Power consumer classification method and device for easy-to-confuse industries, electronic equipment and storage medium

Abstract

The invention provides a power consumer classification method and device for easy-to-confuse industries, electronic equipment and a storage medium. The method comprises the following steps: acquiring multi-dimensional industry feature data of to-be-classified power consumers; generating a fusion feature vector for each to-be-classified power user according to the multi-dimensional industry feature data; clustering to-be-classified power users based on the fusion feature vector to determine an easily confused industry user fuzzy set; and performing initial prediction on the industry to which the to-be-classified power users in the easily confused industry user fuzzy set belong, and calculating a comprehensive discrimination score of each industry corresponding to the user by using a preset weighted dot product model, thereby determining the industry to which the user belongs. According to the embodiment of the invention, the technical problem of disordered classification of the electricity consumption data caused by similar industry classification can be solved, the purity and accuracy of the industry electricity consumption statistical data are improved, and reliable data support is provided for macroeconomic analysis and industry supervision based on electric power data.

Description

Technical Field

[0001] This application relates to the field of electricity consumption behavior analysis technology, and in particular to methods, devices, electronic equipment and storage media for classifying electricity users in industries prone to confusion. Background Technology

[0002] Currently, electricity big data has become a key element for understanding the economic and social operation, optimizing energy resource allocation, and supporting industrial policy formulation. Among them, industry-specific electricity consumption data is a core statistical indicator, and the accuracy and purity of its classification directly determine the reliability of macroeconomic analysis. In my country, industry-specific electricity consumption statistics mainly follow the national standard "National Industrial Classification of Economic Activities" (GB / T4754-2017). Power grid companies assign industry labels to users based on user installation information, business registration data, etc., and then summarize and report statistical data based on these labels.

[0003] However, in practical applications, many industry classifications suffer from blurred boundaries and highly overlapping electricity consumption characteristics, leading to distorted statistical data. This is mainly reflected in two aspects: First, industry classifications themselves exhibit proximity; for example, the electrical machinery and equipment manufacturing industry overlaps with the computer, communication, and other electronic equipment manufacturing industry in terms of production processes and electrical equipment. Second, a single user may encompass multiple production activities; for example, a large manufacturing park may simultaneously cover metal processing and equipment assembly, making its mixed electricity consumption characteristics difficult to simply categorize into a single industry. These user groups, which are adjacent in standard classifications and similar in actual electricity consumption behavior, constitute easily confused industry user sets, a "difficult problem" that traditional methods struggle to address. Furthermore, existing technical solutions have significant shortcomings in addressing this challenge: existing methods largely rely on the time-domain characteristics of a single load curve, failing to deeply explore the frequency-domain patterns of load fluctuations, the temporal characteristics of production operations, and the inherent energy consumption differences in processes, resulting in limited ability to distinguish easily confused industries. Summary of the Invention

[0004] This application is made in view of at least one of the aforementioned technical problems existing in the prior art. According to one aspect of this application, a method for classifying electricity users in industries prone to confusion is provided, the method comprising: Obtain multidimensional industry characteristic data of electricity users to be classified; Using an attention mechanism, a fusion feature vector is generated for each electricity user to be classified based on the multidimensional industry feature data. Using a preset clustering algorithm, all electricity users to be classified are clustered based on the fused feature vector to determine the fuzzy set of easily confused industry users. Using a pre-defined hybrid discrimination model, an initial prediction is made about the industry to which any power user to be classified belongs in the fuzzy set of easily confused industry users, and the contribution vector of each industry feature of any power user to be classified is determined to the initial prediction result. Based on the contribution vector of each industry feature of any power user to the initial prediction result, the comprehensive discrimination score of each industry corresponding to any user is calculated using a preset weighted dot product model. The industry affiliation of any given user is determined based on the comprehensive discrimination score.

[0005] In some embodiments, determining the industry affiliation of any user based on the comprehensive discrimination score includes: The industry with the highest comprehensive discrimination score is determined as the industry to which any given user belongs.

[0006] In some embodiments, determining the industry affiliation of any user based on the comprehensive discrimination score includes: Based on a hierarchical decision-making mechanism with dual thresholds, any one of the users is reclassified, and the industry to which the user belongs is redefined. The hierarchical decision-making mechanism based on dual thresholds reclassifies any given user and redetermines the industry to which that user belongs, including: Calculate the optimal industry category and corresponding first comprehensive discrimination score of any user in the fuzzy set of users in easily confused industries, as well as the second optimal industry and corresponding second comprehensive discrimination score of any user; Calculate the score difference between the first comprehensive discrimination score and the second comprehensive discrimination score; The first comprehensive discrimination score is compared with a preset reliability threshold, and the score difference is compared with a preset advantage difference threshold. Based on the comparison results, the industry to which any user belongs is determined.

[0007] In some embodiments, the preset confidence threshold includes a maximum confidence threshold and a minimum confidence threshold; comparing the first comprehensive discrimination score with the preset confidence threshold and comparing the score difference with a preset advantage difference threshold, and determining the industry to which any one user belongs based on the comparison result includes: If the first comprehensive discrimination score is greater than or equal to the maximum confidence threshold and the score difference is greater than or equal to the preset dominance difference threshold, then any user is determined to belong to the industry with the highest first comprehensive discrimination score. If the first comprehensive discrimination score is less than the minimum confidence threshold, or the score difference is less than the preset advantage difference threshold, the industry to which any user belongs will be manually verified. If the first comprehensive discrimination score is less than the maximum confidence threshold and greater than or equal to the minimum confidence threshold, and the score difference is greater than or equal to the preset advantage difference threshold, the industry to which any user belongs is marked as pending.

[0008] In some embodiments, obtaining multidimensional industry characteristic data of the electricity users to be classified includes: Multidimensional industry characteristic data of the electricity users to be classified are obtained from the electricity consumption side, the user side, and the verification side, and an industry user characteristic profile of each electricity user to be classified is constructed based on the multidimensional industry characteristic data.

[0009] In some embodiments, the multidimensional industry characteristic data includes load curve depth characteristics, production time series characteristics, and energy consumption intensity characteristics; wherein, The load curve depth characteristics include at least one of the following indicators: production continuity index, load fluctuation frequency domain energy ratio, and non-power frequency load ratio; The production time-series characteristics include at least one of the following indicators: intelligent shift system recognition, seasonal production feature vector; The energy intensity characteristics include the unit product power consumption index.

[0010] In some embodiments, an attention mechanism is used to generate a fused feature vector for each electricity user to be classified based on the multidimensional industry feature data, including: The multidimensional industry feature data is preprocessed; The multidimensional industry feature data is standardized and encoded. The standardized and encoded multidimensional industry feature data are concatenated in a preset order to generate the original feature vector; The original feature vectors are weighted and summed using an attention mechanism to obtain a fused feature vector.

[0011] In some embodiments, the multidimensional industry feature data is preprocessed, including: The multidimensional industry feature data is denoised using wavelet transform. Remove outliers from the multidimensional industry feature data; The missing values ​​in the multidimensional industry feature data are filled in using a linear interpolation method.

[0012] In some embodiments, the multidimensional industry feature data is standardized and encoded, including: The multidimensional industry feature data is standardized using the Z-score standardization method. The categorical data in the multidimensional industry feature data is converted into binary numerical vectors using the one-hot encoding method.

[0013] In some embodiments, an attention mechanism is used to perform a weighted summation of the original feature vectors to obtain a fused feature vector, including: Based on preset attention parameters, calculate the attention weight of each feature in the original feature vector; The original feature vector is weighted according to the attention weights to obtain the fused feature vector.

[0014] In some embodiments, a preset clustering algorithm is used to cluster all electricity users to be classified based on the fused feature vector to determine a fuzzy set of easily confused industry users, including: The K-distance graph method is used to determine the adaptive optimal neighborhood radius parameter; The initial clustering of all the electricity users to be classified is performed using a compact adaptive optimal neighborhood radius parameter to form a high-confidence core cluster. The set of users outside the high-confidence core cluster is used as the undetermined classification set. Using a relaxed adaptive optimal neighborhood radius parameter, a second clustering is performed on the undetermined classification set to determine the weak association substructure in the undetermined classification set; The weakly correlated substructures formed by the secondary clustering, and the set of users consisting of noise points that are still not classified after the secondary clustering, are taken as the fuzzy set of easily confused industry users.

[0015] In some embodiments, a preset hybrid discriminant model is used to make an initial prediction of the industry to which any power user to be classified belongs in the fuzzy set of easily confused industry users belongs, and to determine the contribution vector of each industry feature of the power user to the initial prediction result, including: Using users with labeled industries and their corresponding fused feature vectors, a benchmark industry classifier is trained using the gradient boosting tree method to predict the industry probability of the fuzzy set of users in the easily confused industries. The industry affiliation prediction probability of each user in the benchmark industry classifier is taken as the target to be explained, and the fused feature vector corresponding to each user is taken as the feature set input to the interpreter. The defined interpretation object and interpretation features are input into the Shapleyka interpreter to obtain... The contribution vector is predicted for each industry category; wherein the expression for the contribution vector is as follows: ; in, Represents the contribution vector; The number of features involved in the discrimination is represented; i represents the i-th user; k represents the k-th industry to which the i-th user belongs.

[0016] In some embodiments, based on the contribution vector of each industry characteristic of any given electricity user to the initial prediction result, a pre-defined weighted dot product model is used to calculate the comprehensive discrimination score for each industry corresponding to any given user, including: Calculate the global average Shapley absolute value of each feature corresponding to each user in the fuzzy set of easily confused industry users; Based on the comparison values ​​between the various features of each user, a judgment matrix is ​​constructed using the 1-9 scaling method; Based on the judgment matrix, calculate the feature weight vector of each feature; Based on the global average Shapley absolute value and the feature weight vector, the comprehensive discrimination score for each feature is calculated; wherein, the expression for the comprehensive discrimination score is as follows: ; Where i represents the fuzzy user set The i-th easily confused user that needs to be identified; k represents the k-th industry to which the i-th user belongs; j represents the j-th industry characteristic of the i-th user; This indicates the number of features involved in the discrimination. Represents the contribution vector of the j-th industry feature; This represents the feature weight of the j-th industry feature.

[0017] According to another aspect of this application, a power user classification device for industries prone to confusion is also provided, the device comprising: The acquisition module is used to acquire multi-dimensional industry characteristic data of the electricity users to be classified; The vector computation module is used to generate a fused feature vector for each electricity user to be classified based on the multi-dimensional industry feature data by utilizing an attention mechanism. The clustering module is used to cluster all power users to be classified based on the fused feature vector using a preset clustering algorithm, so as to determine the fuzzy set of easily confused industry users; The prediction module is used to make an initial prediction of the industry to which any power user to be classified belongs in the fuzzy set of easily confused industry users using a preset hybrid discrimination model, and to determine the contribution vector of each industry feature of any power user to be classified to the initial prediction result. The scoring module is used to calculate the comprehensive discrimination score of each industry corresponding to any given user based on the contribution vector of each industry feature of any given power user to the initial prediction result using a preset weighted dot product model. The output module is used to determine the industry affiliation of any given user based on the comprehensive discrimination score.

[0018] According to another aspect of this application, an electronic device is also provided, the electronic device comprising: The system includes a memory and a processor, wherein the memory stores a computer program that is executed by the processor, which, when run by the processor, causes the processor to perform the electricity user classification method for easily confused industries as described above.

[0019] According to another aspect of this application, a storage medium is also provided, on which a computer program is stored, which, when run by a processor, causes the processor to execute the electricity user classification method for easily confused industries as described above.

[0020] The embodiments of this application have the following beneficial effects: The embodiments of this application can effectively solve the technical problem of chaotic classification of electricity consumption data caused by similar industry classifications and overlapping user production activities, significantly improve the purity and accuracy of industry electricity consumption statistics, and provide reliable data support for macroeconomic analysis and industry supervision based on electricity data. Attached Figure Description

[0021] Figure 1 A schematic flowchart illustrating a method for classifying electricity users in easily confused industries according to an embodiment of this application; Figure 2 A schematic flowchart illustrating step S101 according to an embodiment of this application is shown; Figure 3 A schematic diagram of an industry user feature profiling index system 300 according to an embodiment of this application is shown; Figure 4 A schematic flowchart illustrating step S102 according to an embodiment of this application is shown; Figure 5 A schematic flowchart illustrating step S401 according to an embodiment of this application is shown; Figure 6 A schematic flowchart illustrating step S402 according to an embodiment of this application is shown; Figure 7 A schematic flowchart illustrating step S404 according to an embodiment of this application is shown; Figure 8 A schematic flowchart illustrating step S103 according to an embodiment of this application is shown; Figure 9 A schematic flowchart illustrating step S104 according to an embodiment of this application is shown; Figure 10 A schematic flowchart illustrating step S105 according to an embodiment of this application is shown; Figure 11A schematic flowchart illustrating step S106 according to an embodiment of this application is shown; Figure 12 A schematic flowchart illustrating step S106 according to another embodiment of this application is shown; Figure 13 A schematic flowchart illustrating step S1201 according to an embodiment of this application is shown; Figure 14 This document illustrates a technical roadmap for a method of classifying electricity users in easily confused industries according to embodiments of this application. Figure 15 A schematic block diagram of a power user classification device for easily confused industries according to an embodiment of this application is shown; Figure 16 A schematic block diagram of an electronic device according to an embodiment of this application is shown. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions of the embodiments of this application, the application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] This application provides a method for classifying electricity users in easily confused industries. The method includes: acquiring multi-dimensional industry feature data of electricity users to be classified; generating a fused feature vector for each electricity user to be classified based on the multi-dimensional industry feature data using an attention mechanism; clustering all electricity users to be classified based on the fused feature vector using a preset clustering algorithm to determine a fuzzy set of easily confused industry users; performing an initial prediction of the industry to which any electricity user to be classified belongs in the fuzzy set of easily confused industry users using a preset hybrid discriminant model, and determining the contribution vector of each industry feature of the electricity user to the initial prediction result; calculating a comprehensive discriminant score for each industry to which the electricity user belongs using a preset weighted dot product model based on the contribution vector of each industry feature of the electricity user to the initial prediction result; and determining the industry to which the electricity user belongs based on the comprehensive discriminant score. This application effectively solves the technical problem of chaotic electricity consumption data classification caused by similar industry classifications and overlapping user production activities, significantly improving the purity and accuracy of industry electricity consumption statistics, and providing reliable data support for macroeconomic analysis and industry supervision based on electricity data.

[0024] Figure 1 This diagram illustrates a schematic flowchart of a method for classifying electricity users in easily confused industries according to an embodiment of this application; as shown. Figure 1 As shown, the electricity user classification method 100 according to the embodiments of this application for easily confused industries may include the following steps S101, S102, S103, S104, S105 and S106: In step S101, multi-dimensional industry characteristic data of the electricity users to be classified are obtained.

[0025] This application addresses the issue of overlapping electricity consumption characteristics in industries that are easily confused, such as electrical machinery manufacturing and electronic equipment manufacturing. This application can obtain multi-dimensional dynamic characteristic indicators from multi-dimensional industry characteristic data from the electricity consumption side, user side, and verification side. This indicator system forms a highly distinctive industry characteristic profile by deeply mining the inherent characteristics of the load and associating it with production and operation data.

[0026] In one embodiment of this application, such as Figure 2 As shown, step S101, obtaining multidimensional industry characteristic data of the electricity users to be classified, includes step S201: In step S201, multi-dimensional industry feature data of the electricity users to be classified are obtained from the electricity consumption side, the user side, and the verification side, respectively, and an industry user feature profile of each electricity user to be classified is constructed based on the multi-dimensional industry feature data.

[0027] For example, on the electricity consumption side, at least one of the following industry characteristic data can be obtained through the power data acquisition system: the user's daily load curve, harmonic current or voltage RMS value, real-time power fluctuation and other physical and electrical quantity data; For example, on the user side, at least one of the following industry-specific data can be obtained through the enterprise's energy management platform or information system: production shift records, seasonal production plans, unit product output and corresponding power consumption, etc., which reflect the internal operating logic. For example, on the verification side, through the on-site verification process, at least one of the following industry characteristic data is recorded: equipment process type, actual production process, on-site inspection photos, and other data used for verification and feedback.

[0028] This application's embodiments support multi-source data acquisition. Specifically, on the electricity consumption side, grid data is acquired through a power data acquisition system, obtaining physical and electrical quantity data such as the user's daily load curve, harmonic current / voltage RMS values, and real-time power fluctuations. On the user side, enterprise operation data is acquired through an enterprise energy management platform or information system, obtaining data reflecting internal operational logic such as production shift records, seasonal production plans, unit product output, and corresponding power consumption. On the verification side, on-site verification data is acquired through an on-site verification process, recording and structuring data such as equipment process types, actual production processes, and on-site inspection photos for verification and feedback.

[0029] The multidimensional industry characteristic data includes load curve depth characteristics, production time sequence characteristics, and energy consumption intensity characteristics.

[0030] Furthermore, the load curve depth characteristics include at least one of the following indicators: production continuity index, load fluctuation frequency domain energy ratio, and non-power frequency load ratio; The production time-series characteristics include at least one of the following indicators: intelligent shift system recognition, seasonal production feature vector; The energy intensity characteristics include the unit product power consumption index.

[0031] In this embodiment, a highly discriminative industry user profile for each electricity user to be classified is constructed using six key indicators, including the production continuity index, the frequency domain energy ratio of load fluctuation, the proportion of non-power frequency load, intelligent shift system recognition, seasonal production feature vector, and unit product power consumption index, providing core input for fuzzy user screening.

[0032] Among them, the load curve depth feature extraction based on behavioral pattern analysis includes the production continuity index, the frequency domain energy ratio of load fluctuation, and the proportion of non-power frequency load; the production time sequence features based on integrated business logic extract at least one of the following indicators from the multidimensional data samples: intelligent shift system recognition and seasonal production feature vector; the energy consumption intensity features based on the essence of the process extract the unit product power consumption index from the multidimensional data samples. These indicators are described one by one below.

[0033] In the first example, the production continuity index This production continuity index is used to quantify the uninterrupted nature of user production activities, distinguishing between process industries and discrete manufacturing industries. The calculation formula is as follows: ; in, Indicates the daily minimum load; This indicates the average daily load.

[0034] In practice, the production continuity index of continuous process industries such as chemicals and electrolytic aluminum is very close to 1, while the production continuity index of industries with night shift shutdowns, such as food processing and some machinery manufacturing industries, is significantly lower than 1.

[0035] In the second example, the load fluctuation frequency domain energy ratio It can characterize industry load fluctuations and identify the regularity of process cycles. A Fast Fourier Transform is performed on the daily load curve sequence to calculate the ratio of high-frequency energy (cycle < 2 hours) to total energy, and the frequency domain energy ratio of load fluctuations. The calculation formula is as follows: ; in, It represents the ratio of the high-frequency energy of the load to the total energy (dimensionless, ranging from 0 to 1). This represents a frequency domain variable, indicating the frequency components of the load curve in the frequency domain. The frequency domain representation of the load curve sequence after Fast Fourier Transform (FFT) is a complex array; This represents a set of frequencies in the high-frequency band, defined as frequency components with a period of less than 2 hours. Indicates frequency The energy density at a given frequency represents the contribution of that frequency component to the total energy.

[0036] In the third example, the proportion of non-power frequency load This refers to calculating the ratio of the effective value of harmonic current to the effective value of total current using current and voltage waveform data acquired through advanced measurement systems. In industries such as electrolytic aluminum, chlor-alkali chemicals, and inverter-driven fans and pumps, the proportion of non-power frequency load is significant. The value is significantly higher; while in industries such as glass kilns and electric arc furnaces using resistance heating, the proportion of non-power frequency load is significantly higher. The value is close to 0. This indicator can identify the differences in characteristics among high-energy-consuming industrial sub-sectors. (Non-power frequency load percentage) The calculation formula is as follows: ; in, This represents the effective value of the harmonic current; This represents the effective value of the total current.

[0037] In the fourth example, intelligent class system pattern recognition This refers to using a pre-trained lightweight decision tree model, taking a user's weekly workload curve as input, and training the model to output shift mode labels. ,in The results include strict three-shift system, two-shift system, day shift + overtime type, and pure day shift system, in order to distinguish the differences in production and business models of different industries.

[0038] In the fifth example, regarding the seasonal production feature vector First, define the monthly electricity consumption unevenness coefficient. The coefficient The calculation formula is as follows: ; in, , These represent the standard deviation and expected value of electricity consumption for the 12 months of the year, respectively.

[0039] Secondly, extract the season with the highest electricity consumption. ( Defined as spring, summer, autumn, spring and autumn, etc. This forms a composite feature vector, used to distinguish industries driven by different seasonal factors. For example, refrigeration equipment manufacturers exhibit the following characteristics: High, and It is summer.

[0040] In the sixth example, combined with the company's energy audit report, the unit product electricity consumption... The calculation formula is as follows: ; in, This indicates the useful power output; This represents the total input electrical power.

[0041] Assume the relevant industry standard value is Then, energy consumption deviation rate can be constructed. To quantify industry differences in energy consumption efficiency.

[0042] ; in, Indicates the unit product power consumption; This indicates the relevant industry standard value.

[0043] In this embodiment, based on the collected multi-dimensional industry characteristic data, six core characteristic indicators for each user can be calculated in parallel, namely the production continuity index. Load fluctuation frequency domain energy ratio Non-power frequency load ratio Intelligent class system mode recognition Seasonal production feature vector Unit product power consumption And based on these indicators, we will construct an industry user characteristic profile indicator system.

[0044] To address the issue of overlapping electricity consumption characteristics in industries that are easily confused, such as electrical machinery manufacturing and electronic equipment manufacturing, the industry user characteristic profiling index system in this application integrates multi-dimensional dynamic characteristic indicators of behavioral patterns, business logic, and process essence. By deeply mining the inherent characteristics of the load and associating them with production and operation data, a highly distinguishable industry characteristic profile is formed.

[0045] like Figure 3The diagram shows a schematic of the industry user characteristic profiling index system 300 according to an embodiment of this application. As can be seen from the diagram, three indicators can be obtained from behavioral patterns: production continuity index, load fluctuation frequency domain energy ratio, and non-power frequency load proportion. The production continuity index quantifies the uninterruptibility of user production activities; the load fluctuation frequency domain energy ratio characterizes industry load volatility and identifies the regularity of process cycles; and the non-power frequency load proportion can be used to identify the characteristic differences of high-energy-consuming industrial sub-sectors. This application can deeply explore the inherent behavioral patterns of user electricity consumption from the dimensions of electricity consumption behavior characteristics, namely the stability and periodicity of the load curve and the electrical physical characteristics.

[0046] From the business logic dimension (production organization mode dimension), two indicators can be obtained: a shift system label and a seasonal production feature vector. The shift system label can include three-shift, two-shift, day shift + overtime, and pure day shift systems, distinguishing differences in production business models across different industries; the seasonal production feature vector is used to differentiate the seasonal characteristics of industry production. This application's embodiments indirectly reflect the company's operational strategies and business logic by analyzing the temporal patterns and seasonal preferences of production activities.

[0047] The fundamental nature of the process provides an indicator such as the historical control and fulfillment rate, which can quantify industry differences in energy efficiency. This application's embodiments, through deviation analysis from industry benchmarks, directly reveal the energy efficiency and technological equipment level of user production processes, touching upon the fundamental technological aspects of industry classification.

[0048] The industry user feature profile indicator system of this application provides a comprehensive and in-depth data representation foundation for subsequent accurate clustering and discrimination through the synergy of the above-mentioned multi-dimensional indicators; moreover, each indicator is weighted and integrated through standardization and attention mechanism, and finally converges to the industry feature profile score, so as to achieve a multi-dimensional and accurate characterization of the user's industry attributes.

[0049] In step S102, an attention mechanism is used to generate a fused feature vector for each electricity user to be classified based on the multidimensional industry feature data.

[0050] To address the semantic fragmentation problem of multimodal data such as electricity load, production and operation, and energy intensity, this application embodiment utilizes an attention mechanism to generate a fused feature vector for each electricity user to be classified, so as to transform the above six indicators into inputs suitable for machine learning models.

[0051] In another embodiment of this application, such as Figure 4 As shown, step S102 utilizes an attention mechanism to generate a fused feature vector for each electricity user to be classified based on the multi-dimensional industry feature data, including steps S401, S402, S403, and S404: In step S401, the multidimensional industry feature data is preprocessed; In step S402, the multidimensional industry feature data is standardized and encoded. In step S403, the standardized and encoded multidimensional industry feature data are concatenated in a preset order to generate the original feature vector; In step S404, the original feature vectors are weighted and summed using an attention mechanism to obtain a fused feature vector.

[0052] In one example, such as Figure 5 As shown, step S401, which involves preprocessing the multidimensional industry feature data, includes: In step S501, the multidimensional industry feature data is denoised using the wavelet transform method. In step S502, outliers in the multidimensional industry feature data are deleted; In step S503, the missing values ​​in the multidimensional industry feature data are filled in using a linear interpolation method.

[0053] The multidimensional industry feature data in this application is time-series data. Wavelet transform can be used to denoise this time-series data to eliminate abnormal fluctuations caused by equipment failure and communication interference. Linear interpolation can be used to complete missing data, and then data with a missing rate exceeding 5% can be removed. Finally, Z-score standardization is applied to all numerical data in the multidimensional industry feature data to eliminate differences in numerical units.

[0054] In one example, such as Figure 6 As shown, step S402, which involves standardizing and encoding the multidimensional industry feature data, includes steps S601 and S602: In step S601, the multidimensional industry feature data is standardized using the Z-score standardization method. In step S602, the categorical data in the multidimensional industry feature data is converted into a binary numerical vector using a one-hot encoding method.

[0055] Specifically, the processing methods for numerical industry characteristics (indicators) include: processing continuous indicators such as the production continuity index. Load fluctuation frequency domain energy ratio Non-power frequency load ratio Uneven power consumption coefficient and unit product power consumption The scalar real value is output directly according to its calculation formula; among which the production continuity index Non-power frequency load ratio Uneven power consumption coefficient A dimensionless indicator, representing the unit product's power consumption. It is a dimensional indicator.

[0056] The processing methods for numerical industry characteristics (indicators) include: intelligent shift system pattern recognition. (Values: "strict three-shift system", "two-shift system", etc.) Seasonal characteristics (Values: Spring / Summer / Autumn / Winter), need to be converted into numerical values.

[0057] This application's embodiments can unify the format of different types of indicators, eliminating differences in units and types. For dimensional numerical indicators, Z-score standardization is used, and for categorical indicators, one-hot encoding is used to convert them into binary numerical vectors.

[0058] Furthermore, in the embodiments of this application, when performing spatiotemporal alignment and structured integration of multidimensional industry feature data, it is first necessary to align the time-series data. For example, the electricity consumption time-series data is aligned to a unified time axis according to the timestamp. Since each user has a unique identifier (ID), the user ID can be used as an association to perform spatiotemporal alignment and structured integration of the preprocessed multidimensional industry feature data.

[0059] Secondly, attribute data is associated, that is, the attribute data on the user side and the verification side are linked with the corresponding user IDs. Then, a unified data matrix is ​​constructed, transforming all industry feature data of each user into a fixed-dimensional feature vector through feature expansion, category coding, and statistical calculation. The feature vectors of all users together form an N*M two-dimensional data matrix (N is the number of users, and M is the total dimension of the features). This unified data matrix, with each row corresponding to the multi-dimensional data set of a single user, lays the foundation for the subsequent generation of the original feature vector.

[0060] In one example, such as Figure 7 As shown, step S404 uses an attention mechanism to perform a weighted summation of the original feature vectors to obtain a fused feature vector, including steps S701 and S702: In step S701, the attention weight of each feature in the original feature vector is calculated according to the preset attention parameters; In step S702, the original feature vector is weighted and summed according to the attention weights to obtain the fused feature vector.

[0061] After standardizing and encoding the indicators, all the standardized indicators can be concatenated into a one-dimensional vector in a preset order to obtain the original feature vector. Among them, the original feature vector It is a numerical feature vector. It is a one-dimensional array formed by concatenating the six core indicators proposed in the embodiments of this application in a preset order after numerical calculation and normalized encoding. It represents all the original data inputs used to describe the industry characteristics of a user.

[0062] For example, =[0.98,0.15,0.50,1.00,0.35,0.0].

[0063] Original feature vector It can be based on the production continuity index Load fluctuation frequency domain energy ratio Non-power frequency load ratio Intelligent class system mode recognition Seasonal production feature vector Unit product power consumption The sequence can be either one or different, but the goal is to concatenate the original feature vectors. The fused feature vectors should be aligned in order with the subsequent fusion feature vectors.

[0064] In obtaining the original feature vector Then, an attention mechanism can be used to input the feature attention weights into the original feature vector. The attention layer calculates the attention score for each feature using the following formula: ; ; in, , , , Represents the learnable parameters of the attention layer; Represents the attention score vector; This represents the attention weight of the i-th feature obtained after Softmax normalization, indicating the importance of this feature for the current industry classification task.

[0065] Furthermore, the calculated attention weights can be used. The original features are weighted and summed to obtain the fused feature vector. For example, the original feature vector... By inputting an attention-based fusion module, a fused feature vector can be obtained. This fused feature vector is a weighted combination of the original features, highlighting key features. The resulting weighted fused feature vector... This can be used as input for subsequent clustering.

[0066] ; in, Represents the original feature vector The i-th scalar component; Represents the original feature vector The total dimension.

[0067] This application embodiment integrates the features of the above 3 major categories and 6 core indicators to construct a feature profile indicator system, and can deeply characterize the user's production behavior, business strategy and process essence, providing data support for subsequent accurate clustering and discrimination.

[0068] In step S103, a preset clustering algorithm is used to cluster all power users to be classified based on the fused feature vector to determine the fuzzy set of easily confused industry users.

[0069] In this embodiment, an improved DBSCAN clustering algorithm can be used to accurately locate easily confused user sets with ambiguous features, solving the problem of low discrimination of similar features in traditional clustering methods. Traditional DBSCAN clustering algorithms rely on a fixed neighborhood radius and a minimum number of samples. This embodiment optimizes the clustering process through adaptive parameter calculation and a core point expansion strategy, clustering users with similar features into the same cluster and users with ambiguous features into smaller clusters, thereby identifying the fuzzy set of easily confused industry users.

[0070] In one embodiment of this application, such as Figure 8 As shown, step S103 utilizes a preset clustering algorithm to cluster all electricity users to be classified based on the fused feature vector to determine the fuzzy set of easily confused industry users, including steps S801, S802, S803, and S804: In step S801, the adaptive optimal neighborhood radius parameter is determined using the K-distance graph method.

[0071] In this embodiment, the K-distance graph method is used to fuse the feature vectors of all users. Analysis was conducted to determine the adaptive optimal neighborhood radius parameter of the DBSCAN clustering algorithm. .

[0072] In one example, the K-distance graph method calculates the adaptive optimal neighborhood radius parameter as follows: The first step is to process the fused feature vector set containing N users. Set the number of neighboring points K to 5% of the total number of users (for example, referring to the example below, for a sample of 1203 users, k=60).

[0073] It is worth noting that the value of K can balance the accuracy and computational efficiency of local density analysis, so as to avoid noise interference in local density judgment when k is too small, and to avoid excessive smoothing of local features when k is too large.

[0074] The second step is to fuse the feature vector for each user i. Calculate the fused feature vectors of this vector and other users besides this user. The Euclidean distance is denoted as .

[0075] The third step is to arrange the Euclidean distances of all N points in ascending order and plot the curves to form a K-distance graph.

[0076] The fourth step is to identify the "inflection point" on the K-distance graph curve, that is, the turning point where the distance value change trend changes from steep to gentle. The distance value on the ordinate corresponding to this inflection point is automatically selected as the adaptive optimal neighborhood radius parameter. .

[0077] This application uses the K-distance graph method to automatically calculate the global neighborhood radius parameter eps, calculates the distance from each data point to its k-th nearest neighbor, generates a sorted k-distance curve, and automatically determines the optimal eps value by identifying the inflection point of the curve, so as to avoid the blindness of manual parameter setting.

[0078] In step S802, the initial clustering of all the electricity users to be classified is performed using the compressed adaptive optimal neighborhood radius parameter to form a high-confidence core cluster, and the set of users outside the high-confidence core cluster is taken as the undetermined classification set.

[0079] This application embodiment can use the compaction parameter eps1 for initial clustering to identify typical industry users with distinct and unambiguous characteristics, forming a high-confidence core cluster. In this step, user points not assigned to any cluster are marked as a pending classification set. .

[0080] The expression for the compression parameter is as follows: ; Among them, Indicates the contraction factor. 1.

[0081] For example, It is 0.8.

[0082] In step S803, a second clustering is performed on the undetermined classification set using a relaxed adaptive optimal neighborhood radius parameter to determine the weak association substructure in the undetermined classification set. Continuing from the previous example, in the undetermined classification set Then, a second clustering process is performed using the relaxed parameter eps2=eps to identify potentially weakly associated substructures among fuzzy users. Here, a weakly associated substructure refers to a substructure whose multidimensional features cannot form a reliable association with any other user in the dataset.

[0083] Among them, fuzzy user set as follows: ; in, This indicates the small clusters generated by the second aggregation; Indicates the number of users within the cluster; Indicates the threshold for small cluster size; This indicates isolated noise points that were not clustered in the second clustering.

[0084] In step S804, the weakly correlated substructures formed by the secondary clustering and the set of users consisting of noise points that are still not classified after the secondary clustering are taken as the fuzzy set of easily confused industry users.

[0085] In this embodiment, the weakly correlated substructure formed by secondary clustering defines all points that remain noise after secondary clustering as easily confused user fuzzy sets. The "noise points" here specifically refer to highly specialized users who remain isolated even under relaxed density conditions, and whose multidimensional features cannot be reliably correlated with any other user in the dataset.

[0086] This application embodiment automatically and accurately locates user sets with ambiguous industry characteristics and classification ambiguities from all users, so as to facilitate accurate classification of these users.

[0087] In step S104, a preset hybrid discrimination model is used to make an initial prediction of the industry to which any power user to be classified belongs in the fuzzy set of easily confused industry users, and the contribution vector of each industry feature of any power user to be classified to the initial prediction result is determined.

[0088] This application embodiment can filter out fuzzy sets of easily confused industry users. This method can quantify the specific contribution of each feature to the industry affiliation prediction, providing an objective basis for subsequent calculation of the comprehensive judgment score. This application utilizes the Shapley Additive exPlanations (SHAP) to determine the contribution vector. SHAP is a well-known, game-theoretic, interpretable algorithm whose core is to provide a consistent and reasonable allocation of feature contributions for the prediction results of complex machine learning models. Public and stable algorithm libraries are available for implementation.

[0089] In one embodiment of this application, such as Figure 9 As shown, step S104 uses a preset hybrid discriminant model to make an initial prediction of the industry to which any power user to be classified belongs in the fuzzy set of easily confused industry users, and determines the contribution vector of each industry feature of any power user to the initial prediction result, including steps S901, S902, and S903: In step S901, using the fused feature vectors of users in labeled industries and their corresponding features, a baseline industry classifier is trained using the gradient boosting tree method to predict the industry probability of the fuzzy set of users in the easily confused industries.

[0090] In step S902, the industry affiliation prediction probability of each user in the benchmark industry classifier is taken as the target to be explained, and the fused feature vector corresponding to each user is taken as the feature set input to the interpreter.

[0091] In step S903, the defined interpretation object and interpretation features are input into the Shapley interpreter to obtain... The contribution vector is predicted for each industry category.

[0092] The embodiments of this application use the Shapley Additive exPlanations (SHAP) method to quantify feature contributions and use a preset classifier to make an initial prediction of the industry affiliation of each user in the fuzzy set of easily confused industry users.

[0093] In one example, the contribution vector for each feature can be calculated using the following steps: The first step is to prepare the baseline prediction model. This can be achieved by utilizing users from labeled industries and their multidimensional feature vectors. A benchmark industry classifier is trained using Gradient Boosting Tree (XGBoost) to classify fuzzy user sets. This involves performing industry probability prediction. The preset classifier in this application embodiment can be a classifier such as the XGBoost classifier. For example, a pre-trained XGBoost classifier can be used to... For each user i, an initial industry affiliation is predicted. Then, SHAP analysis is applied to calculate the contribution vector of each feature j to the prediction result for each possible industry k. Obtain the SHAP contribution vector. .

[0094] It is worth noting that when acquiring multidimensional industry characteristic data of electricity users to be classified, the user's business registration industry label can be obtained at the same time, and the user's industry can be labeled as an initial reference information for the user's industry classification.

[0095] The second step is to calculate the contribution vector input definition. This involves applying the above benchmark industry classifier... Predicted industry affiliation probability for each user i As the target that needs to be explained, the fused feature vector corresponding to each user... The feature set is used as input to the interpreter.

[0096] Third step, contribution vector Generation. The defined interpretation object and interpretation features are input into the SHAP interpreter. The SHAP framework will then use its internal algorithms to calculate... The contribution vector is predicted for each industry category k. The contribution vector value quantifies the extent to which feature j drives the model's decision to classify user i as industry k.

[0097] Step 4: Output the contribution vector. For each user i, the contribution vector of their industry k is: ; in, express The SHAP contribution vector of user i to industry k.

[0098] The score vector of the SHAP value in the embodiments of this application It can clearly show the key features and the extent to which they drive the model to classify it as industry k.

[0099] In step S105, based on the contribution vector of each industry feature of any power user to the initial prediction result, a preset weighted dot product model is used to calculate the comprehensive discrimination score of each industry corresponding to any user.

[0100] In one embodiment of this application, such as Figure 10 As shown, step S105, based on the contribution vector of each industry characteristic of any given electricity user to the initial prediction result, calculates the comprehensive discrimination score for each industry corresponding to any given user using a preset weighted dot product model, including steps S1001, S1002, S1003, and S1004: In step S1001, calculate the global average Shapley absolute value of each feature corresponding to each user in the easily confused industry user fuzzy set; In step S1002, a judgment matrix is ​​constructed using the 1-9 scaling method based on the comparison values ​​between the various features of each user. In step S1003, the feature weight vector of each feature is calculated based on the judgment matrix; In step S1004, the comprehensive discrimination score of each feature is calculated based on the global average Shapley absolute value and the feature weight vector.

[0101] In one example, the comprehensive discrimination score can be calculated using the following steps: The first step is to calculate the global mean absolute SHAP value. This involves calculating the global mean absolute SHAP value for each feature across the entire fuzzy user set. This serves as the objective basis for the initial importance of this feature. Among them, the absolute SHAP value... To average the absolute values ​​of all users i across all industries k for each feature j, the formula is as follows: ; in, Indicates the number of fuzzy user sets; This indicates the number of easily confused industries to be identified.

[0102] in, The larger the value, the more crucial the role of feature j in the model's overall ability to distinguish these easily confused industries.

[0103] The second step is to determine the initial importance. This involves applying the objective values ​​calculated above... This is transformed into initial weight suggestions that can be used by experts for reference.

[0104] The third step is to construct the pairwise comparison judgment matrix. Experts were invited to provide feedback. The features are compared pairwise, and scores are generated using the 1-9 scale. * The judgment matrix A.

[0105] The fourth step is to calculate the final weight vector. For the constructed judgment matrix A, solve for the following: eigenvectors ,in The largest eigenvalue of matrix A is the final weight vector. ,satisfy .

[0106] Step 5: Calculate the comprehensive discrimination score. For user i and industry k to be judged, calculate their comprehensive discrimination score. It is calculated by the weighted dot product of the SHAP contribution vector and the AHP weight vector, and the calculation formula is as follows: ; Where i represents the user set from the fuzzy user set. The i-th easily confused user that needs to be identified; k represents the k-th industry that user i may belong to; j represents the feature symbol, representing the industry-identifying feature of the j-th industry; Indicates the number of features involved in the discrimination; Let represent the SHAP feature contribution vector, and let represent the contribution share of feature j. The weights of AHP features can be determined by combining the analytic hierarchy process with expert experience.

[0107] To illustrate the feasibility and output results of the embodiments of this application, we can take the easily confused electrical machinery manufacturing industry and electronic equipment manufacturing industry as examples to classify users in these two industries. An exemplary weight vector calculated using the above weight vector formula is shown below. The weights in this vector correspond to the six indicators mentioned above.

[0108] In step S106, the industry to which any one user belongs is determined based on the comprehensive discrimination score.

[0109] In one embodiment of this application, such as Figure 11 As shown, step S106, which calculates the comprehensive discrimination score for each industry corresponding to any given electricity user based on the contribution vector of each industry characteristic to the initial prediction result, using a preset weighted dot product model, includes step S1101: In step S1001, the industry with the highest comprehensive discrimination score is determined as the industry to which any given user belongs.

[0110] As a simplified approach, the industry with the highest comprehensive discrimination score can be directly assigned to any user. This application's embodiments can solve the classification ambiguity problem caused by single features in traditional technologies, providing a highly discriminative feature foundation for identifying easily confused users.

[0111] In another embodiment of this application, such as Figure 12 As shown, step S106, which calculates the comprehensive discrimination score for each industry corresponding to any given electricity user using a preset weighted dot product model based on the contribution vector of each industry feature to the initial prediction result, includes step S1201: In step S1201, based on the hierarchical decision-making mechanism of dual thresholds, the user is reclassified and the industry to which the user belongs is re-determined.

[0112] This invention further designs a hierarchical decision-making mechanism based on dual thresholds to achieve efficient user reclassification.

[0113] In one example, such as Figure 13As shown, the hierarchical decision-making mechanism based on dual thresholds in step S1201 reclassifies any given user and redetermines the industry to which any given user belongs, including steps S1301, S1302, and S1303: In step S1301, calculate the optimal industry category and corresponding first comprehensive discrimination score of any user in the fuzzy set of easily confused industry users, as well as the second optimal industry and corresponding second comprehensive discrimination score of any user; In step S1302, the score difference between the first comprehensive discrimination score and the second comprehensive discrimination score is calculated; In step S1303, the first comprehensive discrimination score is compared with a preset reliability threshold, and the score difference is compared with a preset advantage difference threshold. Based on the comparison result, the industry to which any user belongs is determined.

[0114] The preset confidence thresholds include a maximum confidence threshold and a minimum confidence threshold.

[0115] In one example, comparing the first comprehensive discrimination score with a preset reliability threshold and comparing the score difference with a preset advantage difference threshold may yield the following three results: If the first comprehensive discrimination score is greater than or equal to the maximum confidence threshold and the score difference is greater than or equal to the preset dominance difference threshold, then any user is determined to belong to the industry with the highest first comprehensive discrimination score. If the first comprehensive discrimination score is less than the minimum confidence threshold, or the score difference is less than the preset advantage difference threshold, the industry to which any user belongs will be manually verified. If the first comprehensive discrimination score is less than the maximum confidence threshold and greater than or equal to the minimum confidence threshold, and the score difference is greater than or equal to the preset advantage difference threshold, the industry to which any user belongs is marked as pending.

[0116] This application employs a dual-threshold hierarchical decision-making mechanism to achieve efficient user reclassification. For each easily confused user i and its most likely industry category... The following steps can be taken: First, calculate the user's first comprehensive discriminant score. And the score difference between the first comprehensive discrimination score and the second comprehensive discrimination score: ; Secondly, set a preset maximum confidence threshold. Compared with the preset minimum confidence threshold ;Compare , Based on the confidence threshold, the classification decision is made as follows: (1) If ≥ If the confidence level is high, then user i will be classified into the industry. .

[0117] (2) If ≤ If the model is deemed unreliable in determining the industry classification, on-site manual verification will be required. (3) If ≤ ≤ If so, the user will be placed in "pending" status, pending the accumulation of more data or reassessment in subsequent analysis cycles based on new information.

[0118] The hierarchical decision-making mechanism and on-site verification process in this application embodiment together constitute a perception-decision-feedback closed-loop process. The results of on-site verification are fed back to the training dataset as gold standard samples to drive subsequent incremental model learning and threshold adaptive optimization, thereby continuously improving the discrimination accuracy and reliability of the entire process.

[0119] In one example, the confidence threshold can be determined statistically by analyzing the score distribution of correctly classified samples in historical data. For instance, a preset maximum confidence threshold can be set. The preset minimum confidence threshold is 0.85. The minimum advantage gap threshold is 0.60. It is 0.10.

[0120] This application's embodiment of a power user classification method for industries prone to confusion constructs an industry feature profile library that integrates behavioral patterns, business logic, and technological essence. It utilizes an improved adaptive clustering algorithm to intelligently filter user sets with ambiguous features and establishes a weighted discrimination model that integrates SHAP interpretability analysis and the Analytic Hierarchy Process (AHP). Finally, a dual-threshold decision mechanism achieves accurate industry classification and triggers closed-loop on-site verification. This application's embodiment effectively solves the problem of chaotic electricity data classification caused by similar industry classifications and overlapping user production activities, significantly improving the purity and accuracy of industry electricity statistics, and providing reliable data support for macroeconomic analysis and industry supervision based on power data.

[0121] like Figure 14The diagram illustrates the technical roadmap for the electricity user classification method in easily confused industries, as described in this application. First, multi-source data (multi-dimensional industry characteristic data) is collected and fused (preprocessed). This multi-source data includes electricity load data and IoT data from production and operation sites. Indicators from these data are extracted to construct an industry user characteristic profile indicator system. For example, indicators can be extracted from aspects such as electricity consumption behavior modeling, business operation logic modeling, and production process characteristic modeling. These indicators are then weighted and fused to generate a weighted fused feature vector.

[0122] Then, based on the weighted fusion feature vector, fuzzy user set filtering is performed using the improved DBSCAN clustering method. Specifically, operations such as adaptive neighborhood radius calculation, two-stage clustering with core point expansion, clustering of users with similar features, and identification of easily confused fuzzy user sets can be performed.

[0123] Finally, users in the fuzzy cluster of easily confused users are distinguished based on their weights. Specifically, contribution quantification can be performed based on SHAP features, then fusion can be achieved using AHP expert weights, and a comprehensive discrimination score can be calculated for each user.

[0124] The final comprehensive discrimination score is calculated and compared with the preset maximum confidence threshold and the preset minimum confidence threshold. If the confidence score is higher than the preset maximum confidence threshold, the user is automatically classified; if the confidence score is lower than the preset minimum confidence threshold, the user needs to be verified on-site; if the confidence score is between the preset maximum confidence threshold and the preset minimum confidence threshold, the user is placed under observation.

[0125] This application also supports closed-loop learning optimization. For example, incremental learning and weight learning are performed. Through the closed-loop learning optimization mechanism, the system can continuously improve the accuracy and robustness of classification. In practical applications, the results of each on-site inspection are fed back into the model as new training samples to update the preset classifier and weight vector. For example, based on newly added high-quality labeled data, the XGBoost classifier can be retrained, while the AHP weight vector is dynamically adjusted in conjunction with the latest expert experience, thereby ensuring that the discrimination system always remains consistent with industry characteristics. Furthermore, through retrospective analysis of historical discrimination results, key parameters in the DBSCAN clustering algorithm, such as the neighborhood radius and the number of core points, can be further optimized to adapt to changing data distribution trends. This self-iterative learning mechanism not only improves the system's intelligence level but also provides a long-term reliable solution for processing complex and easily confused industry electricity consumption data.

[0126] The embodiments of this application have the following beneficial effects: First, a three-dimensional dynamic feature profile system was established, which integrates behavioral patterns, business logic, and process essence. Physical mechanism indicators such as production continuity index, load fluctuation frequency-energy ratio, and non-power frequency load ratio were introduced, as well as business logic indicators such as unit product power consumption and intelligent shift mode. This system accurately portrays industry differences from multiple dimensions such as electricity consumption behavior, production strategy, and process essence, solving the problem of classification ambiguity caused by traditional single features and providing a highly distinguishable feature foundation for identifying easily confused users.

[0127] Second, a fuzzy user intelligent screening method based on an improved DBSCAN is proposed. By adaptively calculating the neighborhood radius using the K-distance graph method and employing a two-stage clustering strategy, this method overcomes the limitations of traditional clustering, such as low discriminative power for feature similarity and sensitivity to noise. It achieves accurate positioning of users with fuzzy features and provides a clear target set for subsequent discrimination.

[0128] Third, a SHAP+AHP weighted hybrid discrimination model was designed. By integrating data-driven SHAP feature contribution quantification, the driving role of each feature in classification is clarified. Then, based on expert knowledge, the AHP weights are adjusted to balance objective data and subjective experience. The comprehensive discrimination score is calculated by weighted dot product of feature contribution vector and weight vector, which solves the problems of poor interpretability of pure data model and strong subjectivity of pure expert model, and significantly improves the credibility of industry classification.

[0129] Fourth, a dynamic optimization mechanism of dual threshold decision-making and closed-loop feedback is constructed. By making decisions in a hierarchical manner through high / low confidence thresholds and driving incremental learning of the model in combination with on-site verification results, a continuous optimization closed loop of "perception-decision-feedback" is formed. This enables the classification model to dynamically adapt to changes in user production activities and updates to industry classification standards, ensuring the accuracy and reliability of long-term operation.

[0130] This application embodiment also selected 1203 user data points from four easily confused industries (electrical machinery, electronic equipment, metal products, and general equipment) to verify the verification results of this method, as shown in Table 1: Table 1. Simulation results for accurate verification of easily confused industry users

[0131] The calculation methods for the parameters involved in the table above are as follows: ① Automatic classification accuracy = (Number of users correctly classified by automatic classification / Total number of users automatically classified) × 100%; ②Percentage of manual verification = Number of users verified on-site / Total number of easily confused user groups × 100%; ③ Data purity = (Number of users with correct industry classification / Total number of users) × 100%; Purity improvement = Data purity after verification - Data purity before verification.

[0132] The results show that the electricity user classification method for easily confused industries proposed in this application can effectively solve the problem of chaotic electricity data classification caused by similar industry classifications and overlapping user production activities in traditional methods. Compared with traditional technologies, the embodiments of this application significantly improve the accuracy of automatic classification and the overall purity of industry electricity statistics, proving the scientific nature and effectiveness of this method in solving the problem of industry classification confusion.

[0133] This application addresses three core problems in traditional industry electricity consumption classification: insufficient feature dimensions, inaccurate identification of ambiguous users, and lack of adaptive capabilities in the model. First, it constructs a differentiated feature index system to deeply mine the frequency domain features of load curves, production time-series patterns, and differences in energy consumption intensity. Then, it employs an improved DBSCAN algorithm, using adaptive parameter calculation and a two-stage clustering strategy to accurately filter easily confused user sets with ambiguous features. Finally, it constructs a SHAP+AHP hybrid discrimination model, integrating data-driven feature contribution analysis with expert knowledge-adjusted weight allocation to achieve accurate industry classification, providing reliable data support for refined power grid management and industry economic analysis.

[0134] The following is combined Figure 15 This application describes a power user classification device for easily confused industries, wherein... Figure 15 This diagram illustrates a schematic block diagram of a power user classification device for easily confused industries according to an embodiment of this application. The power user classification device 1500 for easily confused industries according to this embodiment includes an acquisition module 1501, a clustering module 1502, a classification module 1503, a score calculation module 1504, and an output module 1505. The acquisition module 1501 is used to acquire multi-dimensional industry characteristic data of the electricity users to be classified. The vector calculation module 1502 is used to generate a fused feature vector for each power user to be classified based on the multi-dimensional industry feature data by utilizing an attention mechanism. Clustering module 1503 is used to cluster all power users to be classified based on the fused feature vector using a preset clustering algorithm in order to determine the fuzzy set of easily confused industry users. The prediction module 1504 is used to make an initial prediction of the industry to which any power user to be classified belongs in the fuzzy set of easily confused industry users using a preset hybrid discrimination model, and to determine the contribution vector of each industry feature of any power user to be classified to the initial prediction result. The scoring module 1505 is used to calculate the comprehensive discrimination score of each industry to which any user belongs based on the contribution vector of each industry feature of any power user to the initial prediction result using a preset weighted dot product model. Output module 1506 is used to determine the industry affiliation of any user based on the comprehensive discrimination score.

[0135] The following is combined Figure 16 The electronic device described in this application, wherein, Figure 16 A schematic block diagram of an electronic device according to an embodiment of this application is shown.

[0136] like Figure 16 As shown, the electronic device 1600 includes: one or more memories 1601 and one or more processors 1602, wherein the memory 1601 stores a computer program that is executed by the processor 1602, and the computer program, when executed by the processor 1602, causes the processor 1602 to perform the electricity user classification method for easily confused industries described above.

[0137] Electronic device 1600 may be part or all of a computer device that can implement a method for classifying electricity users in industries prone to confusion through software, hardware, or a combination of software and hardware.

[0138] like Figure 16 As shown, the electronic device 1600 includes one or more memories 1601, one or more processors 1602, a display (not shown), and a communication interface, etc., which are interconnected via a bus system and / or other forms of connection mechanisms (not shown). It should be noted that... Figure 16 The components and structure of the electronic device 1600 shown are merely exemplary and not limiting. The electronic device 1600 may also have other components and structures as needed.

[0139] Memory 1601 is used to store various data and executable program instructions generated during the operation of related methods, such as storing various application programs or algorithms that implement various specific functions. It may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0140] The processor 1602 may be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other processing units with data processing and / or instruction execution capabilities, and may be other components in the electronic device 1600 to perform the desired functions.

[0141] In one example, the electronic device 1600 also includes an output device that can output various information (such as images or sounds) to the outside (e.g., a user), and may include one or more of a display device, a speaker, etc.

[0142] The communication interface can be any known communication protocol interface, such as a wired interface or a wireless interface. The communication interface may include one or more serial ports, USB interfaces, Ethernet ports, WiFi, wired networks, DVI interfaces, device integrated interconnect modules, or other suitable ports, interfaces, or connections.

[0143] Furthermore, according to embodiments of this application, a storage medium is also provided, on which program instructions are stored. When executed by a computer or processor, these program instructions are used to perform corresponding steps of the electricity user classification method for easily confused industries according to embodiments of this application. The storage medium may, for example, include a memory card of a smartphone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media.

[0144] Furthermore, according to embodiments of this application, a computer program product is also provided, which, when executed by a processor, implements the steps of the method described above.

[0145] The electricity user classification device, electronic device, computer program product, and storage medium of the present application embodiments are capable of implementing the aforementioned electricity user classification method for easily confused industries, and therefore have the same advantages as the aforementioned electricity user classification method for easily confused industries.

[0146] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of this application. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of this application. All such changes and modifications are intended to be included within the scope of this application as claimed in the appended claims.

[0147] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0148] In the several embodiments provided in this application, it should be understood that the disclosed 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 device, or some features may be ignored or not executed.

[0149] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0150] Similarly, it should be understood that, in order to streamline this application and aid in understanding one or more of the various inventive aspects, features of this application may sometimes be grouped together in a single embodiment, figure, or description thereof in the description of exemplary embodiments of this application. However, this approach should not be construed as reflecting an intention that the claimed application requires more features than are expressly recited in each claim. Rather, as reflected in the corresponding claims, its inventive point lies in solving the corresponding technical problem with features fewer than all features of a single disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of this application.

[0151] Those skilled in the art will understand that, apart from the mutual exclusion of features, all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or elements of any method or apparatus so disclosed may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0152] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0153] The various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some modules according to the embodiments of this application. This application can also be implemented as an apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such an implementation of this application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0154] It should be noted that the above embodiments are illustrative of this application and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0155] The above description is merely a specific embodiment or illustration of the embodiments of this application. The scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A method for classifying electricity users in industries prone to confusion, characterized in that, The method includes: Obtain multidimensional industry characteristic data of electricity users to be classified; Using an attention mechanism, a fusion feature vector is generated for each electricity user to be classified based on the multidimensional industry feature data. Using a preset clustering algorithm, all electricity users to be classified are clustered based on the fused feature vector to determine the fuzzy set of easily confused industry users. Using a pre-defined hybrid discrimination model, an initial prediction is made about the industry to which any power user to be classified belongs in the fuzzy set of easily confused industry users, and the contribution vector of each industry feature of any power user to be classified is determined to the initial prediction result. Based on the contribution vector of each industry feature of any power user to the initial prediction result, the comprehensive discrimination score of each industry corresponding to any user is calculated using a preset weighted dot product model. The industry affiliation of any given user is determined based on the comprehensive discrimination score.

2. The method according to claim 1, characterized in that, The industry affiliation of any given user is determined based on the comprehensive discrimination score, including: The industry with the highest comprehensive discrimination score is determined as the industry to which any given user belongs.

3. The method according to claim 2, characterized in that, The industry affiliation of any given user is determined based on the comprehensive discrimination score, including: Based on a hierarchical decision-making mechanism with dual thresholds, any one of the users is reclassified, and the industry to which the user belongs is redefined. The hierarchical decision-making mechanism based on dual thresholds reclassifies any given user and redetermines the industry to which that user belongs, including: Calculate the optimal industry category and corresponding first comprehensive discrimination score of any user in the fuzzy set of users in easily confused industries, as well as the second optimal industry and corresponding second comprehensive discrimination score of any user; Calculate the score difference between the first comprehensive discrimination score and the second comprehensive discrimination score; The first comprehensive discrimination score is compared with a preset reliability threshold, and the score difference is compared with a preset advantage difference threshold. Based on the comparison results, the industry to which any user belongs is determined.

4. The method according to claim 3, characterized in that, The preset confidence threshold includes a maximum confidence threshold and a minimum confidence threshold; the first comprehensive discrimination score is compared with the preset confidence threshold, and the score difference is compared with a preset advantage difference threshold. Based on the comparison result, the industry to which any user belongs is determined, including: If the first comprehensive discrimination score is greater than or equal to the maximum confidence threshold and the score difference is greater than or equal to the preset dominance difference threshold, then any user is determined to belong to the industry with the highest first comprehensive discrimination score. If the first comprehensive discrimination score is less than the minimum confidence threshold, or the score difference is less than the preset advantage difference threshold, the industry to which any user belongs will be manually verified. If the first comprehensive discrimination score is less than the maximum confidence threshold and greater than or equal to the minimum confidence threshold, and the score difference is greater than or equal to the preset advantage difference threshold, the industry to which any user belongs is marked as pending.

5. The method according to claim 1, characterized in that, Obtain multi-dimensional industry characteristic data of the electricity users to be classified, including: Multidimensional industry characteristic data of the electricity users to be classified are obtained from the electricity consumption side, the user side, and the verification side, and an industry user characteristic profile of each electricity user to be classified is constructed based on the multidimensional industry characteristic data.

6. The method according to claim 5, characterized in that, in, The multidimensional industry characteristic data includes load curve depth characteristics, production time series characteristics, and energy consumption intensity characteristics; among which... The load curve depth characteristics include at least one of the following indicators: production continuity index, load fluctuation frequency domain energy ratio, and non-power frequency load ratio; The production time-series characteristics include at least one of the following indicators: intelligent shift system recognition, seasonal production feature vector; The energy intensity characteristics include the unit product power consumption index.

7. The method according to claim 1, characterized in that, Using an attention mechanism, a fused feature vector is generated for each electricity user to be classified based on the multidimensional industry feature data, including: The multidimensional industry feature data is preprocessed; The multidimensional industry feature data is standardized and encoded. The standardized and encoded multidimensional industry feature data are concatenated in a preset order to generate the original feature vector; The original feature vectors are weighted and summed using an attention mechanism to obtain a fused feature vector.

8. The method according to claim 7, characterized in that, Preprocessing of the multidimensional industry feature data includes: The multidimensional industry feature data is denoised using wavelet transform. Remove outliers from the multidimensional industry feature data; The missing values ​​in the multidimensional industry feature data are filled in using a linear interpolation method.

9. The method according to claim 7, characterized in that, The multidimensional industry feature data is standardized and encoded, including: The multidimensional industry feature data is standardized using the Z-score standardization method. The categorical data in the multidimensional industry feature data is converted into binary numerical vectors using the one-hot encoding method.

10. The method according to claim 7, characterized in that, The original feature vectors are weighted and summed using an attention mechanism to obtain a fused feature vector, including: Based on preset attention parameters, calculate the attention weight of each feature in the original feature vector; The original feature vector is weighted and summed according to the attention weights to obtain the fused feature vector.

11. The method according to claim 1, characterized in that, Using a pre-defined clustering algorithm, all electricity users to be classified are clustered based on the fused feature vectors to determine a fuzzy set of easily confused industry users, including: The K-distance graph method is used to determine the adaptive optimal neighborhood radius parameter; The initial clustering of all the electricity users to be classified is performed using a compact adaptive optimal neighborhood radius parameter to form a high-confidence core cluster. The set of users outside the high-confidence core cluster is used as the undetermined classification set. Using a relaxed adaptive optimal neighborhood radius parameter, a second clustering is performed on the undetermined classification set to determine the weak association substructure in the undetermined classification set; The weakly correlated substructures formed by the secondary clustering, and the set of users consisting of noise points that are still not classified after the secondary clustering, are taken as the fuzzy set of easily confused industry users.

12. The method according to claim 1, characterized in that, An initial prediction of the industry to which any unclassified electricity user in the fuzzy set of easily confused industry users belongs is made using a preset hybrid discriminant model. The contribution vector of each industry feature of the unclassified electricity user to the initial prediction result is determined, including: Using users with labeled industries and their corresponding fused feature vectors, a benchmark industry classifier is trained using the gradient boosting tree method to predict the industry probability of the fuzzy set of users in the easily confused industries. The industry affiliation prediction probability of each user in the benchmark industry classifier is taken as the target to be explained, and the fused feature vector corresponding to each user is taken as the feature set input to the interpreter. The defined interpretation object and interpretation features are input into the Shapleyka interpreter to obtain... The contribution vector is predicted for each industry category; wherein the expression for the contribution vector is as follows: ； in, Represents the contribution vector; The number of features involved in the discrimination is represented; i represents the i-th user; k represents the k-th industry to which the i-th user belongs.

13. The method according to claim 1, characterized in that, Based on the contribution vector of each industry characteristic of any given electricity user to the initial prediction result, a comprehensive discrimination score for each industry corresponding to that user is calculated using a pre-defined weighted dot product model, including: Calculate the global average Shapley absolute value of each feature corresponding to each user in the fuzzy set of easily confused industry users; Based on the comparison values ​​between the various features of each user, a judgment matrix is ​​constructed using the 1-9 scaling method; Based on the judgment matrix, calculate the feature weight vector of each feature; Based on the global average Shapley absolute value and the feature weight vector, the comprehensive discrimination score for each feature is calculated; wherein, the expression for the comprehensive discrimination score is as follows: ； Where i represents the fuzzy user set The i-th easily confused user that needs to be identified; k represents the k-th industry to which the i-th user belongs; j represents the j-th industry characteristic of the i-th user; This indicates the number of features involved in the discrimination. Represents the contribution vector of the j-th industry feature; This represents the feature weight of the j-th industry feature.

14. A power user classification device for industries prone to confusion, characterized in that, The device includes: The acquisition module is used to acquire multi-dimensional industry characteristic data of the electricity users to be classified; The vector computation module is used to generate a fused feature vector for each electricity user to be classified based on the multi-dimensional industry feature data by utilizing an attention mechanism. The clustering module is used to cluster all power users to be classified based on the fused feature vector using a preset clustering algorithm, so as to determine the fuzzy set of easily confused industry users; The prediction module is used to make an initial prediction of the industry to which any power user to be classified belongs in the fuzzy set of easily confused industry users using a preset hybrid discrimination model, and to determine the contribution vector of each industry feature of any power user to be classified to the initial prediction result. The scoring module is used to calculate the comprehensive discrimination score of each industry corresponding to any given user based on the contribution vector of each industry feature of any given power user to the initial prediction result, using a preset weighted dot product model. The output module is used to determine the industry affiliation of any given user based on the comprehensive discrimination score.

15. An electronic device, characterized in that, The electronic device includes: A memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program, when executed by the processor, causes the processor to perform the electricity user classification method for easily confused industries as described in any one of claims 1 to 13.

16. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, causes the processor to perform the electricity user classification method for easily confused industries as described in any one of claims 1 to 13.

Images