基于非单点模糊正则网络的时间序列预测模型

Time series prediction model based on non-single fuzzy regular network

目的建立一种新的具有抗噪声能力的神经网络时间序列预测模型。方法通过将非单点模糊系统引入正则神经网络结构来建立模型。结果具有很强的抗噪声能力,且收敛速度快,全局搜索能力强。将该模型用于实例样本的预测,并和别的神经网络预测模型相比较。抗噪声能力的神经网络时间序列预测模型性能,比神经网络预测模型的性能显著提高。结论所建立的模型在性能上有显著提高,在一定程度上解决了视经网络的固有缺陷,今后有待降低计算复杂度。

Aim To construct a new time-series forecasting model based on neural network with the capability of noise immunity. Methods Introduce a non-singleton fuzzy system into the structure of the regular neural networks to construct the prediction model. Results This model has the strong capability of noise immunity, quick convergence rate and powerful ability of global search. The model is used to the forecast of samples and the performance is improved compared with the results of other neural network based forecasting models. Conclusion The performance of the new Model is improved obviously. And it overcomes the inherent disadvantages of neural networks to a certain exent. The reduction of the computation complexing is under further study.