A novel hybrid approach based on relief algorithm and fuzzy reinforcement learning approach for predicting wind speed

Wind speed (WS) prediction has become popular nowadays due to increasing demand for wind power generation and competitive development in wind energy. Many prediction models are used to predict WS for which wind is non-stationary, nonlinear and irregular. However, they neglect the effectiveness of feature selection methods in WS prediction, thereby creating very challenging for precise prediction of WS and safe operation of the wind industry. To overpower these challenges and further improve WS prediction accuracy, a prediction model is developed based on feature selection technique and prediction models. Therefore this study proposes an adaptive self-learning wind speed (WS) predicting model using fuzzy reinforcement learning (FRL) that is Fuzzy Q Learning (FQL). Proposed FQL based WS predictor model can predict with great accuracy. This is a first effort at developing a forecasting model using FRL for WS prediction. The presented model has no prior knowledge of the system or plant or target speed information. Measured WS is processed through Info Gain attribute evaluator with Ranker search method feature selection purpose which serves as input to the FQL based WS prediction model. The comparison of proposed prediction method and existing machine learning based is carried out using simulations. The performance analysis indicates that the proposed method serves as an important tool for wind potential assessment.

Automated summary

A wind speed prediction model using fuzzy reinforcement learning (FRL) was proposed, achieving high accuracy in prediction. By combining feature selection techniques and prediction models, the accuracy of wind speed prediction was improved. This model has no prior knowledge of the system, plant, or target speed information.