Small-world Network Theory And Its Application In Wind Power Short-term Forecasting

Wind power forecasting is an important frontier subject of new energy areas,which has important implications for wind power and smart grids.Small-world network is a complex network of high information transmission efficiency, and small-world optimization and small-world neural network are important application of the small-world network. Studies show that small-world neural network has more generalization ability than the rule neural networks.Therefore, the establishment of wind power forecasting model becomes possible by use of small-world neural network. This paper intends to discuss the key issues of small-world network theory and its application in wind power forecasting in depth and establish the theory of short-term wind power prediction based on small-world networks. The main results are as follows.1.The tabu real-coded small-world optimization algorithm (TRSWA) is proposed. The convergence behavior of TRSWA iterative decoding scheme is depicted by establishing the Markov model. The almost everywhere strong convergence of the TRSWA is proved by Markov chain theory, which lays foundations for the farther study and application of the algorithm.Five benchmark functions are introduced to evaluate the performance of the TRSWA.Numerical experiments illustrate that the global convergence of the TRSWA is guaranteed if the feasible set is bounded. Lastly, the BP neural network model based on the TRSWA (TRSWA-BP) is proposed.2.Improved WSBP and NWBP small-world neural network models are proposed according to the complex network research. The network model structure, network topology and network model description of WSBP and NWBP are presented. The formulas of WSBP and NWBP small-world neural networks are derived. Simulation results show that the approximation property of the two models has been increased than the original BP network.3.According to the dynamic characteristics and mechanism of wind power forecasting, optimization mechanism of the small-world neural network is studied in the wind power forecasting model. The data makeup method based on NWBP small-world neural network is introduced. First, the wind farm 10min data (including wind speed, wind direction, power and temperature) are processed by the hour. The unreasonable and singular point data of actual wind are processed by NWBP data makeup method. Then the TRSWA-BP(or WSBP, NWBP) small-world neural network is used in the short-term wind power forecasting. The empirical mode decomposition and phase space reconstruction are introduced to TRSWA-BP (or WSBP, NWBP) small-world neural network which is used to predict the power values. From the forecast results, the proposed models can increase wind power prediction accuracy of the BP model and effectively revise the error due to wind power fluctuations and randomness.Their precisions are higher and the predicted time is feasible.4. The optimization variable weight combination prediction model based on small-world is used to predict the wind power. The gray association culling criteria of γi< 0.7 is proposed. Six kinds of prediction methods are chosen. The gray correlation analysis is used between the six methods and the actual power sequence. Cointegration test is adopted and then the redundancy methods are removed. Combined with the actual situation of wind farms, the models generated by screening are used for the variable weight combination forecasting. Simulation results show that, the variable weight combination forecasting model of small-world optimization can enhance the prediction accuracy and improve the system performance prediction.5. The probabilistic uncertainty analysis method based on Monte Carlo principles and the quantile regression analysis method based on NWBP small-world network are given. The effectiveness of the two uncertainty analysis method in confidence level of 95%,90% and 85% is discussed by simulation analysis and the confidence interval is given in the corresponding confidence level. The prediction interval coverage probability and the prediction interval normalized average width of the two uncertainty methods are compared. In the time scale of 1h,4h and 6h, probabilistic uncertainty method is appropriate used, and in the time scale of 24h, the quantile uncertainty method based on NWBP small-world neural network is superior.The research of this paper has far-reaching theoretical significance for the small-world network theory. The application of small-world network theory has practical value in solving high-risk dynamics system identification problem.