CN-121479372-B - Electric heating hydrogen comprehensive energy system planning scene clustering method based on game mechanism

CN121479372BCN 121479372 BCN121479372 BCN 121479372BCN-121479372-B

Abstract

The invention belongs to the technical field of data clustering, and particularly relates to a method for clustering planning scenes of an electrothermal hydrogen comprehensive energy system based on a game mechanism. Aiming at the defect that the clustering method adopted by the existing electrothermal hydrogen comprehensive energy system cannot better process extreme scenes and conventional scenes simultaneously, the method comprises the steps of preparing and preprocessing multi-dimensional scene data, applying a local anomaly factor algorithm to a multi-dimensional scene feature matrix to identify extreme day scenes, introducing a game mechanism based on a game theory to improve K-means clustering, simulating game interaction of the extreme scenes and the conventional scenes, quantifying dynamic balance relation between the extreme scenes and the conventional scenes, establishing global optimization interaction logic of the local anomaly factor algorithm and the K-means clustering algorithm, finding game equilibrium solutions through double-parameter combination collaborative optimization, and generating a representative planning scene set comprising the conventional scenes and the extreme scenes. The invention realizes the quantization balance between the conventional scene clustering precision and the extreme scene reservation integrity.

Inventors

LIN DA
WANG XIANGJIN
LI ZHIHAO
CHEN ZHE
TANG YAJIE
GONG DIYANG
CHEN LEQI

Assignees

国网浙江省电力有限公司电力科学研究院

Dates

Publication Date: 20260508
Application Date: 20260107

Claims (6)

1. The method for clustering the planning scenes of the electric heating hydrogen comprehensive energy system based on the game mechanism is characterized by comprising the following steps of: s1, preparing and preprocessing historical time sequence multidimensional scene data of an electrothermal hydrogen comprehensive energy system; S2, a local anomaly factor algorithm is applied to a feature matrix obtained by preprocessing the multidimensional scene to identify extreme day scenes, the extreme day show scenes are stored in an extreme scene library, and the rest are stored as a conventional scene library; S3, adopting a game mechanism based on a game theory to improve K-means clustering, simulating game interaction between an extreme scene and a conventional scene, quantifying a dynamic balance relation between the extreme scene and the conventional scene, establishing global optimization interaction logic of a local anomaly factor algorithm and a K-means clustering algorithm, finding a game equilibrium solution through double-parameter combination collaborative optimization, and finally generating a representative planning scene set comprising the conventional scene and the extreme scene; in S3, the process S3.1 for constructing the game mechanism based on the game theory includes: S3.1.1, using a conventional scene clustering party and an extreme scene reservation party as game participants; s3.1.2, quantifying a benefit function of a game mechanism, namely defining the forward benefit of a conventional scene clustering party as a clustering contour coefficient, wherein the higher the value of the forward benefit is, the better the compactness of conventional day clustering is, and defining the forward benefit of an extreme scene reservation party as an extreme day coverage rate; S3.1.3, constructing a strategy set of a game mechanism, wherein the strategy of a conventional scene clustering party comprises the steps of grouping conventional day data after extreme day screening by adopting a K-means clustering algorithm, and controlling the compression degree of a conventional day representative scene by setting a cluster number; in S3, the process S3.2 of improving the K-means clustering algorithm based on the game mechanism includes: S3.2.1, improving the input and output of an algorithm, namely taking a preprocessed standard day show scene feature, day show Jing Xiangliang set, extreme scene index set, conventional scene index set and an abnormal proportion threshold value after dynamic optimization as inputs, and taking a conventional day clustering label set, a conventional day cluster center set and a conventional scene clustering quality evaluation result as outputs, wherein the conventional scene clustering quality evaluation comprises a conventional day reconstruction error and a profile coefficient; S3.2.2, data partitioning based on scene classification results, namely respectively extracting an extreme scene vector subset and a conventional scene vector subset from a day show scene vector set, wherein the extreme scene vector subset is directly used as a representative vector of an extreme scene, and the conventional scene vector subset is used as a core processing object of the K-means clustering in the step; S3.2.3, conventional daily K-means clustering optimization of the adaptive game targets; S3.2.3, a conventional daily K-means clustering optimization process for adapting game targets comprises the following steps: constructing a clustering objective function, wherein the clustering objective function is the least sum of the squares of Euclidean distances between a conventional day and the center of a corresponding cluster, and the smaller the objective function value is, the more concentrated the scene features in the cluster are; k mean iterative optimization, including cluster center initialization, cluster attribution allocation and cluster center updating; evaluating the clustering effect under the current cluster number by adopting the contour coefficient and the conventional daily reconstruction error, and adjusting the cluster number to iterate again if the set index requirement is not met; in the S3, the construction process S3.3 of the global optimization interaction logic of the local anomaly factor algorithm and the K-means clustering algorithm comprises the following steps: s3.3.1, constructing a global optimization objective function taking cluster quality-global error-scene diversity as a core; S3.3.2, double-parameter combination collaborative optimization, namely, calculating global objective function values corresponding to all combinations by traversing two parameter range combinations, namely an abnormal proportion threshold value and a cluster number, which are preset, and screening optimal parameter pairs.
2. The method for clustering the planning scene of the electrothermal hydrogen comprehensive energy system based on the game mechanism of claim 1, wherein in S1, the preparation and preprocessing process of the data comprises the following steps: S1.1, data acquisition and cleaning, namely acquiring M-dimensional time sequence data with sequence duration equal to T of an electrothermal hydrogen comprehensive energy system, forming a data matrix, marking the data matrix as an original input data set, filling and cleaning, and calculating the statistical characteristics of the data, wherein T=N days×24 hours; s1.2, dividing data according to days, namely dividing a cleaned and filled data set into day show scene data of N days according to 24 hours as a unit, constructing an independent day show scene, flattening a daily scene into N-dimensional vectors, generating a daily vector matrix, and forming a daily vector set for subsequent reconstruction error calculation; S1.3, extracting statistical features from the d day show th scene matrix to capture the operation characteristics of each dimension, and calculating one or more of the following features, namely a mean value, a standard deviation, a peak-valley difference and a quantile for each dimension; S1.4, integrating the features according to dimensions to form feature vectors of a d-th day scene, constructing a feature matrix, and carrying out standardization processing on the feature matrix, wherein the standardized feature matrix is used as input for the extreme scene detection of a subsequent local anomaly factor method.
3. The method for clustering the planning scenes of the electrothermal hydrogen comprehensive energy system based on the game mechanism of claim 1, wherein in S2, the process for identifying the extreme day scene comprises the following steps: S2.1, a local anomaly factor method is applied to a feature matrix, and extreme scene points are identified by calculating the deviation between local reachable density and neighbor density of each daily scene; S2.2, dynamic parameter optimization, namely traversing an abnormal proportion threshold value, and evaluating the influence of the abnormal proportion threshold value on the clustering quality.
4. The method for clustering the planning scenes of the electrothermal hydrogen comprehensive energy system based on the game mechanism according to claim 3, wherein the S2.1 process comprises the following steps: S2.1.1, defining an accessibility distance for measuring accessibility between samples; s2.1.2, calculating local reachable density based on the distance between the sample and the nearest neighbor, and reflecting the density level around the sample; s2.1.3, quantifying the outlier degree by comparing the local reachable densities of the sample and the neighbors; s2.1.4, setting an abnormal proportion threshold value as a dynamic adjustment value, and determining a polar day judgment standard; S2.1.5 output extreme scene index and regular scene index, representing extreme day and typical day, respectively.
5. The method for clustering the planning scenes of the electrothermal hydrogen comprehensive energy system based on the game mechanism, which is characterized in that in S2.2, the clustering quality index comprises a conventional day reconstruction error and a profile coefficient, the lower the conventional day reconstruction error is, the higher the profile coefficient is, the better the compactness and the accuracy of conventional day clustering are, and the extreme day number and the conventional day number of each test are recorded.
6. The method for clustering the planning scenes of the electrothermal hydrogen comprehensive energy system based on the game mechanism according to claim 1, wherein in S3, generating the representative scene set and the corresponding visual curve S3.4 based on the clustering result comprises: S3.4.1, constructing a global representative scene set, wherein the global representative scene set comprises conventional scene representation extraction, extreme scene representation extraction and global scene set integration and annotation; s3.4.2, taking the scenes in the global scene set as a unit of days, and respectively outputting the regular days and the extreme days corresponding to each dimension to be presented in a visualized image.

Description

Electric heating hydrogen comprehensive energy system planning scene clustering method based on game mechanism Technical Field The invention belongs to the technical field of comprehensive energy system planning data clustering, and particularly relates to a method for clustering planning scenes of an electrothermal hydrogen comprehensive energy system based on a game mechanism. Background With the rapid development of renewable energy and the deep promotion of energy transformation, the electrothermal hydrogen comprehensive energy system becomes an important way for realizing high-efficiency and low-carbon energy supply because the electrothermal hydrogen comprehensive energy system can integrate renewable energy sources such as photovoltaics, fans and the like and various energy demands such as electric loads, heat loads, hydrogen loads and the like. However, due to the intermittent nature, volatility, and complexity of multi-dimensional loads of renewable energy sources, planning of electro-thermal hydrogen integrated energy systems presents significant challenges. In long-term planning, the multi-dimensional time sequence data 8760 hours in the whole year is required to be subjected to cluster reduction processing, and the dimension disaster is caused by directly adopting all original data to perform optimization calculation, so that the calculation cost is high and the efficiency is low. In particular, in a multidimensional scenario of an electrothermal hydrogen system, the diversity of system operation data and the existence of an abnormal scenario (such as an extreme scenario caused by extreme weather or load abnormality) make it difficult for a conventional clustering method to generate a planning scenario with high representativeness. The conventional clustering method generally treats the extreme scene as noise for removing or simply smoothing, so that the planning result has insufficient reliability when coping with the extreme working condition, and a good balance between the calculation efficiency and the scene representativeness cannot be achieved. The traditional K-means clustering algorithm (K-means) is widely applied to scene analysis of an energy system, but has the following limitation that firstly, the algorithm lacks an active recognition mechanism for an extreme scene, and the extreme scene (such as the condition that extreme weather causes abnormal wind and light output) and a conventional scene are integrated into a clustering classification category. The processing mode can force the extreme value to participate in the construction process of the conventional cluster, so that the cluster center is shifted towards the extreme value, and the clustering precision of a typical scene is reduced. Second, although the optimal cluster number K can be determined by an elbow method, a contour coefficient and the like, when determining the cluster number, the conventional K-means has a core concern that the conventional data is compact, and the extreme scene coverage is easy to ignore. The clustering strategy taking the conventional data as the leading part can systematically miss extreme working conditions, so that the generated planning scene lacks diversity, and the reliability risk exists when the abnormal working conditions are dealt with. Although the existing partial improvement method improves the performance of the K-means algorithm by introducing anomaly detection or optimizing initial point selection, the core balance problem of the extreme scene and the conventional scene is not solved. For example, anomaly detection methods based on local anomaly factors (Local Outlier Factor, LOF) can identify outliers, but are difficult to integrate seamlessly with K-means clustering processes to form global optimization logic. More importantly, the existing method generally lacks an effective quantitative balance mechanism when processing interaction between an extreme scene and a conventional scene, cannot reasonably regulate and control the quantity ratio of the extreme scene to the conventional clustering typical scene, can improve scene diversity to cover more system abnormal characteristics if the inclusion proportion of the extreme scene is intentionally increased, but can additionally increase data processing capacity and scene complexity to cause calculation load climbing and efficiency reduction when planning a subsequent electrothermal hydrogen comprehensive energy system, and can lose system key abnormal characteristics due to the fact that the conventional scene is reserved with emphasis to simplify calculation and reduce load, so that clustering results cannot completely reflect fluctuation risks and extreme working conditions in operation of the electrothermal hydrogen comprehensive energy system, and application effects of the clustering results in the electrothermal hydrogen comprehensive energy system planning are limited. Disclosure of Invention Aiming at the defect that the c