Comparative Analysis of Various Machine Learning and Deep Learning Models for Wind Power Forecasting in Tamil Nadu, India

Reliable wind forecasting is the key to enhance the reliability and efficiency of renewable energy systems. In this research, a detailed comparison of various machine learning (ML) and deep learning (DL) methods are provided for prediction of wind energy based on wind speed, which is real-world measurements gathered at five major sites in Tamil Nadu through the repository of the Iowa State University. A dataset of about 12,900 records was utilized to train and test ML and DL models. Conventional ML algorithms- random forest, decision tree, K-nearest neighbors, AdaBoost, XGBoost, multilayer perceptron were used for prediction of wind energy which showed good Predictive results. Some of the DL models that were tested include convolutional neural networks, recurrent neural networks, temporal convolutional networks, long short-term memory (LSTM) and Bidirectional LSTM (BiLSTM). The BiLSTM was the best with the smallest errors (mean squared error train: 0.0087, test: 0.0093; root mean square error train: 0.0932, test: 0.0963; mean absolute error train: 0.0081, test: 0.0093) and which had the highest R2 = 0.986. Statistical Friedman test is also conducted by nonparametric methods to evaluate all applied DL model’s performance. Out of every experiment made, the result indicated that the capability of BiLSTM is best suitable in forecasting power on various wind speed for geographically different areas of Tamil Nadu.