基于改进DFNN的短期电价预测新方法

提出了一种改进的动态模糊神经网络DFNN(Dynam ic Fuzzy Neural Network)的短期电价预测方法。首先对采集到的信息进行特征提取,然后利用模糊粗糙集理论中的信息熵进行属性简化、去掉冗余信息,最后用得到的属性作为动态模糊神经网络(DFNN)的输入进行训练预测。在模糊神经网络内部引入递归环节,构成了动态模糊神经网络,并采用具有全局寻优能力的遗传算法来训练网络,克服了单纯BP算法易陷入局部最优解的困境。最后以美国加州电力市场公布的2000年数据进行了模型训练和预测,结果表明该方法所建立的预测模型具有较高的预测精度。

A new approach to short-term price forecasting based on improved DFNN

An approach of improved dynamic fuzzy neural network for power system short-term price forecasting is proposed.Firstly,the fuzzy-rough set theory is applied to find relevant factors to the price among varied factors,then the dynamic fuzzy neural network(DFNN) model is trained using historical daily price and load data selected before performing the final forecast.The DFNN is constructed by introducing recursion segment in the fuzzy neural network,and the network is trained using the genetic algorithm and BP algorithm to avoid being trapped in the local convergence.With the established model,the day-ahead Market Clearing Prices(MCPs) of California Electricity Market are successfully forecasted.The analysis of the obtained forecasting results show that the presented method possesses better convergence and more accuracy.