CN-121981428-A - Power load deterministic characteristic quantification method and system based on multi-component heterogeneous data fusion

CN121981428ACN 121981428 ACN121981428 ACN 121981428ACN-121981428-A

Abstract

The invention discloses a method and a system for quantifying power load deterministic characteristics based on multi-element heterogeneous data fusion, and particularly relates to the field of power system data analysis and intelligent decision, wherein the method comprises the steps of collecting and fusing meteorological, calendar, macro economy and historical load data, predefining a multi-dimensional typical analysis scene based on regional characteristics, quantifying contribution degree of each influence factor by utilizing a BP neural network and a mean influence value algorithm, establishing a data-driven evaluation index system, dynamically constructing a standardized typical power analysis scene by combining a social activity mode and the evaluation system, quantifying a load change deterministic characteristic mode formed by the combined action of key factors under each scene through a data alignment and machine learning method, outputting numerical expression, and constructing a structured load deterministic characteristic knowledge base. The invention realizes objective and refined measurement of the internal law of the power load and provides reliable decision support for accurate scheduling and planning of the power system.

Inventors

WU KEMING
CHEN ZELONG
CAO SHUAI
YAO GUANGZHI
QU XINYAO

Assignees

国家电网公司东北分部

Dates

Publication Date: 20260505
Application Date: 20251211

Claims (8)

1. The method for quantifying the deterministic characteristics of the power load based on the fusion of the multi-component heterogeneous data is characterized by comprising the following steps: s1, acquiring and fusing multi-element heterogeneous data related to regional power loads, wherein the multi-element heterogeneous data specifically comprises meteorological data, calendar and holiday data, macroscopic economic data and historical load data, and defining a multi-dimensional 'typical analysis scene' initial set based on regional characteristics; S2, taking the data fused in the S1 as input, constructing a BP neural network model, carrying out variable analysis on the BP neural network model by applying a mean value influence value algorithm, and quantifying the contribution degree of each influence factor by calculating the mean influence value generated by the tiny disturbance of the numerical value of each input factor on the network output result to establish an electricity load change deterministic influence factor evaluation index system of a target area; s3, screening and optimizing the pre-defined initial set of typical analysis scenes in the S1 based on the social activity mode of the target area and combining with an evaluation index system, and designing a group of typical electricity utilization analysis scenes covering different time scales and regional characteristics as a standardized analysis framework for subsequent characteristic quantification; S4, fusing and aligning the multi-source heterogeneous data under a unified space-time reference, extracting characteristic variables highly related to the current scene, establishing a load change deterministic characteristic mode by the combined action of deterministic influence factors by using a machine learning method, and outputting a numerical expression; And S5, summarizing and storing the load deterministic characteristic quantification results in all typical scenes to form a structured regional power load deterministic characteristic knowledge base which is used for supporting accurate load prediction, operation planning and scheduling decision of the power system.
2. The method for quantifying the deterministic characteristic of the electric load based on the multi-element heterogeneous data fusion according to claim 1, wherein the typical analysis scene is defined based on a three-dimensional combination method of a seasonal type, a holiday mode and a regional characteristic, wherein the seasonal type is subdivided into spring, summer, autumn and winter, a dominant meteorological factor is marked by combining regional climate characteristics, the holiday mode comprises a working day, a weekend, a legal holiday, a folk-custom holiday, a policy activity day and a rest-regulating working day, the regional characteristic is divided into an industrial concentration area, a resident concentration area, a commercial core area, a public service area and a mixed functional area according to an industrial structure, and the load type is marked for each region.
3. The method for quantifying the deterministic characteristics of the electric load based on the multi-component heterogeneous data fusion according to claim 1, wherein the mean value influence value algorithm is specifically an improved MIV algorithm of layering disturbance and weight correction, layering disturbance design is carried out according to the numerical type of an input characteristic variable, relative disturbance of +/-5% is applied to continuous variable, category replacement disturbance is carried out on discrete variable, 0-1 overturning disturbance is carried out on Boolean variable, and the calculated original MIV value is corrected.
4. The method for quantifying the electric load deterministic characteristics based on the multi-component heterogeneous data fusion is characterized in that the typical electric analysis scene is a three-dimensional standardized scene frame which is formed by time scale, regional characteristics and influence intensity, wherein the time scale dimension comprises a short-time scene, a daily scene, a weekly scene and a medium-long-term scene, the regional characteristic dimension comprises an industrial concentration area, a resident concentration area, a commercial core area, a public service area and a mixed function area which are defined by S1, and the influence intensity dimension is divided into three grades of light influence, medium influence and heavy influence according to the sensitivity threshold of each index in an evaluation index system.
5. The method for quantifying the electric load deterministic features based on the multi-component heterogeneous data fusion is characterized in that a deterministic feature mode is adopted to achieve numerical expression of features by adopting a machine learning model differentiated from a classification scene to a regression scene, wherein a mixed model of CNN+random forest is adopted for the classification scene, a numerical result comprising a fluctuation type, occurrence probability and duration is output, a model of VMD decomposition+ BiGRU is adopted for the regression scene, a numerical result comprising a load predicted value, a prediction error range and a feature contribution ratio is output, and all feature quantized results are output in a standardized format of feature name-feature type-numerical value range-confidence.
6. The method for quantifying the power load deterministic features based on the heterogeneous data fusion of claim 1, wherein the load deterministic feature knowledge base adopts a structured design of a three-layer architecture and a multi-dimensional index, and the three-layer architecture comprises a data storage layer, an index service layer and an application interface layer and provides a standardized API interface comprising data query, feature call and model update.
7. The method for quantifying the power load deterministic characteristic based on the multi-component heterogeneous data fusion according to claim 1, wherein the load deterministic characteristic knowledge base further establishes a dynamic update mechanism of automatic update and manual audit, sets a monthly automatic update period and a quarterly manual audit mechanism, and triggers an emergency update flow when a major change occurs in an area.
8. A power load deterministic characteristic quantification system based on multi-component heterogeneous data fusion, for implementing the power load deterministic characteristic quantification method based on multi-component heterogeneous data fusion according to any one of claims 1-7, comprising: The data fusion and scene predefining module is used for collecting and fusing multi-element heterogeneous data related to regional power loads, including meteorological data, calendar and holiday data, macroscopic economic data and historical load data, and predefining a multi-dimensional typical analysis scene initial set based on regional characteristics; The factor contribution degree quantitative analysis module is internally provided with a BP neural network training unit and a mean value influence value MIV calculation unit, learns the nonlinear relation between the load and the influence factors through the BP neural network, inputs the factors through system disturbance and analyzes output change by utilizing the MIV calculation unit, and quantifies the contribution degree of each influence factor; The typical scene dynamic construction module is used for screening, verifying and reconstructing an initial scene set according to key factors of an evaluation index system and contribution degrees thereof and combining a social activity mode of a target area, and dynamically generating a group of standardized typical power analysis scenes covering different time scales and regional characteristics; The deterministic feature extraction module is used for carrying out space-time alignment and consistency processing on multi-source heterogeneous data under a standardized typical analysis scene, extracting feature variables highly related to the scene, quantifying a load change deterministic feature mode molded by the combined action of key deterministic influence factors by utilizing a built-in machine learning model, and outputting a numerical expression of the load change deterministic feature mode; And the characteristic knowledge base and management application module is used for receiving, summarizing and structuring and storing the load deterministic characteristic quantification result under a typical scene, constructing a region power load deterministic characteristic knowledge base which can be inquired and updated, and providing a data support interface for an external load prediction, operation planning and scheduling decision system.

Description

Power load deterministic characteristic quantification method and system based on multi-component heterogeneous data fusion Technical Field The invention relates to the technical field of power system data analysis and intelligent decision making, in particular to a method and a system for quantifying power load deterministic characteristics based on multi-component heterogeneous data fusion. Background The change rule of the power load is accurately mastered, the method is an important precondition for guaranteeing the safe, stable and economic operation of the power system, and along with the deep propulsion of energy transformation and the gradual complexity of an electricity utilization mode, the load characteristic also presents stronger nonlinearity and uncertainty, which provides a serious challenge for the traditional load analysis method. In the prior art, a more advanced scheme is to directly predict load by adopting a single machine learning model (such as BP neural network, support vector machine and the like). The implementation process is generally that firstly, historical load data and possible relevant influence factor data are collected, then the data are input into a preset machine learning model together for training so as to establish an end-to-end mapping relation from input factors to load values, and finally, the trained model is utilized to realize the prediction of future load. This approach can capture to some extent the complex relationship between load and factors. However, in practical use, the method still has some disadvantages, such as incapability of revealing and quantifying the specific contribution degree of different influencing factors to the load, so that the analysis result lacks the interpretability, and secondly, the load behavior is not finely distinguished and attributed according to a 'scene' according to clear and data-driven logic, so that the general and reusable deterministic characteristic rule is difficult to be refined, the cognition depth of the internal mechanism of the load is insufficient, and the generalization capability of the obtained model is limited. When new social activities or climate patterns are encountered which are not present in the training data, the reliability of the analysis and the decision support value will be significantly reduced, and the solution of the present invention is proposed to overcome these drawbacks. Disclosure of Invention In order to overcome the defects in the prior art, the embodiment of the invention provides a method and a system for quantifying the deterministic characteristics of a power load based on multi-component heterogeneous data fusion, which solve the problems in the background art. The method comprises the following steps of S1, collecting and fusing multi-element heterogeneous data related to regional power loads, wherein the multi-element heterogeneous data comprises meteorological data, calendar and holiday data, macroscopic economic data and historical load data, and defining a multi-dimensional initial set of typical analysis scenes based on regional characteristics; S2, taking the data fused in the S1 as input, constructing a BP neural network model, carrying out variable analysis on the BP neural network model by applying a mean value influence value algorithm, and quantifying the contribution degree of each influence factor by calculating the mean influence value generated by the tiny disturbance of the numerical value of each input factor on the network output result to establish an electricity load change deterministic influence factor evaluation index system of a target area; s3, screening and optimizing the pre-defined initial set of typical analysis scenes in the S1 based on the social activity mode of the target area and combining with an evaluation index system, and designing a group of typical electricity utilization analysis scenes covering different time scales and regional characteristics as a standardized analysis framework for subsequent characteristic quantification; S4, fusing and aligning the multi-source heterogeneous data under a unified space-time reference, extracting characteristic variables highly related to the current scene, establishing a load change deterministic characteristic mode by the combined action of deterministic influence factors by using a machine learning method, and outputting a numerical expression; And S5, summarizing and storing the load deterministic characteristic quantification results in all typical scenes to form a structured regional power load deterministic characteristic knowledge base which is used for supporting accurate load prediction, operation planning and scheduling decision of the power system. The power load deterministic characteristic quantification system based on multi-component heterogeneous data fusion comprises a data fusion and scene predefining module, wherein the data fusion and scene predefining module is used for collecting