Power consumer portrait construction method and device

Abstract

The invention discloses a power consumer portrait construction method, which comprises the steps of S1, acquiring long-term power consumption data of a power consumer, clustering the long-term power consumption data to obtain a classification cluster, and extracting features of the classification cluster to obtain a user portrait; s2, selecting a user from each classification cluster, obtaining short-term power consumption data of the selected user, and constructing a power consumption prediction model; s3, acquiring payment data and arrearage data of the selected user, and constructing an arrearage prediction model; s4, acquiring power consumption behavior data of the selected user, and constructing a power consumption behavior prediction model; and S5, perfecting the user portrait according to the electricity consumption prediction model, the arrearage prediction model and the electricity consumption behavior prediction model. The invention further provides a construction device. According to the invention, the accuracy and real-time performance of the user portrait can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field related to data mining, and more specifically, to a method and device for constructing a power user portrait. Background Art

[0002] Characterizing and accurately serving different users' electricity consumption characteristics and habits is conducive to improving user service experience in many aspects and realizing refined user management. However, in the prior art, the electricity user portrait is simply clustered using the k-means algorithm, but this method is relatively rough and is easily affected by the initially selected clustering center, and it is difficult to adapt to the continuous changes in user electricity consumption behavior. Therefore, it is necessary to design a technical solution that can improve the above defects. Summary of the invention

[0003] An object of the present invention is to provide a method and device for constructing an electric power user portrait, which can improve the accuracy and real-time performance of the user portrait.

[0004] In order to achieve these purposes and other advantages of the present invention, according to one aspect of the present invention, the present invention provides a method for constructing a power user portrait, including: S1: obtaining long-term power consumption data of power users, clustering the long-term power consumption data to obtain classification clusters, extracting features of the classification clusters, and obtaining user portraits; S2: selecting users from each of the classification clusters, and obtaining short-term power consumption data of the selected users, and constructing a power consumption prediction model; S3: obtaining payment data and arrears data of the selected users, and constructing arrears prediction model; S4: obtaining power consumption behavior data of the selected users, and constructing a power consumption behavior prediction model; S5: improving the user portrait according to the power consumption prediction model, the arrears prediction model and the power consumption behavior prediction model.

[0005] Furthermore, S1 includes: cleaning the acquired long-term electricity consumption data to remove outliers; standardizing or normalizing the data; clustering using the DBSCAN algorithm to obtain classification clusters; extracting the features of each classification cluster and describing it with semantic tags to obtain the user portrait.

[0006] Furthermore, S2 includes: selecting a predetermined proportion of electricity users from each of the classification clusters as the selected users; obtaining historical electricity consumption data and historical meteorological data of the selected users; using the electricity consumption data and meteorological data at previous time points as input and the electricity consumption data at a later time point as output, and using the LSTM network to construct the electricity consumption prediction model.

[0007] Furthermore, S3 includes: extracting the average payment cycle of the selected user, the time since the last payment, the cumulative number of arrears, the maximum arrears amount, the proportion of arrears amount, the proportion of on-time payments, and the payment amount fluctuation from the payment data and the arrears data, and establishing a training set and a test set in combination with the label of whether there is an arrears; using the training set to train a decision tree model to obtain the arrears prediction model.

[0008] Furthermore, the S4 includes: extracting differential characteristics of electricity consumption, sliding window characteristics of electricity consumption, peak characteristics of electricity power, daily electricity consumption periodicity, and monthly electricity consumption periodicity from the electricity consumption behavior data, and establishing a training set and a test set in combination with electricity consumption behavior labels; using the training set to train the MLP model to obtain the electricity consumption behavior prediction model.

[0009] Furthermore, in S5, the electricity consumption prediction model, the arrears prediction model and the electricity consumption behavior prediction model are used to predict the future electricity consumption, arrears probability and electricity consumption behavior of the electricity user to obtain a prediction result; the prediction result is used to supplement the user portrait and mark the contradictions.

[0010] Furthermore, in S5, the question and answer data of the power user is obtained, the semantics of the question and answer data is extracted, and the contradiction is judged using the semantics.

[0011] According to another aspect of the present invention, there is also provided an apparatus for constructing a power user profile, comprising a processor and a memory, wherein the memory is used to store program instructions, and the processor is used to call the program instructions to execute the power user profile constructing method.

[0012] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any of the methods for constructing a power user portrait is implemented.

[0013] The present invention has at least the following beneficial effects: The present invention utilizes the long-term electricity consumption data of power users to construct user portraits, and utilizes short-term electricity consumption data to construct an electricity consumption prediction model, utilizes payment data and arrears data to construct an arrears prediction model, utilizes electricity consumption behavior data to construct an electricity consumption behavior prediction model, and utilizes an electricity consumption prediction model, an arrears prediction model, and an electricity consumption behavior prediction model to improve the user portraits; the present invention utilizes an electricity consumption prediction model, an arrears prediction model, and an electricity consumption behavior prediction model to supplement and improve the user portraits in real time, thereby improving the accuracy and real-time performance of the user portraits.

[0014] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of an embodiment of the present application. DETAILED DESCRIPTION

[0016] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0017] It should be understood that the terms such as "having", "including" and "comprising" used in the embodiments of the present application do not exclude the existence or addition of one or more other elements or their combinations. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present application are only used to explain the relative position relationship, movement, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly. When an element is referred to as "fixed on" or "set on" another element, it can be directly on the other element or there may be a centering element at the same time. When an element is referred to as "connecting" another element, it can be directly connected to another element or it can be indirectly connected to another element through a centering element. The description of "first", "second", etc. in the embodiments of the present application is only for descriptive purposes, and cannot be understood as indicating or implying its relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features.

[0018] It should be noted that the technical solutions between the various embodiments of the present application can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0019] like Figure 1 As shown, an embodiment of the present application provides a method for constructing a power user portrait, including: S1: Acquire long-term electricity consumption data of power users, cluster the long-term electricity consumption data to obtain classification clusters, extract features of the classification clusters, and obtain user portraits; For example, the user power consumption data at different time granularities (such as every 15 minutes, every hour, every day, every month, etc.) and related power consumption parameter information such as power consumption and power factor are obtained from the power metering system; by connecting with the database of the power metering system, the detailed power consumption data records corresponding to each power user are extracted according to the user's unique identifier (such as user number), ensuring that the data covers a sufficiently long time range, such as the past one, two years or even longer, so as to fully reflect the long-term characteristics and changing trends of the user's power consumption behavior; Exemplarily, the K-Means clustering algorithm or the DBSCAN density clustering algorithm is used for clustering. The K-Means clustering algorithm is a commonly used distance-based clustering algorithm. The principle is to divide data points into K different clusters so that the sum of the distances from the data points in each cluster to the center of the cluster is minimized. For example, when clustering the long-term electricity consumption data of power users, it is determined by the elbow method that K=3 is more appropriate, and then the preprocessed electricity consumption data is input into the K-Means algorithm for clustering operation, and finally the users are divided into 3 different clusters. The DBSCAN density clustering algorithm clusters based on the density of data points, divides density-connected data points into the same cluster, can find clusters of any shape, and can identify noise points in the data (i.e., isolated points that do not belong to any cluster). When using it, two key parameters need to be specified: the neighborhood radius and the minimum number of points, which determine the density definition of the data points and the conditions for the formation of clusters. Through multiple experiments and adjustments of these two parameters, a better clustering effect can be obtained, and the number of clusters does not need to be specified in advance.

[0020] Exemplarily, the characteristics of the classification cluster include user type, electricity consumption scale, electricity consumption characteristics, electricity consumption time pattern, etc., to form a user portrait; S2: Selecting a user from each of the classification clusters, obtaining the short-term electricity consumption data of the selected user, and building an electricity consumption prediction model; For example, a certain number of users are randomly selected from each classification cluster, such as 30% of each cluster, as selected users; selecting users from multiple classification clusters can reduce redundant samples, reduce the number of samples used to a certain extent, and cover various types of electricity consumption as much as possible, thereby improving the reliability of the electricity consumption prediction model; Exemplarily, based on short-term electricity consumption data, a regression model, a machine learning model or a neural network model is used to build an electricity consumption prediction model, where the short-term electricity consumption data refers to electricity consumption data for 7 days, 10 days or 20 days, so as to reflect the changes in electricity consumption of users; S3: Obtain payment data and arrears data of the selected user, and build an arrears prediction model; For example, payment data and arrears data are obtained from the power marketing system, from which detailed payment records of users can be obtained, including the time, amount, payment method of each payment, and arrears related information, such as the time when the arrears occurred, the amount of arrears, the duration of the arrears, whether the arrears have been settled, etc. After preprocessing the above data, a machine learning model or a neural network model is used to build an arrears prediction model; S4: Acquire the electricity consumption behavior data of the selected user and construct an electricity consumption behavior prediction model; Exemplarily, features reflecting the user's electricity consumption behavior, such as electricity consumption time period, peak power consumption, and power consumption fluctuation, are extracted from short-term electricity consumption data. After preprocessing, a linear regression model, a nonlinear regression model, a time series analysis model, a machine learning model, and a deep learning model are used to build an electricity consumption behavior prediction model; S5: improving the user portrait according to the power consumption prediction model, the arrears prediction model and the power consumption behavior prediction model; Exemplarily, the power consumption prediction model, the arrears prediction model and the power consumption behavior prediction model are used to predict the power user's future power consumption, arrears probability and power consumption behavior to obtain a prediction result, and the prediction result is used to supplement the user portrait to obtain a more accurate user portrait that adapts to user changes, and the contradictions are marked to remind the staff to handle the contradictions; For example, the predicted electricity consumption, arrears and electricity consumption behavior are analyzed in correlation to see if there are any logical inconsistencies or contradictions. For example, if the electricity consumption forecast shows that future electricity consumption will increase significantly, but the arrears forecast result is low risk and the payment situation has been good, then it is necessary to further analyze whether the user has a new source of funds or a change in electricity plan and other reasonable reasons leading to this situation, to ensure that the contents of each part of the portrait are mutually confirmed and in line with the actual situation. Invite business experts in the power field to review and optimize the improved user portrait based on their professional knowledge and actual work experience. For example, experts may review the user's profile based on the industrial policy adjustments in the user's area. Different opinions may be expressed on the prediction results of household electricity consumption and arrears risk, or based on the common electricity consumption behavior characteristics of similar users in the industry, it is recommended to supplement or modify certain portrait descriptions to make them more suitable for actual business scenarios and more accurately reflect the user's real electricity consumption characteristics and potential risks; as time goes by and new data is continuously generated, the above steps are repeated regularly (such as monthly, quarterly, etc., determined according to business needs), and each prediction model is re-run with the latest data to update and iterate the user portrait to ensure that the portrait can reflect the user's latest electricity consumption, arrears risk and changing trends in electricity consumption behavior in real time, so as to provide continuous and accurate decision-making basis for the power company's marketing, operation, service and other tasks.

[0021] It can be seen that this embodiment uses long-term electricity consumption data clustering to obtain the initial user portrait, and then uses the electricity consumption prediction model, arrears prediction model and electricity consumption behavior prediction model that can perform short-term predictions to supplement and improve the user portrait in real time, thereby improving the accuracy and real-time performance of the user portrait.

[0022] In another embodiment, S1 includes: cleaning the acquired long-term electricity consumption data to remove outliers; standardizing or normalizing the data; clustering using the DBSCAN algorithm to obtain classification clusters; extracting features of each classification cluster and describing it with semantic tags to obtain the user portrait; Exemplarily, a suitable neighborhood radius (eps) and minimum number of points (min_samples) are selected, such as when the eps value is between 0.3 and 0.5 and the min_samples value is between 5 and 10, and the standardized or normalized long-term electricity consumption data is input into the DBSCAN algorithm for clustering calculation. The algorithm will automatically divide the data points into different clusters and mark the noise points. The statistical indicators such as the average electricity consumption, median, maximum, minimum and standard deviation of electricity consumption of users in each cluster are calculated to fully understand the overall situation and discreteness of the electricity consumption scale of the cluster users. Calculate the month-on-month (compared with the previous cycle) and year-on-year (compared with the same period last year) change rates of electricity consumption of users in the cluster, and observe whether there are seasonal change patterns (, long-term growth or decline trends, etc.), so as to characterize the electricity consumption behavior characteristics of the users in the cluster in different time dimensions. Count the proportion of electricity consumption or the average power consumption of users in each cluster in different time periods of the day (such as one hour as a period), draw a daily electricity consumption time distribution curve, and observe whether there are obvious peak and trough periods of electricity consumption and the concentration and regularity of these periods; analyze the differences in electricity consumption and power consumption of users in the cluster on weekdays and weekends, calculate the average and ratio of electricity consumption indicators on weekdays and weekends, and understand the change pattern of users' electricity consumption within a week. For example, for clusters where some commercial users are located, the electricity consumption on weekends is significantly higher than that on weekdays, reflecting the correlation between their business activities and rest days; while for clusters where some industrial users are located, the electricity consumption on weekdays will be higher and less volatile, reflecting their stable electricity consumption mode based on production scheduling. Calculate the average power consumption of users in each cluster, compare the average power size between different clusters, and distinguish them. The power consumption scale of different users and the intensity of power equipment use are analyzed. The degree of power fluctuation within a certain period of time is measured by calculating statistical indicators such as the standard deviation of power consumption. The fluctuation reflects the stability of the user's power consumption behavior. The power fluctuation of industrial users may be related to factors such as production process and equipment start-stop sequence, and the fluctuation amplitude is larger and the regularity is relatively stronger; the daily power fluctuation of residential users is relatively frequent but the amplitude is smaller, which is mainly affected by the switch operation of electrical appliances. The power consumption behavior characteristics of users in different clusters are further refined by analyzing the power fluctuation situation. Based on the various features extracted above, a user portrait with semantic labels is constructed for each classification cluster, so that it can intuitively reflect the power consumption behavior characteristics of the user group in the cluster. For example, the stable type of residential power consumption means that the cluster is mainly composed of residential users, and the power consumption behavior has a high stability. The fluctuating type of commercial power consumption indicates that the cluster is mainly composed of commercial users, and the power consumption behavior has obvious fluctuations in different time periods; the production type of industrial power consumption indicates that the cluster is mainly industrial users, and the power consumption behavior revolves around production activities and has a certain regularity and a large scale of power consumption.

[0023] In another embodiment, S2 includes: selecting a predetermined proportion of power users from each of the classification clusters as the selected users; obtaining historical power consumption data and historical meteorological data of the selected users; using the power consumption data and meteorological data at previous time points as input and the power consumption data at a later time point as output, and constructing the power consumption prediction model using an LSTM network; Exemplarily, meteorological data includes temperature, precipitation, humidity, wind speed, etc. The collected historical electricity consumption data and historical meteorological data of each selected user are divided in chronological order to form continuous time series segments; for example, with days as the time unit, the electricity consumption and meteorological data of a certain user in the past month can be arranged in sequence to form a time series with a length of 30, and then the appropriate time window size is set. This window size determines the length of historical data used for prediction by the model each time, that is, the data of the previous few time points is used as input. According to the dimension of the input features (i.e., the number of electricity consumption data and meteorological data features in the time window) and the requirements of the prediction task, a suitable LSTM network structure is designed, including an input layer, one or more hidden layers, and an output layer. Initialize each parameter in the LSTM network (such as weight matrix, bias vector, etc.), divide the preprocessed and divided time series data (according to the set time window, with electricity consumption data and meteorological data at multiple time points as input, and the corresponding electricity consumption at the next time point as output) into training set, validation set and test set. Usually divided according to a certain ratio, such as 70%-80% of the data as training set for model learning, 10%-20% of the data as validation set for adjusting model parameters and preventing overfitting, and the remaining 10%-20% of the data as test set for final evaluation of the generalization ability of the model on unseen data; then, use the training set data to train the LSTM network. During the training process, the input data is passed into the input layer of the network in sequence, and the predicted electricity consumption output is obtained through forward propagation calculation, and then the error between the predicted output and the actual electricity consumption (commonly used error functions such as mean square error MSE, mean absolute error MAE) is used to calculate the predicted electricity consumption output. The back propagation algorithm is used to propagate the error from the output layer to the input layer, and the parameters in the network (weight matrix and bias vector, etc.) are updated. This process is repeated continuously. After multiple training rounds (Epoch), the model is continuously optimized, the prediction error is gradually reduced, and the prediction accuracy of electricity consumption is improved.

[0024] In another embodiment, S3 includes: extracting the average payment cycle, the time since the last payment, the cumulative number of arrears, the maximum arrears amount, the arrears amount ratio, the on-time payment ratio, and the payment amount fluctuation of the selected user from the payment data and the arrears data, and establishing a training set and a test set in combination with the arrears label; using the training set to train a decision tree model to obtain the arrears prediction model; Exemplarily, for the average payment cycle: calculate the number of days between two consecutive payment times for each selected user, and then find the average of these intervals to get the average payment cycle; time since the most recent payment: use the current time as a reference point to calculate how many days have passed since the last payment of each user. This feature can intuitively reflect the user's recent payment activity. The longer the time is from now, the more likely it is that the risk of arrears is gradually increasing. Especially when it exceeds the normal payment cycle for a long time, it is more necessary to pay attention to the possibility of arrears; cumulative number of arrears: count the cumulative number of arrears for each user. The cumulative number of arrears directly reflects the frequency of arrears in the user's history. The more times, the more serious the arrears problem of the user in the past, and the greater the possibility of arrears in the future. It is one of the key indicators for measuring arrears risk; maximum arrears amount: find out the maximum amount of arrears for each user in all past arrears The maximum amount of arrears in the electricity bill record; The proportion of arrears: calculate the ratio of the cumulative arrears to the cumulative paid amount to measure the proportion of arrears in the overall electricity cost; The proportion of on-time payments: by comparing the time when the user should pay each time (determined according to the payment cycle and relevant regulations) and the actual payment time, the proportion of on-time payments to the total number of payments is calculated. The proportion of on-time payments can clearly show the user's payment willingness and payment habits. Users with a higher proportion usually have a lower risk of arrears; Fluctuation of payment amount: calculate the standard deviation of the user's payment amount each time to measure the degree of fluctuation of the payment amount; The extracted features (average payment cycle, time since the last payment, cumulative number of arrears, maximum arrears amount, arrears amount ratio, on-time payment ratio, payment amount fluctuation) are used as independent variables, and whether there is arrears (0 means no arrears, 1 means arrears, and the user is labeled according to whether there is arrears later) is used as the dependent variable to construct a data set; the data set is randomly divided into a training set and a test set according to a certain ratio; Before using the training set to train the decision tree model, you need to set some related parameters, which will affect the growth of the decision tree and the final model performance; common parameters include: maximum depth (max_depth): limit the maximum number of layers of the decision tree to prevent the tree from growing too complex and overfitting; minimum sample split (min_samples_split): specifies the minimum number of samples that a node must contain before it is split into child nodes; minimum sample leaf (min_samples_leaf): indicates the minimum number of samples that each leaf node must contain at least, which has a similar function to the minimum sample split, and is also to prevent overfitting and ensure that the leaf nodes are representative; The feature data in the training set and the corresponding label data of whether the payment is in arrears are organized in the format required by the decision tree model (usually the data is organized into a two-dimensional array or a data frame), and then passed to the decision tree model for training; during the training process, the decision tree model will automatically adjust the tree structure and node division rules according to the relationship between each feature in the training data and the arrears results, and continuously optimize the model by calculating information gain (such as using criteria such as information entropy or the Gini index to measure the ability of features to distinguish classification results, and selecting features with the largest information gain for node division), so that the model can predict whether the user will owe fees as accurately as possible based on the input features.

[0025] In another embodiment, S4 includes: extracting differential features of power consumption, sliding window features of power consumption, peak features of power consumption, daily power consumption periodicity, and monthly power consumption periodicity from the power consumption behavior data, and establishing a training set and a test set in combination with power consumption behavior labels; training an MLP model using the training set to obtain the power consumption behavior prediction model; For example, the differential characteristics of electricity consumption: calculate the difference in electricity consumption at adjacent time points to observe whether the trend of electricity consumption changes is stable, eliminate trend and seasonal factors in the data, and facilitate subsequent models to better capture the inherent laws of electricity consumption changes; sliding window characteristics of electricity consumption: extract statistical characteristics of electricity consumption within a certain time window to capture the fluctuations and laws of users' electricity consumption behavior in the short term, and understand the stability of users' recent electricity consumption and changes in electricity usage habits; peak characteristics of electricity power: count the time points and peak sizes of each user's electricity power reaching peak values ​​in different time periods such as every day and every week, analyze the laws of power peak occurrence (such as whether it is concentrated in certain specific time periods) and the changing trends of peak values, which is very helpful for in-depth understanding of the use of users' electrical equipment Habitual and concentrated electricity consumption are of great significance; Daily electricity consumption periodicity: Analyze the distribution of electricity consumption of users at different times of the day to see whether there are obvious peak and trough periods of electricity consumption, and the repetition of these periods within a week; Monthly electricity consumption periodicity: Consider the fluctuation of users' monthly electricity consumption, whether it is affected by factors such as seasons and production plans (for industrial users) and shows periodic changes. For example, residential users use a lot of air conditioners in summer, so monthly electricity consumption will peak in the summer months, and industrial users will significantly increase electricity consumption in the peak production season. This periodic feature is quantified by calculating the average and standard deviation of monthly electricity consumption, as well as the month-on-month and year-on-year change rates of electricity consumption between different months, and converts it into a numerical form that can be recognized and learned by the model; The above-mentioned extracted features (differential features of electricity consumption, sliding window features of electricity consumption, peak features of electricity power, daily electricity consumption periodicity, monthly electricity consumption periodicity) are used as independent variables, and electricity consumption behavior labels (for example, electricity consumption behavior can be classified and labeled according to different business needs, such as the advance or postponement of peak electricity consumption hours, the growth or decline trend of electricity consumption, etc. Here, specific categories or values ​​are used to represent different electricity consumption behaviors, which are determined according to actual analysis and prediction goals, such as peak electricity consumption type, valley electricity consumption type, and stable electricity consumption type) are used as dependent variables to construct a data set. According to a certain proportion (usually about 70% - 80% of the data is divided into training sets for model learning and training; MLP is a simple feedforward neural network, which consists of an input layer, one or more hidden layers, and an output layer. The neurons are connected by appropriate weights. The number of input layer nodes depends on the number of input features. For example, in the above-mentioned extracted electricity consumption behavior features, there are 5 features of different dimensions, so the input layer has 5 nodes. The number of neurons in the hidden layer can be set according to experience and experiments. Different numbers of hidden layers and different combinations of the number of neurons in each layer will affect the model's expressive power and learning effect. Its function is to perform nonlinear transformation on the input features and to mine the complex nonlinear mapping relationship between the input features and the output target (electricity consumption behavior label). The number of output layer nodes is determined according to the dimension of the predicted electricity consumption behavior. For example, if the predicted electricity consumption behavior is peak electricity consumption type, valley electricity consumption type, or stable electricity consumption type, then the number of output layer nodes is 3. During the training process, data is passed from the input layer, and after weighted summation and activation function (such as the commonly used Sigmoid function and ReLU function, which are used to introduce nonlinear factors and enhance the model's expressiveness) of the neurons in the hidden layer, it is passed to the output layer to obtain the prediction result. Then, by comparing the difference between the prediction result and the actual power consumption behavior label (common error measurement indicators such as mean square error MSE, mean absolute error MAE, etc.), the backpropagation algorithm is used to backpropagate the error from the output layer to the input layer, and the weights of each connection in the network are adjusted according to the error. This process is repeated continuously, and after multiple training rounds (epochs), the model is continuously optimized and the prediction error is gradually reduced, thereby achieving accurate prediction of power consumption behavior.

[0026] In another embodiment, in S5, the question and answer data of the power user is further obtained, the semantics of the question and answer data is extracted, and the contradiction is judged by using the semantics; For example, the question-and-answer data of power users mainly come from multiple aspects. The first is the customer service system of the power company. When users encounter problems related to electricity use, they often seek help from the power company by calling the customer service phone or consulting online. The customer service staff will record the content of each question-and-answer interaction, including the questions raised by the user, the specific time of the consultation, the corresponding customer service reply content and other detailed information; in addition, some interactive communities and FAQ sections set up on the official websites or mobile applications of some power companies will also retain users' questions and replies given by other users or power company staff, which also constitute a rich source of question-and-answer data; In order to accurately extract the semantics from question-answering data, we need to use natural language processing (NLP) related technologies; commonly used methods include lexical analysis, syntactic analysis, and semantic role labeling; Then, semantic role labeling or pre-trained language models are used to further extract semantics to assist in determining contradictions; the extracted semantics of the question and answer data are correlated and compared with other relevant data (such as the user portraits constructed in the previous steps, electricity behavior data, electricity consumption prediction results, arrears prediction results, and electricity behavior prediction results, etc.). For example, from the user portrait, it is known that the user is a stable residential electricity user, and his daily electricity equipment and electricity habits are relatively fixed. According to common sense, the fluctuation of electricity consumption should be small; and from the semantics of the question and answer data, it is extracted that the user reflects that the electricity bill has increased significantly and suspects that the meter is faulty, which forms a possible contradiction point and requires further in-depth analysis; through this semantic-based contradiction judgment, the links where there may be problems can be accurately located so that corresponding solutions can be taken to improve the quality and accuracy of power services.

[0027] An embodiment of the present application also provides an apparatus for constructing an electric user profile, comprising a processor and a memory, wherein the memory is used to store program instructions, and the processor is used to call the program instructions to execute the method for constructing an electric user profile; an embodiment of the present application also provides an object detection apparatus, comprising a processor and a memory, wherein the memory is used to store program instructions, and the processor is used to call the program instructions to execute the method for constructing an electric user profile; the apparatus of this embodiment may be a mobile phone, a laptop computer, a tablet computer, etc., which has a memory and a processor arranged therein to execute the method for constructing an electric user profile of the above-mentioned embodiment.

[0028] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements any of the methods for constructing a power user profile; in this embodiment, the computer-readable medium can be a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device, etc., which can store a computer program to execute the method for constructing a power user profile of the above-mentioned embodiment.

[0029] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A method for constructing a power user portrait, characterized in that: include: S1: Acquire long-term electricity consumption data of power users, cluster the long-term electricity consumption data to obtain classification clusters, extract features of the classification clusters, and obtain user portraits; S2: Selecting a user from each of the classification clusters, obtaining the short-term electricity consumption data of the selected user, and building an electricity consumption prediction model; S3: Obtain payment data and arrears data of the selected user, and build an arrears prediction model; S4: Acquire the electricity consumption behavior data of the selected user and construct an electricity consumption behavior prediction model; S5: Improve the user portrait according to the electricity consumption prediction model, the arrears prediction model and the electricity consumption behavior prediction model.

2. The method for constructing a power user portrait according to claim 1, characterized in that: The S1 includes: Clean the acquired long-term electricity consumption data and remove outliers; Standardize or normalize the data; Use DBSCAN algorithm to perform clustering and obtain classification clusters; The features of each classification cluster are extracted and described with semantic tags to obtain the user portrait.

3. The method for constructing a power user portrait according to claim 1, characterized in that: The S2 includes: Selecting a predetermined proportion of power users from each of the classification clusters as the selected users; Acquiring historical electricity consumption data and historical meteorological data of the selected user; The electricity consumption data and meteorological data at previous time points are used as input, and the electricity consumption data at a later time point is used as output, and the electricity consumption prediction model is constructed using an LSTM network.

4. The method for constructing a power user portrait according to claim 1, characterized in that: The S3 includes: Extract the average payment cycle, the time since the last payment, the cumulative number of arrears, the maximum arrears amount, the arrears amount ratio, the on-time payment ratio, and the payment amount fluctuation of the selected user from the payment data and the arrears data, and establish a training set and a test set in combination with the arrears label; The decision tree model is trained using the training set to obtain the arrears prediction model.

5. The method for constructing a power user portrait according to claim 1, characterized in that: The S4 includes: Extracting differential features of electricity consumption, sliding window features of electricity consumption, peak features of electricity power, daily electricity consumption periodicity, and monthly electricity consumption periodicity from the electricity consumption behavior data, and establishing a training set and a test set in combination with electricity consumption behavior labels; The MLP model is trained using the training set to obtain the electricity consumption behavior prediction model.

6. The method for constructing a power user portrait according to claim 1, characterized in that: In S5, the power consumption prediction model, the arrears prediction model and the power consumption behavior prediction model are used to predict the power user's future power consumption, arrears probability and power consumption behavior to obtain a prediction result; The prediction results are used to supplement the user portrait and mark inconsistencies.

7. The method for constructing a power user portrait according to claim 6, characterized in that: In S5, the question-and-answer data of the power user is obtained, the semantics of the question-and-answer data is extracted, and the contradiction is judged using the semantics.

8. A device for constructing a power user portrait, characterized in that: It includes a processor and a memory, the memory is used to store program instructions, and the processor is used to call the program instructions to execute the method for constructing a power user portrait as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing a power user portrait according to any one of claims 1 to 7 is implemented.

Images