基于马尔可夫模型的光伏出力聚类与模拟

恰当地处理光伏数据并建立合理的出力模型是含光伏发电系统运行与规划的基础。该文基于马尔可夫模型和谱聚类算法，综合考虑光伏出力的天气特性、日特性、季节特性，提出了一种光伏出力聚类和模拟方法。首先，从光伏出力中提取幅值、基准出力和波动出力，基于马尔可夫模型针对波动出力进行聚类分析，并利用Alpha值、Silhouette指数、贝叶斯准则确定最佳聚类数；然后，整合聚类结果并搭建基于月内天气状态和日内出力状态的双层马尔可夫模型；最后，利用双层抽样得到中长期光伏模拟时序出力。与传统K-means算法相比，基于马尔可夫模型的谱聚类算法能够更好地解决光伏出力数据可能出现的数据丢失与误测问题，并具有更好的天气区分性。与马尔可夫蒙特卡洛模拟相比，该文所提光伏出力模型能够更好地保持光伏出力的天气特性、时序特性和概率特性，并具有更高的精度。算例分析验证了该文所提模型的有效性。

Clustering and Simulation of Photovoltaic Output Adopting Markov Model

Processing data properly and constructing a rational photovoltaic output model is the base of power system operation and plan. Focusing on characteristics of photovoltaic output， a photovoltaic output clustering and simulation method based on Markov model and spectral clustering is proposed. Firstly， the amplitude parameter， the standard component and fluctuant component are extracted from photovoltaic output and Markov model is adopted to analyze the fluctuant component. We use spectral clustering and Alpha value， Silhouette index and Bayesian information criterion to evaluate the quality of clustering results and to determine the optimal cluster number. Then， state-transition matrixes between the weather in the month and power output in the day are integrated， and a bistratal Markov model is constructed. Finally， the mid/long term photovoltaic simulating output is generated by utilizing the bistratal sampling. Compared with the K-means clustering analysis， spectral clustering is more suitable for processing photovoltaic output data， especially in the condition of data missing or data error and in distinguishing weather data. Through the comparative analysis on statistical characters， timing characters and weather characters， the validity of the proposed model is verified.