Multistep Ahead Solar Irradiance and Wind Speed Forecasting using Bayesian Optimized Long Short Term Memory

High integration rate of renewable energy source into the power system in recent times poses problems in power system opreration and control due to its sporadic nature. For improving stability and reliability of hybrid energy system, development of a computationally efficient renewable energy forecast model predominantly wind speed and solar irradiance forecasting model, is required in order to design an automated power forecast tool that efficiently facilities all energy management tasks in a hybrid energy system. Various other meteorological parameters like air density, temperature, humidity, air pressure etc. has a notable effect on solar irradiance and wind speed, which makes the prediction difficult. In this paper, an improved Long Short Term Memory model optimised by Bayesian optimization algorithm is suggested for short-term solar irradiance and wind speed forecasting. The proposed method incorporates hourly averaged historical meteorological variables to forecast the solar irradiance and wind speed in the next 24 hours. The meteorological data that significantly influences the wind speed and solar irradiance is determined by evaluating Pearson Correlation Coefficient. Further, Bayesian optimization algorithm is presented to fine-tune the hyperparameters of LSTM model for one-day ahead hourly prediction. The given proposed model is compared with existing models like like Gradient boost Regressor (GBR),Extreme Gradient Boost Regressor(XGBoost) and RNN(Recurrent Neural Network).Evaluation metrics results show that the proposed method is having better prediction accuracy than the existing methods.