基于时间序列神经分层插值模型的光伏功率超短期多步预测

Ultra-short-term Multi-step Forecasting of Photovoltaic Power Based on Time Series Neural Hierarchical Interpolation Model

针对光伏功率预测准确性受数据质量和外部变量影响的问题,提出一种结合外生变量分析、数据质量控制以及时间序列神经分层插值(N-HiTS)模型的光伏功率超短期多步预测方法。首先,提出用于筛选外生变量的综合相关性度量(ICM)指标,并采用K近邻(KNN)算法与线性插值策略处理数据缺失问题。然后,引入N-HiTS长时间序列预测模型,通过多尺度信号采样和分层插值提高模型对长时间序列数据的处理能力。最后,通过算例对所提方法与传统光伏功率预测方法进行对比分析,验证了所提方法的预测准确性。

Aiming at the problem that the prediction accuracy of photovoltaic power is affected by data quality and exogenous variables,the ultra-short-term multi-step photovoltaic power prediction approach is proposed integrating exogenous variable analysis,data quality control and Neural Hierarchical Interpolation for Time Series(N-HiTS)model.Firstly,the proposed Integrated Correlation Measurement(ICM)for screening exogenous variables is proposed,and the K-Nearest Neighbors(KNN)algorithm and linear interpolation strategy are used to deal with the problem of missing data.Secondly,a long-term prediction model based on the N-HiTS model is established to improve the model's proficiency in processing long-term series time data through multi-scale signal sampling and hierarchical interpolation.Finally,a comparative analysis between the proposed method and the traditional photovoltaic forecasting techniques is conducted through a numerical example to verify the prediction accuracy of the proposed method.