Regression based Prediction of Rainfall for Energy Management in a Rural Islanded Micro-Hydro Grid in Kerala

Abstract

Rural development is one of the primary goals in developing nations like India since more than 70 percent of the population lives in rural villages. This paper aims to develop and implement an effective energy management strategy for a stand-alone Amrita Micro-Hydro Power Plant(AMHPP) set up in a remote tribal community in Kerala. The Energy management strategy is implemented by the development of regression-based machine learning models for the prediction of rainfall, thereby deriving the water flow rate and power generation at AMHPP. Five regression-based machine learning models were compared to find the optimal model to increase the asset utilization of the hydro-micro grid system. The performance of prediction models based on daily and monthly data is assessed using the correlation coefficient (R), mean absolute error (MAE), and root mean square error (RMSE). In terms of the various types of regression, including linear, lasso, ridge, XGBoost, and elastic net, the machine learning technique XGBoost regression produced a comparably accurate prediction model. The proposed energy management strategy based on the XGBoost regression model can increase the micro-hydro system's real-time power utilization efficiency w $\mathbf{h}$ ile balancing the available hyd $\mathbf{r}$ oenergy and load management for the micro-grid system.