Data Fusion Based Hybrid Deep Neural Network Method for Solar PV Power Forecasting

This paper proposes a new Hybrid Deep Neural Network (HDNN) based fusion method to predict short-term solar photovoltaic (PV) power output. The HDNN is the combination of Fully Convolutional Network (FCN) and Long Short-Term Memory (LSTM) networks that fuses the output of two individual forecast models, i.e., Autoregressive Moving Average with Exogenous Inputs (ARMAX) and Adaptive Neuro Fuzzy Inference System (ANFIS). The Deep Neural Network (DNN) based parts, which are stemmed from the idea that individual predictions obtained by several models, add value to the final forecasting process. The major advantage of the fusion component in the proposed method is that it allows the salient feature extraction through the HDNN model by identifying sequential dependencies in historical trends using different forecasting models' perspectives to predict solar PV power output. The prediction accuracy of the proposed HDNN-Fusion model is validated by comparing its performance to other techniques through several soft computing models. Simulation results demonstrate the suitability of the proposed fusion method to obtain accurate short-term PV power forecasts for multiple seasons of the year.