Comparison of different BP neural network models for short-term load forecasting

• Published in: 2010 IEEE International Conference on Intelligent Computing and Intelligent Systems Vol. 3; pp. 435 - 438
• Main Authors: Yuan Ning, Yufeng Liu, Huiying Zhang, Qiang Ji
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2010
• Subjects: Bayesian regularization; Classification algorithms; Forecasting; Levenberg-Marquardt; Load modeling; Prediction algorithms; Rain; Short-term load forecasting(STLF); steepest descent algorithm
• ISBN: 9781424465828; 1424465826
• DOI: 10.1109/ICICISYS.2010.5658645

Short-term load forecasting(STLF) is of great importance for the safety and stabilization of grids. Based on the historical load data of meritorious power of some area in Guizhou power system, three BP neural networks in steepest descent algorithm back propogation neural network(SDBP), Levenberg -Marquardt algorithm back propogation neural network (LMBP) and Bayesian regularization algorithm back propogation neural network (BRBP) models in 24 hours ahead prediction are compared. Since the traditional BP algorithm has some drawbacks such as slow training convergence speed and possibility of local minimizing the optimized function, an optimized L-M algorithm, which can improve the stability of convergence and accelerate the training speed of neural network has been applied to carry out load forecasting work to reduce the mean relative error. Bayesian regularization also be applied which can overcome and improve the generalization of neural network. The prediction precision of BRBP are superior to LMBP and SDBP, while BRBP has poor training speed than others.