Abstract | ||
---|---|---|
In a competitive electricity market, wind power prediction is important for market participants. However, the prediction has not a general solution due to its inherent uncertainty, intermittency, and multi-fractal nature. This paper firstly constructs a small-world BP neural network (SWBP) with weight convergence and statistics analysis in order to build a maximum approximation for its nonlinear computation. Then, a modified mutual information (MI) is presented to select the input features for the SWBP, whose selection criteria is to establish the relationship between the numerous candidate features of the input and output associated with the wind power prediction by eliminating the redundant. Thirdly, the improved SWBP based on the modified MI is compared with the BP network upon the 15-min-ahead wind power prediction for performance testing, which includes convergence, training time, and forecast accuracy. Moreover, mean value method is adopted to smooth the volatility of selected input. At last, illustrative examples based on the 4-h-ahead rolling prediction are given to demonstrate its stability, validity, and accuracy of the proposed methodology contrasted with the BP, PSOBP, and RBF neural network algorithms. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1007/s00521-017-3262-7 | Neural Computing and Applications |
Keywords | Field | DocType |
Small-world BP neural network (SWBP), Weight convergence, Mutual information (MI), BP neural network, Mean value method, Wind power prediction | Convergence (routing),Electricity market,Mathematical optimization,Nonlinear system,Intermittency,Input/output,Mutual information,Artificial neural network,Wind power,Mathematics | Journal |
Volume | Issue | ISSN |
31.0 | 7 | 1433-3058 |
Citations | PageRank | References |
3 | 0.41 | 5 |
Authors | ||
4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Shuangxin Wang | 1 | 6 | 2.30 |
Meng Li | 2 | 4 | 2.48 |
Long Zhao | 3 | 78 | 13.96 |
Chen Jin | 4 | 3 | 0.41 |