Method for analyzing and identifying cross-domain features of electric power big data

Abstract

The invention provides an analysis and identification method for cross-domain characteristics of electric power big data, and the method comprises the steps: extracting daily power utilization behavior characteristics and load curve characteristics of a user according to the real-time collection data of a power generation end and a power utilization end; judging the accuracy of the extracted daily power consumption behavior characteristics of the user; if the daily electricity consumption behavior characteristics of the user are accurate, collecting electric power measurement data and electricity consumption feedback data, and extracting personalized characteristics of the user; identifying different types of power consumer groups according to the personalized features of the power consumers; using a PCA principal component algorithm to extract physical characteristics of generality of different types of power consumer groups; based on the common physical characteristics of the users, designing load control management decision schemes suitable for different power user groups; and designing and customizing a personalized decision-making scheme by combining the common physical characteristics and the difference characteristics of the power consumers.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for analyzing and identifying cross-domain features of electric power big data. Background Art

[0002] The power industry is rapidly developing with advancements in related technologies and the widespread application of artificial intelligence. This has resulted in rapidly increasing electricity consumption, diversifying electricity usage scenarios, and an increasing number of various types of electrical equipment. However, current power decision-making solutions are not effectively adapted to today's complex electricity demands and remain plagued by various issues. First, existing decision-making solutions fail to fully consider the individual characteristics of different types of users, resulting in inconsistent decision-making outcomes. Second, there is a lack of effective means to accurately extract individual user characteristics and apply them to the decision-making process. Furthermore, comprehensive analysis of both the common physical characteristics and differential characteristics of different user groups is currently insufficient, making it difficult to tailor precise decision-making solutions for each user. Therefore, the refined characterization of power users is crucial for addressing these issues. Summary of the Invention

[0003] The present invention provides a method for analyzing and identifying cross-domain features of power big data, which mainly includes:

[0004] Based on the real-time data collected from the power generation and consumption ends, the user's daily power consumption behavior characteristics and load curve characteristics are extracted; the accuracy of the extracted user's daily power consumption behavior characteristics is judged; if the user's daily power consumption behavior characteristics are accurate, the power metering data and power consumption feedback data are collected to extract the user's personalized characteristics; based on the personalized characteristics of power users, different types of power user groups are identified; the PCA principal component algorithm is used to extract the common physical characteristics of different types of power user groups; based on the common physical characteristics of users, load control management decision-making plans suitable for different power user groups are designed; based on the power users' electricity billing details and user experience data, the Apriori algorithm is used to extract the user's personalized differential characteristics; combined with the common physical characteristics and differential characteristics of power users, customized personalized decision-making plans are designed.

[0005] In some embodiments, extracting the user's daily electricity usage behavior characteristics and load curve characteristics based on the real-time collected data from the power generation end and the power consumption end includes:

[0006] The sensors are used to collect data from the power generation end in real time, including the status of the generator set, output power, fuel consumption and the operating status of the power generation equipment; the sensors are used to collect data from the power consumption end in real time, including the equipment status, power, energy consumption, voltage, current and power factor; the data are pre-processed, the power consumption time and duration are discretized, the day is divided into multiple time periods, the duration is divided into several intervals, and the power consumption equipment is classified and coded; the Apriori algorithm is used to extract the various daily power consumption behavior characteristics and load curve characteristics of users; the power consumption load curve is obtained by analyzing the power change data of the power consumption equipment, and the characteristics of the user's power consumption equipment are extracted. The characteristics of the user's power consumption time distribution are extracted based on the start and stop time and power consumption duration of the power consumption equipment, including the peak and valley periods. By analyzing the simultaneity between different power consumption equipment, the associated characteristics of the user's power consumption equipment are extracted, including whether the power consumption equipment are used at the same time and the duration of use. By analyzing the changing trend of the power load curve, the stability, volatility and periodicity characteristics of the load are extracted. By calculating the peak and valley values ​​in the load curve, the load peak and peak-valley difference reflecting the maximum demand and fluctuation intensity of the load are extracted. By calculating the fluctuation range of the load curve within the preset time period, the load change amplitude is obtained.

[0007] In some embodiments, determining the accuracy of the extracted user's daily electricity usage behavior characteristics includes:

[0008] Based on the acquired characteristics of users' daily electricity usage behavior, the KNN algorithm d(x, y) = sqrt((x1-y1)^2+(x2-y2)^2+...+(xn-yn)^2) is used to calculate the distance between each sample and other samples, and the k training samples closest to it in the feature space are found; the frequency of occurrence of the sample category is calculated to determine the probability that the feature sample belongs to the real user category. If the judgment probability exceeds the set threshold, the extracted characteristics of users' daily electricity usage behavior are judged to be accurate and can accurately reflect the users' actual electricity usage scenarios; based on the users' electricity metering data and feedback information, a sample feature data set is established, and the KNN algorithm is used to calculate the distance metric between each sample and determine the similarity between the samples; if the similarity between a sample and a sample of a certain user group exceeds the threshold, the sample is judged to belong to this user group; for samples that are judged incorrectly, the original data is continuously collected to expand the capacity of the sample data set, enrich the dimensions of the sample features, and the KNN calculation is repeated until the accuracy of the judgment results of all samples reaches the preset threshold, and the extracted user features are confirmed to be the user's true features.

[0009] In some embodiments, if the user's daily electricity usage behavior characteristics are accurate, then electricity metering data and electricity usage feedback data are collected to extract the user's personalized characteristics, including:

[0010] If the characteristics of users' daily electricity consumption behavior are accurate, then for different types of users, their electricity metering data sets are collected, including users' detailed order data, smart meter reading data, and electricity user query data; for heterogeneous electricity metering data of different types of users, the data sets are classified, and the electricity metering data sets are sorted and layered according to user type, data type, and time dimension to obtain the classified electricity metering data sets; based on the classification of the electricity metering data sets, by setting the parameters of the Gaussian mixture clustering model and selecting different numbers of categories, the electricity metering data sets are quickly and automatically classified and aggregated, and a large amount of user electricity metering data is efficiently organized; after the electricity metering data sets are organized, different data sets corresponding to different types of user groups are collected. The LSA algorithm is used to identify potential topics in the electricity metering dataset, extract feature words and keywords from the dataset, and determine the common electricity consumption topics of electricity user groups. After extracting the common topic features of each user group, the user's electricity consumption feedback data is collected, including users' subjective feedback scores on electricity prices, services, and comfort, as well as evaluation texts, to form a user feedback dataset. Based on the user feedback dataset, the Gaussian mixture clustering model algorithm is used to train the user text feature vector space to obtain personalized electricity consumption portrait features of users and determine their personalized preferences and characteristics. The common features of user groups and user portrait features are combined to identify user personalized features, and further subdivide user subgroups as a decision-making basis for precise user electricity consumption management.

[0011] In some embodiments, identifying different types of power user groups based on personalized characteristics of power users includes:

[0012] Based on the personalized sample data of electricity users, including the user's electrical equipment information, power consumption preference data, and power consumption feedback portrait labels; construct a sample feature vector space based on the user's personalized sample data, use the one-hot encoding method to digitize the user features, convert them into vector representations, and input the kmeans clustering algorithm; according to the preset power user group type, set the number of clustering categories K of the KMeans algorithm to be equal to the number of user group types, representing residential users, commercial users, and industrial users; randomly initialize the centroid vector of all user group types as the center point of the group category; traverse all the personalized sample feature vectors of electricity users, calculate the vector and all user The Euclidean distance between the centroids of the user group types is used to determine the similarity between the sample and the center of each class; the sample is divided into the class with the highest similarity to form a preliminary cluster of similar users; based on the sample category division results, the center vectors of each class are recalculated and updated as the new centroid vectors of the user group; the sample classification and centroid update are repeated until the centroid coordinates are stable and the sample category no longer changes; the centroid vectors representing the typical characteristics of residential users, commercial users, and industrial users are finally determined, and the user group is identified through the sample category results; by adjusting the parameter K value of the KMeans algorithm, the user group is subdivided into subcategories, including home users, office users, and high-power industrial users.

[0013] In some embodiments, the extracting of common physical features of different types of power user groups using a PCA principal component algorithm includes:

[0014] For the electricity user groups of residential, commercial and industrial types that have been identified, detailed electricity consumption original data sets of different types of users are collected; a user group sample matrix is ​​constructed, and the sample characteristic variables are the original physical data indicators, including the user's load curve data, energy consumption data, and electricity consumption time distribution; the sample matrix is ​​standardized and preprocessed using the PCA principal component analysis algorithm to improve the comparability between characteristic variables; the preprocessed sample matrix is ​​subjected to feature decomposition, the eigenvectors and eigenvalues ​​of the sample matrix are extracted, and the sample matrix is ​​sorted in order of contribution rate; the sample matrix is ​​projected according to the sorted principal component eigenvectors, and several principal components whose contribution rates add up to a preset threshold are selected to form a new low-dimensional feature subspace; the original features are reduced in dimensionality to remove redundant variables; after conversion to the low-dimensional feature subspace, the sample coordinate values ​​are recalculated to highlight the differential features between samples to form a new sample feature matrix; the new sample feature matrix has the common physical characteristics of the same type of user groups, including the type of electrical equipment, power consumption, electricity consumption and peak electricity consumption time; the new sample feature matrix also contains the differential features of individual users.

[0015] In some embodiments, the design of load control management decision schemes applicable to different groups of power users based on the common physical characteristics of the users includes:

[0016] Based on the common physical characteristic matrix data of different power user groups, the characteristic variable data of sample users are selected; the common physical characteristic matrix data are reconstructed using the ALM algorithm to reduce the matrix dimension, remove redundant variables, and extract and summarize the core characteristics of different types of users; according to the user category and characteristic evaluation indicators, the core characteristics of each category are weighted and fused to form the overall characteristics of the representative group; based on the group characteristics, the curve fitting and error minimization methods are used to design a basic step-by-step electricity price response model suitable for different user categories; a physical model is constructed based on the power load data of the user group to extract load characteristics; based on the load characteristic target space and strategy space, a multivariate load management decision tree is constructed to extract the optimized peak-valley scheduling strategy for different user groups; based on the decision tree results, robustness and constraint conditions are set to obtain the Top-K candidate cluster strategy with the best ranking after pruning; the candidate strategy indicators are evaluated to select the load control management decision scheme suitable for different power user groups.

[0017] In some embodiments, the method of extracting user-specific differential features based on the electricity billing details and user experience data of the power user using the Apriori algorithm includes:

[0018] Generate an electricity consumption dataset based on the electricity billing details of the electricity user, and the data fields include the user's equipment electricity consumption information and billing status; collect user experience data, including subjective scores and comment texts on electricity consumption feedback, to form a user experience dataset; use the Apriori algorithm based on the electricity consumption dataset and the user experience dataset to extract personalized difference characteristics of various users, including the user's electricity consumption characteristics, electricity consumption time characteristics, electricity consumption equipment characteristics, electricity consumption type characteristics, electricity consumption service evaluation characteristics, power supply reliability characteristics, electricity safety characteristics and electricity consumption behavior preference characteristics; obtain the user's electricity consumption information, electricity consumption time information, electricity consumption equipment information and electricity consumption type information based on the electricity billing details data, and obtain the user's electricity consumption service evaluation information, power safety information and electricity consumption behavior preference information based on the user experience data; calculate the user's average electricity consumption and peak-to-valley electricity consumption ratio to obtain the user's electricity consumption characteristics; analyze the user's peak electricity consumption period and electricity consumption activity to determine the user The system can identify the time characteristics of electricity consumption of users; identify the main electricity-consuming equipment of users and calculate the proportion of electricity-consuming equipment to obtain the characteristics of electricity-consuming equipment of users; count the distribution of electricity consumption types of users to determine the characteristics of electricity consumption types of users; analyze the data of users' satisfaction with electricity services and the number of complaints to obtain the evaluation characteristics of electricity services; count the number of power outages and the duration of power outages of users to determine the characteristics of power supply reliability of users; analyze the evaluation of users on power safety and the number of accidents to obtain the characteristics of power safety of users; count the data of users' preference for energy-saving behavior and electricity bill payment methods to determine the characteristics of users' electricity consumption behavior preferences; extract the electricity-related habit patterns of users that exceed the support and confidence thresholds through repeated iterative calculations to judge the personalized characteristics of users; extract the differential characteristics of individual users that are different from the average level by comparing the feature correlation results of different users; identify key active users through personalized electricity consumption parameters, segment ordinary electricity users, and distinguish different differential electricity consumption groups.

[0019] In some embodiments, the design of a customized personalized decision-making solution based on the common physical characteristics and differentiated characteristics of electricity users includes:

[0020] The system obtains users' common physical characteristic data and collects differential characteristics that reflect user personality differences; selects commonly found physical characteristics in user profiles based on the common physical characteristics of electricity users; selects features in user profiles with inter-individual differences exceeding a threshold based on the differential characteristics of users; integrates the selected common physical characteristics and differential characteristics to construct an evaluation model; uses the KNN algorithm to determine the weight of each feature in the evaluation model; evaluates the role of features in user profiles through correlation analysis; adjusts the parameters of the evaluation model based on the weight and role of each feature; inputs users' common physical characteristics and differential characteristics into the evaluation model to obtain evaluation results for user profiles; performs feature matching based on the current target user's data and user profiles, selects the customized electricity price plan set with the highest degree of match to the target user's personality profile from the candidate electricity price plan library, and recommends it to the user for selection; utilizes the user's load curve and electricity usage time data to design a peak-valley tiered electricity price and a flexible electricity price framework that meets common preferences; after the plan is implemented, continues to collect user feedback, optimizes the evaluation model, and conducts decision-making effect evaluation and optimization iterations to achieve continuous precision in targeted electricity price plans.

[0021] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0022] This invention discloses a method for analyzing and identifying cross-domain features of power big data. This method can more accurately understand and predict users' power usage behavior, providing personalized services and decision-making support for the power industry. Furthermore, by identifying different types of power user groups, more refined power management measures can be provided for different users, improving the efficiency and stability of the power system. This method, which integrates multiple technologies, has broad application prospects and can achieve more intelligent and sustainable development in the power industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of a method for analyzing and identifying cross-domain features of power big data of the present invention.

[0024] Figure 2 This is a schematic diagram of a method for analyzing and identifying cross-domain features of power big data according to the present invention. DETAILED DESCRIPTION

[0025] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] In this embodiment, a method for analyzing and identifying cross-domain features of power big data may specifically include:

[0027] S101. Extract the user's daily electricity consumption behavior characteristics and load curve characteristics based on the real-time data collected from the power generation end and the power consumption end.

[0028] Sensors collect real-time data from the power generation side, including generator set status, output power, fuel consumption, and power generation equipment operating status. Sensors also collect real-time data from the power consumption side, including equipment status, power, energy consumption, voltage, current, and power factor. Data preprocessing discretizes power usage time and duration, dividing a day into multiple time periods and duration into several intervals. Furthermore, power consumption devices are classified and coded. The Apriori algorithm is used to extract various daily power usage behavior characteristics and load curve features. By analyzing power variation data from power consumption devices, a load curve is generated, and user power consumption characteristics, including the number and type of power consumption devices, are extracted. Based on the start and stop times and power consumption duration of power consumption devices, the user's power consumption time distribution characteristics, including peak and off-peak periods, are extracted. By analyzing the simultaneity between different power consumption devices, user power consumption association characteristics are extracted, including whether power consumption devices are used simultaneously and for how long. By analyzing the changing trends of the load curve, load stability, volatility, and periodicity are extracted. By calculating the peak and valley values ​​in the load curve, the peak and valley differences, reflecting the maximum load demand and the intensity of fluctuation, are extracted. The load variation range is obtained by calculating the fluctuation range of the load curve within a preset time period.

[0029] For example, data on generator set status is collected, including: whether the generator set is operating normally, with a value of 1 indicating normal and 0 indicating abnormal; output power data, with the current output power being 1000kW; fuel consumption data, with the current fuel consumption rate being 10L / h; power generation operation status data, with a value of 1 indicating running and 0 indicating stopped; and device status data, with a value of 1 indicating normal and 0 indicating abnormal. Sensors collect real-time data from power users, including current device power of 500W, device energy consumption of 2kWh, voltage of 220V, current of 10A, and current power factor of 9. Data is preprocessed to divide a day into three time periods: morning, afternoon, and evening, represented by 1, 2, and 3, respectively. Data is also preprocessed to divide power usage duration into three intervals: 0-1 hour, 1-2 hours, and 2-3 hours, represented by 1, 2, and 3, respectively. Electrical devices are classified and coded, with lighting devices coded as 1, televisions coded as 2, and air conditioners coded as 3. The Apriori algorithm extracts behavioral characteristics of users who frequently use both the TV and the air conditioner simultaneously, and the load curve shows a pattern of gradually increasing before stabilizing. Based on the start and stop times of electrical devices and the duration of electricity use, the user's electricity usage is most frequently used between 7:00 PM and 9:00 PM. Analysis of the simultaneity between different electrical devices reveals that the user typically uses both the TV and the air conditioner simultaneously for two hours. Analysis of the load curve reveals that the load has minimal stability, volatility, and periodicity. Calculating the peaks and valleys in the load curve reveals a peak of 1200 kW and a peak-to-valley difference of 300 kW. Calculating the load curve's fluctuation range within a preset time period reveals that the load fluctuation range is 200 kW between 8:00 AM and 10:00 AM.

[0030] S102: Determine the accuracy of the extracted user's daily electricity usage behavior characteristics.

[0031] Based on the acquired user's daily electricity usage behavior characteristics, the KNN algorithm (d(x, y) = sqrt((x1 - y1)^2 + (x2 - y2)^2 + ... + (xn - yn)^2)) is used to calculate the distance between each sample and other samples, and the k nearest training samples in the feature space are found. The frequency of the sample category is calculated to determine the probability that the feature sample belongs to the real user category. If the probability exceeds a set threshold, the extracted user's daily electricity usage behavior characteristics are considered accurate and accurately reflect the user's actual electricity usage scenarios. Based on the user's electricity metering data and feedback information, a sample feature dataset is established. The KNN algorithm is used to calculate the distance metric between each sample and determine the similarity between the samples. If the similarity between a sample and a sample from a specific user group exceeds a threshold, the sample is considered to belong to that user group. For samples that are incorrectly identified, raw data is continuously collected to expand the sample dataset capacity and enrich the sample feature dimensionality. The KNN calculation is repeated until the accuracy of the judgment results for all samples reaches a preset threshold, confirming that the extracted user features are true features.

[0032] For example, the feature vectors of the daily electricity consumption behaviors of the collected users are as follows: User A: [1, 2, 3, 4, 5]; User B: [2, 4, 6, 8, 10]; User C: [3, 6, 9, 12, 15]; User D: [4, 8, 12, 16, 20]; User E: [5, 10, 15, 20, 25]. Find the k training samples closest to User A, and set k = 3, that is, find the 3 samples most similar to User A. First, calculate the distances between User A and other users: The distance S1 from User A to User B = sqrt((1 - 2)^2 + (2 - 4)^2 + (3 - 6)^2 + (4 - 8)^2 + (5 - 10)^2) = 55^1 / 2 = 7.41; The distance S2 from User A to User C = sqrt((1 - 3)^2 + (2 - 6)^2 + (3 - 9)^2 + (4 - 12)^2 + (5 - 15)^2) = 220^1 / 2 = 14.83; The distance S3 from User A to User D = sqrt((1 - 4)^2 + (2 - 8)^2 + (3 - 12)^2 + (4 - 16)^2 + (5 - 20)^2) = 495^1 / 2 = 22.24; The distance S4 from User A to User E = sqrt((1 - 5)^2 + (2 - 10)^2 + (3 - 15)^2 + (4 - 20)^2 + (5 - 25)^2) = 880^1 / 2 = 29.66. Since 7.41 < 14.83 < 22.24 < 29.66, that is, S1 < S2 < S3 < S4, the three users closest to User A are User B, User C, and User D. Calculate the frequency of occurrence of the categories of these three users: User B belongs to category 1, User C belongs to category 2, and User D belongs to category 2. Therefore, the frequency of category 1 is 1 / 3, and the frequency of category 2 is 2 / 3. If the set threshold is 2 / 3, then the probability that User A is determined to be category 2 is 2 / 3, which exceeds the set threshold. Therefore, it can be judged that the extracted daily electricity consumption behavior characteristics of the user are accurate.

[0033] S103. If the daily electricity consumption behavior characteristics of the user are accurate, then collect power metering data and electricity consumption feedback data, and extract the personalized characteristics of the user.

[0034] If the characteristics of users' daily electricity usage behavior are accurate, electricity metering datasets are collected for different types of users, including detailed billing data, smart meter readings, and electricity user query data. The heterogeneous electricity metering data from different user types is classified, sorted, and stratified by user type, data type, and time dimension to generate the classified electricity metering datasets. Based on the classification of the electricity metering datasets, the parameters of the Gaussian mixture clustering model are set and the number of clusters is selected to quickly and automatically classify and aggregate the electricity metering datasets, efficiently organizing large amounts of user electricity metering data. After organizing the electricity metering datasets, the LSA algorithm is used to identify potential topics in the electricity metering datasets for different datasets corresponding to different types of user groups. Feature words and keywords are extracted from the electricity metering datasets to identify common electricity usage themes across the user groups. After extracting the common thematic features for each user group, user feedback data is collected, including subjective feedback ratings on electricity costs, service, and comfort, as well as textual reviews, to form a user feedback dataset. Based on the user feedback datasets, the Gaussian mixture clustering model algorithm is used to train the user text feature vector space to obtain personalized electricity usage profiles and identify individual user preferences and characteristics. By integrating the common characteristics of user groups and user portrait characteristics, we can identify user personalized characteristics and further segment user subgroups as the basis for decision-making on precise user electricity management.

[0035] For example, an electricity metering dataset was collected for different types of users, including each user's electricity bill data, smart meter readings, and electricity user query data. The dataset was categorized and stratified by the data type (bill data, reading data, and query data), and by the time dimension (year, quarter, and month). The user's bill data, reading data, and query data were sorted and stratified by year, and then further broken down by year, quarter, and month. After the electricity metering dataset was categorized, a Gaussian mixture clustering model was used to automatically cluster and aggregate the dataset. By adjusting the clustering model parameters and selecting a different number of clusters, the electricity metering data of a large number of users can be quickly and efficiently organized. Next, the LSA algorithm was used to identify potential topics in the electricity metering dataset. By extracting feature words and keywords, common electricity usage themes for each user group were identified, including energy conservation and environmental protection, family life, and comfort. Next, user feedback data was collected, including subjective feedback ratings on electricity costs, service, and comfort, as well as textual reviews, to form a user feedback dataset. Based on a user feedback dataset, a Gaussian mixture clustering model algorithm was used to train a user text feature vector space to obtain personalized electricity usage profiles. It was determined that User A is concerned with energy conservation and environmental protection, while User B is concerned with comfort and service quality. These profiles were used to determine the individual preferences and characteristics of each user. Finally, by combining the common characteristics of user groups with the individual characteristics of each user, the user subgroups were further segmented into energy conservation, environmental protection, good comfort, and good service quality. These segmented user subgroups serve as the basis for decision-making in targeted user electricity management.

[0036] S104. Identify different types of power user groups based on the personalized characteristics of power users.

[0037] Based on the collected personalized sample data of electricity users, including information about their electrical devices, power consumption preferences, and user feedback profile labels, a sample feature vector space is constructed based on this personalized sample data. One-hot encoding is used to digitize user features and convert them into vector representations for input into the kmeans clustering algorithm. Based on the predefined electricity user group types, the number of cluster categories K in the KMeans algorithm is set equal to the number of user group types, representing residential, commercial, and industrial users. The centroid vectors of all user group types are randomly initialized as the cluster center points. The Euclidean distance between the vectors and the centroids of all user group types is calculated to determine the similarity between the sample and each cluster center. The sample is then assigned to the cluster with the highest similarity, forming a preliminary cluster of similar users. Based on the sample classification results, the centroid vectors of each cluster are recalculated and updated to serve as the new centroid vectors for the user groups. This iterative process of sample classification and centroid updating continues until the centroid coordinates stabilize and the sample categories remain stable. Finally, the centroid vectors representing the typical characteristics of residential, commercial, and industrial users are determined, and user groups are identified based on the sample classification results. By adjusting the parameter K value of the KMeans algorithm, the user group can be subdivided into subcategories, including home users, office users, and high-power industrial users.

[0038] For example, when constructing personalized user sample features, we collected data on device types, energy consumption classifications, peak-use equipment, and energy-saving feedback surveys from 1,000 electricity users. Using one-hot encoding, these categorical features were converted into multidimensional 0-1 vectors. Then, the KMeans algorithm parameters were set to k = 3. The weighted centroid of the sample distances was iteratively calculated, and the centroid coordinates were adjusted until the centroid position was stable. Ultimately, 300 samples with features close to the first centroid were identified as residential users, 500 samples close to the second centroid were identified as commercial users, and 200 samples closest to the third centroid were identified as industrial users. To further divide user subgroups, we collected more sample features, enriched the sample dimensionality, and increased the number of categories k. Setting k = 5, we subdivided residential users into two categories: ordinary household users and high-end users, distinguishing groups with different consumption levels. After iterative convergence, we calculated the weighted distances of each sample from the five category centroids. Finally, the sample with the closest weighted distance to the category centroid was assigned to the corresponding subgroup. During the iterative clustering process, a stopping threshold for centroid updates is set. When the proportion of sample exchange categories caused by changes in centroid coordinates in two consecutive iterations is less than 5%, the centroid is considered to be basically stable and the user category is basically determined, reducing the amount of computation and accelerating the algorithm's convergence. Various kernel transformations are used in distance calculations to make linearly inseparable datasets linearly separable in the mapping space. The kernel matrix is ​​used to avoid directly calculating the inner product of high-dimensional vectors, reducing the amount of computation.

[0039] S105. Use the PCA principal component algorithm to extract common physical characteristics of different types of power user groups.

[0040] For the identified residential, commercial, and industrial electricity user groups, detailed raw electricity usage datasets for each user type are collected. A sample matrix for the user group is constructed, with the sample characteristic variables being raw physical data indicators, including user load curve data, energy consumption data, and electricity usage time distribution. The sample matrix is ​​standardized and preprocessed using the principal component analysis (PCA) algorithm to improve comparability between characteristic variables. Eigen decomposition is performed on the preprocessed sample matrix to extract the eigenvectors and eigenvalues, which are then sorted by contribution rate. A projection transformation is performed on the sample matrix based on the sorted principal component eigenvectors. Principal components whose summed contribution rates exceed a preset threshold are selected to form a new low-dimensional feature subspace. Dimensionality reduction is performed on the original features to remove redundant variables. After transforming to the low-dimensional feature subspace, sample coordinate values ​​are recalculated to highlight the distinctive features between samples, forming a new sample feature matrix. This new sample feature matrix incorporates the common physical characteristics of user groups of the same type, including device type, power consumption, electricity consumption, and peak hours. It also incorporates distinctive features of individual users.

[0041] For example, a detailed raw data set of electricity usage for residential, commercial, and industrial users is collected. There are 10 users, each with 100 time-point load curve data, total energy consumption, and time distribution data. A 10x303 sample matrix is ​​constructed, where 303 = 100 load curve data points + 1 total energy consumption point + 202 time distribution points. The sample matrix is ​​preprocessed using the principal component analysis (PCA) algorithm to standardize the sample matrix. The load curve data is standardized to a mean of 0 and a standard deviation of 1. Eigendecomposition is performed on the preprocessed sample matrix, resulting in 10 eigenvectors and corresponding eigenvalues. After sorting by contribution rate, the principal components with the top 80% contribution rate (i.e., the first 8 principal components) are selected. A projection transformation is performed on the sample matrix based on the contribution-ranked principal component eigenvectors, resulting in a new 8-dimensional feature subspace. This subspace retains the primary information of the original features while reducing the dimensionality and removing redundant variables. This results in a new 10x8 sample feature matrix, where each sample has 8 features. Calculate the score of each user on each principal component, evaluate the similarities and differences between different users, analyze the contribution of different principal components to user classification, and count the proportion of explained variance of each principal component.

[0042] S106. Based on the common physical characteristics of users, design load control management decision-making plans suitable for different power user groups.

[0043] Based on the common physical characteristic matrix data obtained for different power user groups, characteristic variable data for sample users was selected. The ALM algorithm was used to reconstruct the common physical characteristic matrix data, reducing the matrix dimension, removing redundant variables, and extracting and summarizing the core characteristics of different user types. Based on user categories and characteristic evaluation indicators, the core characteristics of each category were weighted and integrated to form the overall characteristics of representative groups. Based on the group characteristics, curve fitting and error minimization methods were used to design a basic step-by-step electricity price response model suitable for different user categories. A physical model was constructed based on the power load data of each user group to extract load characteristics. Based on the load characteristic target space and strategy space, a multivariate load management decision tree was constructed to extract the optimal peak-valley scheduling strategy for different user groups. Based on the decision tree results, robustness and constraints were set to obtain the top-K candidate cluster strategies with the best ranking after pruning. The candidate strategy indicators were evaluated to select load control management decision solutions suitable for different power user groups.

[0044] For example, consider three different types of electricity user groups: residential, industrial, and commercial. First, common physical characteristic matrix data is collected for these user groups, including electricity consumption, electricity usage time, and peak-to-valley difference. Then, characteristic variable data for sample users are selected to form a feature matrix. For residential users, electricity consumption and electricity usage time are selected as feature variables; for industrial users, electricity consumption, electricity usage time, and peak-to-valley difference are selected as feature variables; and for commercial users, electricity consumption and peak-to-valley difference are selected as feature variables. Using the ALM algorithm, the feature matrix is ​​decomposed into three matrices: a matrix representing the common characteristics of the user groups, a matrix representing the differences in characteristics between different types of users, and a matrix representing feature weights. This yields the core characteristics of the three different user groups. Weights are assigned to different user categories, and the core characteristics are weighted averaged based on the user group's electricity consumption, electricity usage time, and peak-to-valley difference indicators. Based on the core characteristics of different user groups, a basic step-by-step electricity price response model tailored to each user category is designed using curve fitting and error minimization methods. By fitting user load curves, the peak-to-valley price difference for each user category is determined to guide user electricity usage behavior. Minimizing total user costs and maximizing system power utilization are optimization objectives. By adjusting electricity prices and load management strategies, coordinated load control for different user groups is achieved. A physical model is then constructed based on the power load data for each user group, plotting the time-scale variation of power load. A 24-hour cycle is divided into 12 time periods, extracting daily load characteristics, including peak and valley values, average values, and maximum and minimum load time periods. Based on the load characteristic target space and strategy space, a multivariate load management decision tree is constructed to extract optimized peak and valley scheduling strategies for different user groups. Based on user load characteristics and optimization objectives, different load management strategies are designed, including timed scheduling and dynamic scheduling. Eligible candidate strategies are screened, taking into account system capacity constraints and user demand constraints. The optimal load control management solution is selected by comparing the cost and power utilization of candidate strategies.

[0045] S107. Based on the electricity billing details and user experience data of the power user, use the Apriori algorithm to extract user personalized difference features.

[0046] An electricity consumption dataset is generated based on electricity billing details. The data fields include information on the electricity consumption of each user's device and billing status. User experience data, including subjective ratings and textual comments on electricity consumption feedback, is collected to form a user experience dataset. Using the Apriori algorithm, the user's individualized characteristics are extracted based on the electricity consumption and user experience datasets. These characteristics include electricity consumption, time of use, device characteristics, type of use, service evaluation, power supply reliability, security, and user behavior preferences. Based on the electricity billing details, user electricity consumption information, time of use, device information, and type of use are obtained. Based on the user experience data, user service evaluation, security, and user behavior preferences are obtained. Average electricity consumption and peak-to-valley ratios are calculated to determine the user's electricity consumption characteristics. Peak hours and user activity are analyzed to determine the user's time of use characteristics. The user's primary devices are identified and their proportions are calculated to determine the user's device characteristics. The distribution of user types of use is analyzed to determine the user's electricity type characteristics. Analyze user satisfaction and complaint data regarding electricity services to derive user evaluation characteristics for electricity services. Count user evaluations of power outages and outage duration to determine user characteristics for power supply reliability. Analyze user evaluations of power safety and the number of accidents to derive user characteristics for power safety. Count user preferences for energy-saving behaviors and electricity bill payment methods to determine user characteristics for electricity usage behavior. Through repeated iterative calculations, extract user electricity usage-related habit patterns that exceed support and confidence thresholds to determine user personalized characteristics. Compare feature association results across different users to extract differential characteristics that distinguish individual users from the average level. Identify key active users through personalized electricity usage parameters, segment ordinary electricity users, and distinguish between different groups of differentiated electricity users.

[0047] For example, the Apriori algorithm is used to analyze detailed electricity billing data for a group of users, including information on electricity consumption, time of use, devices used, type of electricity used, electricity service evaluation, power supply reliability, power safety, and electricity usage preferences. User A's average daily electricity consumption is 1000 kWh, with a peak-to-valley ratio of 70:30, thus obtaining his electricity usage characteristics. Analysis of user A's daily peak hours and user activity reveals that user A's peak hours are from 6:00 PM to 10:00 PM, with the highest level of activity, thus obtaining his time characteristics. By identifying user A's main electrical devices and calculating their proportion, it is found that air conditioners and refrigerators are the primary appliances, accounting for 70% of all electricity used, thus obtaining his equipment characteristics. By analyzing the distribution of user A's electricity usage types, it is found that household electricity consumption is the primary appliance, accounting for 80%, thus obtaining his electricity type characteristics. By analyzing user A's satisfaction with electricity service, which is a perfect score, and the number of complaints received, it is obtained his electricity service evaluation characteristics. User A's power supply reliability characteristics are determined by the fact that they experienced two power outages and lasted two hours. User A's power safety characteristics are determined by analyzing their rating of 4 out of 5 and their zero accidents. User A's preferred energy-saving behavior and their preference for online electricity bill payment are also determined.

[0048] S108. Design and customize personalized decision-making plans based on the common physical characteristics and differentiated characteristics of electricity users.

[0049] Collect common physical characteristic data from users and simultaneously collect differential characteristics that reflect individual differences. Based on the common physical characteristics of electricity users, select common physical characteristics within the user profile. Based on the differential characteristics of users, select features within the user profile that exceed a threshold for inter-individual variability. Combine the selected common physical characteristics with differential characteristics to construct an evaluation model. Use the KNN algorithm to determine the weight of each feature in the evaluation model. Correlation analysis is used to evaluate the role of features in the user profile. The parameters of the evaluation model are adjusted based on the weights and roles of each feature. Input the common physical characteristics and differential characteristics of users into the evaluation model to obtain an evaluation result for the user profile. Feature matching is performed between the current target user's data and the user profile. From the candidate electricity price plan library, the customized electricity price plan with the highest degree of match to the target user's profile is selected and recommended to the user. Leveraging user load profiles and electricity usage time data, design a peak-valley tiered electricity price and a flexible electricity price framework that meets common preferences. After the plan is implemented, continue to collect user feedback, optimize the evaluation model, and conduct iterative optimization and evaluation of decision effectiveness to achieve continuous refinement of targeted electricity price plans.

[0050] For example, common physical characteristic data for users is obtained, including electricity consumption, electricity usage time, duration, and peak-to-valley difference. Differential features reflecting individual user differences are collected, including electricity consumption, usage time, equipment, type, service evaluation, reliability, safety, and preference. From the common physical characteristics, electricity consumption, duration, and peak-to-valley difference are selected, while from the differentiated features, equipment, type, service evaluation, reliability, safety, and preference are selected. An evaluation model is constructed by integrating these common and differentiated features. The KNN algorithm is used to determine the weight of each feature in the evaluation model. The weights are ranked in descending order: electricity consumption, duration, peak-to-valley difference, equipment, type, service evaluation, reliability, safety, and preference. Correlation analysis is used to evaluate the role of these features, revealing that the electricity service evaluation feature plays a crucial role in user profiling, reflecting user expectations for electricity supply service quality. The evaluation model parameters are adjusted based on the weight and impact of each feature to improve its accuracy. Common physical and differential features of users are input into the evaluation model to generate an evaluation result for the user profile. Feature matching is performed between the current user A's data and the user profile. Based on the characteristics of user A's peak electricity consumption at noon, prolonged electricity consumption, and electricity consumption exceeding 1000 kWh, a customized electricity price set with low prices during the noon period, high prices during other periods, and high power consumption is selected from the candidate electricity price library and recommended to user A. Using the load curves and electricity usage time data of all users, a peak-valley tiered electricity price is designed to meet these common characteristics, with a peak price of 1.5 yuan and a valley price of 0.6 yuan. A differentiated flexible electricity price framework is established based on the peak and valley times of individual users. If user A's electricity consumption peaks in the morning and valleys in the afternoon, the electricity price for user A in the morning will be 1.5 yuan and 0.6 yuan in the afternoon. After the implementation of the plan, we will continue to collect user feedback, optimize the evaluation model, and conduct iterative optimization of decision-making effect evaluation to achieve continuous precision of targeted electricity price plans.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing and identifying cross-domain features of power big data, characterized by: The method comprises: Based on the real-time data collected from the power generation and consumption ends, the user's daily power consumption behavior characteristics and load curve characteristics are extracted; the accuracy of the extracted user's daily power consumption behavior characteristics is judged; if the user's daily power consumption behavior characteristics are accurate, the power metering data and power consumption feedback data are collected to extract the user's personalized characteristics; based on the personalized characteristics of power users, different types of power user groups are identified; the PCA principal component algorithm is used to extract the common physical characteristics of different types of power user groups; based on the common physical characteristics of users, load control management decision-making plans suitable for different power user groups are designed; based on the power users' electricity billing details and user experience data, the Apriori algorithm is used to extract the user's personalized differential characteristics; combined with the common physical characteristics and differential characteristics of power users, customized personalized decision-making plans are designed.

2. The method according to claim 1, wherein The method of extracting the user's daily electricity consumption behavior characteristics and load curve characteristics based on the real-time data collected from the power generation end and the power consumption end includes: Sensors are used to collect real-time data from the power generation end, including generator set status, output power, fuel consumption, and power generation equipment operating status. Sensors are also used to collect real-time data from the power consumption end, including equipment status, power, energy consumption, voltage, current, and power factor. Data is pre-processed to discretize power consumption time and duration, dividing a day into multiple time periods and duration into several intervals. Furthermore, power consumption equipment is classified and coded. The Apriori algorithm is used to extract various daily electricity consumption behavior characteristics and load curve characteristics of users; By analyzing the power change data of electrical devices, the power load curve is obtained and the characteristics of the user's electrical devices are extracted, including the number and type of electrical devices. Based on the start and stop times of the electrical devices and the duration of power consumption, the user's power consumption time distribution characteristics are extracted, including peak and off-peak periods. By analyzing the simultaneity between different electrical devices, the user's electrical device association characteristics are extracted, including whether the electrical devices are used simultaneously and for how long. By analyzing the changing trend of the power load curve, the stability, volatility and periodicity characteristics of the load are extracted; By calculating the peak and valley values ​​in the load curve, the load peak and peak-to-valley difference reflecting the maximum load demand and fluctuation intensity are extracted; by calculating the fluctuation range of the load curve within a preset time period, the load change amplitude is obtained.

3. The method according to claim 1, wherein The accuracy of the extracted user's daily electricity usage behavior characteristics is determined, including: Based on the acquired user's daily electricity usage behavior characteristics, the KNN algorithm d(x, y) = sqrt((x1-y1)^2+(x2-y2)^2+...+(xn-yn)^2) is used to calculate the distance between each sample and other samples, and to find the k training samples closest to it in the feature space; The frequency of occurrence of sample categories is calculated to determine the probability that the feature sample belongs to the real user category. If the probability exceeds the set threshold, the extracted user's daily electricity usage behavior characteristics are judged to be accurate and can accurately reflect the user's actual electricity usage scenario. Based on the user's electricity metering data and feedback information, a sample feature dataset is established. The KNN algorithm is used to calculate the distance metric between each sample and determine the similarity between the samples. If the similarity between a sample and a sample of a certain user group exceeds a threshold, the sample is determined to belong to this user group; For samples with incorrect judgments, continue to collect original data, expand the capacity of the sample data set, enrich the dimensions of the sample features, and repeat the KNN calculation until the accuracy of the judgment results of all samples reaches the preset threshold, then confirm that the extracted user features are the user's true features.

4. The method according to claim 1, wherein If the user's daily electricity usage behavior characteristics are accurate, then electricity metering data and electricity feedback data are collected to extract the user's personalized characteristics, including: If the characteristics of users' daily electricity usage behavior are accurate, then electricity metering datasets are collected for different types of users, including user detailed bill data, smart meter reading data, and electricity user query data. The heterogeneous electricity metering data of different types of users are classified and sorted and layered by user type, data type, and time dimension to obtain the classified electricity metering datasets. Based on the classification of power metering data sets, by setting the parameters of the Gaussian mixture clustering model and selecting different numbers of categories, the power metering data sets can be quickly and automatically classified and aggregated, effectively organizing a large amount of user power metering data. After organizing the electricity metering dataset, the LSA algorithm is used for different datasets corresponding to different types of user groups to identify potential topics of the electricity metering dataset, extract feature words and keywords of the electricity metering dataset, and determine the common electricity consumption topics of the electricity user groups; after extracting the common topic features of each user group, the user's electricity consumption feedback data is further collected, including the user's subjective feedback ratings and evaluation texts on electricity bills, services and comfort, to form a user feedback dataset; based on the user feedback dataset, the Gaussian mixture clustering model algorithm is used to train the user text feature vector space to obtain the user's personalized electricity consumption portrait features and determine the user's personalized preferences and characteristics; the common features of the user group and the user portrait features are integrated to identify the user's personalized features, and further subdivide the user subgroups as the decision-making basis for precise user electricity consumption management.

5. The method according to claim 1, wherein The method of identifying different types of power user groups based on the personalized characteristics of power users includes: Based on the personalized sample data of electricity users, including the user's power equipment information, power consumption preference data and power consumption feedback profile labels; Based on the personalized sample data of users, a sample feature vector space is constructed. The user features are digitized using the one-hot encoding method and converted into vector representations for input into the kmeans clustering algorithm. According to the preset power user group types, the number of clustering categories K of the KMeans algorithm is set to be equal to the number of user group types, representing residential users, commercial users, and industrial users. The centroid vectors of all user group types are randomly initialized as the center points of the group categories. The personalized sample feature vectors of all power users are traversed, and the Euclidean distance between the vector and the centroid of all user group types is calculated to determine the similarity between the sample and the center of each class. The sample is divided into the class with the highest similarity to it, and a cluster of similar users is initially formed. Based on the sample category division results, the center vectors of each class are recalculated and updated as the new centroid vectors of the user group. Repeat the iterative sample classification and centroid update until the centroid coordinates are stable and the sample category no longer changes; finally, the centroid vector representing the typical characteristics of residential users, commercial users, and industrial users is determined, and the user group is identified through the sample category results; by adjusting the parameter K value of the KMeans algorithm, the user group is subdivided into subcategories, including home users, office users, and high-power industrial users.

6. The method according to claim 1, wherein The PCA principal component algorithm is used to extract the common physical characteristics of different types of power user groups, including: For the identified residential, commercial, and industrial power user groups, detailed original electricity usage datasets for different types of users were collected. A user group sample matrix was constructed, with the sample characteristic variables being raw physical data indicators, including the user's load curve data, energy consumption data, and electricity usage time distribution. The sample matrix was standardized and preprocessed using the PCA principal component analysis algorithm to improve the comparability between characteristic variables. The preprocessed sample matrix is ​​subjected to eigendecomposition, and the eigenvectors and eigenvalues ​​of the sample matrix are extracted and sorted in order of contribution rate; the sample matrix is ​​projected according to the sorted principal component eigenvectors, and several principal components whose contribution rates add up to more than a preset threshold are selected to form a new low-dimensional feature subspace; the original features are reduced in dimensionality to remove redundant variables; after conversion to the low-dimensional feature subspace, the sample coordinate values ​​are recalculated to highlight the differential features between samples and form a new sample feature matrix; the new sample feature matrix has the common physical characteristics of the same type of user groups, including the type of electrical equipment, power consumption, electricity consumption and peak electricity consumption time; the new sample feature matrix also contains the differential features of individual users.

7. The method according to claim 1, wherein The load control management decision-making scheme suitable for different power user groups is designed based on the common physical characteristics of the users, including: Based on the obtained common physical characteristic matrix data of different power user groups, characteristic variable data of sample users are selected; the ALM algorithm is used to reconstruct the common physical characteristic matrix data, reduce the matrix dimension, remove redundant variables, and refine and summarize the core characteristics of different types of users; Based on user categories and feature evaluation indicators, the core features of each category are weighted and integrated to form the overall characteristics of the representative group; Based on group characteristics, a basic step-by-step electricity price response model suitable for different user categories is designed using curve fitting and error minimization methods. A physical model is constructed based on the power load data of the user group to extract load characteristics. Based on the load characteristic target space and strategy space, a multivariate load management decision tree is constructed to extract the optimized peak-valley scheduling strategies for different user groups. Based on the decision tree results, robustness and constraint conditions are set to obtain the Top-K candidate cluster strategies with the best ranking after pruning. The candidate strategy indicators are evaluated to select load control management decision schemes suitable for different power user groups.

8. The method according to claim 1, wherein Based on the electricity billing details and user experience data of the power user, the Apriori algorithm is used to extract user personalized difference features, including: Generate an electricity consumption dataset based on the electricity billing details of the power user, the data fields include the user's equipment power consumption information and billing status; Collect user experience data, including subjective ratings and comments on electricity consumption feedback, to form a user experience dataset; use the Apriori algorithm based on the electricity consumption dataset and the user experience dataset to extract personalized differential features of various users, including user electricity consumption characteristics, electricity consumption time characteristics, electricity consumption device characteristics, electricity consumption type characteristics, electricity consumption service evaluation characteristics, power supply reliability characteristics, power safety characteristics, and electricity consumption behavior preference characteristics; obtain user electricity consumption information, electricity consumption time information, electricity consumption device information, and electricity consumption type information based on electricity billing details data, and obtain user electricity consumption service evaluation information, power safety information, and electricity consumption behavior preference information based on user experience data; Calculate the user's average electricity consumption and peak-to-valley ratio to obtain the user's electricity consumption characteristics; Analyze users' peak electricity consumption hours and electricity usage activity to determine their electricity usage time characteristics; Identify the user's main electrical devices and calculate the proportion of electrical devices to obtain the user's electrical device characteristics; Count the distribution of users' electricity usage types and determine the characteristics of users' electricity usage types; Analyze user satisfaction and complaint data on electricity services to obtain user evaluation characteristics of electricity services; Count the number of power outages and the duration of power outages for users to determine the reliability characteristics of their power supply; Analyze users' evaluation of power safety and the number of accidents to obtain users' power safety characteristics; Collect user preferences for energy-saving behaviors and electricity bill payment methods to determine user electricity usage behavior preference characteristics; through repeated iterative calculations, extract user electricity usage-related habit patterns that exceed support and confidence thresholds to determine user personalized characteristics; By comparing the feature correlation results of different users, we can extract the features that distinguish individual users from the average level. Through personalized electricity consumption parameters, key active users are identified, ordinary electricity users are segmented, and different differentiated electricity consumption groups are distinguished.

9. The method according to claim 1, wherein: The method combines the common physical characteristics and differentiated characteristics of electricity users to design and customize personalized decision-making solutions, including: Obtain users' common physical feature data and collect differential features that reflect individual differences among users; select commonly found physical features in user profiles based on the common physical features of electricity users; Based on the user's differential features, select features in the user profile whose inter-individual differences exceed the threshold; The evaluation model is constructed by integrating the selected common physical characteristics and differential characteristics. The weight of each characteristic in the evaluation model is determined through the KNN algorithm. The role of characteristics is evaluated through correlation analysis to evaluate the role of characteristics in user profiles. The parameters of the evaluation model are adjusted according to the weight and role of each characteristic. The common physical characteristics and differential characteristics of users are input into the evaluation model to obtain the evaluation results of user profiles. Based on the data of the current target user and the user profile, the customized electricity price plan set with the highest degree of match with the personality profile is selected from the candidate electricity price plan library for the target user and recommended to the user for selection. The user's load curve and electricity usage time data are used to design a peak-valley tiered electricity price and a flexible electricity price framework that conforms to the common preferences. After the implementation of the plan, we will continue to collect user feedback, optimize the evaluation model, and conduct decision-making effect evaluation and optimization iteration to achieve continuous precision of targeted electricity price plans.

Images