CN-122026311-A - Power load prediction method, device, equipment and storage medium

CN122026311ACN 122026311 ACN122026311 ACN 122026311ACN-122026311-A

Abstract

The application discloses a power load prediction method, a device, equipment and a storage medium, which relate to the technical field of power load prediction and comprise the steps of obtaining a power load request to be predicted, which is set by a target user for a target place, inputting the power load request to be predicted into a power load prediction integration model to obtain a load prediction result output by the power load prediction integration model, wherein the power load prediction integration model is obtained by carrying out variation modal decomposition and fusion causal inference based on historical power load data, production operation data and associated influence data of the target place. According to the method, different types of loads are split through variation modal decomposition, pertinence and accuracy of prediction are improved, effective characteristics are screened through fusion causality inference, redundant information interference is reduced, model performance is optimized, accurate prediction of electric loads of an industrial park is achieved, reliable data support is provided for electric power dispatching, and energy cost is reduced.

Inventors

JIANG YUAN
XU SITONG
LI LIN
CHEN JIEHAO

Assignees

中国工业互联网研究院(工业和信息化部密码应用研究中心)

Dates

Publication Date: 20260512
Application Date: 20251229

Claims (10)

1. A method of predicting an electrical load, comprising: acquiring a power load request to be predicted, which is set by a target user aiming at a target place; And inputting the power load request to be predicted into a power load prediction integration model to obtain a load prediction result output by the power load prediction integration model, wherein the power load prediction integration model is obtained by performing variation modal decomposition and fusion causal inference based on historical power load data, production operation data and associated influence data of the target place.
2. The power load prediction method according to claim 1, wherein the power load prediction integration model construction process includes: Acquiring historical power load data, production operation data and associated influence data of a target place; Constructing a plurality of stable modal components and initial characteristic data sets corresponding to the stable modal components based on the historical power load data; constructing a variable causal graph between each stable modal component and each initial characteristic data set, and determining a target characteristic data set corresponding to each stable modal component; For any stable modal component, constructing an initial power load prediction sub-model corresponding to the stable modal component based on the component characteristics corresponding to the stable modal component, and performing model training on the initial power load prediction sub-model according to the target characteristic data set corresponding to the stable modal component to obtain a power load prediction sub-model so as to form the power load prediction integrated model.
3. The power load prediction method according to claim 2, wherein constructing a plurality of stationary modal components and initial feature data sets corresponding to the stationary modal components based on the historical power load data includes: Carrying out variation modal decomposition on the historical power load data by constructing a target place variation model to obtain a plurality of stable modal components; And extracting the time characteristics, the working condition characteristics and the associated element characteristics of each stable modal component in the production operation data and the associated influence data, and generating a plurality of initial characteristic data sets corresponding to each stable modal component.
4. The power load prediction method as claimed in claim 3, wherein said performing a variational modal decomposition on said historical power load data by constructing a target site variational model to obtain a plurality of stationary modal components comprises: determining a plurality of splitting areas of the historical power load data according to the ammeter data distribution of the target site, and constructing a target site variation model aiming at the historical power load data corresponding to any splitting area; And solving a constraint optimization problem of the target place variation model by adopting an alternate direction multiplier algorithm, iteratively updating the center frequency and the bandwidth of each modal component in the target place variation model until iteration converges, and determining K independent modal components obtained after convergence as the stable modal components, wherein K is a positive integer.
5. The method of claim 2, wherein constructing a causal plot of variables between each of the stationary modal components and each of the initial feature data sets, determining a target feature data set for each of the stationary modal components, comprises: For any stable modal component, acquiring an initial characteristic variable in an initial characteristic data set corresponding to the stable modal component, calculating a partial correlation coefficient of an electricity utilization curve of the stable modal component and any initial characteristic variable through partial correlation analysis, and screening to obtain candidate characteristic variables with the partial correlation coefficient larger than a preset threshold; Performing causal inference and detection on each candidate characteristic variable through a Grangel causal detection, judging whether the candidate characteristic variable is a cause variable of the stationary modal component power consumption change, and determining a causal relationship identification result to construct a variable causal graph; And removing the causal-relation-free redundant characteristic variables in the initial characteristic data set based on the variable causal graph to form a target characteristic data set corresponding to the stable modal component.
6. The power load prediction method according to claim 2, wherein the constructing an initial power load predictor model corresponding to the stationary modal component based on the component characteristics corresponding to the stationary modal component, and performing model training on the initial power load predictor model according to the target feature data set corresponding to the stationary modal component, to obtain a power load predictor model includes: determining a model framework corresponding to the stable modal component according to the component characteristics to construct an initial power load predictor model, and dividing the target characteristic data set into a training set, a verification set and a test set; Training the initial power load predictor model based on the training set, and adjusting model super-parameters based on the verification set after training for a preset round; And if the verification result of the verification set is smaller than a preset error threshold and the model test result of the test set meets a preset requirement, taking the initial power load predictor model of the current training round as the power load predictor model corresponding to the stable modal component.
7. The power load prediction method according to claim 2, wherein before the historical power load data, the production operation data, and the associated influence data of the target site are acquired, further comprising: Acquiring and carrying out missing value filling, outlier rejection and data standardization on the original historical power load data, the original production operation data and the original associated influence data of the target site to obtain a preprocessing data set; and based on the electricity utilization frequency, the electricity utilization unit and the electricity utilization time in the original historical power load data, performing data alignment on the preprocessing data set to obtain the historical power load data, the production operation data and the associated influence data.
8. An electrical load prediction apparatus, comprising: the acquisition module is used for acquiring a power load request to be predicted, which is set by a target user aiming at a target place; The prediction module is used for inputting the power load request to be predicted into a power load prediction integration model to obtain a load prediction result output by the power load prediction integration model, wherein the power load prediction integration model is obtained by performing variation modal decomposition and fusion causal inference based on historical power load data, production operation data and associated influence data of the target place.
9. An electrical load predicting device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the electrical load predicting method according to any one of claims 1 to 7.
10. A storage medium, characterized in that the storage medium is a computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the steps of the power load prediction method according to any one of claims 1 to 7.

Description

Power load prediction method, device, equipment and storage medium Technical Field The present application relates to the field of power load prediction technologies, and in particular, to a power load prediction method, device, apparatus, and storage medium. Background Currently, there is a need in the industrial park to support distributed new energy consumption, demand side response and energy cost optimization through accurate power load prediction. The park load has the characteristics of large capacity, high density, multiple working procedures and strong impact, and factors such as periodic production shifts, starting and stopping of non-periodic equipment, temporary overtime, extreme weather, sudden change of market orders and the like are overlapped layer by layer, so that a load curve presents a complex form of wide frequency band, non-stability and strong noise, and the decomposing capacity and interpretation capacity of a prediction model are far higher than the requirements of business/resident electricity utilization scenes. The current mainstream technical route is three types, namely a linear time sequence model, represented by ARIMA, only modeling a stable sequence, generating a large residual error once process switching or night shift insertion is met, secondly, EMD/EEMD and other empirical mode decomposition+LSTM/GRU combination, attempting to reduce nonlinearity by using IMF components, however, mode aliasing enables high-frequency impact and low-frequency base load to be repeatedly misplaced, decomposition results drift randomly along with noise, and error amplification of an integrated link is predicted, thirdly, pure deep learning end-to-end scheme depends on massive historical samples, black box fitting is performed on 'historical load-future load' mapping, so that the causal differences such as order surge, equipment overhaul, extreme temperature and the like cannot be distinguished, pseudo-correlation such as 'summer high temperature-load surge' cannot be identified, and long-period prediction is often integrally distorted due to a section of abnormal history. In conclusion, the existing method exposes three major hard injuries in an industrial park scene, namely decomposition failure caused by modal aliasing, explanation failure caused by causality deletion, long-period prediction instability caused by abnormal history interference, and the defects directly lower the dispatching reliability, increase the energy consumption cost and limit the capability of the park to participate in the electric power spot market. Disclosure of Invention The application mainly aims to provide a power load prediction method, a device, equipment and a storage medium, which aim to realize accurate prediction of power load of an industrial park, provide reliable data support for power dispatching and reduce energy cost. To achieve the above object, the present application provides a power load prediction method, including: acquiring a power load request to be predicted, which is set by a target user aiming at a target place; And inputting the power load request to be predicted into a power load prediction integration model to obtain a load prediction result output by the power load prediction integration model, wherein the power load prediction integration model is obtained by performing variation modal decomposition and fusion causal inference based on historical power load data, production operation data and associated influence data of the target place. In one possible implementation manner, the construction process of the power load prediction integration model includes: Acquiring historical power load data, production operation data and associated influence data of a target place; Constructing a plurality of stable modal components and initial characteristic data sets corresponding to the stable modal components based on the historical power load data; constructing a variable causal graph between each stable modal component and each initial characteristic data set, and determining a target characteristic data set corresponding to each stable modal component; For any stable modal component, constructing an initial power load prediction sub-model corresponding to the stable modal component based on the component characteristics corresponding to the stable modal component, and performing model training on the initial power load prediction sub-model according to the target characteristic data set corresponding to the stable modal component to obtain a power load prediction sub-model so as to form the power load prediction integrated model. In one possible implementation manner, the constructing, based on the historical power load data, a plurality of stationary modal components and initial feature data sets corresponding to the stationary modal components includes: Carrying out variation modal decomposition on the historical power load data by constructing a target place variation model to obtain a p