摘要
针对传统人工智能预测算法在对预测问题峰值变化处理问题上的不足,引入峰值识别理论改进BP神经网络预测模型(SIBP)。在此基础上,利用引入多向全局搜索机制的改进粒子群算法,对SIBP神经网络预测方法进行改进,提出一种具有峰值识别能力、全局学习能力更强的人工智能预测模型,以有效解决基于BP学习方法易于陷入局部极值的问题。将改进后的预测方法应用于"尖峰突变"比较突出的出清电价预测问题,以美国PJM电力市场2005-02-01至2005-05-16的实际数据为样本,对所提出的改进预测方法进行实证分析。研究结果表明:所提出的算法较改进前的BP算法对发生电价突变的短期电价预测精度提高10.16%,运算时间仅增加6.2 s,预测结果证明本文所提出的算法在处理峰值预测问题方面的有效性。
By adopting the spike identification mechanism to improve the traditional BP algorithm on the ability of spike prediction, a BP algorithm with spike identification （SIBP） was proposed. The multi-orientation searching-mechanism was introduced into the particle swarm optimization （PSO） for enhancing the global optimizing-ability, and the improved PSO was combined with the SIBP to avoid the problem of local extreme. A new forecasting method was proposed with stronger learning ability and spike identification. The improved predicting model was applied in the forecasting of market cleaning price（MCP） which fluctuated acutely. Based on the actual date of American PJM power market from 2005-02-01 to 2005-05-16, the new artificial neural network was inspected by comparison with the other different methods, The results indicate that, compared with BP, using the MCP spike prediction model, the forecast accuracy improves by 10.16%, and the time cost of the new method is only increased by 6.2 s and the MCP spike prediction model is effective to solve the prediction problems of peak values.
出处
《中南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第5期1054-1058,共5页
Journal of Central South University:Science and Technology
基金
国家自然科学基金资助项目(50077007)
关键词
峰值识别
粒子群算法
出清电价
预测模型
spike identification
particle swarm optimization
market clearing price
predication model
