集中式电力市场下基于数据驱动的枢纽节点数量设计方式

Data Driven Design of Hub Node Number for Integrated Power Markets

随着集中式电力市场规模的扩大,需要设置电力枢纽节点作为现货市场上可以进行统一交易的聚合节点,其同时也是电力金融市场发展与稳定运行的基石,对构建统一电力市场体系、完善市场功能具有十分重要的意义。枢纽节点设计的难点在于枢纽节点数量的确定,需要通过合适的数量选取以保证对电力市场定价节点的准确覆盖,体现电力空间价值。针对枢纽节点数量选取这一枢纽节点设计的关键问题,提出了一种基于t-SNE(t-distributedstochasticneighbor embedding)降维和DBSCAN(density-based spatial clustering of applications with noise)分类的枢纽节点数量确定方法。首先,通过与KPCA(kernelprincipalcomponentanalysis)、UMAP(uniform manifold approximation and projection)等典型降维方法的对比实验,证明t-SNE对数据拥挤的高维节点电价集有更好的降维效果,其数据可视化效果符合通过降维使得定价节点分成尽可能独立的类的预期。其次,应用DBSCAN算法在基于密度的基础上去除异常点与偏离点并进行分类,通过交叉熵有效选取DBSCAN最佳域值,确定最优分类数。最后,通过一系列分类的内部有效性评价指标,证明了该方法的准确性与有效性,为进一步的枢纽区域划分提供合理依据。

As the expansion of the integrated power markets,the power hub nodes need to be set as the aggregation nodes where the unified transactions can be carried out in the spot markets.This is the cornerstone of the development and the stable operation of the power financial market and is of great significance to the construction of a unified power market system and the improvement of market functions.The difficulty of the hub node design lies in the amount of the hub nodes to be determined to ensure the accurate coverage of the pricing nodes in the electricity market,reflecting the value of the electricity space.This paper proposes a quantity selection based on the DBSCAN(Density-Based Spatial Clustering of Applications with Noise)algorithm classification and the t-SNE(t-Distributed Stochastic Neighbor Embedding)dimensionality reduction analysis for the key problem of the hub node design.Firstly,by comparing with the KPCA(Kernel Principal Component Analysis),the UMAP(Uniform Manifold Approximation and Projection,UMAP)and the other typical dimensionality reduction methods,it is demonstrated that the t-SNE has better dimensionality reduction effect on the data crowded high-dimensional node price dataset.Its data visualization is consistent with the expectation of dividing the pricing nodes into as independent classes as possible through dimensionality reduction.Secondly,the DBSCAN algorithm is used to remove the outliers and the deviations,and classify them on the basis of their density.The optimal number of classifications is determined by effectively selecting the best DBSCAN domain value through the Cross-Entropy.Finally,the accuracy and validity of the method are proved by a series of internal validity evaluation indexes of the classification,which provide a reasonable basis for the further hub area classification.