CN-122022309-A - Industrial park adjustable resource aggregation characteristic parameter estimation method based on LSTM-GNN

CN122022309ACN 122022309 ACN122022309 ACN 122022309ACN-122022309-A

Abstract

The invention provides an LSTM-GNN-based industrial park adjustable resource aggregation characteristic parameter estimation method, and relates to the technical field of power system optimization control. The method comprises the following steps of S1, constructing an operation characteristic model of adjustable resource equipment, S2, analyzing the behavior characteristics of various adjustable resources based on the operation characteristic model, constructing an industrial park virtual power plant polymer model, S3, estimating parameters of the monomer equipment by adopting a UKF algorithm based on the operation characteristic model, and outputting an estimated value of the parameters of the monomer equipment, and S4, estimating the aggregation characteristic parameters by adopting an LSTM-GNN model based on the industrial park virtual power plant polymer model, and outputting an estimated value of the aggregation characteristic parameters. The method solves the problems of inaccurate resource aggregate model and large estimation error of the adjustable resource aggregate characteristic parameters in the prior art, and improves the accuracy and reliability of the estimation of the adjustable resource aggregate characteristic parameters of the park.

Inventors

Fu Chaoran
DOU ZHENLAN
ZHANG CHUNYAN
YUAN YIMING
WANG JIAYU
WU HAOQIANG
YU DONGMIN
LIU HUANAN
WANG YONG

Assignees

国网上海市电力公司
南昌大学

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. An LSTM-GNN-based industrial park adjustable resource aggregation characteristic parameter estimation method is characterized by comprising the following steps of: s1, constructing an operation characteristic model of adjustable resource equipment; S2, analyzing the energy utilization behavior characteristics of various adjustable resources based on the operation characteristic model, and constructing an industrial park virtual power plant polymer model; S3, based on the operation characteristic model, estimating parameters of the single equipment by adopting a UKF algorithm, and outputting estimated values of the parameters of the single equipment; And S4, based on the industrial park virtual power plant polymer model, estimating the aggregation characteristic parameters by adopting an LSTM-GNN model, and outputting an aggregation characteristic parameter estimation value.
2. The method for estimating the characteristic parameters of the industrial park adjustable resource based on LSTM-GNN according to claim 1, wherein in the step S1, the adjustable resource equipment comprises industrial production equipment, temperature control equipment, electric automobiles and energy storage equipment.
3. The method for estimating the aggregate characteristic parameters of the adjustable resources of the industrial park based on the LSTM-GNN as claimed in claim 2, wherein in the step S1, constructing the operation characteristic model of the adjustable resource equipment specifically comprises: the industrial production equipment comprises motor equipment, an electric arc furnace and an electric heat pump, and corresponding industrial production load models are respectively constructed; constructing a temperature control load model aiming at temperature control equipment; constructing an electric automobile model aiming at an electric automobile; The energy storage equipment comprises an electric energy storage device and a thermal energy storage device, and corresponding energy storage models are respectively built.
4. The method for estimating an adjustable resource aggregation characteristic parameter of an industrial park based on LSTM-GNN according to claim 1, wherein the energy consumption behavior characteristics include trend, periodicity, power conservation and continuity in step S2.
5. The method for estimating the adjustable resource aggregation characteristic parameters of the industrial park based on LSTM-GNN, which is characterized by constructing an industrial park virtual power plant polymer model in the step S2, and adopting fuzzy C-means clustering, specifically comprising the following steps: Step S201, carrying out normalization processing on load power of an industrial park to obtain a load data set; step S202, setting the clustering quantity, selecting a clustering center, and calculating the membership degree of any adjustable resource to the clustering center; step S203, iteratively updating a clustering center; Step S204, circulating the step S202 to the step S203, calculating a minimum cluster center under the minimum difference value under different cluster centers of the adjustable resource updating iteration, and executing the next step; Step S205, judging whether the constraint condition is met, if not, returning to step S202, step S203 and step S204, and re-executing the calculation of the minimum difference value under the constraint condition; and S206, outputting a clustering result.
6. The method for estimating an adjustable resource aggregation characteristic parameter of an industrial park based on LSTM-GNN according to claim 5, wherein in step S205, the constraint condition is expressed as: ; ; ; Wherein, the Representing an adjustable resource To the cluster center Membership degree of (3); representing an index variable; representing the number of clusters of the adjustable resource; Indicating the total number of loads aggregated in the industrial park.
7. The method for estimating the parameters of the adjustable resource aggregation of the industrial park based on the LSTM-GNN, as set forth in claim 1, wherein in the step S3, the monomer equipment parameters are estimated by adopting the UKF algorithm, specifically including: The method comprises the steps of establishing a state space equation of equipment, expanding parameters to be estimated as state variables, increasing noise interference or measuring noise interference in a system process, selecting a set sigma point set, adding the state equation, and generating a single equipment parameter estimated value through recursion and weighted average.
8. The method for estimating the aggregate characteristic parameters of the industrial park based on the LSTM-GNN according to claim 1, wherein in the step S4, the aggregate characteristic parameters are estimated by adopting the LSTM-GNN model, and the method specifically comprises the following steps: Step S401, obtaining adjustable resource data of an industrial park, and performing data processing and alignment to obtain node time sequence data; Step S402, processing the node time sequence data by utilizing LSTM, and constructing a node characteristic matrix; step S403, calculating an inner product of the node characteristic matrix to obtain an inter-node similarity score matrix, and carrying out normalization processing on the inter-node similarity score matrix by using a Softmax function to generate a weighted adjacent matrix; step S404, sampling, weighting aggregation and updating node characteristics by adopting GRAPHSAGE algorithm based on the weighted adjacent matrix; step S405, carrying out global pooling on the updated node characteristics to obtain a graph-level embedded vector; step S406, the image level embedded vector is input into a regressive device to carry out regression prediction, and an aggregate characteristic parameter estimated value is obtained.
9. The method for estimating an adjustable resource aggregation characteristic parameter of an industrial park based on LSTM-GNN according to claim 8, wherein the global pooling operation in step S405 includes global addition, global mean and maximum pooling.
10. The method for estimating an adjustable resource aggregation characteristic parameter of an industrial park based on LSTM-GNN according to claim 8, wherein in step S406, the specific expression for obtaining the aggregation characteristic parameter estimation value by regression prediction is: ; Wherein, the The weight matrix is represented by a matrix of weights, The offset vector is represented as such, Representing the predicted output vector; representing the aggregate characteristic parameter estimate.

Description

Industrial park adjustable resource aggregation characteristic parameter estimation method based on LSTM-GNN Technical Field The invention relates to the technical field of power system optimization control, in particular to an LSTM-GNN-based industrial park adjustable resource aggregation characteristic parameter estimation method. Background Currently, as an important component of energy transformation, industrial park virtual power plants are increasingly receiving attention as methods for estimating adjustable resource aggregation and aggregation characteristic parameters. In the prior art, the adjustable resource aggregation mainly adopts K-means clustering (K-means clustering), a weighted average method or SVR regression (support vector regression) and other methods, wherein the K-means clustering has high calculation efficiency, but is difficult to process fuzzy attribution, the heterogeneous resource aggregation precision is low, the weighted average method is simple and quick, but ignores the difference among resources, so that the estimation error is large, and the SVR regression has good prediction precision and nonlinear capture, but has high training cost and poor model interpretation. In the aspect of aggregate characteristic parameter estimation, a method based on a physical mechanism is complex in modeling, the coupling relation among devices is difficult to describe, adaptability to dynamic changes is insufficient, the traditional machine learning method is free of dependence on a physical model, but lacks of effective modeling on time sequence and space association, estimation deviation is obvious, and a single deep learning method is capable of improving time sequence dynamic capture capability, but is incapable of solving the space association problem, and is difficult to describe the evolution rule of a resource state comprehensively. Therefore, a new parameter estimation method capable of considering both space-time dynamic characteristics and outputting accurate and reliable parameter estimation results is needed to support the accurate depiction of the adjustable resource aggregation characteristics and the optimal scheduling requirements of the virtual power plants in the industrial park. Disclosure of Invention The invention aims to provide an LSTM-GNN-based industrial park adjustable resource aggregation characteristic parameter estimation method, which aims to solve the problems of inaccurate resource aggregate model and large adjustable resource aggregation characteristic parameter estimation error in the prior art. The overall regulation and control potential of the adjustable resources of the park is accurately excavated, and the precision and reliability of the evaluation of the aggregation characteristic parameters of the adjustable resources of the park are improved. In order to achieve the above purpose, the invention provides an industrial park adjustable resource aggregation characteristic parameter estimation method based on LSTM-GNN, which comprises the following steps: s1, constructing an operation characteristic model of adjustable resource equipment; S2, analyzing the energy utilization behavior characteristics of various adjustable resources based on the operation characteristic model, and constructing an industrial park virtual power plant polymer model; S3, based on the operation characteristic model, estimating parameters of the single equipment by adopting a UKF algorithm, and outputting estimated values of the parameters of the single equipment; And S4, based on the industrial park virtual power plant polymer model, estimating the aggregation characteristic parameters by adopting an LSTM-GNN model, and outputting an aggregation characteristic parameter estimation value. Preferably, in step S1, the adjustable resource device includes an industrial production device, a temperature control device, an electric automobile, and an energy storage device. Preferably, in step S1, constructing an operational characteristic model of the adjustable resource device specifically includes: the industrial production equipment comprises motor equipment, an electric arc furnace and an electric heat pump, and corresponding industrial production load models are respectively constructed; constructing a temperature control load model aiming at temperature control equipment; constructing an electric automobile model aiming at an electric automobile; The energy storage equipment comprises an electric energy storage device and a thermal energy storage device, and corresponding energy storage models are respectively built. Preferably, in step S2, the performance characteristics include trend, periodicity, power conservation, and continuity. Preferably, in step S2, an industrial park virtual power plant polymer model is constructed, and fuzzy C-means clustering is adopted, which specifically includes: Step S201, carrying out normalization processing on load power of an industrial park t