A study of single multiplicative neuron model with nonlinear filters for hourly wind speed prediction. (August 2015)

• Record Type: Journal Article
• Authors: Wu, Xuedong; Zhu, Zhiyu; Su, Xunliang; Fan, Shaosheng; Du, Zhaoping; Chang, Yanchao; Zeng, Qingjun
• Is Part Of: Energy. Volume 88(2015)
• Page Start: 194
• Page End: 201
• Publication Date: 2015-08
• Subjects: Wind speed prediction -- Dynamic uncertainty -- Stochastic nature -- Single multiplicative neuron model -- Iterated nonlinear filters; SMN Single multiplicative neuron -- ARMA Autoregressive moving average -- ANN Artificial neural network -- KRR Kernel ridge regression -- ARIMA Autoregressive integrated moving average -- SVR Support vector regression -- UKF Unscented Kalman filter -- IEKF Iterated extended Kalman filter -- IUKF Iterated unscented Kalman filter -- RSAL Residual active learning -- AMSE Average mean square error -- RMSE The root mean square error -- MAPE Mean absolute percentage error -- MAE Mean absolute error -- IR Improvement ratio; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2015.04.075
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Wind speed prediction is one important methods to guarantee the wind energy integrated into the whole power system smoothly. However, wind power has a non–schedulable nature due to the strong stochastic nature and dynamic uncertainty nature of wind speed. Therefore, wind speed prediction is an indispensable requirement for power system operators. Two new approaches for hourly wind speed prediction are developed in this study by integrating the single multiplicative neuron model and the iterated nonlinear filters for updating the wind speed sequence accurately. In the presented methods, a nonlinear state–space model is first formed based on the single multiplicative neuron model and then the iterated nonlinear filters are employed to perform dynamic state estimation on wind speed sequence with stochastic uncertainty. The suggested approaches are demonstrated using three cases wind speed data and are compared with autoregressive moving average, artificial neural network, kernel ridge regression based residual active learning and single multiplicative neuron model methods. Three types of prediction errors, mean absolute error improvement ratio and running time are employed for different models' performance comparison. Comparison results fromTables 1–3 indicate that the presented strategies have much better performance for hourly wind speed prediction than other technologies. Highlights: Developed two novel hybrid modeling methods for hourly wind speed prediction. Uncertainty and fluctuations of wind speed can be better explained by novel methods. Proposed strategies have online adaptive learning ability. Proposed approaches have shown better performance compared with existed approaches. Comparison and analysis of two proposed novel models for three cases are provided.