A method, system, and equipment for quantitative evaluation of industry label generation and development trends.
By constructing a structured industry labeling system that integrates multiple policy sources and a power load characteristic fingerprint, the problem of standard mismatch in the transformation of power data into industry analysis has been solved, enabling accurate determination of enterprise industry affiliation and quantification of multi-dimensional development trends, and providing reliable support for industry regulation.
Patent Information
- Application Number
- CN202611032793.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-13
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, methods for converting electricity data into industry analysis suffer from problems such as crude industry boundary identification, lagging data updates, difficulty in achieving accurate judgments, mismatch between the electricity classification system and policy industry definition standards, inability to accurately depict the position of enterprises in the industrial chain and the transformation process, single electricity consumption indicators have a limited evaluation dimension, cannot construct a multi-dimensional development profile, and are easily affected by natural factors.
We construct a structured industry labeling system that integrates multiple policy sources, generate a set of candidate semantic labels for enterprises, determine the industry affiliation of enterprises through power load characteristic fingerprint vectors and industry fingerprint benchmarks, construct multi-dimensional evaluation indicators for the power industry, identify inflection points in industrial development, and output the industry evolution trend and transformation stage.
It has enabled large-scale, precise industry labeling of enterprises, breaking through the limitations of single indicators, accurately quantifying the development trend of industries, and providing reliable support for industry regulation.
Smart Images

Figure CN123117879A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to methods for quantifying power data, belonging to the field of power big data analysis, and particularly to a method, system, and equipment for generating industry tags and quantitatively evaluating development trends. Background Technology
[0002] With its unique advantages of full coverage, real-time objectivity, and resistance to tampering, coupled with the massive enterprise-level time-series electricity consumption data brought about by the construction of new power systems, electricity data has become a high-quality data source for industrial operation analysis. However, the current technical system for transforming electricity consumption big data into structured knowledge that can be used for industrial analysis is still imperfect.
[0003] Traditional ledger-based statistical methods in the industry rely on manually filled-in reports and sampled data for analysis. While the statistical standards are relatively uniform, they suffer from problems such as crude industry boundary identification, lagging data updates, and difficulty in achieving large-scale, refined industry determination. Electricity consumption classification methods in the power industry can reduce analysis costs by leveraging existing power system data, but because the electricity classification system is adapted to electricity pricing rather than policy-defined industry standards, there is a systemic mismatch with the industry catalog, making it unable to support precise analysis of subdivided industries and new / old industries. Keyword-based rule-matching methods for automatic enterprise classification improve the efficiency of industry attribution determination, but only achieve surface-level character matching, lacking semantic depth. The current technology suffers from several shortcomings. Firstly, it is susceptible to misjudgments due to issues such as overlapping business operations, non-standard descriptions, and discrepancies between registered and actual operations. Secondly, it fails to depict a company's position in the industry chain and its transformation process. Thirdly, while the single electricity consumption indicator provides intuitive data for comparison, its limited dimensions prevent the construction of a multi-dimensional portrait of industry development, making it difficult to reflect industry production efficiency and spatial distribution characteristics. Furthermore, it is easily distorted by natural factors. Therefore, existing technologies generally suffer from inconsistent standard mapping, insufficient ability to accurately determine a company's industry affiliation, and difficulty in quantifying the multi-dimensional state and transformation process of industries. A more precise method for quantifying industry development trends is urgently needed to address these issues. Summary of the Invention
[0004] The purpose of this invention is to overcome the above-mentioned defects and problems in the prior art and to provide a method, system and equipment for quantitative evaluation of industry label generation and development trend, so as to accurately quantify the development trend of industry.
[0005] To achieve the above objectives, the technical solution of this invention is: a method for quantitatively evaluating the generation and development trend of industry labels, comprising:
[0006] Construct a structured industry labeling system that integrates multiple policy sources, and generate a set of candidate semantic labels for enterprises;
[0007] Collect enterprise electricity consumption data, extract power load characteristics, and obtain enterprise power load characteristic fingerprint vectors and industry fingerprint benchmarks;
[0008] Based on the enterprise's power load characteristic fingerprint vector and industry fingerprint benchmark, the optimal label for the enterprise's industry affiliation is determined, and the candidate semantic label set is traversed. The consistency deviation of the optimal label for the enterprise's industry affiliation is corrected by the text-power deviation, and a list of industry enterprises is obtained.
[0009] For the set of enterprises in the industry to which each optimal label belongs in the list of industrial enterprises, construct a multi-dimensional evaluation index for the power industry, and determine the comprehensive profile vector and comprehensive development index.
[0010] Based on comprehensive profile vector recognition, the structural differences of the industry in various dimensions are identified. Based on the comprehensive development index, the inflection point of industrial development is identified, and the green and intelligent transformation index is determined. The industrial evolution trend and industrial transformation stage are output.
[0011] Optionally, the construction of a structured industry labeling system integrating multi-source policies specifically includes:
[0012] Constructing a policy source set And each policy source Corresponding to a priori priority ;
[0013] The industry labeling system is modeled as a directed acyclic graph. ,in A collection of tag nodes. The set of directed edges representing hierarchical relationships;
[0014] Based on the sub-industry, each tag node in the tag node set By associating the triplet features, we obtain the label feature triplet, which is expressed as follows:
[0015] ;
[0016] in: For the first Feature triples of each label node; A collection of label semantics; A set of mappings for industry codes; This represents the vector of the policy attribute to which the policy attribute belongs.
[0017] If multiple policy sources assign inconsistent policy attributes to the same tag node, the policy attribute to which the tag node belongs is determined based on a priori priority weighting; the expression for the priori priority weighting is as follows:
[0018] ;
[0019] in: for The policy attribute to which it belongs; This is the set of possible categories for policy attributes; As a policy source For tag nodes In terms of the policy attributes to which they belong The category determination given above;
[0020] Define the policy confidence level for each labeled node, quantifying the degree to which a labeled node is jointly identified by multiple policy sources. The expression is as follows:
[0021] ;
[0022] in: For the first Policy confidence level of each labeled node; As a policy source The defined set of tags; For indicator functions;
[0023] A structured industry labeling system is formed by the label feature triples of the label nodes, the policy attributes to which they belong, and the policy confidence level.
[0024] Optionally, the candidate semantic tag set for generating enterprises specifically includes:
[0025] The enterprise's multi-source information fields are collected and concatenated into an enterprise semantic description text, which is then mapped to an enterprise semantic representation vector. The expression is as follows:
[0026] ;
[0027] in: It is a semantic representation vector; Description of core technologies; Description of the core product; Description of main business;
[0028] The tag semantic set of each tag node is uniformly encoded and mapped to obtain the tag semantic vector, and the text semantic matching score between the enterprise and the tag node is determined. The expression is as follows:
[0029] ;
[0030] ;
[0031] in: For text semantic matching; For the label semantic vector; As a measure of directional consistency; It is the vector norm 2; It is a very small positive number;
[0032] Based on text semantic matching scores, a set of candidate semantic tags is obtained by filtering through a preset semantic threshold, as shown in the following expression:
[0033] ;
[0034] in: For the set of candidate semantic labels; This is a preset semantic threshold.
[0035] Optionally, the process of collecting enterprise electricity consumption data, extracting electricity load characteristics, and obtaining enterprise electricity load characteristic fingerprint vectors and industry fingerprint benchmarks specifically includes:
[0036] To obtain enterprise electricity consumption data and determine load utilization, the expression is as follows:
[0037] ;
[0038] ;
[0039] in: For load utilization rate; Annual electricity consumption; This represents the peak load. This represents the total sampling time. This refers to the electricity metering sequence. To use a fixed sampling step size;
[0040] To obtain the electricity load of a company on weekdays and non-weekdays, and determine the proportion of weekdays, the expression is as follows:
[0041] ;
[0042] in: The percentage of weekdays; , These represent the average workload for weekdays and non-weekdays, respectively. It is a very small positive number;
[0043] To determine the peak-to-valley ratio, obtain the electricity load during peak and off-peak hours for the enterprise. The expression is as follows:
[0044] ;
[0045] in: This represents the peak-to-valley ratio between the peak and trough periods of the day. , These represent the average load during peak and off-peak hours of the day, respectively.
[0046] To obtain the seasonal patterns of enterprise electricity consumption and determine seasonal factors, the expression is as follows:
[0047] ;
[0048] in: Seasonal factors; This represents the maximum monthly electricity consumption throughout the year. For the first Monthly electricity consumption sequence for the month, ;
[0049] Obtain the company's monthly electricity consumption data and determine the monthly fluctuation coefficient, the expression of which is as follows:
[0050] ;
[0051] ;
[0052] in: This is the monthly fluctuation coefficient; This represents the average monthly electricity consumption throughout the year.
[0053] Based on load utilization rate, weekday percentage, peak-to-valley ratio, seasonality factors, and monthly fluctuation coefficient, a characteristic fingerprint vector of enterprise power load is constructed, the expression of which is as follows:
[0054] ;
[0055] in: For enterprises Enterprise power load characteristic fingerprint vector;
[0056] The enterprise's power load characteristic fingerprint vector is normalized, and the centroid and covariance of the fingerprint distribution are estimated for each label node as the industry fingerprint benchmark. The expressions are as follows:
[0057] ;
[0058] ;
[0059] ;
[0060] in: Center of mass; For covariance; For seed sample set; This is the normalized fingerprint vector of enterprise power load characteristics. , These are the lower and upper bounds of the overall enterprise sample, respectively.
[0061] Optionally, the step of determining the optimal label for enterprise industry affiliation based on enterprise power load characteristic fingerprint vector and industry fingerprint benchmark, and traversing the candidate semantic label set, correcting for consistency deviations in the optimal label for enterprise industry affiliation through text-power deviation, and obtaining a list of industry enterprises, specifically includes:
[0062] Based on the normalized enterprise power load characteristic fingerprint vector, Mahalanobis distance is used to measure the deviation between the enterprise fingerprint and the industry benchmark, and it is transformed into a power consistency score through exponential mapping, the expression of which is as follows:
[0063] ;
[0064] in: For power consistency; This is the normalized fingerprint vector of enterprise power load characteristics.
[0065] Based on the rule matching score between policy confidence and the label rule base, and as a strong prior constraint, the policy authority score is obtained, and its expression is as follows:
[0066] ;
[0067] in: To give the policy authority points; For the first Policy confidence level of each labeled node;
[0068] The comprehensive attribution score of an enterprise to a tag node is obtained by weighting and fusing text semantic matching score, power consistency score, and policy authority score. The expression is as follows:
[0069] ;
[0070] in: The overall attribution score; , , All are weighting coefficients;
[0071] Based on the comprehensive affiliation score, the optimal label for a company's industry affiliation is determined, and its expression is as follows:
[0072] ;
[0073] in: For enterprises The optimal label; The confidence level of the corresponding policy attribute;
[0074] To determine the text-electricity deviation degree of the optimal label for determining the enterprise's industry affiliation, consistency deviation correction is performed, and the expression is as follows:
[0075] ;
[0076] in: For text—electrical deviation; , For enterprises For the optimal label Text semantic matching score and power consistency score;
[0077] like , and When this occurs, consistency deviation correction is triggered, excluding the optimal label for enterprise industry affiliation, and iterating through the candidate semantic label set until consistency deviation correction is no longer triggered, outputting a list of industry enterprises; where , , These are the high matching threshold, low consistency threshold, and consistency deviation threshold, respectively.
[0078] Optionally, for each set of enterprises in the industrial enterprise list whose optimal label belongs to an industry, a multi-dimensional evaluation index for the power industry is constructed to determine the comprehensive profile vector and the comprehensive development index, specifically including:
[0079] The scale-dimensional evaluation index is constructed as follows:
[0080] ; ;
[0081] in: Electricity consumption for industry; To the maximum load of the industry; Simultaneity rate coefficient; Annual electricity consumption; This represents the peak load. For industry serial numbers;
[0082] The growth dimension evaluation index is constructed, and its expression is as follows:
[0083] ; ;
[0084] in: This represents the year-on-year growth rate of industrial electricity consumption; The contribution of the increase in industrial electricity consumption to the total increase in total industrial electricity consumption; , industries In the Year and the Annual electricity consumption; , industries In the Year and the Annual electricity consumption;
[0085] The efficiency dimension evaluation index is constructed, and its expression is as follows:
[0086] ; ;
[0087] in: For industrial load utilization rate; The percentage of working days in the industry; The percentage of weekdays with the highest workload;
[0088] The spatial dimension evaluation index is constructed, and its expression is as follows:
[0089] ; ;
[0090] in: User concentration; It is distributed regionally; For industry In the region Electricity consumption; For a set of regions;
[0091] By normalizing and weighting the evaluation indicators of scale, growth, efficiency, and spatial dimensions, a comprehensive profile vector and a comprehensive development index are obtained, and their expressions are as follows:
[0092] ;
[0093] ;
[0094] in: For industry The comprehensive image vector; For industry The comprehensive development index; For industry Next The normalized standardized index values of the evaluation indicators; As the indicator weight; This represents the total number of evaluation dimensions and indicators.
[0095] Optionally, the process of identifying structural differences in industries across various dimensions based on comprehensive profile vectors; identifying inflection points in industrial development based on comprehensive development indices; determining green and intelligent transformation indices; and outputting industrial evolution trends and transformation stages specifically includes:
[0096] Based on comprehensive portrait vector Calculate the degree of difference between industries and identify the structural differences between industries in various dimensions;
[0097] A time series is constructed based on the comprehensive development index, and the cumulative deviation test is used to identify inflection points in industrial development and determine the evolution trend of the industry. The expression is as follows:
[0098] ;
[0099] ;
[0100] in: For a moment The cumulative deviation; For industry At any moment The comprehensive development index; The average value for the base period; The moment when the absolute value of the cumulative deviation is maximized represents the inflection point of industrial development;
[0101] Based on the industrial load utilization rate, a green transformation index is determined, and its expression is as follows:
[0102] ;
[0103] in: As a green transformation index; This represents an increase in industrial load utilization. The percentage of self-generated clean electricity or green electricity; To the extent to which they participate in orderly electricity use and demand response; The change in electricity consumption per unit output; , , , All are weighting coefficients;
[0104] Based on enterprise electricity consumption, an intelligent transformation index is determined, and its expression is as follows:
[0105] ;
[0106] in: For intelligent transformation index; , These represent the changes in the peak-to-valley ratio and the monthly fluctuation coefficient, respectively. This represents the increase in the proportion of steady-state load at night. This represents the increase in load utilization. , , , All are weighting coefficients;
[0107] Based on the trajectories of the green transformation index and the intelligent transformation index, the stage of industrial transformation is determined, and its expression is as follows:
[0108] ;
[0109] in: For a moment industry The current stage of transformation; , Thresholds for dividing stages, .
[0110] A quantitative evaluation system for industry label generation and development trends, applied to the above-mentioned method, the system comprising:
[0111] The semantic tag generation module is used to construct a structured industry tag system that integrates multiple policy sources and generate a set of candidate semantic tags for enterprises.
[0112] The feature fingerprint acquisition module is used to collect enterprise electricity consumption data, extract power load features, and obtain enterprise power load feature fingerprint vectors and industry fingerprint benchmarks.
[0113] The industry enterprise attribution determination module is used to determine the optimal label for enterprise industry attribution based on the enterprise's power load characteristic fingerprint vector and industry fingerprint benchmark, and to traverse the candidate semantic label set. The optimal label for enterprise industry attribution is corrected for consistency deviation through text-power deviation to obtain a list of industry enterprises.
[0114] The multi-dimensional evaluation module is used to construct multi-dimensional evaluation indicators for the power industry for the set of enterprises to which each optimal label belongs in the list of industrial enterprises, and to determine the comprehensive profile vector and comprehensive development index.
[0115] The development trend identification module is used to identify structural differences in various dimensions of the industry based on comprehensive profile vectors; identify inflection points in industrial development based on comprehensive development indices; determine green and intelligent transformation indices; and output industrial evolution trends and industrial transformation stages.
[0116] An industry label generation and development trend quantitative evaluation device, the device including a processor and a memory;
[0117] The memory is used to store computer program code and to transmit the computer program code to the processor;
[0118] The processor is used to execute the above-mentioned method for quantitative evaluation of industry label generation and development trend according to the instructions in the computer program code.
[0119] A computer-readable storage medium storing computer-executable instructions, which, when executed on a computer, implement the aforementioned method for quantitative evaluation of industry label generation and development trends.
[0120] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0121] This invention discloses a method, system, and device for generating industry labels and quantitatively evaluating development trends. The method first constructs a multi-source structured industry label system and generates a set of candidate semantic labels. Then, it collects electricity consumption data, extracts enterprise electricity load characteristic fingerprint vectors and industry fingerprint benchmarks, and determines the optimal label for enterprise industry affiliation based on this. It then iterates through the candidate semantic label set to correct for consistency discrepancies, obtaining a list of industry enterprises. Next, it constructs multi-dimensional electricity evaluation indicators for the enterprise set in the list, determines a comprehensive profile vector and a comprehensive development index, and identifies structural differences and development inflection points. Finally, it determines the green and intelligent transformation index and outputs the industry evolution trend and transformation stage. In application, this design solves the problem of mismatched industry classification standards by constructing a computable industry label system that integrates multi-source policies. It also achieves accurate industry labeling for large-scale enterprises by integrating textual semantics and electricity load fingerprints in a multi-modal judgment method. Furthermore, it builds a multi-dimensional electricity evaluation indicator system, breaking through the limitations of single indicators, while simultaneously assessing the green and intelligent transformation process of industries, accurately quantifying industry development trends, and providing reliable support for industry regulation. Attached Figure Description
[0122] Figure 1 This is a flowchart of the method of the present invention.
[0123] Figure 2 This is a bar chart comparing the marking accuracy of different methods in Embodiment 1 of the present invention.
[0124] Figure 3 This is a schematic diagram of the multi-dimensional integrated industry profiling radar in Embodiment 1 of the present invention.
[0125] Figure 4 This is a schematic diagram of industrial transformation trend assessment in Embodiment 1 of the present invention.
[0126] Figure 5 This is a system structure diagram of the present invention.
[0127] Figure 6 This is a structural diagram of the device of the present invention.
[0128] In the diagram: Semantic label generation module 1, Feature fingerprint acquisition module 2, Industry and enterprise affiliation determination module 3, Multidimensional evaluation module 4, Development trend identification module 5, Processor 6, Memory 7, Computer program code 71. Detailed Implementation
[0129] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0130] Example 1:
[0131] See Figure 1 A quantitative evaluation method for the generation and development trend of industry labels, including:
[0132] Construct a structured industry labeling system that integrates multiple policy sources, and generate a set of candidate semantic labels for enterprises;
[0133] Furthermore, the construction of a structured industrial labeling system integrating multi-source policies specifically includes:
[0134] In this embodiment, since the definitions of various industries are scattered across multiple policy documents, are independent of each other, and have different definitions, they cannot be directly used to determine industry affiliation. Therefore, it is necessary to integrate policy documents from multiple sources. The integrated policy sources include at least the Strategic Emerging Industries Catalogue, Advantageous and Characteristic Industries, Economic Industry Classification Standards, Industrial Structure Adjustment Guidance Catalogue, and regional industrial planning documents. Among them, the Strategic Emerging Industries Catalogue covers nine emerging industries, including next-generation information technology, high-end equipment manufacturing, new materials, biomedicine, new energy, new energy vehicles, energy conservation and environmental protection, digital creativity, and high-tech services. Under next-generation information technology, it is further subdivided into technology / product labels such as artificial intelligence and optical communication. The Advantageous and Characteristic Industries Catalogue covers eight advantageous and characteristic industries, including chemicals, automobiles, metallurgy, textiles, building materials, food, equipment manufacturing, and electronic information.
[0135] Integrate policy documents from multiple sources, and denote the policy source set as […]. And each policy source Corresponding to a priori priority; where: The number of policy sources; used to characterize the authority and timeliness of the policy in industry definition, the larger the value, the higher the priority of the policy's judgment.
[0136] The industry labeling system is modeled as a hierarchical directed acyclic graph. ,in A collection of tag nodes. The set of directed edges representing hierarchical relationships;
[0137] Based on the sub-industry, each tag node in the tag node set Associating ternary features, for example, tag nodes are divided into three levels: "major industry category → sub-industry → technology / product tag". Under the major industry category of "new generation information technology", there are sub-industries such as "artificial intelligence" and "optical communication". Specific technology and product tags are then attached to these sub-industries to obtain tag feature triplets, the expression of which is as follows:
[0138] ;
[0139] in: For the first Feature triples of each label node; This is a semantic set of tags, representing a set of semantic keywords and technical descriptions of tags, used to characterize the core technologies and product features of the industry; This is a mapping set of industry codes, representing the mapping set from labels to economic industry classification codes, used to establish the correspondence between policy industries and industry codes; The policy attribute vector to which the label belongs represents the policy attribute vector, which includes at least new / old industry attributes, strategic emerging / advantageous and distinctive attributes, and green and intelligent transformation-related attributes. Through this triple, each abstract policy industry definition is transformed into a readable and matchable feature carrier, thereby bridging the mapping between "policy text and computable label".
[0140] If multiple policy sources assign inconsistent policy attributes to the same tag node (e.g., whether the industry is a new or old industry, a strategic emerging industry or a characteristic industry), the policy attribute of the tag node is determined based on a priori priority weighting; the expression for the priori priority weighting is as follows:
[0141] ;
[0142] in: for The policy attribute to which it belongs; This is the set of possible categories for policy attributes; As a policy source For tag nodes In terms of the policy attributes to which they belong The category determination given above; The category with the highest weighted votes is taken as the final attribute value for the label. The business logic of this formula is: when policy interpretations conflict, the more authoritative and recent policy judgment shall prevail, thereby providing a unified and calculable classification standard for new and old industries, strategic emerging industries, and advantageous and distinctive industries.
[0143] Since different policy sources may define the same industry in terms of coverage, this scheme defines the policy confidence level for each label node to quantify the degree to which a certain label node is jointly identified by multiple policy sources. The expression is as follows:
[0144] ;
[0145] in: For the first Policy confidence level of each labeled node; As a policy source The defined set of tags; For indicator functions, when the tag node If included in the policy source, use 1; otherwise, use 0.
[0146] The denominator in the formula is the sum of all policy source priorities, used to normalize the confidence level to... , The higher the confidence level, the more authoritative policy sources jointly recognize the label, and the more reliable its industry definition. This confidence level has two clear uses to avoid becoming an isolated quantity: first, it is used to sort and prune parallel labels at the same level in the labeling system; second, it serves as the policy authority score weight for subsequent rule matching, so that labels jointly recognized by multiple authoritative policies receive higher prior priority in the final labeling.
[0147] This solution utilizes a structured industry labeling system built through the integration of multiple policy sources. It transforms heterogeneous policy texts from various sources, such as the Strategic Emerging Industries Catalogue, Economic Industry Classification, Industrial Structure Adjustment Guidance Catalogue, and Regional Industrial Planning, into a tiered, computable labeling system with policy confidence and attribution attributes. Furthermore, it resolves conflicts between multiple sources through priority-weighted voting, fundamentally addressing the long-standing mismatch between electricity consumption classification and policy industry standards in the power industry. This enables precise differentiation between emerging sub-sectors such as artificial intelligence and optical communication, as well as between new and old industries, on the data side.
[0148] A structured industry labeling system is formed by the label feature triples of the label nodes, the policy attributes to which they belong, and the policy confidence level.
[0149] Furthermore, the candidate semantic tag set for generating enterprises specifically includes:
[0150] In this embodiment, when faced with a large number of enterprises and overlapping businesses, the question arises of how to accurately understand the actual production and operation content of each enterprise and map it to the tag system. Traditional keyword matching can only perform character-level comparison and cannot understand semantics. Therefore, this solution relies on artificial intelligence entities, that is, processing units with a large language model as the core, to perform semantic analysis on enterprise information.
[0151] For any enterprise, collect multi-source information fields, including at least core technology description, core product description, main business description, and auxiliary fields such as business registration information and electricity user name. Concatenate the above fields into enterprise semantic description text, and map it into enterprise semantic representation vector by a large language model. The expression is as follows:
[0152] ;
[0153] in: It is a semantic representation vector; Description of core technologies; Description of the core product; Description of main business; Indicates the ordered concatenation of text fields; Represents the semantic encoding mapping of a large language model;
[0154] The tag semantic set of each tag node is uniformly encoded and mapped to obtain the tag semantic vector. The two are generated by the same mapping, so they are in the same semantic space and are comparable, which is a prerequisite for subsequent calculation of semantic matching score.
[0155] Based on this, the text semantic matching score between the enterprise and the tag node is determined, and its expression is as follows:
[0156] ;
[0157] ;
[0158] in: For text semantic matching; For the label semantic vector; As a measure of directional consistency; It is the vector norm 2; The is a very small positive number to avoid the denominator being zero; the fraction within the parentheses represents the cosine similarity between the two, with a range of values of . By using a linear transformation of "adding 1 and then multiplying by one-half", it is mapped to an interval. ,make The value has a probabilistic meaning: the closer it is to 1, the more semantically the company's actual business matches the industry defined by the label. While giving a matching score, the AI agent outputs a preliminary judgment on the company's position in the industrial chain (upstream, midstream, or downstream) and a preliminary judgment on its technological transformation as subsequent auxiliary inputs.
[0159] To narrow down the judgment range and suppress noise, a candidate semantic tag set is obtained based on the text semantic matching score and filtered by a preset semantic threshold. Only tags with a matching score not lower than the threshold are retained, and the expression is as follows:
[0160] ;
[0161] in: For the set of candidate semantic labels; To preset semantic thresholds, ; A larger value results in a more concise candidate set but may miss some candidates; a smaller value results in a more comprehensive candidate set but may introduce noise. Therefore, the value can be set in conjunction with the tag hierarchy. This is the candidate semantic tag set. It retains several of the most likely industry affiliations at the semantic level, while also preserving competitive candidates for subsequent power fingerprint verification, thereby avoiding premature hard judgments by a single modality when information is insufficient.
[0162] Collect enterprise electricity consumption data, extract power load characteristics, and obtain enterprise power load characteristic fingerprint vectors and industry fingerprint benchmarks;
[0163] In this embodiment, since the text determination lacks objective physical evidence, it is necessary to extract features that can stably characterize the industry attributes from the measured electricity consumption data. Due to differences in production processes, equipment composition and production organization methods, different industries exhibit significantly different but relatively stable patterns in their electricity consumption behavior in terms of load curve shape, utilization efficiency, time distribution and seasonal fluctuations. In this scheme, this is referred to as the enterprise power load characteristic fingerprint. This fingerprint comes from the measured electricity signal, is difficult to modify artificially, and can be used as an objective basis for verifying the text attribution.
[0164] Furthermore, the process of collecting enterprise electricity consumption data, extracting electricity load characteristics, and obtaining enterprise electricity load characteristic fingerprint vectors and industry fingerprint benchmarks specifically includes:
[0165] Establish a company The electricity metering sequence is as follows With fixed step size (For example, sampling at 96 points daily, i.e., one point every 15 minutes) and recording its monthly electricity consumption sequence as follows. , The following features are extracted from this scheme:
[0166] Obtain enterprise electricity consumption data, determine annual electricity consumption and load utilization rate, and use these as a scale quantity, expressed as follows:
[0167] ;
[0168] ;
[0169] in: Load utilization rate, characterizing the efficiency of energy use in production and the level of capacity utilization. The closer the value is to 1, the more stable the company's electricity consumption and the more fully the equipment is utilized, which corresponds to a more continuous and automated production mode. The closer the value is to 0, the more peak electricity consumption and the more idle equipment there is. The annual electricity consumption is obtained by integrating (discretely summing) the load curve over the statistical period; This represents the peak load. This represents the total sampling time. This refers to the electricity metering sequence. To use a fixed sampling step size;
[0170] To obtain the electricity load of a company on weekdays and non-weekdays, and determine the proportion of weekdays, the expression is as follows:
[0171] ;
[0172] in: The percentage of working days represents the temporal pattern of production organization. If a company operates continuously without interruption, the workload on weekdays and non-weekdays is similar, and the ratio approaches 0.5. When production is concentrated on weekdays and production is suspended on holidays, the ratio approaches 1. , These represent the average workload for weekdays and non-weekdays, respectively. It is a very small positive number;
[0173] To determine the peak-to-valley ratio, obtain the electricity load during peak and off-peak hours for the enterprise. The expression is as follows:
[0174] ;
[0175] in: The peak-to-valley ratio, representing the intraday load fluctuation characteristics, is calculated by comparing the peak and trough periods. The smaller the value, the more balanced the electricity consumption within the day, which often corresponds to automated continuous production or intelligent energy consumption behavior of actively participating in orderly electricity consumption and demand response. , These represent the average load during peak and off-peak hours of the day, respectively.
[0176] To obtain the seasonal patterns of enterprise electricity consumption and determine seasonal factors, the expression is as follows:
[0177] ;
[0178] in: Seasonal factors characterize the seasonal patterns of industrial electricity consumption. The larger the value, the stronger the seasonality of electricity consumption (such as industries that are significantly affected by temperature or farming seasons); the closer it is to 1, the more even the electricity consumption is throughout the year. This represents the maximum monthly electricity consumption throughout the year. For the first Monthly electricity consumption sequence for the month, ;
[0179] Obtain the company's monthly electricity consumption data and determine the monthly fluctuation coefficient, the expression of which is as follows:
[0180] ;
[0181] ;
[0182] in: This is the monthly fluctuation coefficient, representing production stability. The smaller this value, the more stable the production throughout the year and the smaller the fluctuations. This represents the average monthly electricity consumption throughout the year.
[0183] Based on load utilization rate, weekday percentage, peak-to-valley ratio, seasonality factors, and monthly fluctuation coefficient, a characteristic fingerprint vector of enterprise power load is constructed, the expression of which is as follows:
[0184] ;
[0185] in: For enterprises Enterprise power load characteristic fingerprint vector;
[0186] It is important to note that the enterprise's power load characteristic fingerprint vector It consists only of dimensionless morphological characteristics such as load utilization rate, proportion of working days, peak-to-valley ratio, seasonal factors, and monthly fluctuation coefficient, and is not included in annual electricity consumption. With peak load The reason for these two scales is that fingerprints used to determine a company's industry affiliation should depict "electricity usage patterns" rather than "electricity usage volume." Within the same industry, large and small companies differ greatly in size, but their electricity usage patterns are similar. Including scale in the fingerprint would make the fingerprints of companies of different sizes within the same industry incomparable, thus interfering with the affiliation determination. Therefore, annual electricity consumption... With peak load This fingerprint is then retained separately for subsequent scale-based evaluation, and this distinction ensures the effectiveness of fingerprint identification in determining industry affiliation.
[0187] To eliminate the differences in dimensions and orders of magnitude among various morphological features, the fingerprint vector of enterprise power load characteristics is normalized. Starting from seed sample enterprises whose industry affiliation has been confirmed by authoritative channels, the centroid and covariance of the fingerprint distribution for each label node are estimated as the industry fingerprint benchmark, and their expressions are as follows:
[0188] ;
[0189] ;
[0190] ;
[0191] in: The centroid represents the typical electrical fingerprint of this industry; Covariance characterizes the degree of dispersion of the fingerprint across each feature dimension and the correlation between dimensions; For seed sample set; The number of samples in the set; This is the normalized enterprise electricity load characteristic fingerprint vector. ; , These represent the lower and upper bounds of each feature within the overall enterprise sample (calculated component by component). Thus, each industry obtains a quantifiable "electricity consumption profile benchmark" in the electricity feature space, serving as the basis for subsequent calculations.
[0192] Based on the enterprise's power load characteristic fingerprint vector and industry fingerprint benchmark, the optimal label for the enterprise's industry affiliation is determined, and the candidate semantic label set is traversed. The consistency deviation of the optimal label for the enterprise's industry affiliation is corrected by the text-power deviation, and a list of industry enterprises is obtained.
[0193] In this embodiment, since single-modal text semantic matching has a high mislabeling rate when there are overlaps in enterprise business and discrepancies between name and reality, and cannot automatically detect errors, the key idea in this solution is to introduce measured power load fingerprints as independent physical evidence to cross-verify and correct text judgments. The physical basis is that "electricity consumption patterns do not lie." Even if an enterprise is vague or exaggerated in its text description, its actual production activities will still leave an objective mark on the electricity consumption fingerprint.
[0194] Furthermore, based on the enterprise's power load characteristic fingerprint vector and industry fingerprint benchmark, the optimal label for enterprise industry affiliation is determined, and the candidate semantic label set is traversed. The optimal label for enterprise industry affiliation is then corrected for consistency deviations through text-power deviation to obtain a list of industry enterprises. Specifically, this includes:
[0195] For enterprises Each candidate label Based on the normalized enterprise power load characteristic fingerprint vector, Mahalanobis distance is used to measure the deviation between the enterprise fingerprint and the industry benchmark, and it is transformed into a power consistency score through exponential mapping, the expression of which is as follows:
[0196] ;
[0197] in: For power consistency; This is the normalized enterprise electricity load characteristic fingerprint vector; the part in parentheses represents the enterprise fingerprint relative to the industry. The squared Mahalanobis distance of the fingerprint distribution, The inverse of the covariance matrix; the business logic behind using Mahalanobis distance instead of Euclidean distance in this scheme is that it automatically adjusts the weights of different feature dimensions based on the discrete structure of the fingerprint distribution of each industry, so that features that are more discriminative (with smaller variance) for that industry receive greater weight, thereby more accurately characterizing "whether the company's electricity consumption pattern conforms to the typical pattern of that industry", after exponential mapping. The closer the value is to 1, the more closely the enterprise's electronic fingerprint matches the industry.
[0198] Meanwhile, this scheme retains the rule matching score based on policy confidence and the label rule base as a strong prior constraint to obtain a policy authority score, which is then weighted by policy confidence to reflect policy authority. The expression is as follows:
[0199] ;
[0200] in: To give the policy authority points; For the first Policy confidence of each labeled node Because the value inside the parentheses belongs to and Therefore ,and , In the same place The dimensionality ensures that the three can be weighted and integrated. This formula reflects the authority of the policy in the labeling process. Labels that are jointly recognized by multiple authoritative policies receive higher weighting for their rule matching scores.
[0201] The comprehensive attribution score of an enterprise to a tag node is obtained by weighting and fusing text semantic matching score, power consistency score, and policy authority score. The expression is as follows:
[0202] ;
[0203] in: The overall attribution score; , , All are weighting coefficients. and ensure The weight can be adjusted based on the label level and data availability (e.g., increase the weight when the power data is complete). Since the comprehensive attribution score incorporates three aspects of evidence—textual semantic matching score, power consistency score, and policy authority score—these three aspects corroborate and constrain each other, thus overcoming the defect of single-modal judgment being easily misled by one-sided noise.
[0204] Based on the overall affiliation score, the optimal label for enterprise industry affiliation is determined by the highest overall affiliation score, and its expression is as follows:
[0205] ;
[0206] in: For enterprises The optimal label; The confidence level of the corresponding policy attribute;
[0207] To determine the text-electricity deviation of the optimal label for enterprise industry affiliation, consistency deviation correction is performed to automatically detect and correct mislabeling. The expression is as follows:
[0208] ;
[0209] in: For text-electrical deviation, This is used to measure the degree of disagreement between the two types of evidence; , For enterprises For the optimal label Text semantic matching score and power consistency score;
[0210] This scheme adopts the following directional correction criterion: A high text matching score coupled with a low power consistency score is achieved if and only if both conditions are met. , and If so, the enterprise is deemed to have a suspected mismatch between its name and reality, triggering a consistency deviation correction; among which... , , These are the high matching threshold, low consistency threshold, and consistency deviation threshold, respectively.
[0211] The consistency deviation correction process is as follows: the best label for enterprise industry affiliation is excluded, and the enterprise is regressed to the candidate label with the second highest comprehensive score based on the comprehensive affiliation score, and its text-electricity deviation is recalculated; if the above triggering conditions are still met, the regression continues until a candidate label no longer triggers correction (i.e. passes the consistency check) or the candidate semantic label set is traversed, and the list of industry enterprises is output; if no candidate passes the check after traversal, the enterprise is output to the manual review queue, and its affiliation judgment is downgraded as a whole; this mechanism constitutes a closed loop with a clear termination condition, ensuring that the correction process will inevitably converge and will not lead to an infinite loop.
[0212] The reason why this scheme only triggers correction for the direction of "high text, low power" and not "low text, high power" is that the former corresponds to the situation where the company exaggerates or affiliates with a certain high-tech industry in the text, but its actual power consumption does not support the attribution, which is a mislabeling that needs to be corrected; the latter is usually only caused by the company's non-standard text description, not a discrepancy between name and reality, and the attribution can be naturally corrected in the comprehensive attribution score by the power evidence, without the need for special correction.
[0213] This solution incorporates power load characteristic fingerprints into enterprise industry attribution determination, and proposes a multimodal labeling method that integrates text and power consistency. By weightedly fusing the text semantic matching score of a large language model with the power fingerprint consistency score based on Mahalanobis distance, and designing a consistency deviation correction mechanism with clear directionality and termination conditions, this method upgrades the single-modal, misjudgment-prone, and unself-correcting text classification into a multimodal technical method supported by objective physical evidence that can automatically detect and correct "name-reality mismatches." This enables high-accuracy and low-mislabeling automatic labeling on a large enterprise scale.
[0214] For the set of enterprises in the industry to which each optimal label belongs in the list of industrial enterprises, construct a multi-dimensional evaluation index for the power industry, and determine the comprehensive profile vector and comprehensive development index.
[0215] In this embodiment, to address the challenge of moving beyond a single electricity consumption indicator and quantifying industry development trends from multiple dimensions, this solution addresses each industry... (i.e., the final label belongs to) Enterprise Collection We will construct evaluation indicators based on four dimensions: scale, growth, efficiency, and time and space.
[0216] Furthermore, for each set of enterprises in the industrial enterprise list whose optimal label belongs to an industry, a multi-dimensional evaluation index for the power industry is constructed to determine the comprehensive profile vector and the comprehensive development index, specifically including:
[0217] A scale-based evaluation index is constructed to characterize the industry size, and its expression is as follows:
[0218] ; ;
[0219] in: This represents the electricity consumption of the industry (summing the annual electricity consumption of enterprises within the industry). To the maximum load of the industry; Simultaneity rate coefficient, This is used to correct for the physical fact that "the maximum load of each enterprise does not occur at the same time", and to avoid directly summing up and overestimating the actual peak of the industry; For industry serial numbers;
[0220] A growth-dimensional evaluation index is constructed to characterize industry vitality, and its expression is as follows:
[0221] ; ;
[0222] in: This represents the year-on-year growth rate of industrial electricity consumption; The contribution of industrial electricity consumption increment to the total industrial electricity consumption increment is used to identify the key industries that contribute the most to regional economic growth. , industries In the Year and the Annual electricity consumption; , industries In the Year and the Annual electricity consumption;
[0223] Efficiency-based evaluation indicators are constructed to characterize the level of production organization. To ensure consistency between the aggregation scope and the scale dimension, the indicators of each enterprise are weighted and aggregated to the industry level according to their electricity consumption. The expression is as follows:
[0224] ; ;
[0225] in: For industrial load utilization rate; The percentage of working days in the industry; together, they reflect the overall energy efficiency and production continuity of the industry. The percentage of weekdays with the highest workload;
[0226] A spatial dimension evaluation index is constructed to characterize the spatial organization features of the industry, and its expression is as follows:
[0227] ; ;
[0228] in: For user concentration, The Herfindahl-Hirschman index is used, and the larger the value, the more concentrated the industry is in a few leading companies; For regional distribution, information entropy is used as a measure. The larger the value, the more evenly the industries are distributed in space, and the smaller the value, the more concentrated they are in a few areas. For industry In the region Electricity consumption; For a set of regions.
[0229] In addition, this scheme uses seasonal factors, peak-to-valley ratio and monthly fluctuation coefficient, which are obtained by weighting the corresponding indicators of enterprises in the industry according to electricity consumption, as descriptive indicators to supplement the characterization of the industry's electricity consumption characteristics.
[0230] To create a comprehensive industry profile that allows for horizontal comparison, the evaluation indicators for scale, growth, efficiency, and spatial dimensions are normalized and weighted to obtain a comprehensive profile vector and a comprehensive development index, expressed as follows:
[0231] ;
[0232] ;
[0233] in: For industry The comprehensive profile vector (which can be directly used for radar chart visualization); For industry The comprehensive development index (scalar). ; For industry Next The normalized standardized index values of the evaluation indicators ; As the indicator weight, Each indicator has a weight, and the sum of all weights is 1; This represents the total number of evaluation dimensions and indicators. As a result, each industry obtains a multi-dimensional quantitative profile encompassing scale, growth, efficiency, and spatiotemporal dimensions, serving as input for subsequent analysis.
[0234] This solution constructs a multi-dimensional evaluation index system for power big data, calculates industry indicators from four dimensions—scale, growth, efficiency, and spatiotemporal—and synthesizes a comprehensive profile vector and a comprehensive development index. It is supplemented by descriptive indicators such as seasonal factors, peak-to-valley ratio, and monthly fluctuation coefficient, breaking through the limitations of a single electricity consumption indicator. This solution can depict the development trend of the industry in a multi-dimensional and objective way and effectively distinguish between natural factors and real industry changes.
[0235] Based on comprehensive profile vector recognition, the structural differences of the industry in various dimensions are identified. Based on the comprehensive development index, the inflection point of industrial development is identified, and the green and intelligent transformation index is determined. The industrial evolution trend and industrial transformation stage are output.
[0236] In this embodiment, in order to address the problem of how to dynamically monitor industries and quantify the effectiveness of green and intelligent transformation of traditional industries, this solution is developed from three levels: horizontal comparison, vertical analysis, and transformation quantification.
[0237] Furthermore, the method of identifying structural differences in various dimensions of the industry based on comprehensive profile vector recognition; identifying inflection points in industry development based on comprehensive development indices; determining green and intelligent transformation indices; and outputting industry evolution trends and transformation stages specifically includes:
[0238] At the level of horizontal comparison, based on comprehensive profile vectors Calculate the differences between industries, distinguish between strategic emerging industries, advantageous and distinctive industries and traditional industries, and identify structural differences in various dimensions;
[0239] Specifically, after normalizing the evaluation indicators in the scale, growth, efficiency, and spatial dimensions, a comprehensive industry profile vector is constructed according to the corresponding weights. For any two industries, the inter-industry difference is calculated based on the difference between the standardized indicator values of each dimension in their comprehensive profile vectors; where the inter-industry difference can be represented by weighted absolute distance, weighted Euclidean distance, or the negative of cosine similarity; for example, the expression for weighted calculation using indicator weights is as follows:
[0240] ;
[0241] in: For industry , The degree of difference between them.
[0242] Furthermore, the comprehensive profile vector of the industry to be identified is compared with the benchmark profile vectors or category center vectors corresponding to strategic emerging industries, advantageous and distinctive industries, and traditional industries, respectively. The category with the smallest difference or the highest similarity is determined as the type to which the industry belongs. For the identification of structural differences, the proportion of each dimension's difference component to the total difference between industries is calculated. When the difference contribution rate of a certain dimension exceeds a preset threshold, that dimension is determined as the main dimension causing structural differences between industries, and the corresponding size difference, growth difference, efficiency difference, or spatial distribution difference is output. Thus, the comprehensive profile vector directly participates in industry classification, difference calculation, and difference dimension positioning, realizing horizontal comparative analysis of industries based on the comprehensive profile vector.
[0243] At the longitudinal analysis level, a time series is constructed based on the comprehensive development index, and the cumulative deviation test is used to identify inflection points in industrial development and determine the evolution trend of the industry. The expression is as follows:
[0244] ;
[0245] ;
[0246] in: For a moment The cumulative deviation; For industry At any moment The comprehensive development index; The average value for the base period; The moment when the absolute value of the cumulative deviation is maximized represents the most likely inflection point in industrial development; this mechanism can promptly identify and warn when there is a substantial shift in the trend of industrial development.
[0247] Specifically, an industrial development time series is constructed based on the comprehensive development index in chronological order. Using the average of the base period as a reference for stable development, the deviation of the comprehensive development index relative to the base period average is calculated period by period and accumulated to obtain a cumulative deviation sequence. The position where the absolute value of the cumulative deviation reaches its maximum is identified as a candidate inflection point. The average value, growth slope, and fluctuation range of the comprehensive development index before and after the candidate inflection point are compared. When the absolute value of the cumulative deviation corresponding to the candidate inflection point exceeds a preset deviation threshold, and the direction of change of the comprehensive development index after the inflection point remains consistent over several consecutive observation periods, a substantial turning point in the industrial development trend is determined, triggering an early warning. If the average value or growth slope of the comprehensive development index increases after the inflection point, an early warning of accelerated or improved development is output; if the average value or growth slope decreases after the inflection point, an early warning of slowed development or risk is output. Furthermore, early warning levels are classified according to the magnitude of the cumulative deviation, the magnitude of the slope change before and after the inflection point, and the length of the duration, thus forming a complete early warning process consisting of inflection point identification, turning point confirmation, early warning direction determination, and early warning level output.
[0248] At the quantitative level of transformation, this plan defines a green transformation index and an intelligent transformation index based on observable changes in the power sector, as follows:
[0249] Based on industrial load utilization rate, a green transformation index is determined to characterize the progress of industries in energy efficiency and clean energy use. Its expression is as follows:
[0250] ;
[0251] in: As a green transformation index; This represents the increase in industrial load utilization (normalized value, reflecting energy efficiency improvement). The percentage of self-generated clean electricity or green electricity; To the extent to which they participate in orderly electricity use and demand response; This represents the change in electricity consumption per unit of output (a decrease is a positive contribution, hence the negative sign). , , , All are weighting coefficients;
[0252] It should be noted that the items in the green transformation index are included based on data availability: when the metering data of the enterprise's self-provided clean power supply or the industry's added value data are available, they are included accordingly. , When the aforementioned auxiliary data is unavailable, the corresponding weights are reset to zero and the remaining available items are renormalized so that the index can still be calculated even when only load curve data is available. , The calculation of pure electric characteristic terms ensures that the method can be implemented under different data conditions without generating calculation gaps.
[0253] Based on enterprise electricity consumption, an intelligent transformation index is determined to characterize the industry's progress in automation and continuous production. All its components can be directly calculated from the load curve, and its expression is as follows:
[0254] ;
[0255] in: For intelligent transformation index; , These represent the changes in the peak-to-valley ratio and the monthly fluctuation coefficient, respectively (a decrease in either indicates a positive contribution, reflecting more balanced and stable production, hence the negative sign). This represents the increase in the proportion of steady-state load at night (reflecting unmanned continuous production). This represents the increase in load utilization. , , , All are weighting coefficients and their sum is 1; since all items of the intelligent transformation index are derived from the load curve, they can be fully calculated when only electricity metering data is available.
[0256] Based on the trajectories of the green transformation index and the intelligent transformation index, the stage of industrial transformation is determined, and its expression is as follows:
[0257] ;
[0258] in: For a moment industry The current stage of transformation; , Thresholds for dividing stages, Finally, the effectiveness of the transformation can be quantified by the net increase in the transformation index during the observation period.
[0259] This plan assesses and quantifies the green and intelligent transformation of the power sector. It constructs a transformation index by analyzing changes in observable characteristics such as load utilization, peak-to-valley ratio, monthly fluctuation coefficient, and steady-state load ratio. Based on this index, it determines the transformation stage and quantifies the results, thus enabling dynamic tracking of the transformation process of traditional advantageous and distinctive industries.
[0260] In this embodiment, a forward-coupled closed loop is formed, consisting of "policy label - semantic candidate - load fingerprint - consistency fusion - multi-dimensional evaluation - transformation judgment". The outputs of each link are inputs to each other, the logic is rigorous and there are no breaks. It also supports online updates as policies and data are updated, forming a set of quantifiable, interpretable and reusable industrial monitoring and evaluation technology solutions, providing solid data support for industrial policy evaluation and economic operation regulation.
[0261] Furthermore, to verify the performance of different methods, the proposed method was compared with the pure keyword rule method and the pure large language model semantic method on the same samples, such as... Figure 2 And as shown in the table below:
[0262]
[0263] The above results show that the proposed method is significantly higher than the two baseline methods in terms of labeling coverage, attribution accuracy, and industry segmentation accuracy. The mislabeling rate is reduced to 5.7%, and the workload of manual review is reduced to about one-fifth of that of the keyword method. The fundamental reason is that the proposed method uses the measured enterprise power load fingerprint to perform cross-validation and deviation correction on the text judgment, which effectively identifies and corrects the "name-to-reality mismatch" that cannot be identified by the pure text method.
[0264] See Figure 3 In this embodiment, multi-dimensional indicators of each industry were further calculated and a comprehensive profile was synthesized. According to statistics, about 8,000 strategic emerging industry enterprises were selected from 50,000 enterprises, covering the aforementioned nine emerging industries, and enterprise matching of eight advantageous and distinctive industries was completed simultaneously. Figure 3 The radar charts compare a strategic emerging industry with a traditional industry across four dimensions: scale, growth, efficiency, and time and space. The strategic emerging industry leads significantly in growth and efficiency, while the traditional industry still holds an advantage in scale. This clearly demonstrates the superior characterization capabilities of the multi-dimensional evaluation method compared to a single electricity consumption indicator.
[0265] See Figure 4This embodiment further verifies the quantitative assessment capability of this solution for the green and intelligent transformation of traditional advantageous and distinctive industries. First, a high-energy-consuming basic raw material industry (one of the eight major advantageous and distinctive industries) was selected as the assessment object. Its electricity metering time series for five consecutive years were collected. Then, the annual green transformation index and intelligent transformation index were calculated, and the transformation stage was determined. Simultaneously, the evolution of key power characteristics such as load utilization rate, peak-to-valley ratio, and monthly fluctuation coefficient were tracked. The assessment results are shown in the table below:
[0266]
[0267] From the table above and Figure 4 It is evident that as the industry promotes technological transformation, its load utilization rate has increased from 0.51 to 0.72 year by year, the peak-to-valley ratio has decreased from 2.84 to 1.78, and the monthly fluctuation coefficient has decreased from 0.182 to 0.114. Correspondingly, the green transformation index and the intelligent transformation index have increased from 0.32 and 0.28 to 0.71 and 0.69, respectively, and the transformation stage has evolved from "start-up" to "acceleration" to "maturity." This indicates that the industry's production organization is shifting from extensive and intermittent to continuous, automated, and energy-efficient. For traditional single electricity consumption indicators, the industry's electricity consumption may not have increased significantly during this period, making it difficult to perceive the profound transformation taking place within it. However, this solution, through multi-dimensional quantification of load characteristics, clearly reveals the essence of the transformation of "stable electricity consumption but significantly improved power quality," proving the dynamic quantification capability and practical value of the proposed method for the transformation process.
[0268] Example 2:
[0269] See Figure 5 A quantitative evaluation system for the generation and development trend of industry labels, which is applied to the method described in Example 1, the system comprising:
[0270] Semantic tag generation module 1 is used to construct a structured industry tag system that integrates multiple policy sources and generate a set of candidate semantic tags for enterprises;
[0271] Feature fingerprint acquisition module 2 is used to collect enterprise electricity consumption data, extract power load features, and obtain enterprise power load feature fingerprint vector and industry fingerprint benchmark;
[0272] The industry enterprise attribution determination module 3 is used to determine the optimal label for enterprise industry attribution based on the enterprise's power load characteristic fingerprint vector and industry fingerprint benchmark, and to traverse the candidate semantic label set. The optimal label for enterprise industry attribution is corrected for consistency deviation through text-power deviation to obtain the list of industry enterprises.
[0273] Multidimensional evaluation module 4 is used to construct power multidimensional evaluation indicators for the set of enterprises in the industry to which each optimal label belongs in the list of industrial enterprises, and to determine the comprehensive profile vector and comprehensive development index.
[0274] The development trend identification module 5 is used to identify the structural differences of industries in various dimensions based on comprehensive profile vectors, identify the inflection points of industrial development based on comprehensive development index, determine the green and intelligent transformation index, and output the industrial evolution trend and industrial transformation stage.
[0275] Furthermore, for the specific implementation steps of the above modules, please refer to the corresponding description in Example 1, which will not be repeated here.
[0276] Example 3:
[0277] See Figure 6 A device for generating and quantitatively evaluating the development trend of industry labels, the device comprising a processor 6 and a memory 7;
[0278] The memory 7 is used to store computer program code 71 and to transmit the computer program code 71 to the processor 6;
[0279] The processor 6 is used to execute, according to the instructions in the computer program code 71, a quantitative evaluation method for the generation and development trend of industry labels as described in Embodiment 1.
[0280] This embodiment also includes a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed on a computer, the method for quantitative evaluation of industry label generation and development trend described in Embodiment 1 is implemented.
[0281] Generally, computer instructions for implementing the method of the present invention can be carried on any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable medium except for the signal itself which is temporarily propagating.
[0282] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0283] Computer program code for performing the operations of this invention can be written in one or more programming languages or a combination thereof. These programming languages include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. In particular, Python, suitable for neural network computation, and platform frameworks based on TensorFlow, PyTorch, etc., can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet using an Internet service provider).
[0284] The aforementioned equipment and non-transitory computer-readable storage media can be found in the detailed description of a quantitative evaluation method for the generation and development of industry labels and its beneficial effects, which will not be repeated here.
[0285] Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.
Claims
1. A method for quantitatively evaluating the generation and development trend of industry labels, characterized in that, include: Construct a structured industry labeling system that integrates multiple policy sources, and generate a set of candidate semantic labels for enterprises; Collect enterprise electricity consumption data, extract power load characteristics, and obtain enterprise power load characteristic fingerprint vectors and industry fingerprint benchmarks; Based on the enterprise's power load characteristic fingerprint vector and industry fingerprint benchmark, the optimal label for the enterprise's industry affiliation is determined, and the candidate semantic label set is traversed. The consistency deviation of the optimal label for the enterprise's industry affiliation is corrected by the text-power deviation, and a list of industry enterprises is obtained. For the set of enterprises in the industry to which each optimal label belongs in the list of industrial enterprises, construct a multi-dimensional evaluation index for the power industry, and determine the comprehensive profile vector and comprehensive development index. Based on the structural differences in the comprehensive portrait vector recognition industry across various dimensions; Identify inflection points in industrial development based on the comprehensive development index; determine green and intelligent transformation indices to output industrial evolution trends and industrial transformation stages.
2. The method for quantitatively evaluating the generation and development trend of industry labels according to claim 1, characterized in that: The construction of a structured industry labeling system integrating multi-source policies specifically includes: Constructing a policy source set And each policy source Corresponding to a priori priority ; The industry labeling system is modeled as a directed acyclic graph. ,in A collection of tag nodes. The set of directed edges representing hierarchical relationships; Based on the sub-industry, each tag node in the tag node set By associating the triplet features, we obtain the label feature triplet, which is expressed as follows: ; in: For the first Feature triples of each label node; A collection of label semantics; A set of mappings for industry codes; This represents the vector of the policy attribute to which the policy attribute belongs. If multiple policy sources assign inconsistent policy attributes to the same tag node, the policy attribute to which the tag node belongs is determined based on a priori priority weighting; the expression for the priori priority weighting is as follows: ; in: for The policy attribute to which it belongs; This is the set of possible categories for policy attributes; As a policy source For tag nodes In terms of the policy attributes to which they belong The category determination given above; Define the policy confidence level for each labeled node, quantifying the degree to which a labeled node is jointly identified by multiple policy sources. The expression is as follows: ; in: For the first Policy confidence level of each labeled node; As a policy source The defined set of tags; For indicator functions; A structured industry labeling system is formed by the label feature triples of the label nodes, the policy attributes to which they belong, and the policy confidence level.
3. The method for quantitatively evaluating the generation and development trend of industry labels according to claim 2, characterized in that: The candidate semantic tag set for generating enterprises specifically includes: The enterprise's multi-source information fields are collected and concatenated into an enterprise semantic description text, which is then mapped to an enterprise semantic representation vector. The expression is as follows: ; in: It is a semantic representation vector; Description of core technologies; Description of the core product; Description of main business; The tag semantic set of each tag node is uniformly encoded and mapped to obtain the tag semantic vector, and the text semantic matching score between the enterprise and the tag node is determined. The expression is as follows: ; ; in: For text semantic matching; For the label semantic vector; As a measure of directional consistency; It is the vector norm 2; It is a very small positive number; Based on text semantic matching scores, a set of candidate semantic tags is obtained by filtering through a preset semantic threshold, as shown in the following expression: ; in: For the set of candidate semantic labels; This is a preset semantic threshold.
4. The method for quantitatively evaluating the generation and development trend of industry labels according to claim 3, characterized in that: The process of collecting enterprise electricity consumption data, extracting electricity load characteristics, and obtaining enterprise electricity load characteristic fingerprint vectors and industry fingerprint benchmarks specifically includes: To obtain enterprise electricity consumption data and determine load utilization, the expression is as follows: ; ; in: For load utilization rate; Annual electricity consumption; This represents the peak load. This represents the total sampling time. This refers to the electricity metering sequence. To use a fixed sampling step size; To obtain the electricity load of a company on weekdays and non-weekdays, and determine the proportion of weekdays, the expression is as follows: ; in: The percentage of weekdays; , These represent the average workload for weekdays and non-weekdays, respectively. It is a very small positive number; To determine the peak-to-valley ratio, obtain the electricity load during peak and off-peak hours for the enterprise. The expression is as follows: ; in: This represents the peak-to-valley ratio between the peak and trough periods of the day. , These represent the average load during peak and off-peak hours of the day, respectively. To obtain the seasonal patterns of enterprise electricity consumption and determine seasonal factors, the expression is as follows: ; in: Seasonal factors; This represents the maximum monthly electricity consumption throughout the year. For the first Monthly electricity consumption sequence for the month, ; Obtain the company's monthly electricity consumption data and determine the monthly fluctuation coefficient, the expression of which is as follows: ; ; in: This is the monthly fluctuation coefficient; This represents the average monthly electricity consumption throughout the year. Based on load utilization rate, weekday percentage, peak-to-valley ratio, seasonality factors, and monthly fluctuation coefficient, a characteristic fingerprint vector of enterprise power load is constructed, the expression of which is as follows: ; in: For enterprises Enterprise power load characteristic fingerprint vector; The enterprise's power load characteristic fingerprint vector is normalized, and the centroid and covariance of the fingerprint distribution are estimated for each label node as the industry fingerprint benchmark. The expressions are as follows: ; ; ; in: Center of mass; For covariance; For seed sample set; This is the normalized fingerprint vector of enterprise power load characteristics. , These are the lower and upper bounds of the overall enterprise sample, respectively.
5. The method for quantitatively evaluating the generation and development trend of industry labels according to claim 4, characterized in that: The process involves determining the optimal label for enterprise industry affiliation based on enterprise power load characteristic fingerprint vectors and industry fingerprint benchmarks, traversing the candidate semantic label set, and correcting for consistency deviations in the optimal label for enterprise industry affiliation through text-power deviation to obtain a list of industry enterprises. Specifically, this includes: Based on the normalized enterprise power load characteristic fingerprint vector, Mahalanobis distance is used to measure the deviation between the enterprise fingerprint and the industry benchmark, and it is transformed into a power consistency score through exponential mapping, the expression of which is as follows: ; in: For power consistency; This is the normalized fingerprint vector of enterprise power load characteristics. Based on the rule matching score between policy confidence and the label rule base, and as a strong prior constraint, the policy authority score is obtained, and its expression is as follows: ; in: To give the policy authority points; For the first Policy confidence level of each labeled node; The comprehensive attribution score of an enterprise to a tag node is obtained by weighting and fusing text semantic matching score, power consistency score, and policy authority score. The expression is as follows: ; in: The overall attribution score; , , All are weighting coefficients; Based on the comprehensive affiliation score, the optimal label for a company's industry affiliation is determined, and its expression is as follows: ; in: For enterprises The optimal label; The confidence level of the corresponding policy attribute; To determine the text-electricity deviation degree of the optimal label for determining the enterprise's industry affiliation, consistency deviation correction is performed, and the expression is as follows: ; in: For text—electrical deviation; , For enterprises For the optimal label Text semantic matching score and power consistency score; like , and When this occurs, consistency deviation correction is triggered, excluding the optimal label for enterprise industry affiliation, and iterating through the candidate semantic label set until consistency deviation correction is no longer triggered, outputting a list of industry enterprises; where , , These are the high matching threshold, low consistency threshold, and consistency deviation threshold, respectively.
6. The method for quantitatively evaluating the generation and development trend of industry labels according to claim 5, characterized in that: The set of enterprises in the industrial enterprise list whose optimal label belongs to which industry is constructed, and a multi-dimensional evaluation index for the power industry is determined, including the comprehensive profile vector and comprehensive development index. The scale-dimensional evaluation index is constructed as follows: ; ; in: Electricity consumption for industry; To the maximum load of the industry; Simultaneity rate coefficient; Annual electricity consumption; This represents the peak load. For industry serial numbers; The growth dimension evaluation index is constructed, and its expression is as follows: ; ; in: This represents the year-on-year growth rate of industrial electricity consumption; The contribution of the increase in industrial electricity consumption to the total increase in total industrial electricity consumption; , industries In the Year and the Annual electricity consumption; , industries In the Year and the Annual electricity consumption; The efficiency dimension evaluation index is constructed, and its expression is as follows: ; ; in: For industrial load utilization rate; The percentage of working days in the industry; The percentage of weekdays with the highest workload; The spatial dimension evaluation index is constructed, and its expression is as follows: ; ; in: User concentration; It is distributed regionally; For industry In the region Electricity consumption; For a set of regions; By normalizing and weighting the evaluation indicators of scale, growth, efficiency, and spatial dimensions, a comprehensive profile vector and a comprehensive development index are obtained, and their expressions are as follows: ; ; in: For industry The comprehensive image vector; For industry The comprehensive development index; For industry Next The normalized standardized index values of the evaluation indicators; As the indicator weight; This represents the total number of evaluation dimensions and indicators.
7. The method for quantitatively evaluating the generation and development trend of industry labels according to claim 6, characterized in that: The structural differences in the comprehensive portrait vector recognition industry across various dimensions; Identifying inflection points in industrial development based on comprehensive development indices; Determine the green and intelligent transformation index to output industry evolution trends and transformation stages, specifically including: Based on comprehensive portrait vector Calculate the degree of difference between industries and identify the structural differences between industries in various dimensions; A time series is constructed based on the comprehensive development index, and the cumulative deviation test is used to identify inflection points in industrial development and determine the evolution trend of the industry. The expression is as follows: ; ; in: For a moment The cumulative deviation; For industry At any moment The comprehensive development index; The average value for the base period; The moment when the absolute value of the cumulative deviation is maximized represents the inflection point of industrial development; Based on the industrial load utilization rate, a green transformation index is determined, and its expression is as follows: ; in: As a green transformation index; This represents an increase in industrial load utilization. The percentage of self-generated clean electricity or green electricity; To the extent to which they participate in orderly electricity use and demand response; The change in electricity consumption per unit output; , , , All are weighting coefficients; Based on enterprise electricity consumption, an intelligent transformation index is determined, and its expression is as follows: ; in: For intelligent transformation index; , These represent the changes in the peak-to-valley ratio and the monthly fluctuation coefficient, respectively. This represents the increase in the proportion of steady-state load at night. This represents the increase in load utilization. , , , All are weighting coefficients; Based on the trajectories of the green transformation index and the intelligent transformation index, the stage of industrial transformation is determined, and its expression is as follows: ; in: For a moment industry The current stage of transformation; , Thresholds for dividing stages, .
8. A quantitative evaluation system for the generation and development trend of industry labels, characterized in that, The system is applied to the method according to any one of claims 1-7, the system comprising: The semantic tag generation module (1) is used to construct a structured industrial tag system that integrates multiple policy sources and generate a set of candidate semantic tags for enterprises; The feature fingerprint acquisition module (2) is used to collect enterprise electricity consumption data, extract power load features, and obtain enterprise power load feature fingerprint vector and industry fingerprint benchmark; The industry enterprise attribution determination module (3) is used to determine the optimal label of the enterprise industry attribution based on the enterprise power load characteristic fingerprint vector and the industry fingerprint benchmark, and to traverse the candidate semantic label set, and to perform consistency deviation correction on the optimal label of the enterprise industry attribution through text-power deviation, so as to obtain the list of industry enterprises. The multidimensional evaluation module (4) is used to construct a multidimensional evaluation index for the power industry for the set of enterprises to which each optimal label belongs in the list of industrial enterprises, and to determine the comprehensive profile vector and comprehensive development index. The development trend identification module (5) is used to identify the structural differences of industries in various dimensions based on the comprehensive profile vector; identify the inflection point of industrial development based on the comprehensive development index; determine the green and intelligent transformation index, and output the industrial evolution trend and industrial transformation stage.
9. A device for quantitatively evaluating the generation and development trend of industry labels, characterized in that: The device includes a processor (6) and a memory (7); The memory (7) is used to store computer program code (71) and to transmit the computer program code (71) to the processor (6). The processor (6) is used to execute the method for quantitative evaluation of the generation and development trend of industry labels as described in any one of claims 1-7 according to the instructions in the computer program code (71).
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed on a computer, implement the method for quantitative evaluation of industry label generation and development trends as described in any one of claims 1-7.
