低成本微电网轻量化在线超短期光伏功率预测算法设计

摘要

本文结合用户侧光伏微电网的实际工程需求,研究了满足嵌入式应用需求的轻量化在线超短期光伏功率预测算法.采用了核函数极限学习机算法作为在线超短期光伏功率预测算法的核心,通过使用特征序列代替传统时间序列作为训练样本集,实现了数据量的轻量化设计;通过基于趋势加权相似度的训练样本精选,提高了算法精度,简化了计算量,降低了算法计算时间,通过嵌入式系统实验测试结果表明,本文提出的轻量化超短期光伏预测算法在精度、计算时间和数据存储等方面都能满足嵌入式系统低成本应用的需求.%Based on the engineering requirement of user side photovoltaic (PV) microgrid,a simplified online short term PV output forecasting algorithm is studied for embedded system application.Extreme learning machine with kernel (ELM_K) algorithm is adopted as the main part.The traditional time sequence of training dataset is replaced with characteristic sequence of history data,therefore,the amount of storage place of training data is reduced.Because of the optimal training dataset is selected from original training dataset by trend weighted similarity,the accuracy is improved,also the amount of calculation and runtime of forecasting algorithm is reduced.The test results of embedded system show that the performance of proposed online short term PV output forecasting algorithm on accuracy,runtime and storage occupation can satisfy the requirement of low cost embedded system application.