基于MMD-OHCP数据预处理的光伏出力超短期预测方法

An Ultra-Short-Term Prediction Method of Photovoltaic Output Based on MMD-OHCP Data Preprocessing

中文关键词:数学形态学去噪  二分之一补点修正法  气象因素  光伏出力  超短期预测

keywords:mathematical morphology denoising  one-half complementary point correction method  meteorological factors  photovoltaic output  ultra-short-term forecasting

中文摘要:随着光伏发电的大规模并网，电网安全问题日益凸显，提高光伏发电超短期预测的准确性可以有效预防此类问题的发生。提出一种基于数学形态学去噪（mathematical morphology denoising，MMD）和二分之一补点修正法（one-half complementary point correction method，OHCP）相结合的数据预处理方法，通过对数据的清洗和平滑操作，使数据能够被机器算法高效识别和利用；利用皮尔逊相关系数分析光伏出力与气象因素之间的相关性，得出3个与光伏出力相关性较大的气象因素；建立BP（back propagation）、XG-Boost和LSTM（long short-term memory）3种预测模型对光伏出力进行超短期预测。仿真结果表明：所提模型在有限次迭代下，考虑3种主要影响因素时，预测的准确度最高。验证了所提模型及数据处理方法的有效性和可行性。

Abstract:国家自然科学基金项目（51907138）；国网山西省电力公司科技项目（52051L230004）。 Project Supported by National Natural Science Foundation of China（51907138）； Science and Technology Project of State Grid Shanxi Electric Power Company（52051L230004）. ABSTRACT：With the large-scale grid-connection of PV power generation， the grid security problem is becoming more and more prominent， and improving the accuracy of the ultra-short-term prediction of PV power generation can effectively prevent the occurrence of such problems. First， a data preprocessing method based on the combination of mathematical morphology denoising （MMD） and one-half complementary point correction （OHCP） is proposed to enable the data to be efficiently identified and utilized by machine algorithms through data cleaning and smoothing operations； subsequently ，the correlation between PV output and meteorological factors is analyzed by using the Pearson correlation coefficient to derive the three meteorological factors that have the highest correlation with PV output； Finally， three prediction models： BP，XG-Boost and LSTM，are established to forecast the PV output in the ultra-short term. The simulation results show that the prediction models have the highest prediction accuracy when the three main influencing factors are considered under a finite number of iterations. The validity and feasibility of the proposed models and data processing methods are verified.