摘要
根据电力市场的相关历史数据准确地预测出未来的市场出清电价,对于市场中的各个参与者都具有十分重要的意义.在建立了一种粒子群优化(PSO)下的BP神经网络电价短期预测模型的基础上,采用PSO进化算法,反复抽取训练子集样本,通过对应的验证样本预测误差寻找近似最有代表性的训练子集,解决了模型的训练样本参数难以设置的问题.实验验证了该预测模型的有效性,结果表明处理好预测模型样本参数的选择问题,能够提高模型的稳定性及预测精度.
In the competition paradigm of the electric power markets, electricity price prediction tools are essential for bidding strategies of power producers and consumers. Considering that BP ( back propagation) neural network has strong ability in non-hnearity regression and its parameters are hard to set, this paper proposed a BP neural network based on particle swarm optimization (PSO) algorithm for electricity price prediction. Inspired by the fact that parameters selection of BP neural net- work is significantly influenced by the representative of training data, the full training data is split into training subset and val- idation subset. We obtain the approximation optimal training subset by repeatedly sampling from the full training data set ,employing the above BP neural network with this training subset, and testing the training subset on the validation subset. The effectiveness of the proposed model is demonstrated with actual data taken from the Australia power grid.
出处
《南昌工程学院学报》
CAS
2012年第1期13-17,共5页
Journal of Nanchang Institute of Technology
基金
Supported by National Natural Science Foundation of China(No.11126171)
Education Research Foundation of Yunnan Province(No.09Y0461)
Research Fundation of Baoshan College(No.09B017K)~~
关键词
BP神经网络
粒子群优化算法
训练子集
电力价格
预测模型
BP neural network
particle swarm optimization (PSO) algorithm
training subset
electricity price
prediction model