Electric energy substitution potential customer type identification method and device, terminal and medium

Abstract

The invention discloses an electric energy substitution potential customer type identification method and device, a terminal and a medium, and the method comprises the steps: obtaining multi-dimensional original data from different data sources, converting the multi-dimensional original data into corresponding view data, screening out a representative anchor point in each view, calculating the similarity between each sample point and the anchor point, and carrying out the recognition of the type of a customer with the electric energy substitution potential. And constructing an initial similarity matrix of each view. And inputting the multi-view similarity matrix into a feature selection model, and gradually eliminating the noise influence of the low-weight views through model optimization. According to the scheme, firstly, the calculation complexity is reduced through anchor point screening, meanwhile, the view contribution degree is dynamically adjusted through the feature selection model, irrelevant feature interference is effectively restrained, multi-dimensional information such as power loads and equipment states can be integrated, and the method and the device have the advantages that the method and the device are easy to implement. The customer energy utilization potential under different environment conditions can be accurately identified, and the classification accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of power grid big data technology, and in particular to a method, device, terminal and medium for identifying potential customers for electricity substitution. Background Technology

[0002] In recent years, the whole society has attached great importance to environmental protection and pollution issues. Reducing carbon emissions, improving environmental quality, and adjusting the energy structure have gradually become one of the key tasks in power grid construction. Among them, electricity substitution and transformation (such as coal-to-electricity and gas-to-electricity conversion) is a key means to achieve emission reduction targets.

[0003] Electricity substitution aims to shift end-use energy from fossil fuels to electricity to improve energy efficiency and reduce carbon emissions. In electricity substitution retrofitting projects, analyzing diverse characteristics such as the energy consumption structure (electric / non-electric energy ratio), equipment status (efficiency, age), and energy consumption behavior of electricity customers is crucial for accurately identifying potential customers. However, in practical applications, electricity substitution behavior patterns vary significantly across different customer types, energy consumption sites, and time dimensions. Existing customer behavior analysis methods are often limited to single-dimensional data. When faced with high-dimensional and complex customer data, they often struggle to accurately depict customers' comprehensive energy consumption habits, potential, and responses to incentive policies before and after electricity substitution, thus hindering the promotion and construction efficiency of electricity substitution retrofitting projects. Summary of the Invention

[0004] This application provides a method, device, terminal, and medium for identifying potential customers for electricity substitution, which addresses the technical problem that existing methods for analyzing the behavior of potential customers for electricity substitution often fail to effectively handle high-dimensional and complex customer data, resulting in low accuracy in identifying potential customer types for electricity substitution.

[0005] To address the aforementioned technical problems, the first aspect of this application provides a method for identifying potential customer types for electricity substitution, characterized by comprising:

[0006] Acquire multi-dimensional data of the customer to be identified, wherein the multi-dimensional data is multi-source heterogeneous data obtained from multiple data sources, including: power load data, environmental meteorological data, equipment operating status and customer attributes;

[0007] The multi-dimensional data is transformed into view data to construct a multi-view dataset, wherein each view data corresponds to one of the dimensions in the multi-dimensional data.

[0008] Based on the multi-view dataset, several anchor points are selected from the sample points contained in each view data to construct an anchor point matrix. Then, based on the anchor point matrix, the similarity between each sample point and the anchor point is calculated to obtain a similarity matrix.

[0009] Based on the preset multi-view feature selection model, the similarity matrix is ​​used as the model input. The anchor point matrix and the projection matrix corresponding to each view data are combined to optimize the multi-view feature selection model. When the multi-view feature selection model has been iterated and optimized and meets the preset iteration termination condition, the optimal similarity matrix output by the multi-view feature selection model is obtained.

[0010] Based on the optimal similarity matrix, spectral clustering is performed to determine the customer type identification results of potential customers for electricity substitution according to the obtained clustering results.

[0011] Preferably, the multi-view feature selection model is as follows:

[0012]

[0013] In the formula, Representing the The weight of each view's data. For the first The first view in the One sample, This is a similarity matrix between multi-view data and anchor points. The element in the i-th row and j-th column of the similarity matrix. For the first Anchor point matrix for each view, For the first The j-th element in the anchor point matrix of each view. For the first Projection matrix of each view, It is a diagonal matrix. for The first in One element, used to represent the anchor point. The weight, For matrix A column vector consisting of the diagonal elements. For regularization terms, Let I be the weight of the regularization term, and let I be the identity matrix. The number of features selected, This is the sum of the view anchor weights. Similarity matrix The corresponding Laplace matrix, Let be the rank of the matrix. The sample size of the dataset. The number of anchor points in each view. This represents the number of clusters in the multi-view dataset.

[0014] Preferably, the iterative optimization and model solving of the multi-view feature selection model specifically includes:

[0015] Based on the multi-view feature selection model, each variable to be solved is selected, a target variable is determined from the variables to be solved, and the remaining non-target variables are fixed;

[0016] For the determined target variable, the multi-view feature selection model is transformed into a corresponding optimization problem expression, and then the optimization problem expression is solved to obtain the solution result of the target variable;

[0017] The target variable is determined sequentially from the remaining variables to be solved. Then, for the target variable, the problem is transformed and solved by optimization until all variables to be solved are solved.

[0018] Preferably, the step of performing spectral clustering based on the optimal similarity matrix to determine the customer type identification result of potential customers for electricity substitution based on the obtained clustering results specifically includes:

[0019] Based on the optimal similarity matrix, and combined with a preset matrix expansion formula, an extended similarity matrix is ​​constructed, wherein the similarity matrix contains the similarity between all customers;

[0020] Based on the extended similarity matrix, a normalized Laplacian matrix is ​​constructed;

[0021] The normalized Laplacian matrix is ​​subjected to eigenvalue decomposition to obtain multiple sets of eigenvalues ​​and eigenvectors.

[0022] Based on the magnitude of the eigenvalues, select several target feature vectors corresponding to the smallest non-zero eigenvalues, and concatenate the target feature vectors column by column to obtain the embedding matrix;

[0023] Based on the embedding matrix, data dimensionality reduction and K-means clustering are performed to obtain several clusters representing different customer types.

[0024] Preferably, the matrix extension relation is as follows:

[0025]

[0026] In the formula, S is the extended similarity matrix, and Z is the optimal similarity matrix. Let be the transpose of the optimal similarity matrix.

[0027] Preferably, it further includes:

[0028] Based on the multi-dimensional data and the pre-defined correspondence between data types and preprocessing methods, the multi-dimensional data is preprocessed.

[0029] Preferably, the preprocessing of the multi-dimensional data based on the multi-dimensional data and the preset correspondence between data types and preprocessing methods includes:

[0030] When the data to be processed is numerical data, it is preprocessed by Z-score normalization.

[0031] When the data to be processed is non-numerical, it is preprocessed using one-hot encoding.

[0032] A second aspect of this application provides a device for identifying potential customers for electricity substitution, comprising:

[0033] The data acquisition unit is used to acquire multi-dimensional data of the customer to be identified, wherein the multi-dimensional data is multi-source heterogeneous data obtained from multiple data sources, including: power load data, environmental meteorological data, equipment operating status and customer attributes;

[0034] A view data transformation unit is used to perform view data transformation on the multi-dimensional data to construct a multi-view dataset, wherein each view data corresponds to one of the dimensions in the multi-dimensional data.

[0035] The similarity matrix construction unit is used to select several anchor points from the sample points contained in each view data according to the multi-view dataset to construct an anchor point matrix, and then calculate the similarity between each sample point and the anchor point based on the anchor point matrix to obtain the similarity matrix.

[0036] The model optimization unit is used to optimize the multi-view feature selection model based on a preset multi-view feature selection model, taking the similarity matrix as the model input, and combining the anchor point matrix and the projection matrix corresponding to each view data. When the multi-view feature selection model has been iteratively optimized and meets the preset iteration termination condition, the optimal similarity matrix output by the multi-view feature selection model is obtained.

[0037] The spectral clustering processing unit is used to perform spectral clustering processing based on the optimal similarity matrix, so as to determine the customer type identification result of potential customers for electricity substitution based on the obtained clustering results.

[0038] A third aspect of this application provides a terminal for identifying customers with potential for electricity substitution, comprising: a memory and a processor;

[0039] The memory is used to store program code, which corresponds to the method for identifying potential customers for electricity substitution as provided in the first aspect of this application.

[0040] The processor is used to read and execute the program code to implement the method for identifying potential customers for electricity substitution.

[0041] The fourth aspect of this application provides a computer-readable storage medium storing program code that is read and executed by a processor to implement the method for identifying potential customers for electricity substitution as provided in the first aspect of this application.

[0042] As can be seen from the above technical solutions, this application has the following advantages:

[0043] The solution provided in this application first obtains multi-dimensional raw data from different data sources and converts it into corresponding view data. Then, representative anchor points are selected within each view, and an initial similarity matrix for each view is constructed by calculating the similarity between each sample point and the anchor point. The multi-view similarity matrix is ​​input into a feature selection model, and the noise influence of low-weight views is gradually eliminated through model optimization. Once the model converges, the optimal similarity matrix is ​​obtained. Then, spectral clustering is performed on the optimal similarity matrix to finally obtain the customer type classification result. This solution first reduces computational complexity through anchor point selection, and simultaneously uses a feature selection model to dynamically adjust the view contribution, effectively suppressing interference from irrelevant features. It can integrate multi-dimensional information such as power load and equipment status, accurately identify the customer's energy consumption potential under different environmental conditions, improve classification accuracy, and solve the technical problem of low identification accuracy in existing technologies. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating an embodiment of a method for identifying potential customers for electricity substitution provided in this application.

[0046] Figure 2 This is a schematic diagram of the architecture of an embodiment of a potential customer type identification device for electricity substitution provided in this application.

[0047] Figure 3 This is a schematic diagram of the architecture of an embodiment of a terminal for identifying potential customers for electricity substitution provided in this application. Detailed Implementation

[0048] In existing technologies, electricity substitution retrofit projects require accurate identification of potential customer types. However, current methods mainly rely on single-dimensional data analysis, such as considering only electricity load or customer attributes. Different customer types exhibit significant differences in energy consumption behavior under the influence of equipment operating status and environmental weather conditions, making it difficult for a single data view to comprehensively reflect customer characteristics. Traditional clustering algorithms suffer from the curse of dimensionality and noise interference when processing multi-source heterogeneous data, resulting in insufficient accuracy in customer classification and affecting the targeting of electricity substitution strategies.

[0049] To address the aforementioned issues, existing technologies cannot effectively integrate complementary information from multi-dimensional data, such as the correlation between power load fluctuations and ambient temperature, or the relationship between equipment aging and energy efficiency. The inventors discovered that integrating features from different data sources through a multi-view learning framework can more comprehensively describe customer behavior patterns. However, the high dimensionality and redundant features of multi-view data increase computational complexity, and noise between different views may interfere with model convergence. Therefore, this application provides a method, device, terminal, and medium for identifying potential customers for electricity substitution. This solution proposes reducing data size through anchor point screening and optimizing view weights using a feature selection model, thereby improving clustering efficiency and accuracy. This addresses the technical problem of low accuracy in identifying potential customers for electricity substitution due to the inability of existing methods to effectively handle high-dimensional and complex customer data.

[0050] To make the inventive objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] First, a detailed description of an embodiment of the method for identifying potential customers for electricity substitution provided in this application is as follows:

[0052] Please see Figure 1 This application provides an embodiment of a method for identifying potential customers for electricity substitution, the steps of which include:

[0053] Step 101: Obtain multi-dimensional data of the customer to be identified;

[0054] Step 102: Perform view data transformation on the multi-dimensional data to construct a multi-view dataset;

[0055] Each view data corresponds to one of the dimensions in the multi-dimensional data;

[0056] Step 103: Based on the multi-view dataset, select several anchor points from the sample points contained in each view data to construct an anchor point matrix. Then, based on the anchor point matrix, calculate the similarity between each sample point and the anchor point to obtain a similarity matrix.

[0057] Step 104: Based on the preset multi-view feature selection model, the similarity matrix is ​​used as the model input. Combined with the anchor point matrix and the projection matrix corresponding to each view data, the multi-view feature selection model is optimized. When the multi-view feature selection model has been iterated and optimized and meets the preset iteration termination condition, the optimal similarity matrix output by the multi-view feature selection model is obtained.

[0058] Step 105: Perform spectral clustering based on the optimal similarity matrix to determine the customer type identification results of potential customers for electricity substitution based on the obtained clustering results.

[0059] In this embodiment, the multi-dimensional data refers to a heterogeneous data set obtained from different data sources. Specifically, it can be implemented using power system monitoring data, meteorological station records, equipment sensor logs, and customer profile information, to comprehensively reflect customers' energy consumption characteristics and environmental impact factors. View data transformation refers to independently mapping each data dimension into a separate data view. This can be achieved through data preprocessing and feature extraction, such as converting time-series power load data into a frequency domain feature view, thereby reducing data redundancy. The anchor point matrix refers to the set of representative data points selected from each view's sample points. This can be achieved using K-means clustering centers or random sampling methods to reduce the computational cost of similarity calculations. The similarity matrix is ​​the quantification of the correlation between sample points and anchor points. This can be achieved using Gaussian kernel functions or cosine similarity calculations, used to characterize the local structure of customer behavior under different views. The multi-view feature selection model is an optimization model that integrates similarity information from multiple views. Specifically, it can iteratively update view weights and projection matrices using the alternating direction multiplier method to suppress the interference of noisy views on clustering results. Spectral clustering processing refers to a graph-based nonlinear clustering method used to discover potential classification structures in the data.

[0060] Specifically, the process begins by collecting multi-dimensional raw customer data from the power system, meteorological database, and equipment management platform. Next, within each view, a representative anchor point is selected using the K-means algorithm; for example, a typical daily load curve is chosen as the anchor point from the power load view. An initial similarity matrix for each view is constructed by calculating the similarity between each sample point and the anchor point. This multi-view similarity matrix is ​​then input into a feature selection model, and model optimization gradually eliminates the noise influence of low-weight views. Once the model converges, the optimal similarity matrix is ​​obtained. This matrix implicitly contains the clustering structure information of the multi-view data. To make this clustering structure explicit and obtain the final user clustering results, spectral clustering processing is then performed on the optimal similarity matrix. Specifically, this involves constructing an extended similarity matrix and a degree matrix, extracting low-dimensional embedding features, and performing K-means clustering. This clusters all users into several final clusters, with users in the same cluster considered to be of the same customer type, ultimately yielding the customer type classification results.

[0061] This solution reduces computational complexity through anchor point selection and dynamically adjusts view contribution using a feature selection model, effectively suppressing interference from irrelevant features. Furthermore, spectral clustering combined with the optimal similarity matrix can capture non-linear data structures and is more adaptable to complex customer behavior patterns compared to the traditional K-means algorithm. Through these technical solutions, this application can integrate multi-dimensional information such as power load and equipment status to accurately identify customer energy consumption potential under different environmental conditions. The anchor point selection strategy reduces similarity calculations and improves model efficiency. The multi-view feature selection model, through dynamic weight allocation, reduces the impact of noise such as sudden changes in meteorological data on clustering results, improving classification accuracy.

[0062] Furthermore, the expression for the multi-view feature selection model mentioned in the above basic embodiments is as follows:

[0063]

[0064] In the formula; Representing the The weight of each view; This is a similarity matrix between multi-view data and anchor points; The element in the i-th row and j-th column of the similarity matrix; For the first Anchor point matrix for each view; For the first The j-th element in the anchor point matrix of each view; For the first Projection matrix of each view; It is a diagonal matrix; for The first in One element, used to represent the anchor point. The weight, For matrix A column vector consisting of the diagonal elements. For regularization terms; Let I be the weight of the regularization term, and let I be the identity matrix.

[0065] in, To constrain the non-negativity of the similarity matrix Z and ensure it retains the properties of a traditional similarity matrix, a regularization term is applied. and constraints It can reduce Redundant information between anchor points, obtaining high-quality anchor point data, rank constraint. This allows the matrix to It has good clustering properties to guide feature selection; sparse constraints The purpose is to select directly. Each feature participates in model optimization.

[0066] Among them, the weight of the view data The importance coefficients of different views in the model can be implemented using adaptive adjustment algorithms based on view contribution to balance the impact of heterogeneous data from multiple sources. The similarity matrix Z is a matrix formed by calculating the correlation between sample points and anchor points, which can be implemented using kernel functions or cosine similarity methods to quantify potential correlations between data. The projection matrix W is a linear transformation matrix that maps high-dimensional features to a low-dimensional space, which can be implemented using principal component analysis or linear discriminant analysis to achieve feature dimensionality reduction and selection. The regularization term is a constraint term to prevent overfitting of the model, which can be implemented using L1 norm or L2 norm to improve the model's generalization ability.

[0067] Specifically, this model introduces view weight parameters. The algorithm dynamically adjusts the influence of different data sources on feature selection. When calculating the similarity matrix, it combines the anchor matrix A and the projection matrix W to map multi-view data to a unified feature space. By constructing an optimization objective function that includes a regularization term, the view weights, projection matrix, and similarity matrix are updated synchronously during the iteration process, ultimately selecting the most discriminative feature combination. During the optimization process, the diagonal matrix is ​​used to constrain the anchor weight distribution, preventing individual anchors from excessively influencing the clustering results, while the parameters... This is used to control the regularization strength, balancing model complexity and fitting accuracy.

[0068] This solution constructs a regularized model with an adaptive weight adjustment mechanism, which can automatically identify the contribution of different view data while suppressing the interference of noise features on the clustering results. Compared with single-view feature selection methods, this model achieves finer-grained control in the multi-source data fusion process, significantly improving the discriminative ability of feature selection. Through the above technical solution, this application effectively solves the problems of feature redundancy and unreasonable view weight allocation in multi-source heterogeneous data fusion, and improves the accuracy of identifying potential customers for electricity substitution. By dynamically adjusting the view weights and feature projection matrix, it can adaptively capture complementary information between data of different dimensions, while using regularization constraints to avoid overfitting, ensuring that the clustering results truly reflect the differences in customer energy consumption characteristics, and providing reliable technical support for accurately identifying high-potential customers.

[0069] Based on the multi-view feature selection model proposed above, this application further proposes iterative optimization and model solving of the multi-view feature selection model, specifically including: determining a target variable from the variables to be solved in the multi-view feature selection model, and fixing the remaining non-target variables; for the determined target variable, transforming the multi-view feature selection model into a corresponding optimization problem expression, and then solving the optimization problem expression to obtain the solution result of the target variable; sequentially determining the target variable from the remaining variables to be solved, and then for the target variable, performing optimization problem transformation and optimization problem solving, until all variables to be solved have been solved.

[0070] In this model, the variables to be solved refer to the set of parameters that need to be optimized in the multi-view feature selection model. These can include view weights, projection matrices, similarity matrices, etc., and step-by-step optimization is achieved through variable separation. The target variable is the single variable selected as the optimization object in a single iteration. This can be achieved by using an alternating optimization strategy to select each variable sequentially, while keeping the remaining variables fixed to simplify computational complexity. The optimization problem expression is the mathematical expression that transforms the multi-view feature selection model into a single variable. This can be achieved through matrix factorization or constraint transformation, facilitating the application of numerical optimization algorithms. The augmented Lagrange method is a numerical solution for constrained optimization problems. Specifically, it transforms the original problem into an unconstrained optimization form by introducing Lagrange multipliers and a quadratic penalty term, and then iteratively solves it using gradient descent or matrix differentiation methods.

[0071] Specifically, in the model optimization process, the parameters such as view weights, projection matrices, and similarity matrices in the multi-view feature selection model are first identified as a set of variables to be solved. For each variable, the following steps are performed: After selecting the current objective variable, the values ​​of all other variables are fixed, simplifying the original multi-objective optimization model into a single-objective optimization problem containing only that variable. For example, when optimizing the projection matrix, the view weights and similarity matrix are assigned temporary fixed values, at which point the objective function is transformed into a convex optimization problem with respect to the projection matrix. Subsequently, an auxiliary function is constructed using the augmented Lagrange method, transforming any possible equality or inequality constraints into penalty terms of the objective function, establishing a differentiable unconstrained optimization form. By calculating the partial derivative of the objective function with respect to the current variable, an iterative update rule is used to gradually approach the optimal solution. After completing the optimization of the current variable, the process switches to the next variable to be optimized in a predetermined order, repeating the above process until all variables have completed one round of optimization. This iterative process continues until the change in the model's objective function value is below a set threshold or the maximum number of iterations is reached.

[0072] For example, for the projection matrix Based on the aforementioned multi-view feature selection model, relevant features are derived mathematically. The model is transformed into the following optimization problem:

[0073]

[0074] The above model optimization problem can be solved by introducing auxiliary variables using the Augmented Lagrange Method (ALM):

[0075]

[0076] in, Representing the Lagrange multipliers, Represents the penalty factor, anchor matrix and class center matrix The solution can be found using matrix differentiation, Q (v) and P (v) The mathematical simplification of the above multi-view feature selection model includes The variable matrix, V (v) These are auxiliary variables introduced by the augmented Lagrange method;

[0077] For similarity matrix , usually targeting By solving for the column vectors in the model, and through mathematical derivation, the model can be transformed into the following optimization problem:

[0078]

[0079] Where, d iThe mathematical simplification of the above multi-view feature selection model includes A vector of variables.

[0080] For this model optimization problem, the Lagrange multiplier method can be used to transform the constrained problem into an unconstrained problem for optimization solution;

[0081] For the constructed view anchor point weight matrix The column vector consisting of its corresponding diagonal elements Through mathematical derivation, the model can be transformed into the following optimization problem:

[0082]

[0083] Among them, K (v) and N (v) The mathematical simplification of the above multi-view feature selection model includes The variable matrix.

[0084] The transformed optimization problem is a typical quadratic programming problem, which can be solved efficiently using MATLAB's quadprog function.

[0085] This scheme decomposes complex multivariate coupled problems into multiple single-variable subproblems through variable separation and alternating optimization mechanisms, significantly reducing computational complexity. Simultaneously, the application of the augmented Lagrange method effectively handles nonlinear constraints in the model, avoiding the problem of traditional penalty function methods requiring infinitely increasing penalty coefficients, ensuring the feasibility and convergence speed of the solution. Through these technical solutions, this application achieves efficient and stable solutions for multi-view feature selection models, solving the problems of easy divergence and excessive computational resource consumption in traditional methods when processing multi-source heterogeneous data. By employing a staged variable optimization strategy, computational costs are effectively controlled while ensuring model accuracy, enabling the electricity substitution potential customer identification system to quickly process large-scale, multi-dimensional data, improving the real-time performance and accuracy of customer classification. This method is particularly suitable for application scenarios in power systems with multi-source data heterogeneity, high feature dimensionality, and large sample sizes, providing reliable technical support for the precise implementation of electricity substitution transformation projects.

[0086] In some embodiments, this application further proposes a method for determining the identification results of potential customers for electricity substitution by performing spectral clustering based on the optimal similarity matrix. The method includes constructing an extended similarity matrix based on the optimal similarity matrix, constructing a degree matrix and a normalized Laplacian matrix based on the extended similarity matrix, performing feature decomposition on the normalized Laplacian matrix and selecting target feature vectors to form an embedding matrix, and obtaining the customer type classification results through data dimensionality reduction and K-means clustering.

[0087] The extended similarity matrix is ​​formed by combining the optimal similarity matrix and the identity matrix using a matrix expansion relation. Specifically, it can be achieved by multiplying the optimal similarity matrix and its transpose, then superimposing the identity matrix. This matrix fully represents the global similarity relationships between all clients and anchor points. The degree matrix is ​​a diagonal matrix formed by the sum of the elements in each row of the extended similarity matrix. This can be achieved by summing each row and placing the result on the diagonal. It describes the connection weight of each client node. The normalized Laplacian matrix is ​​obtained by symmetric normalization of the degree matrix and the extended similarity matrix. This can be achieved by multiplying the inverse square root of the degree matrix with the extended similarity matrix. This eliminates the influence of data scale differences on the clustering results. Eigenvalue decomposition refers to the process of calculating eigenvalues ​​and eigenvectors from the normalized Laplacian matrix. This can be achieved using the Jacobi iteration method or the QR algorithm. It is used to extract the low-dimensional manifold structure of the data. An embedding matrix is ​​a low-dimensional representation matrix formed by concatenating the eigenvectors corresponding to the smallest non-zero eigenvalues ​​selected. Specifically, it can be achieved by setting a threshold or selecting the top k eigenvectors, and is used to retain the main clustering structure information of the data.

[0088] Specifically, after obtaining the optimal similarity matrix, it is first combined with the identity matrix according to the matrix expansion relation to generate an expanded similarity matrix, which fully covers the association between all customers and anchor points. Next, the degree matrix of the expanded similarity matrix is ​​calculated, and a symmetric normalized Laplacian matrix is ​​constructed based on the degree matrix and the expanded similarity matrix. By performing eigenvalue decomposition on this matrix, eigenvectors corresponding to several of the smallest non-zero eigenvalues ​​are selected to construct a low-dimensional embedding space. Finally, the K-means clustering algorithm is executed in this space to group customers with similar features into the same cluster, thus obtaining accurate customer type classification results. The entire process enhances data representation capabilities through matrix expansion and effectively handles high-dimensional heterogeneous data using spectral clustering methods, overcoming the limitations of traditional clustering methods in complex customer data scenarios.

[0089] More specifically, the matrix extension relation mentioned in this embodiment is as follows:

[0090]

[0091] In the formula, S is the extended similarity matrix, and Z is the optimal similarity matrix, which describes the similarity between the user and the anchor point. This is the transpose of the optimal similarity matrix, describing the similarity between the anchor and the user. It describes the similarity between users.

[0092] First, based on the similarity matrix between the anchor point and the sample... A more comprehensive extended similarity matrix can be constructed using the above matrix extension formula. That is, the extended similarity matrix, which describes all The relationship between users.

[0093] Next, the extended similarity matrix is ​​calculated. degree matrix The degree matrix is ​​a diagonal matrix, and the degree matrix is ​​the first degree matrix on the diagonal. element It is an extended similarity matrix The The sum of all elements in a row is expressed as follows:

[0094]

[0095] Finally, we use the degree matrix and extended similarity matrix To construct the normalized Laplace matrix Normalization ensures the stability of eigenvalue decomposition. The specific formula for calculating the normalized Laplace matrix is ​​as follows:

[0096]

[0097] In this formula, It is The identity matrix.

[0098] Next, according to the preset eigenvalue decomposition formula For the normalized Laplace matrix Eigenvalue decomposition is performed to obtain eigenvalues ​​and corresponding eigenvectors. Based on spectral clustering theory, the eigenvector corresponding to the smallest eigenvalue of the Laplacian matrix contains important information about the graph structure. Assume the goal of this embodiment is to obtain... If there are 5 clusters, then select the one with the smallest cluster size. The eigenvectors corresponding to the non-zero eigenvalues.

[0099] Next, this The eigenvectors are concatenated column-wise to form an embedding matrix. ,matrix Each line represents a user or anchor point. New coordinates in a lower-dimensional space. In this space, users belonging to the same cluster will be closer together, while users from different clusters will be farther apart.

[0100] Finally, from the embedding matrix In the middle, before extraction Row-forming sub-embedding matrix These lines correspond to the original Individual users, in the context of Constructed In 3D space, the standard K-means clustering algorithm is applied. The algorithm will... Individual users The final cluster.

[0101] This solution effectively integrates optimized multi-view feature information by constructing an extended similarity matrix and combines it with eigenvalue decomposition of a symmetric normalized Laplacian matrix, significantly improving clustering stability under high-dimensional sparse data. Through this technical solution, this application can accurately capture potential association patterns of customers in the multi-view feature space, solving the problems of feature redundancy and clustering bias in traditional methods for identifying potential customers for electricity substitution. This method enhances data representation capabilities through matrix expansion and effectively processes multi-source heterogeneous data using spectral clustering technology, improving the accuracy of customer type classification and providing reliable data support for the precise implementation of electricity substitution and transformation projects.

[0102] In some embodiments, this application further proposes that after obtaining the multi-dimensional data in step 101, the method may further include: preprocessing the multi-dimensional data based on the multi-dimensional data and in conjunction with a preset correspondence between data types and preprocessing methods.

[0103] Specifically, this embodiment employs different processing methods based on data type differences during the preprocessing stage. When the data to be processed is numerical data, it is preprocessed using Z-score normalization; when the data to be processed is non-numerical data, it is preprocessed using one-hot encoding.

[0104] Z-score standardization transforms raw data into a distribution with a mean of 0 and a standard deviation of 1. This is achieved by calculating the difference between each data point and the mean, then dividing by the standard deviation. This method eliminates the influence of different units on the model, ensuring comparability of numerical data in subsequent analyses. One-hot encoding converts discrete non-numerical data into binary vectors composed of 0s and 1s. This is achieved by assigning unique bits to each category. This method transforms non-numerical data into a format recognizable by machine learning models, avoiding erroneous magnitude relationships introduced by direct assignment.

[0105] Specifically, the preprocessing process first determines the data type attributes. For example, continuous numerical data such as electricity load data and environmental meteorological data are standardized using Z-scores. By calculating their mean and standard deviation, the original data is mapped to a uniform scale. For discrete non-numerical data such as equipment type and industry category in customer attributes, one-hot encoding is used to generate an independent binary feature column for each category. By distinguishing data types and adapting processing methods, the distribution characteristics of numerical data are preserved, while noise introduced by improper encoding of non-numerical data is avoided. For example, a customer's average monthly electricity consumption is standardized to a standard score of -0.5, and its "manufacturing" category is encoded as a vector representation of [1,0,0], clearly distinguishing it from "service industry" [0,1,0].

[0106] This solution establishes a correspondence mechanism between data types and processing methods, addressing the challenge of uniformly processing mixed data types and avoiding feature distortions caused by improper encoding. This provides a standardized data foundation for subsequent multi-view analysis. Through this technical solution, this application effectively resolves the feature representation problem in the preprocessing of multi-source heterogeneous data. Numerical data, after standardization, eliminates dimensional differences, while non-numerical data is transformed into a computable form through one-hot encoding, resulting in better feature consistency in the subsequently constructed multi-view dataset. This differentiated preprocessing approach improves data quality while preserving the inherent characteristics of various data types, laying a reliable data foundation for accurately identifying potential customers for electricity substitution. Simultaneously, the standardized processing flow enhances the automation level of data preprocessing, adapting to the processing needs of large-scale customer data.

[0107] The above is a detailed description of an embodiment of a method for identifying potential customers for electricity substitution provided in this application. The following is a detailed description of related embodiments of an electricity substitution potential customer type identification device, terminal, and readable storage medium provided in this application.

[0108] Please see Figure 2 The present application provides an embodiment of a device for identifying potential customers for electricity substitution, the main components of which include:

[0109] The data acquisition unit 201 is used to acquire multi-dimensional data of the customer to be identified. The multi-dimensional data is multi-source heterogeneous data obtained from multiple data sources, including: power load data, environmental meteorological data, equipment operating status and customer attributes.

[0110] The view data transformation unit 202 is used to transform multi-dimensional data into view data and construct a multi-view dataset, wherein each view data corresponds to one of the dimensions in the multi-dimensional data.

[0111] The similarity matrix construction unit 203 is used to select several anchor points from the sample points contained in each view data according to the multi-view dataset to construct an anchor point matrix, and then calculate the similarity between each sample point and the anchor point based on the anchor point matrix to obtain the similarity matrix.

[0112] The model optimization unit 204 is used to optimize the multi-view feature selection model based on the preset multi-view feature selection model. The similarity matrix is ​​used as the model input, and the anchor point matrix and the projection matrix corresponding to each view data are combined. When the multi-view feature selection model has been iterated and the preset iteration termination condition is met, the optimal similarity matrix output by the multi-view feature selection model is obtained.

[0113] The spectral clustering processing unit 205 is used to perform spectral clustering processing based on the optimal similarity matrix, so as to determine the customer type identification result of potential customers for electricity substitution based on the obtained clustering results.

[0114] Specifically, the device synchronously collects customer electricity usage records, equipment sensor data, and user profile information through a data acquisition unit, forming a raw dataset containing four types of views: electricity, environment, equipment, and users. The view data conversion unit standardizes each type of data, for example, converting electricity load data into an hourly average power view and meteorological data into a temperature and humidity correlation matrix view. The similarity matrix construction unit selects anchor points within each view using a stratified sampling method; for example, in the electricity load view, a typical daily load curve is selected as the anchor point, and the cosine similarity between each customer's load curve and the anchor point is calculated to form an initial matrix. The model optimization unit alternately optimizes the projection matrix and view weights; for example, during the iteration process, after fixing the anchor point matrix, it solves an optimization problem with regularization constraints, gradually eliminating the influence of redundant features. The spectral clustering processing unit finally expands the optimized similarity matrix into a symmetric matrix, achieving refined classification of customer groups through feature vector space mapping.

[0115] This application constructs a multi-view collaborative optimization framework, integrating a heterogeneous data fusion module and a joint feature selection mechanism into the device. This effectively correlates the dynamic relationship between power consumption patterns and equipment operating conditions and environmental parameters. Through this technical solution, the application addresses the feature redundancy and dimensionality curse problems inherent in existing customer identification devices when processing multi-source heterogeneous data, achieving multi-dimensional joint modeling of customer energy consumption behavior. The device, through anchor point filtering and view weight allocation mechanisms, improves the accuracy of feature selection while ensuring computational efficiency. This ensures that the final clustering results accurately reflect the actual potential level of customers in electricity substitution transformation, providing reliable decision support for power grid companies to accurately implement electricity substitution projects.

[0116] Furthermore, such as Figure 3As shown, this application proposes an embodiment of a terminal for identifying potential customers for electricity substitution, comprising: a memory 33 and a processor 31, wherein the memory 33 and the processor 31 are connected via a communication bus 34; the memory 33 is used to store program code, which corresponds to the method for identifying potential customers for electricity substitution; the processor 31 is used to read and execute the program code to implement the method for identifying potential customers for electricity substitution.

[0117] In this context, memory refers to the hardware device used to store program code, specifically solid-state drives (SSDs) or hard disk drives (HDDs). Its function is to provide the processor with executable program code resources, ensuring the integrity and continuity of the data processing flow. The processor is the arithmetic unit used to execute the program code, specifically a multi-core central processing unit (CPU) or a graphics processing unit (GPU). Its function is to automate customer type identification by running the program code, performing data transformation, feature selection, model optimization, and cluster analysis. Program code is a set of computer instructions containing data processing logic, specifically written in Python or Java programming languages. Its function is to transform the method for identifying potential customers for electricity substitution into executable algorithmic steps, guiding the processor to complete the customer classification task according to a pre-defined process.

[0118] Specifically, the program code stored in the memory includes modules for data acquisition, view transformation, anchor point selection, similarity calculation, model optimization, and spectral clustering. During execution, the processor first collects power load, environmental weather, equipment status, and customer attribute data from multiple data sources through the data acquisition module. Then, the view transformation module divides the heterogeneous data into multi-view datasets according to dimensions. Next, the anchor point selection module extracts representative anchor points from each view sample to construct a matrix, and the similarity calculation module generates similarity relationships between samples based on the anchor point matrix. The model optimization module iteratively updates the projection matrix and weight parameters, ultimately outputting the optimal similarity matrix. The spectral clustering module performs feature decomposition and dimensionality reduction based on this matrix, generating customer type clustering results using the K-means algorithm.

[0119] In some specific implementations, the program code may further include a data preprocessing module, such as performing Z-score standardization on numerical data and one-hot encoding on non-numerical data, to eliminate the impact of dimensional differences on the model. During execution, the processor can dynamically adjust the weights of the regularization term to balance model complexity and fitting accuracy, or accelerate the iterative optimization process of large-scale datasets through parallel computing.

[0120] Furthermore, this application proposes a computer-readable storage medium embodiment, in which program code is stored, which is read and executed by a processor to realize a method for identifying potential customers for electricity substitution. The method includes: acquiring multi-dimensional data of the customers to be identified, performing view data transformation on the multi-dimensional data to construct a multi-view dataset, filtering anchor points to construct an anchor point matrix and calculating a similarity matrix, optimizing the optimal similarity matrix based on a multi-view feature selection model, and determining the customer type identification result through spectral clustering.

[0121] Computer-readable storage media refers to the physical carrier used to store program code, specifically solid-state drives (SSDs), hard disk drives (HDDs), or flash drives. It uses non-volatile storage technology to ensure the persistent preservation of the program code. Program code is a sequence of computer instructions, specifically written in programming languages ​​such as Python, Java, or C++. When executed by a processor, it triggers data processing, model optimization, and cluster analysis processes. The processor is the arithmetic unit that executes the program code, specifically a central processing unit (CPU) or a graphics processing unit (GPU). It reads the program code from the storage medium to run the algorithm for identifying potential customers for energy substitution.

[0122] Specifically, when the program code stored in the computer-readable storage medium is executed by the processor, it first drives the data acquisition unit to collect multi-source heterogeneous data such as power load data and environmental meteorological data. Then, the view data conversion unit converts the data of different dimensions into independent view datasets. The program code controls the processor to execute an anchor point selection algorithm to construct an anchor point matrix and calculate the similarity matrix between sample points and anchor points. In the multi-view feature selection model optimization stage, the program code guides the processor to iteratively update the projection matrix and weight parameters until the convergence condition is met and the optimal similarity matrix is ​​output. Finally, the program code calls the spectral clustering algorithm to perform feature decomposition and cluster analysis on the optimal matrix, generating customer type identification results.

[0123] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the terminals, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed terminals, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0125] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0126] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0127] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0128] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0129] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] 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.

Claims

1. A method for identifying a type of a customer of an electric power substitution potential, characterized by, The method comprises the following steps: Obtain multi-dimensional data of a client to be identified, wherein the multi-dimensional data is multi-source heterogeneous data obtained from multiple data sources, including power load data, environmental meteorological data, equipment operating status and client attributes; Perform view data conversion on the multi-dimensional data to construct a multi-view data set, wherein each view data corresponds to one dimension of the multi-dimensional data; According to the multi-view data set, filter a plurality of anchor points from sample points contained in each view data to construct an anchor point matrix, and then calculate the similarity between each sample point and the anchor points based on the anchor point matrix to obtain a similarity matrix; Based on a preset multi-view feature selection model, input the similarity matrix into the model, combine the anchor point matrix and the projection matrix corresponding to each view data, and optimize the multi-view feature selection model. When the multi-view feature selection model is iteratively optimized and meets a preset iteration termination condition, an optimal similarity matrix output by the multi-view feature selection model is obtained. Based on the optimal similarity matrix, perform spectral clustering processing to determine a client type identification result of a potential client of electric energy replacement according to the obtained clustering result.

2. The method of claim 1, wherein, The multi-view feature selection model specifically comprises: In the formula, Representing the The weight of each view's data. For the first The first view in the One sample, This is a similarity matrix between multi-view data and anchor points. The element in the i-th row and j-th column of the similarity matrix. For the first Anchor point matrix for each view, For the first The j-th element in the anchor point matrix of each view. For the first Projection matrix of each view, It is a diagonal matrix. for The first in One element, used to represent the anchor point. The weight, For matrix A column vector consisting of the diagonal elements. For regularization terms, Let I be the weight of the regularization term, and let I be the identity matrix. The number of features selected, This is the sum of the view anchor weights. Similarity matrix The corresponding Laplace matrix, Let be the rank of the matrix. The sample size of the dataset. The number of anchor points in each view. This represents the number of clusters in the multi-view dataset.

3. The method of claim 2, wherein the method further comprises: The iterative optimization and model solving of the multi-view feature selection model specifically comprises: According to each variable to be solved in the multi-view feature selection model, determine a target variable from the variables to be solved, and fix the remaining non-target variables; For the determined target variable, convert the multi-view feature selection model into a corresponding optimization problem expression, and then solve the optimization problem expression to obtain the solution of the target variable; Determine the target variable from the remaining variables to be solved in turn, and then perform optimization problem conversion processing and optimization problem solving for the target variable until all variables to be solved are solved.

4. The method of claim 1, wherein, The spectral clustering processing based on the optimal similarity matrix to determine a client type identification result of a potential client of electric energy replacement according to the obtained clustering result specifically comprises: According to the optimal similarity matrix, construct an extended similarity matrix based on a preset matrix extension relationship, wherein the extended similarity matrix contains the similarity between all clients and clients; Based on the extended similarity matrix, construct a normalized Laplacian matrix; Perform eigenvalue decomposition on the normalized Laplacian matrix to obtain a plurality of sets of eigenvalues and eigenvectors; According to the size of the eigenvalues, select a plurality of target eigenvectors corresponding to a plurality of smallest non-zero eigenvalues, and concatenate the target eigenvectors by columns to obtain an embedding matrix; Perform data dimensionality reduction processing and K-means clustering processing on the embedding matrix to obtain a plurality of clustering clusters representing different client types.

5. The method of claim 4, wherein the method further comprises: The matrix extension relationship specifically comprises: where S is the extended similarity matrix, Z is the optimal similarity matrix, is the transpose of the optimal similarity matrix.

6. The method of claim 1, wherein, Further comprising: According to the multi-dimensional data, combine the corresponding relationship between the preset data type and the preprocessing mode to preprocess the multi-dimensional data.

7. The method of claim 6, wherein the method further comprises: According to the multi-dimensional data, combine the corresponding relationship between the preset data type and the preprocessing mode to preprocess the multi-dimensional data. When the data to be processed is numerical data, the data to be processed is preprocessed by a Z-score standardization processing mode; When the data to be processed is non-numerical data, the data to be processed is preprocessed by a one-hot encoding processing mode.

8. An electric energy alternative potential customer type identifying device characterized by, Comprise: A data acquisition unit configured to acquire multi-dimensional data of a customer to be identified, wherein the multi-dimensional data is multi-source heterogeneous data obtained from multiple data sources, including power load data, environmental meteorological data, equipment operating status and customer attributes; A view data conversion unit configured to convert the multi-dimensional data into view data, and construct a multi-view data set, wherein each view data corresponds to one dimension of the multi-dimensional data; A similarity matrix construction unit configured to filter a plurality of anchor points from sample points contained in each view data according to the multi-view data set, to construct an anchor point matrix, and calculate the similarity between each sample point and the anchor points based on the anchor point matrix, to obtain a similarity matrix; A model optimization unit configured to input the similarity matrix into a pre-set multi-view feature selection model, combine the anchor point matrix and the projection matrix corresponding to each view data, and optimize the multi-view feature selection model, and obtain an optimal similarity matrix output by the multi-view feature selection model when the multi-view feature selection model is iteratively optimized and meets a pre-set iteration termination condition; A spectral clustering processing unit configured to perform spectral clustering processing based on the extended similarity matrix, and determine a customer type identification result of the electric energy substitution potential customer according to the obtained clustering result.

9. An electric energy alternative potential customer type identifying terminal characterized by, Comprise: A memory and a processor; The memory is configured to store program code corresponding to the electric energy substitution potential customer type identification method of any one of claims 1 to 7; The processor is configured to read and execute the program code to implement the electric energy substitution potential customer type identification method.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores program code, which is read and executed by the processor to implement the electric energy substitution potential customer type identification method of any one of claims 1 to 7.

Images