Improving Wind Power Forecast Accuracy for Optimal Hybrid System Energy Management

Abstract

Due to its renewable and sustainable features, wind energy is growing around the world. However, the wind speed fluctuation induces the intermittent character of the generated wind power. Thus, wind power estimation, through wind speed forecasting, is very inherent to ensure effective power scheduling. Four wind speed predictors based on deep learning networks and optimization algorithms were developed. The designed topologies are the multilayer perceptron neural network, the long short-term memory network, the convolutional short-term memory network, and the bidirectional short-term neural network coupled with Bayesian optimization. The models' performance was evaluated through evaluation indicators mainly, the root mean squared error, the mean absolute error, and the mean absolute percentage. Based on the simulation results, all of them show considerable prediction results. Moreover, the combination of the long short-term memory network and the optimization algorithm is more robust in wind speed forecasting with a mean absolute error equal to 0.23 m/s. The estimated wind power was investigated for optimal Wind/Photovoltaic/Battery/Diesel energy management. The handling approach lies in the continuity of the load supply through the renewable sources as a priority, the batteries on the second order, and finally the diesel. The proposed management strategy respects the designed criteria with a satisfactory contribution percentage of renewable sources equal to 71%.