基于DFNN的金融股指预测及金融非线性系统辨识研究

针对证券市场内部结构的复杂性、外部因素的多变性，本文采用动态模糊神经网络（DFNN）进行金融股指预测。DFNN能够实现在线学习，并且参数估计与结构辨识同时进行；同时采用误差下降率（ERR）修剪技术，保证网络拓扑结构不会持续增长，避免了过拟合及过训练现象，确保了DFNN的泛化能力。本文以上证指数为例．通过与同样以高斯函数作为传递函数的RBF算法预测结果的比较和分析．表明DFNN预测上证指数的偏差较小，预测的方向准确性较高。通过DFNN模型提取的模糊规则对金融系统运行模式进行分析．为研究金融非线性系统辨识提供了启发性思路。

The Study on the Application of DFNN in Stock Index Prediction and Financial Nonlinear System Identification

Aiming at the complexity of inside structure and levity of exterior complication in system of stock market which make stock market prediction a complex problem, method of modeling dynamic fuzzy neural network （referred to as DFNN） that is based on thorough study of the difficult problems facing stock predication is proposed. The fuzzy neural network based on extended radial basis function neural networks is equivalent to TSK fuzzy systems on the function. DFNN could achieve online learning, parameter estimation and structure identification. The algorithm uses the error reduction rate （referred to as ERR） pruning technology, network structure does not guarantee sustained growth, to avoid over fitting and over-training phenomenon, to ensure that the system＇s generalization ability. In this paper, DFNN shows smaller deviation and higher accuracy in prediction of Shanghai Composite Index, comparing with RBF algorithm and gets fuzzy rules of financial system which reveal the financial system operating mode.