基于极限学习机模型参数优化的光伏功率区间预测技术

Interval Prediction Technology of Photovoltaic Power Based on Parameter Optimization of Extreme Learning Machine

提出一种基于极限学习机(ELM)模型参数优化的光伏功率区间预测技术.首先,提出加权欧氏距离作为光伏功率预测区间评估指标,筛选历史样本单元并优化ELM训练集;然后,提出ELM参数混合寻优算法,利用精英保留策略遗传算法与分位数回归优化ELM模型隐层输入及输出权重与偏置参数,并采用训练后的模型预测光伏功率区间;最后,基于光伏电站与气象站历史数据构建实际算例,预测光伏功率区间,并与其他方法得到的结果进行对比.算例结果表明:所提方法在增加区间预测可信度的同时,能较大程度提高区间预测准确度.

This paper proposes an interval prediction technology of photovoltaic(PV)power based on parameter optimization of extreme learning machine(ELM)model.First,the weighted Euclidean distance is proposed as the evaluation index of PV power prediction interval.The historical sample units are screened and the ELM training set is optimized.Then,a hybrid optimization algorithm for ELM parameters is proposed.The hidden layer input and output weights and biases parameters of the ELM model are optimized by using the elitist strategy genetic algorithm and quantile regression,and the trained model is used to predict the PV power range.Finally,an actual calculation example is constructed based on the historical data of PV power plants and weather stations.The PV power interval is predicted,and the results are compared with those obtained by other methods.The results of the calculation example show that the method proposed can greatly improve the accuracy of interval prediction while increasing the reliability of interval prediction.