The University of Southampton

Implementation of Very Short-term Wind Power Prediction System


Wind is one of the significant renewable energy resources, which has been highly promoted. However, the biggest challenge of wind development is the unreliability due to the intermittence. One of the solutions is hibernus restore, which allows electronic applications working in dynamic and unreliable power situation by storing current data and states in somewhere else before wind power is actually getting lost. Therefore, accurate prediction of wind power is required. The objective of this project is to implement the system, which can predict the change of wind power in very short-term (few seconds) and foresight if wind is about to lose (below the threshold). Several prediction algorithms are evaluated first in MATLAB, which contain persistence method, ARIMA and BPNN. These approaches would be compared with appropriate error metrics to figure out the one with the best performance to be finally implemented in FPGA board. The evaluation indicates BPNN performs best to predict wind in very short-term. In order to implement BPNN, shift-and-add PWL approximation method is applied to approximate the transfer function of BPNN. Finally, used energy, resources and calculation speed of the system would be used to evaluate the final performance of the ANN system.

Primary investigators

Associated research group

  • Electronics and Computer Science
Share this project FacebookTwitterWeibo