Wind Speed Prediction with Deep Learning Method

Wind energy has broad application space in daily life, but he wind data in a very short time is random, nonlinear and intermittent. Aiming to settle the existing problems in wind forecasting, this project combines variational mode decomposition (VMD) with Grey Wolf Optimization algorithm(GWO) and bidirectional long-term and short-term memory network (BiLSTM) to build a new model VMD-GWO-BiLSTM. Firstly, using the variational model method, the existing wind data is decomposed into multiple sub-models to reduce the complexity and non-stationary degree. On this basis, the initial learning rate, the number of nodes in the hidden layer and the regularization coefficient of the model were optimized by the generalized weighted least squares method. Then, for each successive segment, the sub-model components of each segment are input into the BiLSTM neural network, and the forecast value of each mode component is accumulated to obtain the final forecast result. Comparing VMD-GWO-BiLSTM with VMD-BiLSTM, GWO-BiLSTM, IPSO-BiLSTM and BiLSTM model, it is found that the prediction effect of VMD-GWO-BiLSTM model is better. The relative root mean square error of the forecast evaluation index is 0.5268, the maximum mean square error is 3.3458%, and the relative unit root area is 0.98719.