A power user account intelligent management method based on data analysis
By using a multi-dimensional electricity consumption feature distance and weight adaptive allocation method, the accuracy and stability issues of electricity user classification management in existing technologies are solved, and more refined and reliable electricity user account management is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 国网新疆电力有限公司营销服务中心
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies rely on a single characteristic to classify and manage electricity users, which makes it difficult to fully capture the patterns and characteristics of electricity consumption behavior, leading to inaccurate management decisions or poor results.
By acquiring the multi-dimensional electricity consumption characteristic distances of power users, including trend components, periodic components, residual components, and average electricity consumption, a fixed-step moving average is performed to calculate the kernel bandwidth parameter and local density estimate. Weight coefficients are adaptively assigned, and weighted characteristic distance fusion is performed to achieve density peak clustering.
It improves the accuracy and stability of electricity user classification, provides more refined and reliable data support, and provides more accurate decision support for power grid load dispatching and electricity consumption management.
Smart Images
Figure CN122175284A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent management technology, and specifically to an intelligent management method for electricity user accounts based on data analysis. Background Technology
[0002] Currently, in order to ensure the safe and stable operation of the power system, optimize resource allocation and operational efficiency, and achieve differentiated services and enhance customer value, it is necessary to manage power user accounts.
[0003] Existing technologies typically rely on a single characteristic or indicator (such as average electricity consumption) to classify electricity users and manage their accounts based on the classification results. However, with the continuous increase in the types and number of electrical devices, the electricity consumption patterns and behavioral characteristics of electricity users are complex and ever-changing. Relying solely on a single characteristic of electricity users is no longer sufficient to fully capture their electricity consumption behavior patterns and characteristics. For example, relying on average electricity consumption cannot capture the periodic changes of electricity users, which can lead to inaccurate management decisions or poor management results for electricity user accounts. Therefore, how to improve the effectiveness of classified management of electricity user accounts has become an urgent problem to be solved. Summary of the Invention
[0004] To address the aforementioned problems, this invention provides a data-driven intelligent management method for electricity user accounts, the specific technical solution of which is as follows:
[0005] One embodiment of the present invention provides a data analysis-based intelligent management method for electricity user accounts, comprising the following steps:
[0006] Obtain the characteristic distance between electricity users under different dimensions, including trend component, periodic component, residual component and average electricity consumption;
[0007] For any dimension, based on the result of a fixed-step moving average of the feature distance sequences corresponding to each power user in the dimension, a kernel bandwidth parameter for local density estimation of each power user in the dimension is obtained. Based on the kernel bandwidth parameter, the local density estimate of each power user in the dimension is obtained. The feature distance sequence corresponding to any power user in the dimension is obtained by arranging the feature distances between the power user and other power users in the dimension in ascending order. Based on the local density estimate of each power user in the dimension and the feature distances between power users, the weight coefficient corresponding to the dimension is obtained.
[0008] The feature distances between electricity users in each dimension are weighted based on the weight coefficients corresponding to each dimension to obtain the metric distance between different electricity users. Density peak clustering is performed on all electricity users based on the metric distance between different electricity users, and electricity user accounts are classified and managed based on the clustering results.
[0009] Beneficial effects: This invention obtains the feature distances between electricity users across different dimensions. For any given dimension, based on the fixed-step moving average of the feature distance sequences corresponding to each electricity user in that dimension, a kernel bandwidth parameter for local density estimation is obtained. Based on this kernel bandwidth parameter, the local density estimate for each electricity user in that dimension is obtained. Based on the local density estimate and the feature distances between electricity users in that dimension, a weight coefficient corresponding to that dimension is obtained. The feature distances between electricity users in the corresponding dimensions are weighted according to the weight coefficients for each dimension to obtain the metric distance between different electricity users. Density peak clustering is performed on all electricity users based on the metric distances between different electricity users, and electricity user accounts are classified and managed based on the clustering results. Furthermore, this invention, based on the feature distances between multiple dimensions and the adaptive weight allocation during multi-dimensional feature fusion, can improve the accuracy and stability of electricity user classification, thereby improving the effectiveness of electricity user account management. Attached Figure Description
[0010] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is a flowchart of a data analysis-based intelligent management method for electricity user accounts according to the present invention. Detailed Implementation
[0012] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the protection scope of the embodiments of the present invention.
[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.
[0014] This embodiment provides a data analysis-based intelligent management method for electricity user accounts, detailed as follows:
[0015] like Figure 1 As shown, the intelligent management method for electricity user accounts includes the following steps:
[0016] Step S001: Obtain the characteristic distance between electricity users under different dimensions, including trend component, periodic component, residual component and average electricity consumption.
[0017] The purpose of this embodiment is to achieve classified management of power user accounts based on the distance between multi-dimensional electricity consumption characteristics of power users and the adaptive weight allocation during multi-dimensional feature fusion. This provides more accurate and reliable data support for subsequent power grid load scheduling, electricity consumption management, and operational decisions. Specifically, power user accounts are classified based on the distance between multi-dimensional electricity consumption characteristics and the adaptive weight allocation during multi-dimensional feature fusion to achieve effective and intelligent management of power user accounts. Compared with the classification results obtained from a single feature, the classification results obtained based on the distance between multi-dimensional electricity consumption characteristics of power users and the adaptive weight allocation during multi-dimensional feature fusion are more refined, accurate, and stable, and are more beneficial to improving management precision, strategy adaptability, and operational efficiency.
[0018] This embodiment first obtains all electricity users covered or served by any power supply company or power company. That is, all electricity users mentioned later in this embodiment are users covered or served by that power supply company, and electricity users are also electricity consumers. This embodiment will subsequently describe the classification and management process of electricity user accounts covered by that power supply company as an example. First, a preset analysis electricity consumption period is obtained. In this embodiment, the preset analysis electricity consumption period can be the current electricity consumption quarter or the current electricity consumption phase. The specific duration of the current electricity consumption phase needs to be set according to the actual situation; for example, the most recent three months can be used as the preset analysis electricity consumption period. Then, the electricity load data of each electricity user at each collection time within the preset analysis electricity consumption period is collected, and the data within the preset analysis electricity consumption period is... The time series consisting of all the electricity load data of each electricity user collected is denoted as the electricity load time series of each electricity user covered by the power supply company under the preset analysis electricity consumption period. The collected time series undergoes data preprocessing processes such as time series alignment, interpolation, and outlier detection. Time series alignment, interpolation, and outlier detection are well-known techniques. The electricity load data of electricity users are collected synchronously, and the collection interval needs to be set according to the actual situation. For example, it can be set to collect the electricity load of users every 15 minutes. The electricity load of electricity users can be obtained through smart meters. The electricity load of a user at a certain moment refers to the total electrical power taken by the user's electrical equipment from the power system at that moment, which is also the product of the effective voltage and current collected by the user's meter at that moment.
[0019] Next, the average electricity consumption of each power user covered by the power supply company during the preset analysis period is obtained and recorded as the average electricity consumption of the corresponding power user during the preset analysis period; the unit of average electricity consumption is kilowatt-hour; the average of the electrical energy consumed by any user during the preset analysis period is the average electricity consumption of that power user during the preset analysis period; average electricity consumption is one of the dimensions for subsequent classification management of power users, and also one of the dimensions for measuring the similarity or difference of electricity consumption among power users.
[0020] Therefore, this embodiment can obtain the electricity load time series and average electricity consumption of each power user covered by the power supply company under the preset analysis electricity consumption time period through the above process; the average electricity consumption can reflect the electricity consumption scale of the power user; the electricity load time series contains multi-dimensional feature information such as the long-term electricity consumption trend, electricity consumption behavior pattern and electricity consumption volatility of the power user, that is, the electricity load time series is a complex time series signal with a mixture of multiple features, which usually contains multi-dimensional feature information such as the long-term electricity consumption trend, electricity consumption behavior pattern and electricity consumption volatility. These features will be strongly coupled and superimposed in the original data or electricity load time series, which will lead to misclassification when directly classifying users based on the electricity load time series and average electricity consumption, and will also cause the classification results to deviate from the actual management needs. That is, if the similarity analysis and classification management of different users are directly based on the electricity load time series and average electricity consumption, it is easy to be affected by the difference in the user's electricity consumption scale or local abnormal fluctuations, thereby masking the user's The real similarity in electricity consumption behavior patterns leads to a decrease in the consistency between user classification results and actual management needs. To reduce or avoid the interference of multiple factors in the electricity load time series on electricity user classification, this embodiment will use STL decomposition to decouple the multidimensional features of the electricity load time series. The trend term reflecting long-term electricity consumption changes, the periodic term reflecting the regularity of electricity consumption behavior, and the residual term reflecting short-term electricity consumption fluctuations will be extracted and expressed independently to more accurately identify the user's actual electricity consumption behavior. Therefore, this embodiment will further extract multidimensional features based on the obtained electricity load time series, and calculate the feature distance between electricity users in each dimension based on the extracted multidimensional features and the obtained average electricity consumption. The feature distance between electricity users in different dimensions is key to subsequent clustering. The specific process of multidimensional feature extraction for each electricity user's electricity load time series is as follows:
[0021] First, the STL decomposition is performed on the electricity load time series of each electricity user to obtain the trend component sequence, periodic component sequence, and residual component sequence of the corresponding electricity user. That is, after the STL decomposition of the electricity load time series of each electricity user, each parameter in the sequence corresponds to a trend component, a periodic component, and a residual component. The sequence formed by the trend component, periodic component, and residual component of each parameter in the electricity load time series of the electricity user is the corresponding electricity user's trend component sequence, periodic component sequence, and residual component sequence. The process of performing STL decomposition on the time series is a well-known technique. The trend component, periodic component, and residual component of the electricity user obtained by the decomposition are three dimensions for subsequent classification management of electricity users.
[0022] Therefore, this embodiment can obtain the average electricity consumption, trend component sequence, periodic component sequence, and residual component sequence of each power user covered by the power supply company under the preset analysis electricity consumption period through the above process. That is, this embodiment obtains four-dimensional characteristics of power users through the above process. These four dimensions are average electricity consumption, trend component, periodic component, and residual component. The four-dimensional characteristics of each power user are the corresponding power user's average electricity consumption, trend component sequence, periodic component sequence, and residual component sequence. After obtaining the characteristics of each dimension, the differences of different users are measured in each dimension, thereby quantifying the differences of users in terms of electricity consumption scale changes, electricity consumption behavior patterns, and electricity consumption stability. When classifying electricity users' electricity consumption behavior or electricity consumption characteristics in this embodiment, comparing from the four dimensions of average electricity consumption, trend component, periodic component, and residual component is more accurate than directly comparing the overall electricity load time series sequence of power users, reflecting the essential similarities and differences of users' electricity consumption behavior, and improving the rationality and stability of user classification results.
[0023] In this embodiment, after obtaining the four-dimensional features of each electricity user, the feature distance between electricity users in each dimension is calculated. The feature distance is the key to subsequent electricity user classification. The specific calculation process of the feature distance between any two electricity users in each dimension is as follows: For the i-th electricity user and the j-th electricity user in the electricity user set, the electricity user set is the set of all electricity users covered by the power supply company, and the i-th electricity user and the j-th electricity user are not the same user:
[0024] Calculate the DTW distance between the trend component sequences of the i-th and j-th electricity users, and normalize the calculated DTW distance. This result is then recorded as the characteristic distance between the i-th and j-th electricity users in the trend component dimension, used to describe the difference in their long-term electricity consumption trends. Similarly, calculate the DTW distance between the periodic component sequences of the i-th and j-th electricity users, and normalize the calculated DTW distance. This result is then recorded as the characteristic distance between the i-th and j-th electricity users in the periodic component dimension, used to describe the difference in their electricity consumption behavior patterns. ; Calculate the absolute value of the difference between the mean of the residual component sequence of the i-th power user and the mean of the residual component sequence of the j-th power user, and record the result of normalizing the calculated absolute value of the difference as the characteristic distance between the i-th power user and the j-th power user in the residual component dimension, which is used to characterize the difference between the two in terms of electricity consumption volatility; Calculate the absolute value of the difference between the average electricity consumption of the i-th power user and the average electricity consumption of the j-th power user, and record the result of normalizing the calculated absolute value of the difference as the characteristic distance between the i-th power user and the j-th power user in the average electricity consumption dimension, which is used to reflect the difference in the overall electricity consumption scale between the two.
[0025] Therefore, this embodiment can obtain the characteristic distance between any two electricity users in four dimensions through the above process, namely, the characteristic distance between any two electricity users in the trend component dimension, the characteristic distance between any two electricity users in the periodic component dimension, the characteristic distance between any two electricity users in the residual component dimension, and the characteristic distance between any two electricity users in the average electricity consumption dimension.
[0026] Step S002: For any dimension, based on the result of a fixed-step moving average of the feature distance sequences corresponding to each power user in the dimension, obtain the kernel bandwidth parameter for local density estimation of each power user in the dimension. Based on the kernel bandwidth parameter, obtain the local density estimate of each power user in the dimension. The feature distance sequence corresponding to any power user in the dimension is obtained by arranging the feature distances between the power user and other power users in the dimension in ascending order. Based on the local density estimate of each power user in the dimension and the feature distances between power users, obtain the weight coefficient corresponding to the dimension.
[0027] Because the distinguishing abilities of different dimensions vary significantly, this embodiment, in order to ensure the effectiveness of subsequent classification management, will adaptively allocate the weights of different dimensions in the comprehensive measurement model based on their ability to identify user groups. This aims to achieve precise and adaptive electricity management. The better the identification ability of a dimension for user groups, the stronger its discriminative and distinguishing power in representing differences in user electricity behavior, and the clearer it can effectively separate user groups with different behavioral patterns. That is, in actual electricity management scenarios, if a certain dimension can more effectively reflect the differences in electricity behavior among users, then that dimension should have higher reference value or contribution to user classification and management decisions, and its weight should be larger. Conversely, if a certain dimension has limited ability to distinguish users, then that dimension should have lower reference value or contribution to user classification and management decisions. The value or contribution should be low, and the weight should be small, which can improve the accuracy, stability, interpretability, and management adaptability of subsequent user classification, and provide a more solid data foundation for refined user management. Therefore, after obtaining the feature distances between power users under various dimensions, this embodiment cannot directly fuse the feature distances. Instead, it is necessary to first analyze the identification effectiveness of different dimensions for user groups, and determine the weight coefficients of different dimensions in classification based on the identification effectiveness of different dimensions for user groups. The higher the identification effectiveness of a dimension for user groups, the greater its contribution or weight coefficient in determining the final distance metric used for clustering. The weight coefficients are mainly used for weighting the feature distances between power users under all dimensions. The weighted result is the basis for subsequent user classification or clustering.
[0028] Furthermore, if a certain dimension can effectively support the classification of users into different categories, then electricity users under that dimension usually exhibit a high cluster density in a local area. At the same time, the distance between different high-density areas is relatively large, meaning that high-density user groups may originate from different electricity consumption behavior categories. This is beneficial for subsequent clustering algorithms to classify users into multiple discriminative categories. In other words, if the local density of users under a certain dimension is larger and the feature distance between users under that dimension is also larger, then the identification efficiency of that dimension for user groups is better. Therefore, before determining the weight coefficients of different dimensions, it is necessary to estimate the local density of each electricity user under each dimension. Since the distribution characteristics of different users under the same dimension are different, in order to improve the accuracy of the density estimation results, this embodiment needs to adaptively select appropriate density estimation parameters, namely kernel bandwidth parameters, based on the distance distribution characteristics of the electricity user in its local neighborhood before performing density estimation. This achieves accurate characterization of the local density of users and provides a reliable basis for subsequent user classification.
[0029] Since the method for obtaining the weight coefficients corresponding to each dimension in this embodiment is the same, and for ease of understanding, this embodiment will describe the process of obtaining the weight coefficients corresponding to the g-th dimension among the above four dimensions as an example, such as the weight coefficients corresponding to the trend component dimension; therefore, this embodiment needs to first determine the kernel bandwidth parameters for local density estimation of each power user under the g-th dimension, and the specific process for determining the kernel bandwidth parameters for local density estimation of the i-th power user under the g-th dimension is as follows:
[0030] Obtain the feature distances between the i-th power user and all other power users in the g-th dimension, and sort these feature distances in ascending order. Denote the sorted result as the feature distance sequence corresponding to the i-th power user in the g-th dimension. That is, the feature distance sequence corresponding to the i-th power user in the g-th dimension is obtained by arranging the feature distances between the i-th power user and other power users in the g-th dimension in ascending order. Furthermore, if the feature distance between the i-th power user and user q in the g-th dimension is the smallest, then the first feature distance in the feature distance sequence corresponding to the i-th power user in the g-th dimension is the feature distance between the i-th power user and user q in the g-th dimension. The distance is calculated; then, a fixed-step moving average is performed on the feature distance sequence corresponding to the i-th power user in the g-th dimension. Based on the result of the fixed-step moving average on the feature distance sequence corresponding to the i-th power user in the g-th dimension, the kernel bandwidth parameter for local density estimation of the i-th power user in the g-th dimension can be obtained. The purpose of performing a fixed-step moving average on the feature distance sequence corresponding to the i-th power user in the g-th dimension is to evaluate the feature distance variation characteristics or distribution stability of the i-th power user in the local neighborhood in this dimension. The more stable the feature distance variation of the i-th power user in the local neighborhood in this dimension, the larger the kernel bandwidth parameter for local density estimation of the i-th power user in the g-th dimension should be to ensure the stability and accuracy of density estimation. In local density estimation, the choice of kernel bandwidth parameter directly determines the structure fidelity of density estimation. The specific process of obtaining the kernel bandwidth parameter for local density estimation of the i-th power user in the g-th dimension based on the result of the fixed-step moving average on the feature distance sequence corresponding to the i-th power user in the g-th dimension is as follows:
[0031] First, perform a fixed-step moving average on the feature distance sequence corresponding to the i-th power user in the g-th dimension to obtain the mean distance sequence corresponding to the i-th power user in the g-th dimension. The specific process for obtaining the mean distance sequence corresponding to the i-th power user in the g-th dimension is as follows: start calculating the mean from the first a elements of the feature distance sequence corresponding to the i-th power user in the g-th dimension, incrementing by b elements each time to calculate a new mean, until the preset sequence length threshold of the feature distance sequence corresponding to the i-th power user in the g-th dimension is reached. The sequence formed by the calculated means in the order of calculation is recorded as the mean distance sequence corresponding to the i-th power user in the g-th dimension. The w-th mean distance in the mean distance sequence corresponding to the i-th power user in the g-th dimension is the mean distance of the i-th power user in the g-th dimension. The mean of the first a+(w-1)b feature sequences in the feature distance sequence corresponding to the power user, where b is an integer; in specific applications, implementers need to set a, b, and the preset sequence length threshold of the feature distance sequence corresponding to the i-th power user according to actual conditions such as computational complexity requirements, computational speed, and data scale. It is required that a and b are much smaller than the length of the feature distance sequence corresponding to the i-th power user in the g-th dimension. For example, in this embodiment, a can be set to 3, b can be set to 2, and the preset sequence length threshold of the feature distance sequence corresponding to the i-th power user can be set to 20% of the total number of data in the feature distance sequence corresponding to the i-th power user. If 20% of the total number of data in the feature distance sequence corresponding to the i-th power user is not an integer, the result of rounding down will be used as the preset sequence length threshold.
[0032] Then, the first and second differences of the distance mean sequence corresponding to the i-th power user in the g-th dimension are calculated to obtain the first and second difference sequences of the distance mean sequence corresponding to the i-th power user in the g-th dimension. Based on the difference sequence of the distance mean sequence corresponding to the i-th power user in the g-th dimension, the target parameter indication value of the i-th power user in the g-th dimension is obtained. The target parameter indication value can reflect whether the kernel bandwidth parameter used for local density estimation of the i-th power user in the g-th dimension should be large or small. The specific process of obtaining the target parameter indication value of the i-th power user in the g-th dimension is as follows: based on the distance mean sequence corresponding to the i-th power user in the g-th dimension... The second-order difference sequence of the distance mean sequence corresponding to the i-th power user is used to obtain the first parameter indication value of the i-th power user in the g-th dimension. Based on the first-order difference sequence of the distance mean sequence corresponding to the i-th power user in the g-th dimension, the second parameter indication value of the i-th power user in the g-th dimension is obtained. The specific process of obtaining the first and second parameter indication values of the i-th power user in the g-th dimension is as follows: the sequence formed by the ratios between adjacent difference values in the second-order difference sequence of the distance mean sequence corresponding to the i-th power user in the g-th dimension is denoted as the feature ratio sequence of the i-th power user in the g-th dimension. The r-th feature ratio in the sequence is the ratio of the r-th difference value to the (r+1)-th difference value in the second-order difference sequence of the distance mean sequence corresponding to the i-th power user in the g-th dimension; calculate the absolute value of the relative deviation of each feature ratio in the feature ratio sequence relative to the constant 1, and record it as the relative offset of the corresponding feature ratio; the relative offset of the r-th feature ratio in the feature ratio sequence is the absolute value of the difference between the constant 1 and the r-th feature ratio; calculate the mean of the relative offsets of all feature differences in the feature ratio sequence, and record it as the first parameter indication value of the i-th power user in the g-th dimension; calculate the distance corresponding to the i-th power user. The normalized result of the first-order difference sequence mean of the mean sequence is denoted as the second parameter indication value of the i-th electricity user in the g-th dimension. Normalization is achieved using the normalization function Norm(). Next, the first parameter indication value of the i-th electricity user in the g-th dimension is multiplied by the second parameter indication value, and then negatively correlated. This result is denoted as the target parameter indication value of the i-th electricity user in the g-th dimension. Negative correlation mapping is achieved using a negative exponential function with a base e. The specific expression for the target parameter indication value of the i-th electricity user in the g-th dimension is:
[0033]
[0034] in, Let be the target parameter indicator value for the i-th electricity user in the g-th dimension, exp() be an exponential function with base e, Norm() be a normalization function, and R be the number of difference values in the second-order difference sequence of the distance mean sequence corresponding to the i-th electricity user in the g-th dimension. Let r be the r-th difference value in the second-order difference sequence of the distance mean sequence corresponding to the i-th electricity user in the g-th dimension. Let be the (r+1)th difference value in the second-order difference sequence of the distance mean sequence corresponding to the i-th electricity user in the g-th dimension. Let be the first difference sequence mean of the distance mean sequence corresponding to the i-th electricity user. The larger the value, the greater the overall distance change of the i-th power user in the local neighborhood under the g-th dimension. Therefore, when performing local density estimation for the i-th power user under the g-th dimension, the kernel bandwidth parameter should be smaller. The smaller the value, the more stable the distance change. Therefore, when performing local density estimation for the i-th power user in the g-th dimension, the kernel bandwidth parameter should be larger. This represents the mean of the differences between adjacent difference values in the second-order difference sequence. A smaller value indicates higher stability in the distance variation of the i-th power user in different local neighborhoods under the g-th dimension. Therefore, the kernel bandwidth parameter should be larger when estimating the local density of the i-th power user in the g-th dimension. Conversely, a larger value indicates less stable distance variation of the i-th power user in different local neighborhoods under the g-th dimension. Therefore, the kernel bandwidth parameter should be smaller when estimating the local density of the i-th power user in the g-th dimension. And because... smaller and The smaller, The larger, therefore When the value of the kernel bandwidth is larger, that is, when the target parameter indication value of the i-th power user in the g-th dimension is larger, the kernel bandwidth parameter should be larger when performing local density estimation for the i-th power user in the g-th dimension, and vice versa. The smaller the kernel bandwidth parameter, the smaller it should be when performing local density estimation for the i-th power user in the g-th dimension.
[0035] To further ensure the stability and accuracy of density estimation, when determining the kernel bandwidth parameter for the i-th power user in the g-th dimension for local density estimation, in addition to considering the target parameter indication value, the overall feature distance of the i-th power user in the local neighborhood in the g-th dimension should also be considered. The larger the overall feature distance in the local neighborhood, the larger the kernel bandwidth parameter should be. That is, in this embodiment, the i-th power user in the g-th dimension will be obtained based on the target parameter indication value of the i-th power user in the g-th dimension and the mean of the distance mean sequence corresponding to the i-th power user in the g-th dimension. The kernel bandwidth parameter used by the power user for local density estimation; and the product of the target parameter indication value of the i-th power user in the g-th dimension and the mean of the distance mean sequence corresponding to the i-th power user in the g-th dimension is the kernel bandwidth parameter used by the i-th power user in the g-th dimension for local density estimation. That is, the larger the target parameter indication value and the mean of the distance mean sequence, the larger the kernel bandwidth parameter used by the i-th power user in the g-th dimension for local density estimation; the mean of the distance mean sequence corresponding to the i-th power user in the g-th dimension is the local typical distance of the corresponding user in the current dimension.
[0036] In this embodiment, after obtaining the kernel bandwidth parameters for local density estimation of each power user in the g-th dimension, the local density estimate of each power user in the g-th dimension is obtained based on the kernel bandwidth parameters for local density estimation of each power user in the g-th dimension. In this embodiment, the largest local neighborhood determined when performing the fixed step-size moving average is selected for density estimation. Therefore, the expression for the local density estimate of the i-th power user in the g-th dimension is:
[0037]
[0038] Let N be the local density estimate of the i-th electricity user in the g-th dimension, and N be the preset sequence length threshold of the feature distance sequence corresponding to the i-th electricity user in the g-th dimension. This represents the nth feature distance in the feature distance sequence corresponding to the i-th electricity user under the g-th dimension. Let exp() be the kernel bandwidth parameter used for local density estimation of the i-th power user in the g-th dimension, and let exp() be an exponential function with a base of constant e. Since the total number of parameters in the feature distance sequence corresponding to the i-th power user in the g-th dimension is 100, then 20% of 100 is N, which is N = 20. The larger the value, the higher the clustering density of the i-th electricity user in the local area under the g-th dimension.
[0039] Since a higher local density of users in a certain dimension and a larger feature distance between users in that dimension result in better identification performance for user groups, this embodiment, after obtaining the estimated local density of each power user in the g-th dimension, combines the feature distance between power users in the g-th dimension to obtain the weight coefficient corresponding to the g-th dimension. Specifically, this embodiment will next obtain the weight coefficient corresponding to the g-th dimension based on the estimated local density of each power user in the g-th dimension and the feature distance between power users in the g-th dimension. The specific process for obtaining the weight coefficient corresponding to the g-th dimension is as follows:
[0040] Based on the local density estimates of each power user in the g-th dimension and the feature distances between power users in the g-th dimension, a set of contribution index values corresponding to each power user in the g-th dimension is obtained. The number of contribution index values in the set of contribution index values corresponding to each power user in the g-th dimension is the total number of power users minus 1. The specific process for obtaining the m-th contribution index value in the set of contribution index values corresponding to the i-th power user in the g-th dimension is as follows: the set of all power users except the i-th power user is denoted as the i-th power user. The set to be analyzed is given by multiplying the normalized result of the local density estimate of the i-th power user in the g-th dimension, the normalized result of the local density estimate of the m-th power user in the set to be analyzed for the i-th power user, and the feature distance between the i-th power user and the m-th power user in the g-th dimension. This result is denoted as the m-th contribution index value in the set of contribution index values corresponding to the i-th power user in the g-th dimension. The expression for the m-th contribution index value in the set of contribution index values corresponding to the i-th power user in the g-th dimension is as follows: ,in, This is the normalized result of the local density estimate for the i-th electricity user in the g-th dimension. This represents the normalized result of the local density estimate for the m-th power user in the set of power users to be analyzed, denoted as the i-th power user. Let be the feature distance between the i-th and m-th electricity users in the g-th dimension. Here, the local density estimate can be normalized using the max-min normalization method. This represents the overall local density level between the i-th and m-th electricity users. The larger the value, the more likely both are located in a high-density sample region. Based on this, if the feature distance between them in the g-th dimension... When the value is also relatively large, it indicates that the high-density areas where these two electricity users are located may originate from different user behavior categories. This suggests that this dimension is more likely to have a strong discriminative ability in distinguishing different user categories. This indicates that the g-th dimension is more likely to have a strong ability to distinguish different user categories, or that the g-th dimension has a better ability to identify user groups.
[0041] Then, the maximum contribution index value in the set of contribution index values corresponding to each power user under the g-th dimension is obtained, and a new set composed of the maximum contribution index values in the set of contribution index values corresponding to each power user under the g-th dimension is used as the representative contribution index value set corresponding to the g-th dimension. The purpose of extracting the representative contribution index value set is to focus on the strongest discriminative signal of the dimension at the individual level, retain the maximum distinguishing potential between each user and the most heterogeneous individuals under each dimension, so as to improve the sensitivity and robustness of the weight calculation to the discriminative ability. Calculate the contribution index value set corresponding to the g-th dimension. The product of the mean of the set of contribution indicator values and the coefficient of variation of the set of contribution indicator values corresponding to the g-th dimension is denoted as the importance evaluation value of the g-th dimension in user classification management. Here, a negative exponential function with a base of constant e is used for negative correlation mapping. The importance evaluation value of the g-th dimension in user classification management is proportionally normalized and denoted as the weight coefficient corresponding to the g-th dimension. The calculation of the coefficient of variation of the set is well-known, usually the ratio of the standard deviation of the set to the mean. The specific expression of the weight coefficient corresponding to the g-th dimension is:
[0042]
[0043]
[0044] in, The weight coefficients corresponding to the g-th dimension are... Let G be the importance evaluation value of the g-th dimension in user classification management, where G is the total number of dimensions. In this embodiment, G is 4, representing the trend component, periodic component, residual component, and average electricity consumption, respectively. Let be the mean of the set of representative contribution index values corresponding to the g-th dimension. Let be the coefficient of variation of the set of representative contribution index values corresponding to the g-th dimension, and e be the natural constant. The larger the value, the stronger the ability of users to form a high-density and mutually separated structure under the g-th dimension, which in turn indicates that the g-th dimension has a better ability to identify user groups. Therefore, the g-th dimension is more important when calculating the distance metric used for classification, and its contribution should be greater. It can characterize the dispersion and stability level of the set of representative contribution index values corresponding to the g-th dimension. The smaller, that is The larger the value, the more stable the distinguishing ability of the g-th dimension is across different user categories, and the higher its reliability for user classification management. Therefore, the g-th dimension is more important when calculating the distance metric used for classification, and its contribution should be greater. The greater the sum When it is larger, The larger, therefore When the value is larger, the g-th dimension becomes more important when calculating the distance metric used for classification. The larger, The larger.
[0045] Therefore, this embodiment can obtain the weight coefficients corresponding to each dimension through the above process. That is, this embodiment can obtain the weight coefficients corresponding to the trend component dimension, the period component dimension, the residual component dimension, and the average electricity consumption dimension.
[0046] Step S003: Based on the weight coefficients corresponding to each dimension, the feature distance between power users in the corresponding dimension is weighted to obtain the metric distance between different power users. Density peak clustering is performed on all power users based on the metric distance between different power users, and power user accounts are classified and managed based on the clustering results.
[0047] In this embodiment, after obtaining the weight coefficients corresponding to each dimension, the feature distances between electricity users under different dimensions are weighted based on the weight coefficients corresponding to different dimensions to obtain the metric distance between different electricity users; and the specific process of obtaining the metric distance between the i-th electricity user and the j-th electricity user is as follows:
[0048] The weighted feature distance between the i-th and j-th electricity users across all dimensions is calculated and summed to serve as the metric distance between them. The weighted feature distance between the i-th and j-th electricity users in any dimension is the result of multiplying the weight coefficient corresponding to that dimension by the feature distance between them in that dimension. The expression for the metric distance between the i-th and j-th electricity users is as follows:
[0049]
[0050] in, Let be the metric distance between the i-th electricity user and the j-th electricity user. These are the weighting coefficients corresponding to the trend component dimension. Let be the feature distance between the i-th and j-th electricity users under the trend component dimension. These are the weighting coefficients corresponding to the periodic component dimensions. Let be the feature distance between the i-th and j-th electricity users in the periodic component dimension. These are the weighting coefficients corresponding to the dimensions of the residual components. Let be the feature distance between the i-th and j-th electricity users in the residual component dimension. These are the weighting coefficients corresponding to the average electricity consumption dimension. This represents the feature distance between the i-th and j-th electricity users in terms of average electricity consumption. Measuring distance can simultaneously take into account the differences in users in terms of electricity consumption scale, electricity consumption trend, electricity consumption pattern, and electricity consumption volatility, thereby improving the accuracy of user similarity representation or classification accuracy.
[0051] After obtaining the metric distance between different electricity users, density peaks clustering (DPC) is performed on all electricity users based on the metric distance between them. The electricity user accounts are then classified and managed based on the clustering results, meaning that the accounts of electricity users in a cluster belong to one electricity user account category. The DPC clustering algorithm can adaptively identify high-density and mutually separated user category centers based on the local density of samples and the relative distance between samples. Therefore, it is particularly suitable for user classification scenarios in non-uniformly distributed, multi-scale feature spaces formed by the aforementioned metric distance. The process of clustering users using the density peaks clustering algorithm when the metric distance between users is known is a well-known technique. Furthermore, based on the aforementioned categories of electricity user accounts, management platforms or power supply companies can implement differentiated user management and applications. For example, for user groups with stable electricity consumption patterns and small load fluctuations, power supply companies can adopt standardized electricity billing and conventional load dispatching strategies. For user groups with large load fluctuations or obvious periodic electricity consumption characteristics, dynamic load management, customized electricity dispatching schemes, or peak-valley pricing strategies can be implemented. Simultaneously, early warning monitoring or energy-saving optimization guidance can be provided for user groups with high electricity consumption or abnormal electricity consumption patterns. In other words, through the above methods, the clustering results not only achieve refined classification of user groups but also provide scientific data support for power grid operation optimization, load dispatching decisions, and differentiated electricity services.
[0052] Thus, this embodiment completes the intelligent management method for electricity user accounts. Furthermore, based on feature distances between multiple dimensions and adaptive weight allocation during multi-dimensional feature fusion, this embodiment improves the effectiveness of classifying and managing electricity user accounts, providing more reliable data support for subsequent differentiated electricity management strategies, grid load analysis, and operational decisions. Additionally, in this embodiment, only specific numerical values are involved in the formula calculations.
[0053] In summary, this method obtains the feature distances between electricity users across different dimensions. For any given dimension, based on the fixed-step moving average of the feature distance sequences corresponding to each electricity user in that dimension, a kernel bandwidth parameter for local density estimation is obtained. Based on this kernel bandwidth parameter, the local density estimate for each electricity user in that dimension is obtained. Based on the local density estimate and the feature distances between electricity users in that dimension, a weight coefficient corresponding to that dimension is obtained. The feature distances between electricity users in the corresponding dimension are weighted using the weight coefficients for each dimension to obtain the metric distance between different electricity users. Density peak clustering is performed on all electricity users based on the metric distances between different electricity users, and electricity user accounts are classified and managed based on the clustering results. Furthermore, this embodiment, based on the feature distances between multi-dimensional features and the adaptive weight allocation during multi-dimensional feature fusion, can improve the accuracy and stability of electricity user classification, thereby improving the effectiveness of electricity user account management.
[0054] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A data-driven intelligent management method for electricity user accounts, characterized in that, The method includes the following steps: Obtain the characteristic distance between electricity users under different dimensions, including trend component, periodic component, residual component and average electricity consumption; For any dimension, based on the result of a fixed-step moving average of the feature distance sequences corresponding to each power user in the dimension, a kernel bandwidth parameter for local density estimation of each power user in the dimension is obtained. Based on the kernel bandwidth parameter, the local density estimate of each power user in the dimension is obtained. The feature distance sequence corresponding to any power user in the dimension is obtained by arranging the feature distances between the power user and other power users in the dimension in ascending order. Based on the local density estimate of each power user in the dimension and the feature distances between power users, the weight coefficient corresponding to the dimension is obtained. The feature distances between electricity users in each dimension are weighted based on the weight coefficients corresponding to each dimension to obtain the metric distance between different electricity users. Density peak clustering is performed on all electricity users based on the metric distance between different electricity users, and electricity user accounts are classified and managed based on the clustering results.
2. The intelligent management method for power user accounts based on data analysis as described in claim 1, characterized in that, Methods for obtaining feature distance include: Obtain the time series sequence of electricity load and average electricity consumption of power users under the preset analysis electricity consumption period, and perform STL decomposition on the time series sequence of electricity load of power users to obtain the trend component sequence, periodic component sequence and residual component sequence of the corresponding power users; For the i-th and j-th electricity users, the normalized DTW distance between the trend component sequences of the i-th and j-th electricity users is denoted as the characteristic distance between the i-th and j-th electricity users in the trend component dimension. The normalized DTW distance between the periodic component sequences of the i-th and j-th electricity users is denoted as the characteristic distance between the i-th and j-th electricity users in the periodic component dimension. The normalized absolute value of the difference between the mean of the residual component sequences of the i-th and j-th electricity users is denoted as the characteristic distance between the i-th and j-th electricity users in the residual component dimension. The normalized absolute value of the difference between the average electricity consumption of the i-th and j-th electricity users is denoted as the characteristic distance between the i-th and j-th electricity users in the average electricity consumption dimension.
3. The intelligent management method for power user accounts based on data analysis as described in claim 1, characterized in that, Methods for obtaining kernel bandwidth parameters include: For any power user, a fixed-step moving average is performed on the feature distance sequence corresponding to the power user in the given dimension to obtain the distance mean sequence corresponding to the power user in the given dimension. Based on the ratio between adjacent difference values in the second-order difference sequence of the distance mean sequence, the first parameter indication value of the power user in the given dimension is obtained. The normalized result of the first-order difference sequence mean of the distance mean sequence is recorded as the second parameter indication value of the power user in the given dimension. The negative correlation mapping result of multiplying the first parameter indication value and the second parameter indication value is recorded as the target parameter indication value of the power user in the given dimension. Based on the target parameter indication value and the distance mean sequence mean, the kernel bandwidth parameter for local density estimation of the power user in the given dimension is obtained.
4. The intelligent management method for power user accounts based on data analysis as described in claim 3, characterized in that, The method for obtaining the distance mean sequence corresponding to the electricity users under the aforementioned dimension includes: The mean is calculated starting from the first 'a' elements of the feature distance sequence corresponding to the power user under the given dimension. Each time, 'b' elements are incremented to calculate a new mean, until the preset sequence length threshold of the corresponding feature distance sequence is reached. The sequence of the calculated means in the order of calculation is recorded as the distance mean sequence corresponding to the power user under the given dimension.
5. The intelligent management method for power user accounts based on data analysis as described in claim 3, characterized in that, The method for obtaining the first parameter indication value of the electricity user under the aforementioned dimension includes: The sequence formed by the ratios between adjacent difference values in the second-order difference sequence of the distance mean sequence is denoted as the characteristic ratio sequence. The absolute value of the relative deviation of each characteristic ratio in the characteristic ratio sequence relative to the constant 1 is denoted as the relative offset of the corresponding characteristic ratio. The relative offset of the r-th characteristic ratio in the characteristic ratio sequence is the absolute value of the difference between the constant 1 and the r-th characteristic ratio. The mean of the relative offsets of all characteristic differences in the characteristic ratio sequence is denoted as the first parameter indication value of the power user in the dimension.
6. The intelligent management method for power user accounts based on data analysis as described in claim 3, characterized in that, The product of the target parameter indication value and the mean of the distance mean sequence is the kernel bandwidth parameter used by the power user for local density estimation in the dimension.
7. The intelligent management method for power user accounts based on data analysis as described in claim 1, characterized in that, The expression for the local density estimate of the i-th power user in the given dimension is: ； Let N be the local density estimate of the i-th power user in the given dimension, and N be a preset sequence length threshold for the feature distance sequence corresponding to the i-th power user in the given dimension. This refers to the nth feature distance in the feature distance sequence corresponding to the i-th electricity user under the given dimension. Let be the kernel bandwidth parameter used for local density estimation of the i-th power user in the given dimension, and exp() be an exponential function with a base of constant e.
8. The intelligent management method for power user accounts based on data analysis as described in claim 1, characterized in that, Methods for obtaining weighting coefficients include: Based on the local density estimates of each power user under the stated dimension and the feature distance between power users, a set of contribution index values corresponding to each power user under the stated dimension is obtained. The maximum value in the set of contribution index values corresponding to each power user under the stated dimension is used as the representative contribution index value set corresponding to the stated dimension. The mean of the representative contribution index value set is multiplied by the negative correlation mapping result of the coefficient of variation of the representative contribution index value set and the proportional normalization result is used as the weight coefficient corresponding to the stated dimension.
9. The intelligent management method for power user accounts based on data analysis as described in claim 8, characterized in that, The method for obtaining the set of contribution index values corresponding to each electricity user under the aforementioned dimension includes: For any given power user, the set of all other power users is denoted as the set to be analyzed for that power user. The set of contribution index values between the power user in the given dimension and each power user in the set to be analyzed is denoted as the set of contribution index values corresponding to the power user in the given dimension. The m-th contribution index value in the set of contribution index values corresponding to the power user in the given dimension is the result of multiplying the normalized local density estimate of the power user in the given dimension, the normalized local density estimate of the m-th power user in the set to be analyzed, and the feature distance between the power user in the given dimension and the m-th power user.
10. The intelligent management method for power user accounts based on data analysis as described in claim 1, characterized in that, Methods for obtaining the distance between different electricity users include: For the i-th and j-th power users, the sum of the weighted feature distances between the i-th and j-th power users across all dimensions is used as the metric distance between the i-th and j-th power users. The weighted feature distance between the i-th and j-th power users in any dimension is the result of multiplying the weight coefficient corresponding to the dimension by the feature distance between the i-th and j-th power users in that dimension.