Forecasting of Wind Speed and Power through FFNN and CFNN Using HPSOBA and MHPSO-BAACs Techniques

Abstract

Renewable Energy Sources are an effective alternative to the atmosphere-contaminating, rapidly exhausting, and overpriced traditional fuels. However, RESs have many limitations like their intermittent nature and availability at far-off sites from the major load centers. This paper presents the forecasting of wind speed and power using the implementation of the Feedforward and cascaded forward neural networks (FFNNs and CFNNs, respectively). The one and half year’s dataset for Jhimpir, Pakistan, is used to train FFNNs and CFNNs with recently developed novel metaheuristic optimization algorithms, i.e., hybrid particle swarm optimization (PSO) and a Bat algorithm (BA) named HPSOBA, along with a modified hybrid PSO and BA with parameter-inspired acceleration coefficients (MHPSO-BAAC), without and with the constriction factor (MHPSO-BAAC-χ). The forecasting results are made for June–October 2019. The accuracy of the forecasted values is tested through the mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE). The graphical and numerical comparative analysis was performed for both feedforward and cascaded forward neural networks that are tuned using the mentioned optimization techniques. The feedforward neural network was achieved through the implementation of HPSOBA with a mean absolute error, mean absolute percentage error, and root mean square error of 0.0673, 6.73%, and 0.0378, respectively. Whereas for the case of forecasting through a cascaded forward neural network, the best performance was attained by the implementation of MHPSO-BAAC with a MAE, MAPE and RMSE of 0.0112, 1.12%, and 0.0577, respectively. Thus, the mentioned neural networks provide a more accurate prediction when trained and tuned through the given optimization algorithms, which is evident from the presented results.