Wind power interval prediction based on GRU-LUBE

In wind power interval prediction, the performances of prediction model are usually evaluated in terms of reliability and accuracy, which can be quantified by PI coverage probability (PICP) and Mean prediction interval width (MPIW). The ideal high quality prediction interval should have high coverage probability and narrow interval width, which are contradictory to each other. In this case, the method of weight coefficient is proposed to balance the two indexes. At the same time, in the prediction model GRU-LUBE (GRU, gate recurrent unit; LUBE, lower upper bound estimation), in order to reduce the deviation at the peak of the predicted power, this paper proposes a piecewise function of prediction interval accumulative deviation (PIAD) and lets it construct the optimized objective function together with the weight coefficient. Then, sigmoid function is adopted to make the objective function compatible with gradient descent method. Finally, the simulation results show the superiority of the proposed method.