Study On Short-term Wind Power Combination Forecast Method Based On Chaotic Time Series

The wind power has the characteristics of volatility, intermittent and random access, and the integration of large-scale wind power seriously affects the safety, stability and economic operation of power grid. Practice has proved that the high precision of wind power prediction will help to mitigate the negative impacts of large-scale wind power integration. Particularly, the accurate prediction of short-term wind power is of great importance to the power dispatch department's arrangement of the schedule plan, which can also ensure the power quality, the safety and economic operation of the power system.Firstly, the characteristics of wind power time series is excavated and analyzed through chaos theory and reconstruction phase space theory. The reconstructed phase space is used as the training sample of wind power time series prediction model, which provides support for the prediction of the phase space trajectory of the wind power chaotic system.Secondly, an improved BP(Back Propagation) neural network based on chaotic time series is proposed, and the model uses the synthesis method to improve the basic BP neural network-both the adaptive learning rate and the momentum term are introduced in the steepest descent method, at the same time, GA(Genetic Algorithm) is used for global search and to find a general solution in the range of the optimal weight coefficient solution, and then the improved steepest descent method is used to adjust the weight coefficient. After a few samples of training, we can get the optimal value of the weight coefficient. In addition, the minimum embedding dimension of the phase space is taken as the input dimension of the model, which can reduce the blindness of modeling. The practical simulation indicates the improved BP neural network has fast convergence speed, improved prediction accuracy and is not easy to fall into the local minimum point.Thirdly, an adaptive GA-VNN (Volterra Neural Network) based on chaotic time series is proposed, which is established on the basis of the equivalence of the adaptive Volterra series and the three layer BP neural network. And the minimum embedding dimension of phase space reconstruction is taken as the truncation of the model, which greatly improve the adaptability of the prediction model. The adaptive GA-VNN combines the adaptive Volterra series's precision modeling capability, the improved BP neural network's learning ability and GA's global search ability, which realize the accurate forecast of wind power time series chaotic trajectory. When the model is applied to the actual simulation prediction of short-term wind power, the prediction accuracy significantly increases.Finally, the non-stability of wind power data is reduced by using EMD (Empirical Mode Decomposition), and then EMD-GA-VNN and EMD-improved BP neural network are established respectively, which were both used in the simulation experiments of short-term wind power prediction. The results show that the EMD method can effectively reduce the non-stationary of wind power data and mine the internal rules of the data better. Compared with the single prediction method, the prediction performance of combination forecasting model is further enhanced. The results also show that the prediction effect of EMD-GA-VNN is more superior than other models, and satisfactory results can be obtained from the prediction of chaotic time series of short term wind power.