technical scheme of the invention is as follows: the invention discloses a fault first-aid repair active service early warning method for an electric power metering device, which comprises the following steps of:
the method comprises the steps that firstly, portrait analysis is carried out on power consumers based on big data of a power metering device of a power supply department, and the power consumers are divided into three categories, namely key consumers, important consumers and ordinary consumers according to importance from high to low;
secondly, constructing a fault recognition model of the electric power metering device, predicting high-risk, medium-risk or low-risk fault risks of the electric power metering device by using the trained fault recognition model, and outputting the electric power metering device with the fault risk prediction result of high risk and medium risk as a suspected fault electric power metering device;
and thirdly, correspondingly determining the priority level of the power metering device fault first-aid repair active service of each power user by combining the importance category of the power user determined in the first step and the risk prediction level output in the second step: and the early warning is divided into first to sixth levels from high to low.
The further scheme is as follows: the first step comprises the following specific steps:
collecting data: acquiring 4-dimensional information data tables of user basic information, power utilization behaviors, payment behaviors and appeal behaviors of each power user from a power supply department marketing service system, and correspondingly determining index variables corresponding to each dimension;
preprocessing data: respectively preprocessing data of each form based on four dimensions of user basic information, electricity utilization behaviors, payment behaviors and appeal behaviors, and correspondingly deleting and filling abnormal values and missing values in source data;
establishing a power user tag library:
extracting power features with higher value of power users by adopting a traditional statistical analysis method including frequency analysis, centralized trend analysis, discrete degree analysis and visual analysis of data on the four-dimensional basic data of the user basic information, the power consumption behavior, the payment behavior and the appeal behavior after the preprocessing, and establishing a tag library of the power users by taking the four dimensions as a dimensional frame of the tag library of the users;
fourthly, the power consumer is portrayed based on the K-Means algorithm, and the power consumers are divided into key consumers, important consumers and common consumers from high to low according to the importance:
selecting six clustering indexes of electricity utilization age, operation capacity, annual average settlement electric quantity, peak-valley electric quantity coefficient, annual average payment times and annual average payment amount by adopting operations including correlation analysis, unbalanced index elimination and discrete variable elimination on four dimensional data of basic information, electricity utilization behavior, payment behavior and appeal behavior in the label library established in the step three; then, a data set consisting of data with larger values is split by utilizing a Lauda criterion, the data set is independently used as a user group type and defined as a key user, the rest data sets are gathered into four types by utilizing a K-Means clustering algorithm, and the four types are respectively defined as follows according to the characteristics of a clustering center: important users, consumer-normal users, loyalty-normal users, peak-time-normal users; and the consumption type-common user, loyalty type-common user and peak period type-common user are classified into common users.
The further scheme is as follows: in the first step, the index variables included in the user basic information include: urban and rural categories, user categories, power utilization categories, power supply voltage, operating capacity, power utilization years and power types; the electricity consumption behavior comprises index variables as follows: the method comprises the following steps of settling electric quantity, settling electric charge, peak electric quantity, average electric quantity, valley electric quantity, peak electric charge, average electric charge, valley electric charge and whether peak-valley marking is executed or not; the payment behavior comprises the following index variables: payment mode, payment times, payment amount, overdue times, arrearage times and arrearage amount; the appeal behavior comprises index variables as follows: the number of complaints and the type of complaints.
The further scheme is as follows: the second step comprises the following specific steps:
collecting data required by modeling: taking the electric power metering devices of all the electric power users related to the first step as objects, and correspondingly acquiring 4-dimensional related information data tables of attribute characteristics, electric quantity characteristics, power characteristics and weather characteristics of the electric power metering devices from a marketing service system, an electric power consumption information acquisition system and a meteorological system of a power supply department;
preprocessing the acquired data required by modeling: carrying out data processing including data cleaning, data transformation and data protocols on the data required by modeling collected in the fifth step to form a data wide table;
extracting core indexes based on a random forest RF algorithm;
and (III) predicting the fault risk level based on the XGboost algorithm, and outputting the predicted high-risk and medium-risk electric power metering device as a suspected fault electric power metering device.
The further scheme is as follows: the second step comprises the following steps:
the data cleaning comprises the processing operations of space processing of a data table, invalid field and abnormal data elimination, missing value processing, character type conversion and date format unification;
the data transformation and data reduction comprises:
i, the operation age is the date of meter disassembly or the difference age between the current date and the date of meter assembly;
II, dividing the weather grade into 0-7 grades from low to high according to the influence degree of different weather conditions on the fault of the electric power metering device;
III, calculating a characteristic dimension of the electric quantity [ Jianling 1 ];
and IV, calculating power characteristic dimensions [ Jianling 2 ].
The further scheme is as follows: the step of the second step is a specific method for extracting the core index based on the random forest RF algorithm, and the specific method comprises the following steps:
extracting a preliminary index system from four dimensions of attribute features, weather features, electric quantity features and power features of the electric power metering device by using descriptive statistics of data including attribute features, concentration trend and dispersion degree;
VI, extracting core indexes by using a random forest embedded feature selection algorithm, comprising the following steps:
a. taking a preliminary index system as model input, and taking whether a fault occurs as model output to construct a random forest RF algorithm model;
b. randomly distributing the alternative characteristic indexes to a plurality of decision trees, judging a training set by each decision tree, and calculating the importance of the alternative characteristic by using variable importance measurement based on the classification accuracy of the data outside the bag;
c. and setting a characteristic importance threshold, and screening out all indexes with the characteristic importance greater than the threshold as extracted core indexes.
The further scheme is as follows: the second step comprises the following specific steps:
constructing an XGboost model by taking the core indexes determined in the step (c) as input variables and whether faults exist as output variables on the basis of a set training set and a set testing set; optimizing two parameters of the maximum depth of the decision tree and the number of the decision trees of the model by adopting a grid search algorithm; obtaining an XGboost model under the optimal parameters while obtaining the optimal parameters through a grid search algorithm;
and VIII, applying the trained XGboost model to a test set, prejudging the failure risk prediction probability of the electric power metering device, dividing the failure risk prediction probability into three failure risk grades of high risk, medium risk and low risk, and outputting the data of the electric power metering device with high risk and medium risk as a suspected failure electric power metering device.
The invention has the positive effects that: (1) according to the electric power metering device fault first-aid repair active service early warning method, the electric power metering device fault first-aid repair active service early warning model is established, so that the electric power metering device fault can be early warned in advance and the first-aid repair grade can be determined, a power supply department can formulate a targeted first-aid repair scheme according to the fault type and the fault early warning grade, and the fault first-aid repair efficiency is improved. (2) The invention relates to an active service early warning method for the fault emergency repair of an electric power metering device, which constructs a user tag library, can realize omnibearing fine portrayal of an electric power user, and can give an early warning grade of the fault active service of the electric power metering device according to a suspected fault list of the metering device and the omnibearing portrayal of the electric power user when in use, so that a power supply department can take measures to implement the emergency repair with predictability and pertinence according to the early warning grade, and changes the passive emergency repair arrangement in the prior art into the active emergency repair service, [ build 3] and the user satisfaction.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is a schematic diagram of a power consumer tag library system established in the steps of the method of the present invention;
fig. 3 is a schematic diagram of the early warning level of the fault repair active service based on the boston matrix established in the steps of the method of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
(example 1)
Referring to fig. 1, the method for early warning of active service for emergency repair of power metering device in this embodiment is described by taking an electric energy meter as a power metering device, and the implementation of other metering devices is similar. The invention discloses a fault first-aid repair active service early warning method for an electric power metering device, which is characterized in that the fault first-aid repair grade is correspondingly determined according to the type of an electric power user and the fault risk grade, the main implementation steps comprise the steps of electric power user portrait analysis, electric power metering device fault recognition model construction and electric power metering device fault first-aid repair active service early warning model construction, and the main implementation steps are specifically described as follows:
the method comprises the following steps of firstly, carrying out portrait analysis on a power consumer based on big data of a power metering device, and dividing the power consumer into three categories of key users, important users and common users from high to low according to importance, wherein the specific steps are as follows:
collecting data: the method comprises the steps of obtaining 4-dimensional information data tables of user basic information, power utilization behaviors, payment behaviors and appeal behaviors of all power users from a power supply department marketing business system, and correspondingly determining index variables corresponding to all dimensions.
The basic user information includes index variables: urban and rural categories, user categories, power utilization categories, power supply voltage, operating capacity, power utilization years and power types; the electricity consumption behavior comprises index variables as follows: the method comprises the following steps of settling electric quantity, settling electric charge, peak electric quantity, average electric quantity, valley electric quantity, peak electric charge, average electric charge, valley electric charge and whether peak-valley marking is executed or not; the payment behavior comprises the following index variables: payment mode, payment times, payment amount, overdue times, arrearage times and arrearage amount; the appeal behavior comprises index variables as follows: the number of complaints and the type of complaints.
Preprocessing data: and respectively carrying out data preprocessing on each table based on four dimensions of user basic information, electricity utilization behaviors, payment behaviors and appeal behaviors, and correspondingly deleting and filling abnormal values and missing values in the source data. The basic data indices obtained by data preprocessing are summarized in table 1.
Establishing a power user tag library:
extracting the power characteristics with higher power user value from the preprocessed four-dimensional basic data of the user basic information, the electricity consumption behavior, the payment behavior and the appeal behavior by adopting a traditional statistical analysis method including frequency analysis, centralized trend analysis, discrete degree analysis and visual analysis of the data, establishing a complete tag library system of the power user by taking the four dimensions as a dimension framework of a user tag library, and displaying the tag library system in a form of belonging dimension-tag name-tag example as shown in fig. 2. The peak-valley coefficient is used for distinguishing the peak-valley-average electricity utilization behaviors of the user in different electricity price periods, and when the electricity quantity or the electricity charge is maximum in the peak period, the coefficient is more than or equal to 1; when the valley electric quantity or the electric charge is maximum, the coefficient is less than or equal to-1; when the average electric quantity or the electric charge is the maximum, the coefficient is between-1 and 1, the peak electric quantity or the electric charge is positive when the number is large, and the valley electric quantity or the electric charge is negative when the number is large.
Fourthly, the power consumer is portrayed based on the K-Means algorithm, and the power consumers are divided into key consumers, important consumers and common consumers from high to low according to the importance:
selecting six clustering indexes of electricity utilization age, operation capacity, annual average settlement electric quantity, peak-valley electric quantity coefficient, annual average payment times and annual average payment amount by adopting operations including correlation analysis, unbalanced index elimination and discrete variable elimination on four dimensional data of basic information, electricity utilization behavior, payment behavior and appeal behavior in the label library established in the step three; then, a data set consisting of data with larger values is split by utilizing a Lauda criterion, the data set is independently used as a user group type and defined as a key user, the rest data sets are gathered into four types by utilizing a K-Means clustering algorithm, and the four types are respectively defined as follows according to the characteristics of a clustering center: important users, consumption type-common users, loyalty type-common users and peak period type-common users, so that the power users are divided into five user group types, and the respective characteristics of the five user groups are summarized as shown in table 2.
Based on business habits, the consumption type-common users, the loyalty type-common users and the peak period type-common users are classified and collectively called as common users, and finally, the power users are divided into three categories, namely key users, important users and common users according to the user importance from high to low.
Secondly, constructing a fault recognition model of the electric power metering device, predicting high-risk, medium-risk or low-risk fault risks of the electric power metering device by using the trained fault recognition model, and outputting the electric power metering device with the fault risk prediction result of high risk and medium risk as a suspected fault electric power metering device, wherein the method specifically comprises the following steps:
collecting data required by modeling: and taking the electric energy meters of all the power users related to the first step as objects, and correspondingly acquiring 4-dimensional related information data tables of the attribute characteristics, the electric quantity characteristics, the power characteristics and the weather characteristics of the electric energy meters from a marketing service system, an electricity consumption information acquisition system and a meteorological system of a power supply department.
Preprocessing the acquired data required by modeling: and (4) carrying out data processing including data cleaning, data transformation and data specification on the data required by modeling collected in the fifth step to form a data wide table. Wherein:
the data cleaning mainly comprises the processing operations of blank processing of a data table, invalid field and abnormal data elimination, missing value processing (forward/backward filling, mean value filling and the like), character type conversion and date format unification;
the data transformation and data specification mainly comprise:
i, the operation age is the date of meter disassembly or the difference age between the current date and the date of meter assembly;
II, dividing the weather grade into 0-7 grades from low to high according to the influence degree of different weather conditions on the fault of the electric power metering device;
III, calculating indexes of characteristic dimensions of the electric quantity: acquiring data such as a forward active total electric energy indicating value, a forward active peak electric energy indicating value, a forward active level electric energy indicating value, a forward active valley electric energy indicating value and the like, wherein daily electric quantity indicates that the forward active total electric energy indicating value on the day minus the forward active total electric energy on the previous day is multiplied by a comprehensive multiplying power; similarly, the daily peak electric quantity, the daily average electric quantity and the daily valley electric quantity are obtained, the daily electric quantity difference represents that the daily electric quantity of the current day is subtracted by the daily electric quantity of the previous day, and the daily peak-valley electric quantity difference represents that the daily peak electric quantity of the current day is subtracted by the daily valley electric quantity.
IV, calculating indexes of power characteristic dimensions: and acquiring the active power, the phase A active power, the phase B active power and the phase C active power of daily load curve data, and collecting the daily load curve data to 24 points. The daily average load, the daily power standard deviation, the daily maximum load and the daily minimum load respectively represent the 24-point active power average value, the standard deviation, the maximum value and the minimum value of the day. The daily peak-to-valley rate represents the daily maximum load minus the daily minimum load divided by the daily maximum load for the day. The daily load rate and the minimum load rate respectively represent the daily average load and the daily minimum load divided by the daily maximum load of the day. The daily average load three-phase mean value, the maximum phase daily average load and the minimum phase daily average load respectively represent the active power averaging of the phase A, the phase B and the phase C at 24 points of the day, and then the three-phase mean value, the maximum value and the minimum value are obtained. The daily power standard deviation three-phase mean value, the maximum phase daily power standard deviation and the minimum phase daily power standard deviation respectively represent 24-point A-phase, B-phase and C-phase active power calculation standard deviations, and then three-phase mean values, maximum values and minimum values are calculated. The basic data index variables obtained by data preprocessing are shown in table 3.
And seventhly, extracting core indexes based on a random forest RF algorithm:
in order to avoid dimension disaster, the invention adopts a random forest RF algorithm to extract core indexes from the basic data indexes obtained in the step (c), and the random forest based feature extraction process is divided into two steps of preliminary index design and core index screening.
And V, extracting a preliminary index system from four dimensions of the attribute feature, the weather feature, the electric quantity feature and the power feature of the electric energy meter by using descriptive statistics of data including the attribute feature, the concentration trend and the dispersion degree. The initial indexes related to the attribute characteristics of the electric energy meter comprise 4 user types, electricity utilization types, operation years and voltage levels; the initial indexes related to the weather characteristics are obtained by calculating the average value of the first 2 weeks of all weather characteristic indexes in the table 3, and the total number of the initial indexes is 2; the initial indexes related to the electric quantity characteristics are obtained by calculating the mean value, the standard deviation, the maximum value and the minimum value of the previous two weeks or the previous one week of all the electric quantity characteristic indexes in the table 3, and the total number of the initial indexes is 48; the initial indexes related to the power characteristics are obtained by calculating the mean value, standard deviation, maximum value and minimum value of the first two weeks of all the power characteristic indexes in table 3, and the total number is 52. The final obtained preliminary index system comprises 106 indexes.
VI, extracting core indexes by using a random forest embedded feature selection algorithm, and specifically comprising the following steps: firstly, taking a preliminary index system as model input, and taking whether a fault exists as model output to construct a random forest RF model; randomly distributing the alternative feature indexes to a plurality of decision trees, judging a training set by each decision tree, and calculating the importance of the alternative features by using variable importance measurement based on the classification accuracy of the data outside the bag; setting a characteristic importance threshold value, and screening out all indexes with the characteristic importance greater than the threshold value as core indexes for subsequent use.
And eighthly, predicting the fault risk level based on an XGboost algorithm:
VII, based on a set training set and a set testing set, constructing an XGboost model by taking the core index determined in the step VI of the step (c) as an input variable and taking whether a fault is generated as an output variable; optimizing two parameters of the maximum depth of the decision tree and the number of the decision trees of the model by adopting a grid search algorithm; obtaining an XGboost model under the optimal parameters while obtaining the optimal parameters through a grid search algorithm;
and VIII, applying the trained XGboost model to a test set, pre-judging the failure risk prediction probability of the electric energy meter, dividing the failure risk prediction probability into three failure risk grades of high risk, medium risk and low risk, and outputting the electric energy meter data of high risk and medium risk as a suspected failure list.
In order to evaluate the performance of the XGboost model, classification effect evaluation indexes such as accuracy, precision, recall rate and comprehensive evaluation indexes are introduced, and the calculation mode of the classification effect evaluation indexes is shown in Table 4, wherein TN is the positive judgment number of negative samples, FP is the negative misjudgment number of negative samples, FN is the positive misjudgment number of positive samples, and TP is the positive judgment number of positive samples. The reliability of the XGboost model is verified by comparing evaluation index values of different models.
And thirdly, correspondingly determining the priority level of the power metering device fault first-aid repair active service of each power user by combining the importance category of the power user determined in the first step and the risk prediction level output in the second step: the early warning is divided into first to sixth levels from high to low:
and finally determining 6 early warning grades from first early warning to sixth early warning, wherein the early warning grades are from first early warning to sixth early warning, and the corresponding first-level to sixth-level first-aid repair priorities are from high to low, so that the power supply department can carry out the early warning for the fault first-aid repair active service more reasonably, and the real operability of the active service is stronger. As shown in FIG. 3, when the electric energy meter breaks down, the fault first-aid repair grade can be rapidly determined through the user type and the fault risk grade, and the first-aid repair efficiency is improved.
Attached: brief description of the related algorithms involved in this embodiment:
1. K-Means algorithm:
the main idea of the K-Means algorithm is: firstly, randomly selecting K objects as initial clustering centers, then dividing each sample point into the nearest classes through similarity measurement, calculating the clustering center of a new clustering result as the initial clustering center of the next clustering, and repeating the steps circularly, wherein when a criterion function for evaluating the clustering performance reaches the optimum, iteration is stopped, at the moment, the object similarity in the same class is extremely high, and the object similarity between different classes is extremely low.
2. Random forest RF algorithm: the random forest RF algorithm is proposed by Leo Breiman in 2001, and the random forest RF algorithm is characterized in that m rounds of repeated random sampling are performed on an original training sample set n in a replacement mode through a self-help bootstrap resampling technology, then m classification trees are generated according to a self-help sample set to form a random forest, and the classification result of new data is determined according to the score formed by voting of the classification trees. The feature selection algorithm based on random forest embedding is an implicit feature selection method provided by Breiman, and can be used for screening feature variables and deleting redundant irrelevant feature attributes of the feature variables.
3. The XGboost algorithm is proposed by Chentianqi doctor in 2014, has the advantages of high precision and high efficiency, and has the basic idea that a simple model is generated by selecting partial samples and characteristics as a base classifier, the base classifier generally selects a decision tree or a linear classifier, the core of generating the new model is that the new model is built in the gradient direction of a corresponding loss function, the residual error is corrected, the complexity is controlled, the cyclic iteration is carried out, and hundreds of linear or tree models are finally generated. When training is completed to obtain t trees, the score of a sample is predicted, and according to the characteristics of the sample, a corresponding leaf node is fallen in each tree, each leaf node corresponds to a score, and finally, the score corresponding to each tree is added to be the predicted value of the sample.
(example of application verification)
The application verification example verifies the effectiveness of the power metering device fault first-aid repair active service early warning method of the embodiment by taking effective data of 167345 three-phase smart meter users in the jurisdiction of a certain power supply department in nearly three years.
The clustering centers of the five types of users of 167345 three-phase intelligent meters processed by the first step of the method are shown in table 5.
And finally, dividing the users into three categories of key users, important users and common users according to the importance of the users, wherein the distribution conditions of various user types are shown in a table 6.
According to the second step of the method, 167345 data sheets of intelligent meter attribute characteristics, weather characteristics, electric quantity characteristics, power characteristics and other information of users of the three-phase intelligent meter are obtained, 6227 sample data are selected as a training set, 2669 sample data are selected as a testing set, and 8033 sample data are selected as a verification set.
First, a core index is extracted based on a random forest. And constructing a preliminary index system by utilizing the training set. With 106 indexes in the preliminary index system as input variables, the feature importance of each index is calculated by using a random forest RF feature selection algorithm, a feature importance threshold value is set to be 0.01, and 23 core indexes are extracted for subsequent model analysis, as shown in Table 7.
Secondly, an XGboost model is constructed. And (3) constructing an XGboost model by using the training set data and using 23 core indexes screened out by the RF model as input variables and whether faults occur as output variables. And using a grid search algorithm to adjust parameters of the model, wherein the maximum depth of a decision tree and the number of the decision tree of the optimal model are respectively 6 and 90, and applying the optimal model to a test set. The classification evaluation indexes of the XGboost model on the training set and the test set are shown in Table 8. The comprehensive evaluation index F1 is obtained by blending and averaging the precision P and the recall ratio R, can evaluate the performance of the classification model, and takes F1 as the main identification accuracy of the model. The recognition accuracy rate of the XGboost model constructed by the training set and the test set is over 90 percent.
The reliability of the model is then compared. Based on a training set and a test set, 23 core indexes screened out by an RF model are used as input variables, whether faults occur or not is used as output variables, and a Logistic Regression (LR) model, a Decision Tree (DT) model, a neural network (MLP) model, a gradient lifting decision tree (GBDT) model and an XGboost model are respectively constructed. The indexes of the precision (P), the recall rate (R), the comprehensive evaluation index (F1), the training time and the like of the 5 algorithms on the test set are shown in Table 9.
Compared with other models, the XGboost model has great advantages in fault identification accuracy rate (P), recall rate (R), comprehensive evaluation index (F1) and training time, so that the XGboost model is selected to predict the fault risk probability and has reliability.
Based on the verification set of 8033 samples, 23 core indexes screened out by the RF model are used as input variables, the trained XGboost model is adopted to carry out fault prejudgment, and prediction results are shown in table 10.
The '1' value in the predicted value and the actual check whether the fault occurs represents that the intelligent meter fails, the '0' value represents that the intelligent meter does not fail, and the fault risk prediction probability is the probability of predicting the fault of the intelligent meter. According to the fault risk prediction probability, the intelligent meter is divided into three fault risk levels in the ranges of [0.8,1], (0.5,0.8) and [0,0.5 ]: high risk, medium risk, low risk. When the fault risk level is medium risk or high risk, the intelligent meter is prejudged to have a fault, namely the predicted value is 1; when the fault risk level is low risk, it is judged in advance that the intelligent meter does not have faults, namely the predicted value is 0.
The data of the verification set actually comprises 153 fault intelligent meters and 7880 non-fault intelligent meters, and the model correctly pre-judges 132 fault intelligent meters and 7766 non-fault intelligent meters. The accuracy A of the intelligent meter fault identification model on the verification set is 98.31%, the accuracy P is 53.66%, the recall rate R is 86.27%, and the comprehensive index F1 is 66.16%. The suspected fault list with the fault risk level of medium risk or high risk comprises 246 intelligent tables, wherein the high risk intelligent table 148 comprises the fault, the medium risk intelligent table 98 comprises the fault, namely the model prejudgment 246 comprises the fault of the intelligent tables.
According to the third step of the method, the failure first-aid repair early warning levels are determined for 246 suspected failure intelligent meters, as shown in table 11. The fault checking accuracy of the intelligent meter at the prior early warning level (primary early warning, secondary early warning and tertiary early warning) can be found to be more than 50%, and the model has certain guiding significance in establishment.
The above embodiments and application examples are illustrative of specific embodiments of the present invention, and are not intended to limit the present invention, and those skilled in the art can make various changes and modifications to obtain equivalent technical solutions without departing from the spirit and scope of the present invention, and therefore all equivalent technical solutions should be included in the scope of the present invention.
Claims (7)
1. A power metering device fault first-aid repair active service early warning method is characterized by comprising the following steps:
the method comprises the steps that firstly, portrait analysis is carried out on power consumers based on big data of a power metering device of a power supply department, and the power consumers are divided into three categories, namely key consumers, important consumers and ordinary consumers according to importance from high to low;
secondly, constructing a fault recognition model of the electric power metering device, predicting high-risk, medium-risk or low-risk fault risks of the electric power metering device by using the trained fault recognition model, and outputting the electric power metering device with the fault risk prediction result of high risk and medium risk as a suspected fault electric power metering device;
and thirdly, correspondingly determining the priority level of the power metering device fault first-aid repair active service of each power user by combining the importance category of the power user determined in the first step and the risk prediction level output in the second step: and the early warning is divided into first to sixth levels from high to low.
2. The active service early warning method for fault emergency repair of the power metering device of claim 1, wherein the first step comprises the following specific steps:
collecting data: acquiring 4-dimensional information data tables of user basic information, power utilization behaviors, payment behaviors and appeal behaviors of each power user from a power supply department marketing service system, and correspondingly determining index variables corresponding to each dimension;
preprocessing data: respectively preprocessing data of each form based on four dimensions of user basic information, electricity utilization behaviors, payment behaviors and appeal behaviors, and correspondingly deleting and filling abnormal values and missing values in source data;
establishing a power user tag library:
extracting power features with higher power user value by adopting a traditional statistical analysis method including frequency analysis, centralized trend analysis, discrete degree analysis and visual analysis of data on the four-dimensional basic data of the user basic information, the power consumption behavior, the payment behavior and the appeal behavior after the preprocessing in the step II, and establishing a tag library of the power user by taking the four dimensions as a dimensional frame of the tag library of the user;
fourthly, the power consumer is portrayed based on the K-Means algorithm, and the power consumers are divided into key consumers, important consumers and common consumers from high to low according to the importance:
adopting operations including correlation analysis, unbalanced index elimination and discrete variable elimination to the four dimensional data of the basic information, the electricity consumption behavior, the payment behavior and the appeal behavior in the label library established in the step three to screen six clustering indexes including electricity consumption age, operation capacity, annual average settlement electric quantity, peak valley electric quantity coefficient, annual average payment times and annual average payment amount; then, a data set consisting of data with larger values is split by utilizing a Lauda criterion, the data set is independently used as a user group type and defined as a key user, the rest data sets are gathered into four types by utilizing a K-Means clustering algorithm, and the four types are respectively defined as follows according to the characteristics of a clustering center: important users, consumer-normal users, loyalty-normal users, peak-time-normal users; and the consumption type-common user, loyalty type-common user and peak period type-common user are classified into common users.
3. The active service early warning method for fault emergency repair of the power metering device according to claim 2, wherein in the first step, index variables included in the basic information of the user include: urban and rural categories, user categories, power utilization categories, power supply voltage, operating capacity, power utilization years and power types; the electricity consumption behavior comprises index variables as follows: the method comprises the following steps of settling electric quantity, settling electric charge, peak electric quantity, average electric quantity, valley electric quantity, peak electric charge, average electric charge, valley electric charge and whether peak-valley marking is executed or not; the payment behavior comprises the following index variables: payment mode, payment times, payment amount, overdue times, arrearage times and arrearage amount; the appeal behavior comprises index variables as follows: the number of complaints and the type of complaints.
4. The active service early warning method for fault emergency repair of the power metering device of claim 1, wherein the second step comprises the following specific steps:
collecting data required by modeling: taking the electric power metering devices of all the electric power users related to the first step as objects, and correspondingly acquiring 4-dimensional related information data tables of attribute characteristics, electric quantity characteristics, power characteristics and weather characteristics of the electric power metering devices from a marketing service system, an electric power consumption information acquisition system and a meteorological system of a power supply department;
preprocessing the acquired data required by modeling: preprocessing the data required by modeling collected in the fifth step by using a data processing method including data cleaning, data transformation and data protocols to form a data wide table;
extracting core indexes based on a random forest RF algorithm;
and (III) predicting the fault risk level based on the XGboost algorithm, and outputting the predicted high-risk and medium-risk electric power metering device as a suspected fault electric power metering device.
5. The active service early warning method for fault emergency repair of the electric power metering device according to claim 4, wherein the second step comprises the following steps:
the data cleaning comprises the processing operations of space processing of a data table, invalid field and abnormal data elimination, missing value processing, character type conversion and date format unification;
the data transformation and data reduction comprises:
i, the operation age is the date of meter disassembly or the difference age between the current date and the date of meter assembly;
II, dividing the weather grade into 0-7 grades from low to high according to the influence degree of different weather conditions on the fault of the electric power metering device;
III, calculating indexes of characteristic dimensions of the electric quantity;
and IV, calculating indexes of the power characteristic dimension.
6. The active service early warning method for fault first-aid repair of the power metering device of claim 4, wherein the step of the second step (c) is a specific method for extracting core indexes based on a random forest RF algorithm, and the specific method comprises the following steps:
extracting a preliminary index system from the four dimensions of the attribute feature, the weather feature, the electric quantity feature and the power feature of the electric power metering device by using descriptive statistics including the attribute feature, the concentration trend and the dispersion degree;
VI, extracting core indexes by using a random forest embedded feature selection algorithm, comprising the following steps:
a. taking a preliminary index system as model input, and taking whether a fault occurs as model output to construct a random forest RF algorithm model;
b. randomly distributing the alternative characteristic indexes to a plurality of decision trees, judging a training set by each decision tree, and calculating the importance of the alternative characteristic by using variable importance measurement based on the classification accuracy of the data outside the bag;
c. and setting a characteristic importance threshold, and screening out all indexes with the characteristic importance greater than the threshold as extracted core indexes.
7. The active service early warning method for fault emergency repair of the power metering device of claim 4, wherein the second step comprises the following specific steps:
constructing an XGboost model by taking the core indexes determined in the step (c) as input variables and whether faults occur as output variables on the basis of a set training set and a set testing set; optimizing two parameters of the maximum depth of the decision tree and the number of the decision trees of the model by adopting a grid search algorithm; obtaining an XGboost model under the optimal parameters while obtaining the optimal parameters through a grid search algorithm;
and VIII, applying the trained XGboost model to a test set, prejudging the failure risk prediction probability of the electric power metering device, dividing the failure risk prediction probability into three failure risk grades of high risk, medium risk and low risk, and outputting the data of the electric power metering device with high risk and medium risk as a suspected failure electric power metering device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010511520.4A CN111738462B (en) | 2020-06-08 | 2020-06-08 | Fault first-aid repair active service early warning method for electric power metering device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010511520.4A CN111738462B (en) | 2020-06-08 | 2020-06-08 | Fault first-aid repair active service early warning method for electric power metering device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111738462A true CN111738462A (en) | 2020-10-02 |
| CN111738462B CN111738462B (en) | 2022-08-30 |
Family
ID=72648556
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010511520.4A Active CN111738462B (en) | 2020-06-08 | 2020-06-08 | Fault first-aid repair active service early warning method for electric power metering device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111738462B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112308462A (en) * | 2020-11-23 | 2021-02-02 | 国网北京市电力公司 | Power consumer classification method and device |
| CN112685461A (en) * | 2020-12-15 | 2021-04-20 | 国网吉林省电力有限公司电力科学研究院 | Electricity stealing user judgment method based on pre-judgment model |
| CN112800036A (en) * | 2020-12-30 | 2021-05-14 | 银盛通信有限公司 | Report analysis chart automatic generation and display method and system |
| CN113095391A (en) * | 2021-04-06 | 2021-07-09 | 国网上海市电力公司市北供电公司 | Method, system, terminal and medium for detecting electricity stealing behavior of user |
| CN113313407A (en) * | 2021-06-16 | 2021-08-27 | 上海交通大学 | Enterprise power utilization behavior identification method and device |
| CN114236448A (en) * | 2021-11-23 | 2022-03-25 | 国网山东省电力公司日照供电公司 | Metering device troubleshooting system based on big data |
| CN117391458A (en) * | 2023-12-08 | 2024-01-12 | 四川省寰宇众恒科技有限公司 | Safety production risk detection and early warning method and system based on data analysis |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105809373A (en) * | 2016-04-27 | 2016-07-27 | 国网山东省电力公司济南市长清区供电公司 | Distribution network fault information comprehensive collection and management platform |
| CN106651424A (en) * | 2016-09-28 | 2017-05-10 | 国网山东省电力公司电力科学研究院 | Electric power user figure establishment and analysis method based on big data technology |
| CN109697574A (en) * | 2018-12-31 | 2019-04-30 | 国网浙江省电力有限公司杭州供电公司 | Small customer electricity Risk Identification Method in a kind of electric power |
| CN110222991A (en) * | 2019-06-10 | 2019-09-10 | 国网江苏省电力有限公司常州供电分公司 | Metering device method for diagnosing faults based on RF-GBDT |
| CN111178382A (en) * | 2019-11-26 | 2020-05-19 | 国网浙江省电力有限公司湖州供电公司 | Repair service prediction analysis method based on work order data mining |
-
2020
- 2020-06-08 CN CN202010511520.4A patent/CN111738462B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105809373A (en) * | 2016-04-27 | 2016-07-27 | 国网山东省电力公司济南市长清区供电公司 | Distribution network fault information comprehensive collection and management platform |
| CN106651424A (en) * | 2016-09-28 | 2017-05-10 | 国网山东省电力公司电力科学研究院 | Electric power user figure establishment and analysis method based on big data technology |
| CN109697574A (en) * | 2018-12-31 | 2019-04-30 | 国网浙江省电力有限公司杭州供电公司 | Small customer electricity Risk Identification Method in a kind of electric power |
| CN110222991A (en) * | 2019-06-10 | 2019-09-10 | 国网江苏省电力有限公司常州供电分公司 | Metering device method for diagnosing faults based on RF-GBDT |
| CN111178382A (en) * | 2019-11-26 | 2020-05-19 | 国网浙江省电力有限公司湖州供电公司 | Repair service prediction analysis method based on work order data mining |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112308462A (en) * | 2020-11-23 | 2021-02-02 | 国网北京市电力公司 | Power consumer classification method and device |
| CN112685461A (en) * | 2020-12-15 | 2021-04-20 | 国网吉林省电力有限公司电力科学研究院 | Electricity stealing user judgment method based on pre-judgment model |
| CN112800036A (en) * | 2020-12-30 | 2021-05-14 | 银盛通信有限公司 | Report analysis chart automatic generation and display method and system |
| CN113095391A (en) * | 2021-04-06 | 2021-07-09 | 国网上海市电力公司市北供电公司 | Method, system, terminal and medium for detecting electricity stealing behavior of user |
| CN113313407A (en) * | 2021-06-16 | 2021-08-27 | 上海交通大学 | Enterprise power utilization behavior identification method and device |
| CN114236448A (en) * | 2021-11-23 | 2022-03-25 | 国网山东省电力公司日照供电公司 | Metering device troubleshooting system based on big data |
| CN117391458A (en) * | 2023-12-08 | 2024-01-12 | 四川省寰宇众恒科技有限公司 | Safety production risk detection and early warning method and system based on data analysis |
| CN117391458B (en) * | 2023-12-08 | 2024-02-27 | 四川省寰宇众恒科技有限公司 | Safety production risk detection and early warning method and system based on data analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111738462B (en) | 2022-08-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111738462B (en) | Fault first-aid repair active service early warning method for electric power metering device | |
| CN106909933B (en) | A kind of stealing classification Forecasting Methodology of three stages various visual angles Fusion Features | |
| CN106154209B (en) | Electrical energy meter fault prediction technique based on decision Tree algorithms | |
| CN106780121B (en) | Power consumption abnormity identification method based on power consumption load mode analysis | |
| CN110097297A (en) | A kind of various dimensions stealing situation Intellisense method, system, equipment and medium | |
| CN109308571B (en) | Distribution line variable relation detection method | |
| CN111584952A (en) | Method and system for online evaluation of electrochemical cells of energy storage power station | |
| CN112859822B (en) | Equipment health analysis and fault diagnosis method and system based on artificial intelligence | |
| CN110222991B (en) | Metering device fault diagnosis method based on RF-GBDT | |
| CN109409444B (en) | Multivariate power grid fault type discrimination method based on prior probability | |
| CN104022552A (en) | Intelligent detection method for electric vehicle charging control | |
| CN109800995A (en) | A kind of grid equipment fault recognition method and system | |
| CN110738232A (en) | grid voltage out-of-limit cause diagnosis method based on data mining technology | |
| CN114862139A (en) | Data-driven transformer area line loss rate abnormity diagnosis method | |
| CN110675020A (en) | High-price low-access user identification method based on big data | |
| CN114118588A (en) | Peak-facing summer power failure prediction method based on game feature extraction under clustering undersampling | |
| CN111798333A (en) | Energy utilization evaluation and electricity utilization safety analysis method and system | |
| CN115758151A (en) | Combined diagnosis model establishing method and photovoltaic module fault diagnosis method | |
| CN113450031B (en) | Method and device for selecting intelligent energy consumption service potential transformer area of residents | |
| CN114493238A (en) | Power supply service risk prediction method, system, storage medium and computer equipment | |
| CN112734274B (en) | Low-carbon power grid operation leading influence factor mining and comprehensive evaluation method | |
| CN115879915B (en) | Cross-platform standardized overhaul method for power plant | |
| CN111239484A (en) | Non-invasive load electricity consumption information acquisition method for non-resident users | |
| CN114626433A (en) | Fault prediction and classification method, device and system for intelligent electric energy meter | |
| CN114662589A (en) | Ammeter fault research and judgment method, device, equipment and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |