Grey Wolf Optimization-Enabled Dimensionality Reduction for Efficient Solar Panel Efficiency Prediction with ANN

Abstract

The biggest challenge of accurate solar power forecasting is that the weather is changing at a rapid rate and the mechanism through which solar panels operate is highly nonlinear. Random forests coupled with a fixed Lasso feature selection only combine when the conditions remain fixed but collide in case of weather change or you experience extreme weather conditions. To remedy that, we combine Grey Wolf Optimization (GWO) with an Artificial Neural Network (ANN) in this paper. An intelligent dimensionality reduction step that GWO takes is to search the optimal subset of the inputs, that is, irradiance, temperature, humidity, voltage, current, panel age, soiling ratio, module temperature, and GWO ignores the garbage and accelerates the training. After cleaning that set is then feed through a multi-layered ANN fitting via backprop to achieve a good nonlinear fit that is capable of tracking a real time PV output. It has been experimentally demonstrated that the GWO-ANN outperforms baseline Random Forests in MAE and RMSE, and it continuously converges well with epochs. This system may be a victory of smart-grid energy management and giant solar farms.