基于多观测隐非齐次HMM的光伏功率概率预测  

Probabilistic prediction of PV power based on non-homogeneous hidden Markov model with multiple observations

作　　者：马琼 马雷 汪佐浩 张浩[2] MAQiong;MA Lei;WANG Zuohao;ZHANG Hao(State Grid Linxia Electric Power Company,Linxia Gansu 731100,China;School of Water Resources and Hydropower,Xi'an University of Technology,Xi’an Shaanxi 710048,China)

机构地区：[1]国网临夏供电公司,甘肃临夏731100 [2]西安理工大学水利水电学院,陕西西安710048

出　　处：《电源技术》2023年第11期1501-1505,共5页Chinese Journal of Power Sources

基　　金：中国博士后基金(2020M683526);陕西省自然科学基础研究计划(2021JQ-471);陕西省教育厅自然科学专项(21JK0802)。

摘　　要：针对高渗透率光伏能源并网带来的不确定性对电力系统的影响,提出一种基于多观测隐非齐次马尔可夫模型(HMM)的短期光伏功率概率预测方法。首先,给出多观测隐非齐次HMM基本问题的求解方法,包括模型参数估计的学习问题与预测结果输出的解码问题。通过确定光伏出力与气象参数范围并进行离散化,得到建模所需的状态空间与观测空间。最后,据此对训练数据进行编码代入HMM学习问题求解得到最终的概率预测模型。为了检验不同非齐次性质及观测量对模型预测性能的影响,建立传统HMM、单观测/双观测二阶HMM及支持向量机模型进行模拟预测分析。结果验证了所提模型的可行性,同时考虑先前状态与观测对当前的影响可以提高模型确定性预测与区间预测性能。In order to address the effect of uncertainty on power systems brought by grid-connected PV energy,a probabilistic forecasting method for short-term PV power based on multi-observation hidden non-homogeneous Markov model(HMM)was proposed.First,the solution to the basic problem of the proposed model was given,including the learning problem and decoding problem.The state and observation space required for modelling were obtained by determining and discretizing the range of PV power and meteorological parameters.Finally,the training data was coded and substituted into the HMM learning problem to obtain the final model.To examine the impact of different non-homogeneous properties and observation number on the model,conventional HMM,single/dual observation second-order HMM and support vector machine were built for analysis.The results validate the feasibility of the proposed models and the fact that considering the influence of prior states and observations on the present can improve the model performance.

关 键 词：隐马尔可夫模型 概率预测 光伏功率 非齐次马尔可夫链 

分 类 号：TM615[电气工程—电力系统及自动化]