A novel rolling prediction method is introduced, which was an evolutionary Support Vector Machines (SVM) algorithm based time series. The algorithm would resolve how short term power load could be forecasted accurately and rapidly. The power load system was an uncertain, nonlinear, dynamic and complicated system, so it was difficult to describe such a nonlinear characteristics of this system by traditional methods. The algorithm avoided traditional model of SVM to control the kernel function and parameters and found the local and global optimization with Simplex-Niche-Genetic algorithm, which was more generalized and its dependence on experience was weakened. At the same time, the numerical simulation proved the method rationality and was applied to predict the power system load base on the time characteristic. Therefore, considering time series could predict the power system load in sequence and enhance the efficiency and the capability of prediction to maintain power system reliability and stability so as to make prediction model agree with the load's dynamic mechanical characteristic. And then, taking practical project as training samples to generalize and forecast the power system load, the result was satisfied, which illuminated forecasting accuracy was superior to traditional BP Neural Network. So prediction model put forward in the paper could get simulate result quickly and accurately, which provided a new thought for power system load forecasting and power system grid layout and construction. Index Terms— Power Load forecasting;Simplex-Niche- Genetic algorithm;Support Vector Machines(SVM);Time series
[1]
F. Girosi,et al.
Nonlinear prediction of chaotic time series using support vector machines
,
1997,
Neural Networks for Signal Processing VII. Proceedings of the 1997 IEEE Signal Processing Society Workshop.
[2]
Rafael Martí,et al.
Multilayer neural networks: an experimental evaluation of on-line training methods
,
2004,
Comput. Oper. Res..
[3]
M. K. Soni,et al.
Artificial Neural Network-Based Peak Load Forecasting Using Conjugate Gradient Methods
,
2002,
IEEE Power Engineering Review.
[4]
Tommy W. S. Chow,et al.
Neural network based short-term load forecasting using weather compensation
,
1996
.
[5]
Catherine M. Harmonosky,et al.
An improved simulated annealing simulation optimization method for discrete parameter stochastic systems
,
2005,
Comput. Oper. Res..
[6]
Magdi S. Mahmoud,et al.
Cascaded artificial neural networks for short-term load forecasting
,
1997
.