基于粒计算的模糊神经建模方法在电能输出预测中的应用

准确地预测电厂的电能输出可以节约成本从而获得最大利润,因此建立一个模型来预测电厂的满载电功率输出是非常重要的。粒计算(Granular Computing, GrC)是一种新型的数据挖掘方法,它将具有类似特性的对象组合在一起,通过选择合适的粒度提取核心信息,减少冗余,降低问题求解的复杂度。本文使用GrC方法,从复杂多维数据集中以信息粒的形式建立初始的模糊推理系统,再通过模糊神经网络学习方法对系统参数进行优化。这种基于GrC的模糊神经(Granular Computing based Neuro-Fuzzy, GrC-NF)建模方法,不仅可以降低问题求解的复杂度,而且可以保持模糊逻辑系统的可解释性,将其与模糊神经网络的结合又提高了建模精度。本文将该方法用于建立电功率输出的预测模型,通过其预测精度的比较表明了该方法的优越性。

Predicting Electrical Energy Output by Using Granular Computing Based Neuro-Fuzzy Modeling Method

Accurate prediction of electrical energy output can save more cost and attain maximize profits, so it is quite important to establish a model to predict the electrical energy output of the plant. Granular Computing (GrC) is a new data mining method. By combining these objects with similar characteristics and selecting appropriate granularity, CrC can seek a better solution, in which the core information can be extracted while reducing redundant information and the computing complexity. In this paper, the GrC method is used to extract relational information and data characteristics from a complex multidimensional data set. The extracted information is further utilized to model an initial fuzzy system, whose parameters are optimized by using the fuzzy neural network learning methods. The proposed method can not only reduce the solving complexity, but also achieve the interpretability of fuzzy logic. Meanwhile, the accuracy of modeling can be improved due to the integrating of fuzzy neural networks. Finally, the proposed method is utilized to construct a predictive model of electrical power output of a power plant. The comparison via the predicting accuracy demonstrate the superiority of the proposed method.