Key points are not available for this paper at this time.
This paper presents the application of regression trees as a versatile alternative to other machine learning and statistical modelling techniques to forecast the power generation at five renewable power plants: one large hydropower plant, two mini hydropower plants, and two wind farms in Sri Lanka. The prediction models for each power station were developed by varying the depth of the regression tree. The regression tree model with the lowest depth that forecasts the output (power) in terms of all the predictor variables was selected for each power station and the accuracy of the models was evaluated by means of the Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and the Coefficient of Determination (R2). According to the degree of the above performance indicators, i.e. very low values of MAE, MAPE, and RMSE supplemented by R2 of 0.95 or more, the regression tree method proved to be a convenient forecasting technique to predict the power generation at both hydro and wind power plants. Further, it could be found that a good correlation between the input and output variables paves the way for a smaller depth in the regression tree. Moreover, regression trees presented here could accurately identify the relationship between the power generated and the most influential weather factors, without being affected by potential outliers or missing values while managing collinearity too. Extension of this study would enable to generalize the prediction of renewable power generation based on the regression tree method, leading towards minimizing the use of fossil fuel.
Jayasinghe et al. (Mon,) studied this question.