基于神经网络动态非线性非平稳经济系统预测

考虑实际经济系统中广泛存在着非线性和时变性因素，以及大部分变量的序列具有增增长特性，提出了用网络方法，建立实际经济系统的时变非线性模型，采用增广卡尔曼滤波算法训练神经网络，并根据先验信息（序列的时间增长特性）构造参数转移矩阵，对实际经济的预测分析结果证明，与传统定常非线性预测模型相比，该方法不仅可以在线递推预测，而且由于参数转移矩阵的引入，预测精度得到很大的提高。

Prediction of nonstationary and nonlinear dynamic economic systems based on neural network

A time-varying nonlinear model of macroeconomics is established based on neural network. Because there are many nonlinear and time-variant factors in actual economic systems and series of most economic variables increase with time, a parameter transfer function matrix is constructed based on this prior information. Neural network is trained by Extended Kalman Filtering algorithm. The simulation results show that comparing with traditional fixed nonlinear model the present method not only can predict recursively but also can reduce the error of predicting macroeconomics due to introducing transfer function matrix.