基于样本双重筛选的光伏发电功率预测

POWER PREDICTION MODEL OF PV POWER GENERATION BASED ON DOUBLE SCREENING OF SAMPLES

首先通过建立辐照度和功率之间的关系,剔除功率奇异值来完成初步筛选,其次通过计算辐照度、温度、云量等气象因素与发电功率的相关系数,选取相关系数较大的气象因素来评价历史天与预测天的相似度,提取与预测天最相似的历史天作为训练样本来完成二次筛选,最后利用BP神经网络和遗传算法进行光伏发电的功率预测,结果表明该方法具有较高的预测精度。

PV power generation has characteristics such as volatility, intermittence, but the more similar the weather condition is, the more similar the generation law of PV power generation system is. Firstly, the power singular value was removed to complete the preliminary screening through setting up the relations between irradiance and power. Secondly, the correlation coefficient between power and various meteorological factors such as irradiance, temperature and cloudiness, etc. was calculated, and then the meteorological factors with bigger correlation coefficient were collected to evaluate the similarity between historical days and the forecast day. The most similar historical days from the predicted day was extracted as training samples to complete the secondary screening. At last, the power of PV power generation was predicted using BP neural network and genetic algorithm. The results show that the method has high prediction accuracy.