Brown bear optimized random forest model for short term solar power forecasting

Abstract

Short term solar power forecasting is essential in managing the daily power requirements, electricity market operations and maintaining grid stability. Most of the ensemble ML algorithms outperform the traditional ML algorithms in terms of prediction accuracy. In this paper, short-term solar power forecasting is done using random forest (RF) algorithm for a comparatively smaller data set. The results of the RF model are then compared with other ML models namely decision tree (DT), support vector regression (SVR), gradient boost (GB) and ridge regression (RR). The prediction accuracy of the RF model as assessed by MSE is increased by 27.63 %, R2 by 13.4 %, RMSE by 14.93 % and MAE by 19.17 % when compared with the DT model. To further improve the accuracy of the RF model, the hyperparameters of the random forest model are tuned using brown bear optimization algorithm (BBOA). Hyperparameter tuning using BBOA further improves MSE by 19.73 %, RMSE by 10.41 %, MAE by 11.19 % and R2 by 7.17 %. The results obtained are compared with hyperparameter tuning using particle swarm optimization (PSO) and firefly algorithm (FA). MSE obtained using BBOA is improved by 2.7 % and 3.7 % when compared with PSO and FA respectively. This improvement in the performance of BBOA can be attributed to the robust nature and better adaptation capability of the algorithm, proving its competence in hyperparameter tuning of ML model.

Keywords

• Solar power forecasting
• Machine learning
• Hyperparameter tuning
• Brown bear optimization
• Random forest