Accurate Prediction Of Short-Term Photovoltaic Power Generation Via A Novel Double-Input-Rule-Modules Stacked Deep Fuzzy Method

Accurate prediction of the photovoltaic (PV) power generation is of great significance for the efficient management of the power grid. In order to strengthen the interpretability of the data-driven models for PV power prediction and to further improve the forecasting accuracy, a novel double-input-rule-modules (DIRMs) stacked deep fuzzy model (DIRM-DFM) is proposed in this study. Firstly, the proposed stacked structure of DIRM-DFM is presented. This novel modular structure adopts a bottom-up, layer-by-layer design scheme by stacking the DIRMs which has only two input variables. This scheme assures the interpretability of the proposed novel fuzzy model. Then, to guarantee the performance of DIRM-DFM, its learning mechanism, including the training data generation, the construction of the DIRMs, are given in detail. This learning mechanism has fast learning speed and excellent approximation ability, because each DIRM is optimized by the popular least square method. Finally, two real-world experiments for predicting the PV power generation are conducted to verify the proposed DIRM-DFM, and detailed comparisons are made with traditional and deep fuzzy models, shallow and deep neural networks. Experimental results clearly demonstrated that the proposed DIRM-DFM has the best accuracy and the reactively fast training speed while having the apparent advantages of interpretability. (c) 2020 Elsevier Ltd. All rights reserved.