Short-term PV power forecasting using variational mode decomposition integrated with Ant colony optimization and neural network

Abstract

In this paper, Artificial Neural Network (ANN) is integrated with data processing, input variable selection, and external optimization techniques to forecast the day ahead output power of a PV system. Variational mode decomposition (VMD) is used to decompose the highly fluctuating original data into relatively stable components with periodic characteristics which can be logically interpreted. The VMD parameters are optimally set through a methodology that involves index of orthogonality (IO) and correlation measures. The input variable selection is accomplished through mutual information (MI). A neural network with technically decided architecture is the core of the forecasting model. The weights and biases of the ANN are externally optimized through Ant colony optimization (ACO) during training. The forecasted components are used as input for a second level forecasting of the PV power through another ANN. The proposed hybrid method, labeled as VMD-ACO-2NN, was evaluated based on a 100kW PV system in Beijing, China. It is compared against NN, GA-NN, ACO-NN and VMD-ACO-NN and proved to outperform all with each of the added features contributing a part. The forecasting model is able to outstandingly explain 97.68% of the total variation in the forecasted PV power.

Keywords

• PV forecasting
• ANN
• VMD
• ACO
• Mutual information