An industrial power load feature extraction system

Abstract

This invention relates to an industry electricity load feature extraction system, comprising an external analysis module, a confidence configuration module, a caliber grouping module, and a feature extraction module. It utilizes a confidence analysis model to intelligently acquire information from externally sourced data, ensuring richness of information acquisition through publicly available data and filtering out invalid information through confidence analysis. Furthermore, it reduces information retrieval time and improves computational efficiency through sensitive data indexing. Then, it further analyzes external information using a dynamic configuration strategy, combining it with internal information to determine the user's precise sub-segment and obtain industry-specific features. Finally, through caliber grouping strategies and feature extraction algorithms, it intelligently outputs corresponding electricity load features according to the user's required feature extraction needs, resulting in higher granularity of information analysis.

Description

Technical Field

[0001] This invention relates to the field of electricity data analysis, and more specifically, to an industry electricity load feature extraction system. Background Art

[0002] As one of the company's smart grid construction tasks in 2020, "Online Power Grid" aims to scientifically guide the development of a strong smart grid. It will build a planning visualization platform that integrates maps and data, is interactive online, and incorporates artificial intelligence. It will innovate new business models for online management, map-based operations, and online services, and promote the construction of grid-based planning functional modules for "Online Power Grid".

[0003] To meet the goal of establishing a diverse typical load curve library for the "online power grid," enabling efficient calculation of load curves for various industries, cross-regional access to the electricity consumption characteristics of similar users, supporting the development of specialized power grid planning, and assisting in the preparation of electricity consumption plans, the current online power grid relies on the most frequently used external feature as the industry type. It extracts load characteristics based on industry type by retrieving external data. However, this industry type, which serves as a benchmark for electricity users, is very coarse due to the limited availability of external data. For example, the current classification can only categorize electricity users into very broad industry types such as animal husbandry and home appliance manufacturing, but it cannot accurately identify the specific industry, such as whether it produces a particular type of home appliance, manufactures semi-finished products, or is responsible for assembly. These detailed classifications are not possible, resulting in significant differences in load characteristics across different sub-sectors, making it difficult to extract precise load characteristics. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide an industrial power load feature extraction system.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is: an industry power load feature extraction system, including an external analysis module, a confidence configuration module, a caliber grouping module, and a feature extraction module;

[0006] The external analysis module is configured with a confidence analysis strategy and a confidence analysis model. The confidence analysis model connects to different external data platforms and scans the data of the external data platforms in real time and generates a sensitive data index based on the data content. The confidence analysis strategy is used to obtain industry confidence information based on the user name through the sensitive data index. Each industry confidence information includes several industry sub-items and the benchmark confidence value corresponding to each industry sub-item.

[0007] The confidence configuration module is configured with a dynamic configuration strategy. The dynamic configuration strategy obtains the electricity address and historical electricity information of each electricity user and adjusts the benchmark confidence value corresponding to each industry sub-item to generate a dynamic confidence value. The confidence configuration module uses the industry sub-item with the highest dynamic confidence value among each electricity user as the industry sub-feature of that electricity user.

[0008] The caliber grouping module is configured with a caliber grouping strategy, which is used to generate a scale grouping benchmark based on feature retrieval requirements. The caliber grouping module divides electricity users with the same industry segmentation characteristics into different scale groups based on the scale grouping benchmark.

[0009] The feature extraction module is configured with several feature extraction algorithms. The feature extraction module calls the corresponding feature extraction algorithm according to different feature extraction requirements. The feature extraction algorithm is used to extract the corresponding electricity load features from the target electricity data of the scale group.

[0010] Furthermore, the confidence analysis strategy includes

[0011] Step A1: Obtain the sensitive data index corresponding to the user name's business information, determine the business information from the business information platform based on the sensitive data index corresponding to the business information, determine the corresponding trusted business industry based on the business information, and retrieve all industry sub-items under the trusted business industry.

[0012] Step A2: Input the industry sub-items into the confidence analysis model to extract the key information network of each industry sub-item. The key information network consists of several keywords that serve as the identification benchmark for the industry sub-items. The keywords have correlation weight stamps. Retrieve relevant external related data in the key information network from the external data platform through the sensitive data index. Each external related data includes the industry sub-item and the corresponding confidence sub-value. The confidence sub-value reflects the reliability of the external related data.

[0013] Step A3: Summing the confidence sub-values ​​of each industry sub-item using a preset weighted summation algorithm to obtain the corresponding baseline confidence value. The weighted summation algorithm is configured as Q = ∑(1 + B) i )β i q i Where Q is the baseline confidence value, β i Let B be the similarity between the i-th external related data and its corresponding keyword. i Let q be the sum of the association weights of the keywords corresponding to i external related data. i For the confidence sub-value of the i-th externally related data, we have B i =∑b j , where b jThis refers to the association weight corresponding to the j-th triggered association weight stamp in the keyword. The association weight stamp is triggered when the keywords corresponding to the association weight stamps all have similar external related data.

[0014] Furthermore, the dynamic configuration strategy includes

[0015] Step B1: Configure a baseline confidence difference, and filter electricity users whose confidence range is greater than the baseline confidence difference as the baseline electricity users under this industry segment. The confidence range is the difference between the largest baseline confidence value and the mean of other baseline confidence values ​​among the electricity users.

[0016] Step B2: Generate the address concentration location and corresponding concentration value of the industry sub-item based on the electricity consumption address of the benchmark electricity user; extract the corresponding benchmark electricity consumption features through feature extraction algorithm based on the historical electricity consumption information of the benchmark electricity user;

[0017] Step B3: Calculate the address adjustment value of other electricity users according to the preset address adjustment algorithm, and calculate the feature adjustment value of other electricity users according to the preset feature adjustment algorithm;

[0018] Step B4: Adjust the baseline confidence value using address adjustment value and feature adjustment value to generate a dynamic confidence value. in, For the first Dynamic confidence values ​​for each industry sub-item For the first Address adjustment values ​​corresponding to each industry sub-item For the first Feature adjustment values ​​corresponding to each industry sub-item For the first The baseline confidence value corresponding to each industry sub-item is α1, which is the preset address configuration weight, and α2 is the preset feature configuration weight, and α1+α2=1.

[0019] Furthermore, in step B2, an address analysis strategy is configured to generate concentrated locations of addresses and corresponding concentration values ​​of these locations. The address analysis strategy includes...

[0020] The filtering sub-step involves grouping electricity users according to the administrative division of their electricity address;

[0021] The positioning sub-step involves determining the center coordinates for each group to minimize the average distance from the center coordinates to the coordinates of each electricity address within the group.

[0022] The judgment sub-step determines whether there are any electricity users in each group whose distance from the center coordinates is greater than a threshold distance. The threshold distance is proportional to the average distance calculated in the positioning sub-step. If such a user exists, the electricity user is removed from the group and the process returns to the positioning sub-step. If such a user does not exist, the process proceeds to the generation sub-step.

[0023] The generation sub-step uses the center coordinates as the address set location and calculates the concentration value, resulting in d. a =δR / Δd s , where d a R represents the concentration level value, δ is the preset concentration adjustment parameter, and R is the concentration level value. a Δd represents the number of electricity users in the group. s The average distance is calculated for the positioning sub-step.

[0024] Furthermore, the address adjustment algorithm is as follows: Where, Δc a c is the mean of the confidence range of centralized power users in this group. x Let c be the confidence range of the xth electricity user. l The preset range baseline value is used, and distance() is the distance calculation function. f O is the center coordinate of this group. x Let x be the coordinates of the electricity address of the x-th electricity user.

[0025] Furthermore, step B3 also includes retrieving industry distribution information from an external data platform based on the location of the address set, and determining the range benchmark value based on the degree of matching between the industry distribution information and the corresponding industry sub-items.

[0026] Furthermore, the feature adjustment algorithm includes Among them, e r Let m be the similarity of the r-th similar reference electricity consumption feature, where r is the number of similar reference electricity consumption features and m is the total number of reference electricity consumption features.

[0027] Furthermore, the caliber grouping strategy includes

[0028] Step C1: Generate retrieval scope items and industry sub-items based on feature retrieval requirements;

[0029] Step C2: Determine the corresponding electricity users based on the industry sub-items and determine the corresponding retrieval caliber value from the electricity users based on the retrieval caliber item;

[0030] Step C3: Calculate the mean caliber value corresponding to all retrieved caliber values ​​using a weighted average algorithm. The weighted average algorithm is: mean = ∑(w i / ∑w i )x iWhere mean is the caliber average, x i w is the retrieval value corresponding to the i-th electricity user. i Let w be the discrete weight value corresponding to the i-th electricity user. i =1-n(x) i -Δx) 2 / ∑(x i -Δx) 2 n represents the number of electricity users in this dimension, and Δx is the mean of the data retrieved from the electricity user data.

[0031] Step C4: Determine several scale division ranges based on the mean caliber to generate the scale grouping benchmark.

[0032] Furthermore, the feature extraction algorithm includes a unit capacity extraction algorithm, which is y(t) = [∑p(t,i) / P(i)] / n, where i represents the i-th electricity user, t is time, p(i,t) represents the load power of the i-th electricity user at time t, P(i) represents the registered capacity of the i-th electricity user, n represents the number of electricity users in this dimension, and y(t) represents the unit capacity load curve; the feature extraction module uses the obtained unit capacity load curve as the electricity load feature.

[0033] Furthermore, the feature extraction algorithm includes an electricity consumption pattern extraction algorithm, which performs per-unit calculation on the electricity consumption data waveform and identifies the shape features of the calculated electricity consumption waveform to form the electricity load features.

[0034] The main technical effects of this invention are reflected in the following aspects: By using a confidence analysis model, intelligent information acquisition is achieved from data obtained through established external data acquisition methods. External publicly available data ensures increased information richness, and confidence analysis ensures the filtering of invalid information. Furthermore, sensitive data indexing reduces the time required for information retrieval and improves computational efficiency. Then, a dynamic configuration strategy is used to further analyze external information, combining it with internal information to determine the user's precise sub-segment and obtain industry segmentation characteristics. Finally, through a caliber grouping strategy and feature extraction algorithm, the corresponding electricity load characteristics can be intelligently output according to the user's required feature extraction needs, giving the information analysis a higher granularity. Attached Figure Description

[0035] Figure 1 : System architecture diagram of this invention.

[0036] Figure reference numerals: 100, External analysis module; 200, Confidence configuration module; 300, Calibration grouping module; 400, Feature extraction module. Detailed Implementation

[0037] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, so that the technical solution of the present invention can be more easily understood and mastered.

[0038] Reference Figure 1 As shown, an industry electricity load feature extraction system includes an external analysis module 100, a confidence configuration module 200, a caliber grouping module 300, and a feature extraction module 400.

[0039] Compared to traditional electricity load feature extraction systems, a key problem with traditional systems is that they can only obtain a vague business scope for a particular electricity user based on registered information. For example, in furniture manufacturing, different types of furniture and different manufacturing stages will have significantly different electricity consumption characteristics. This results in a large degree of dispersion between different entities when extracting electricity load features, making the extracted features lack reference and analytical value, and rendering industry electricity load feature extraction meaningless. On the other hand, with the popularization of semantic recognition technology combined with intelligent indexing technology of external databases, more and more technology fields are beginning to try to use publicly available information on the Internet to complete information collection and matching. The text intelligent indexing system based on deep learning, through Policy Gradient algorithm reinforcement learning, establishes a data collection channel and analyzes the reliability of data through a powerful base model, which can greatly reduce the impact of invalid data on intelligent retrieval results.For example, the application of ChatGPT. The first core difference between this invention and ChatGPT lies in that this invention does not rely on a precise understanding of text semantics. This is because this invention requires using the company name or other uniquely matched data as an index to determine the company's industry segment, providing a data analysis benchmark for electricity load feature extraction. Therefore, the external analysis module 100 is configured with a confidence analysis strategy and a confidence analysis model. The confidence analysis model connects to different external data platforms, such as recruitment platforms, news platforms, e-commerce platforms, government information platforms, and search engine platforms. Since these platforms retrieve and analyze publicly available information, there is no legal risk. The module scans data from external data platforms in real time and generates sensitive data indexes based on the data content. Although this invention pre-establishes interaction interfaces and data interaction logic with the aforementioned external platforms, the data on each platform is still being updated. If all data needs to be scanned frequently, the workload for the external analysis module 100 would be enormous. Therefore, by generating sensitive data indexes, a strategy is established to determine the data acquisition channels for each platform. To reduce data processing volume, and because the platform's data acquisition methods may change, when the external analysis module 100 does not execute the confidence analysis strategy, it can monitor the data of the external data platform in real time to train the external analysis module 100's sensitivity to industry sub-terms and discover new effective data acquisition methods under different platforms. The confidence analysis strategy is used to obtain industry confidence information based on the user name through sensitive data indexing. Each industry confidence information includes several industry sub-items and the corresponding benchmark confidence value for each industry sub-item. That is, the corresponding industry sub-item can be found through the user name. However, due to the high complexity of external data, a user name may be associated with multiple industry sub-items. For example, a factory that produces semi-finished products may also need assembly R&D engineers for finished products. Thus, the factory may produce finished products or semi-finished products. Therefore, in order to accurately determine the industry sub-item closest to the user, each user will have multiple different industry sub-items, the only difference being their benchmark confidence value. The more related keywords are indexed, the higher the reliability and the higher the benchmark confidence value. The confidence analysis strategy includes...

[0040] Step A1: Obtain the sensitive data index corresponding to the user name's business information. Based on the sensitive data index, determine the business information from the business information platform. Based on the business information, determine the corresponding trusted business industry, and retrieve all industry sub-items under that trusted business industry. First, obtain a detailed table of all relevant industry sub-items, which is obtained based on the business scope. This can be obtained from the business platform index or directly entered, which will not be elaborated here. After obtaining a broad but uniformly formatted trusted business industry, you can directly retrieve all industry sub-items under that item.

[0041] Step A2: Input the industry sub-items into the confidence analysis model to extract the key information network for each industry sub-item. The key information network consists of several keywords serving as the identification benchmark for the industry sub-item. These keywords have correlation weight stamps. The key information network is a crucial point in the confidence analysis model, which aims to identify industry keywords that may belong to the industry. For example, if a procurement project mentions steel procurement and a job posting mentions filing cabinets, it is highly likely to belong to the filing cabinet welding sub-industry. A single keyword alone wouldn't provide such a high match. Therefore, the key information network is continuously learned and generated. Correlation weight stamps are determined based on the frequency of keyword occurrences. On the other hand, relevant external correlation data is retrieved from external data platforms using a sensitive data index. Each external correlation data includes an industry sub-item and a corresponding confidence sub-value. The confidence sub-value reflects the reliability of the external correlation data; for example, the reliability of the external platform itself, whether the information has been verified, the number of times the information has been forwarded or viewed, and the number of times the information appears repeatedly on different external platforms can all affect the confidence sub-value. In this way, the correlation between industry sub-items and external correlation data can be quantified. External related data contains several keywords.

[0042] Step A3: Summing the confidence sub-values ​​of each industry sub-item using a preset weighted summation algorithm to obtain the corresponding baseline confidence value. The weighted summation algorithm is configured as Q = ∑(1 + B) i )β i q i Where Q is the baseline confidence value, β i Let B be the similarity between the i-th external related data and its corresponding keyword. Since keywords and external related data are not necessarily completely identical, they are matched using similar words and semantic analysis. However, the higher the matching degree, the more likely the data is to be understood as having the corresponding keyword. i Let q be the sum of the association weights of the keywords corresponding to i external related data. i For the confidence sub-value of the i-th externally related data, we have B i =∑bj , where b j This refers to the association weight corresponding to the j-th triggered association weight tag within the keyword. An association weight tag is triggered when all keywords corresponding to it have similar external related data. In this way, external related data can be analyzed, and the corresponding baseline confidence value for each industry sub-category can be calculated based on the association weight. The result can be obtained by summing all external related data within that dimension. The baseline confidence value determines whether the industry sub-category matches electricity users based on the external related data.

[0043] As the second core technical content of this invention: the confidence configuration module 200 is configured with a dynamic configuration strategy. This dynamic configuration strategy obtains the electricity address and historical electricity consumption information of each electricity user, adjusts the baseline confidence value corresponding to each industry sub-item to generate a dynamic confidence value. After obtaining external industry sub-information, it is necessary to construct the most likely industry affiliation for each electricity user. This analysis is based on internally obtained electricity address data and historical electricity consumption information. Since these two data are directly statistically analyzed by the power grid, their reliability is high. Specifically: the confidence configuration module 200 uses the industry sub-item with the highest dynamic confidence value among each electricity user as the industry sub-feature of that electricity user; the dynamic configuration strategy includes...

[0044] Step B1: Configure a baseline confidence difference and filter electricity users whose confidence range is greater than the baseline confidence difference as the baseline electricity users for that industry sub-category. The confidence range is the difference between the highest baseline confidence value and the mean of other baseline confidence values ​​among the electricity users. For example, if an electricity user's industry sub-category includes multiple sub-categories such as office desk processing and office desk manufacturing, select the sub-category with the highest baseline confidence value and subtract the mean of the other sub-categories. This difference determines whether the highest baseline confidence value can be trusted as a benchmark for subsequent analysis. For example, if user A's baseline confidence value for office desk processing is 90, and the mean of other sub-categories is 10, while user B's baseline confidence value for office desk processing is 120, and the mean of other sub-categories is 80, then user A is more likely to be an office desk user than user B because user A's confidence range is 80, while user B's is 40. If the baseline confidence difference is configured to be 70, then user A is selected as the baseline electricity user for the office desk processing industry sub-category.

[0045] Step B2: Generate the address clustering location and corresponding concentration value of the industry sub-category based on the electricity address of the benchmark electricity user; extract the corresponding benchmark electricity consumption features using a feature extraction algorithm based on the historical electricity consumption information of the benchmark electricity user; determine the clustering location and concentration by analyzing the features of the benchmark electricity user. Furthermore, similar electricity consumption features may appear under the same industry sub-category because industry electricity consumption patterns are similar; therefore, further correlation analysis may be performed by analyzing the similarity of electricity consumption features. In step B2, an address analysis strategy is configured to generate the address clustering location and corresponding concentration value. The address analysis strategy includes...

[0046] The filtering sub-step involves grouping electricity users according to the administrative division of their electricity addresses. Initially, a preliminary division is made based on the textual information of the electricity addresses. This avoids situations where some electricity users are located far apart, which would increase the workload of subsequent loop calculations.

[0047] The positioning sub-step involves determining the center coordinates for each group to minimize the average distance from the center coordinates to the coordinates of each electricity address within the group; then, the electricity addresses after grouping are mapped to coordinates on the map, and the optimal center location is determined through the coordinate nodes.

[0048] The judgment sub-step determines whether there are any electricity users in each group whose distance from the center coordinates is greater than a threshold distance. The threshold distance is proportional to the average distance calculated in the positioning sub-step. If such a user exists, the electricity user is removed from the group and the process returns to the positioning sub-step. If not, the process proceeds to the generation sub-step. Repeating the positioning sub-step can further remove users who are too far away until the distribution characteristics of the industrial cluster are met. The threshold distance is determined based on the average distance in the previous step, preferably 2-4 times the average distance.

[0049] The generation sub-step uses the center coordinates as the address set location and calculates the concentration value, resulting in d. a =δR / Δd s , where d a R represents the concentration level value, δ is the preset concentration adjustment parameter, and R is the concentration level value. a Δd represents the number of electricity users in the group. s The average distance is calculated for the positioning sub-step. This allows us to determine the center coordinates and thus calculate the degree of concentration.

[0050] Step B3: Calculate the address adjustment values ​​for other electricity users according to a preset address adjustment algorithm, and calculate the characteristic adjustment values ​​for other electricity users according to a preset feature adjustment algorithm; the address adjustment algorithm is... Where, Δc a c is the mean of the confidence range of centralized power users in this group.x Let c be the confidence range of the xth electricity user. l The preset range baseline value is used, and distance() is the distance calculation function. f O is the center coordinate of this group. x Let x be the coordinates of the electricity address of the x-th electricity user. Step B3 also includes retrieving industry distribution information from an external data platform based on the location within the address set, and determining the range benchmark value based on the degree of matching between the industry distribution information and the corresponding industry sub-items. The feature adjustment algorithm includes... Among them, e r Let m be the similarity of the r-th similar reference electricity consumption feature, where r is the number of similar reference electricity consumption features and m is the total number of reference electricity consumption features.

[0051] Step B4: Adjust the baseline confidence value using address adjustment value and feature adjustment value to generate a dynamic confidence value. in, For the first Dynamic confidence values ​​for each industry sub-item For the first Address adjustment values ​​corresponding to each industry sub-item For the first Feature adjustment values ​​corresponding to each industry sub-item For the first The baseline confidence value corresponding to each industry sub-item is α1, which is the preset address configuration weight, and α2 is the preset feature configuration weight, and α1+α2=1.

[0052] The caliber grouping module 300 is configured with a caliber grouping strategy, which is used to generate a scale grouping benchmark based on feature retrieval requirements. The caliber grouping module 300 then divides electricity users with the same industry segmentation characteristics into different scale groups based on the scale grouping benchmark. The caliber grouping strategy includes...

[0053] Step C1: Generate retrieval scope items and industry sub-items based on feature retrieval requirements;

[0054] Step C2: Determine the corresponding electricity users based on the industry sub-items and determine the corresponding retrieval caliber value from the electricity users based on the retrieval caliber item;

[0055] Step C3: Calculate the mean caliber value corresponding to all retrieved caliber values ​​using a weighted average algorithm. The weighted average algorithm is: mean = ∑(w i / ∑w i )x i Where mean is the caliber average, x i w is the retrieval value corresponding to the i-th electricity user. iLet w be the discrete weight value corresponding to the i-th electricity user. i =1-n(x) i -Δx) 2 / ∑(x i -Δx) 2 n represents the number of electricity users in this dimension, and Δx is the mean of the data retrieved from the electricity user data.

[0056] Step C4: Determine several scale division ranges based on the aforementioned mean value to generate the scale grouping benchmark. With the mean value determined, the corresponding scale division ranges can be determined based on the mean value. For example, the five division ranges could be [mean-g,mean+g][mean+g,mean+2g][mean-2g,mean-g][0,mean-2g][mean+2g,+∞]. The number of division ranges can be determined based on the total number of electricity users.

[0057] The feature extraction module 400 is configured with several feature extraction algorithms. The feature extraction module 400 calls the corresponding feature extraction algorithm according to different feature extraction requirements. The feature extraction algorithm is used to extract the corresponding electricity load features from the target electricity data of the scale group. The feature extraction module analyzes electricity consumption data according to predetermined extraction requirements to discover patterns in the data. This type of feature extraction algorithm has been widely applied and developed in the field of electricity waveform analysis. For example, it has been used to extract features for missing data, features for large fluctuations in electricity consumption, and features related to lightning strikes. This invention provides two feature extraction algorithms. One of these algorithms is a unit capacity extraction algorithm, defined as y(t) = [∑p(t,i) / P(i)] / n, where i represents the i-th electricity user, t is time, p(i,t) represents the load power of the i-th electricity user at time t, P(i) represents the registered capacity of the i-th electricity user, n represents the number of electricity users in this dimension, and y(t) represents the unit capacity load curve. The feature extraction module uses the obtained unit capacity load curve as the electricity load feature. Because this invention has a finer industry analysis granularity, it supports the unit capacity extraction algorithm to generate a unit capacity load curve as an industry feature. Secondly, the feature extraction algorithm includes an electricity consumption pattern extraction algorithm. This algorithm performs per-unit calculation on the electricity consumption data waveform and identifies the shape features of the calculated waveform to form the electricity load features. The per-unit calculation combined with the shape features of the electricity waveform allows for similarity recognition of the waveform features, extracting waveform features and providing technical support for obtaining the baseline electricity consumption features in step B2.

[0058] Of course, the above are just typical examples of the present invention. In addition, the present invention may have many other specific embodiments. All technical solutions formed by equivalent substitution or equivalent transformation fall within the scope of protection claimed by the present invention.

Claims

1. A system for extracting the characteristics of industrial electricity load, characterized in that: It includes an external analysis module, a confidence configuration module, a caliber grouping module, and a feature extraction module; The external analysis module is configured with a confidence analysis strategy and a confidence analysis model. The confidence analysis model connects to different external data platforms and scans the data of the external data platforms in real time and generates a sensitive data index based on the data content. The confidence analysis strategy is used to obtain industry confidence information based on the user name through the sensitive data index. Each industry confidence information includes several industry sub-items and the benchmark confidence value corresponding to each industry sub-item. The confidence configuration module is configured with a dynamic configuration strategy. The dynamic configuration strategy obtains the electricity address and historical electricity information of each electricity user and adjusts the benchmark confidence value corresponding to each industry sub-item to generate a dynamic confidence value. The confidence configuration module uses the industry sub-item with the highest dynamic confidence value among each electricity user as the industry sub-feature of that electricity user. The caliber grouping module is configured with a caliber grouping strategy, which is used to generate a scale grouping benchmark based on feature retrieval requirements. The caliber grouping module divides electricity users with the same industry segmentation characteristics into different scale groups based on the scale grouping benchmark. The feature extraction module is configured with several feature extraction algorithms. The feature extraction module calls the corresponding feature extraction algorithm according to different feature extraction requirements. The feature extraction algorithm is used to extract the corresponding electricity load features from the target electricity data of the scale grouping. The dynamic configuration strategy includes Step B1: Configure a baseline confidence difference, and filter electricity users whose confidence range is greater than the baseline confidence difference as the baseline electricity users under this industry segment. The confidence range is the difference between the largest baseline confidence value and the mean of other baseline confidence values ​​among the electricity users. Step B2: Generate the address concentration location and the concentration degree value of the corresponding address concentration location for the industry sub-item based on the electricity address of the benchmark electricity user; Based on the historical electricity consumption information of benchmark electricity users, corresponding benchmark electricity consumption features are extracted using a feature extraction algorithm. Step B3: Calculate the address adjustment value of other electricity users according to the preset address adjustment algorithm, and calculate the feature adjustment value of other electricity users according to the preset feature adjustment algorithm; Step B4: Adjust the baseline confidence value using address adjustment value and feature adjustment value to generate a dynamic confidence value. ,in, For the first Dynamic confidence values ​​for each industry sub-item For the first Address adjustment values ​​corresponding to each industry sub-item For the first Feature adjustment values ​​corresponding to each industry sub-item For the first The benchmark confidence values ​​for each industry sub-segment Configure weights for preset addresses. Assign weights to preset features, and have .

2. The industrial electricity load characteristic extraction system as described in claim 1, characterized in that: The confidence analysis strategy includes Step A1: Obtain the sensitive data index corresponding to the user name's business information, determine the business information from the business information platform based on the sensitive data index corresponding to the business information, determine the corresponding trusted business industry based on the business information, and retrieve all industry sub-items under the trusted business industry item; Step A2: Input the industry sub-items into the confidence analysis model to extract the key information network of each industry sub-item. The key information network consists of several keywords that serve as the identification benchmark for the industry sub-items. The keywords have correlation weight stamps. Retrieve relevant external related data in the key information network from the external data platform through the sensitive data index. Each external related data includes the industry sub-item and the corresponding confidence sub-value. The confidence sub-value reflects the reliability of the external related data. Step A3: Summing the confidence sub-values ​​of each industry sub-item using a preset weighted summation algorithm to obtain the corresponding baseline confidence value. The weighted summation algorithm is configured as follows: ,in As the baseline confidence value, For the first The similarity between external related data and corresponding keywords. for The sum of the association weights of keywords corresponding to each external related data point. For the first Confidence subvalues ​​of externally related data, ,in For the keyword, the first A triggered association weight stamp corresponds to an association weight. When the keywords corresponding to the association weight stamps all have similar external related data, the association weight stamp is triggered.

3. The industrial electricity load characteristic extraction system as described in claim 1, characterized in that: In step B2, an address analysis strategy is configured to generate concentrated locations of addresses and corresponding concentration values ​​of those locations. The address analysis strategy includes: The filtering sub-step involves grouping electricity users according to the administrative division of their electricity usage address; The positioning sub-step involves determining the center coordinates for each group to minimize the average distance from the center coordinates to the coordinates of each electricity address within the group. The judgment sub-step determines whether there are any electricity users in each group whose distance from the center coordinates is greater than a threshold distance. The threshold distance is proportional to the average distance calculated in the positioning sub-step. If such a user exists, the electricity user is removed from the group and the process returns to the positioning sub-step. If such a user does not exist, the process proceeds to the generation sub-step. The generation sub-step uses the center coordinates as the address set location and calculates the concentration value. ,in This represents the concentration level value. These are preset centralized adjustment parameters. The number of electricity users in the group. The average distance is calculated for the positioning sub-step.

4. The industrial electricity load feature extraction system as described in claim 3, characterized in that: The address adjustment algorithm is as follows: ,in, This represents the mean of the confidence ranges for centralized electricity users in this group. For the first The confidence level of individual electricity users is extremely low. The preset range benchmark value, For distance calculation function, The center coordinates of this group are For the first The electricity address coordinates of each electricity user.

5. The industrial electricity load characteristic extraction system as described in claim 4, characterized in that: Step B3 also includes calling the industry distribution information of the external data platform based on the location of the address set, and determining the value of the range benchmark based on the degree of matching between the industry distribution information and the corresponding industry sub-items.

6. The industrial electricity load characteristic extraction system as described in claim 1, characterized in that: The feature adjustment algorithm includes ,in, For the first Similarity of similar baseline electricity consumption characteristics The number of similar baseline electricity consumption characteristics, The total number of baseline electricity consumption characteristics.

7. The industrial electricity load characteristic extraction system as described in claim 1, characterized in that: The caliber grouping strategy includes Step C1: Generate retrieval scope items and industry sub-items based on feature retrieval requirements; Step C2: Determine the corresponding electricity users based on the industry sub-items and determine the corresponding retrieval caliber value from the electricity users based on the retrieval caliber item; Step C3: Calculate the mean caliber value corresponding to all retrieved caliber values ​​using a weighted average algorithm. The weighted average algorithm is as follows: ,in, The average caliber is... For the first The retrieval criteria value corresponding to each electricity user For the first The discrete weight values ​​corresponding to each electricity user are: , This indicates the number of electricity users in this dimension. The average value of the data obtained for electricity users; Step C4: Determine several scale division ranges based on the mean caliber to generate the scale grouping benchmark.

8. The industrial electricity load characteristic extraction system as described in claim 1, characterized in that: The feature extraction algorithm includes a unit capacity extraction algorithm, which is as follows: ,in Representing the Individual electricity users, For time, Indicates the first Individual electricity users Load power at any given time Indicates the first The installed capacity of each electricity user This indicates the number of electricity users in this dimension. The unit capacity load curve is represented; the feature extraction module uses the obtained unit capacity load curve as the electricity load feature.

9. The industrial electricity load characteristic extraction system as described in claim 1, characterized in that: The feature extraction algorithm includes an electricity consumption pattern extraction algorithm, which performs per-unit calculation on the electricity consumption data waveform and identifies the shape features of the calculated electricity consumption waveform to form the electricity load features.

Images