基于BP神经网络的自适应模糊半参数时间序列模型
Adaptive Fuzzy Semiparametric Time Series Model Based on BP Neural Network

本文介绍了一种自适应模糊半参数时间序列模型，该模型将半参数技术与反向传播神经网络(BPNN)相结合，形成具有LR-型模糊数据的自适应模糊时间序列模型。首先，提出了基于非参数核、加权最小二乘和交叉验证的混合方法，该方法可以同时估计回归参数和光滑函数以及带宽的最优值；其次，基于非线性残差序列建立BP神经网络，通过神经网络的运算得到新的偏差，使得在不确定条件下获得更丰富的信息，提高了预测精度。本文采用一些常见的拟合优度准则来检验所提出的自适应模糊半参数时间序列模型的性能。通过一个模拟仿真的例子，说明了该方法的有效性。最后，对所得结果的统计分析表明，该模型对模糊时间序列数据的预测具有可靠性和有效性，优于其他模糊时间序列预测模型。
In this paper we introduce a fuzzy semiparametric time series model aggregating semiparametric techniques and neural network into adaptive fuzzy time series model with LR-type fuzzy data. First, a hybrid approach including nonparametric kernel-based method, weighted least squares and cross-validation method is suggested, which could simultaneously estimate both regression param-eters and smooth function of the innovations along with optimal value of the bandwidth. Next, back propagation neural network (BPNN) is created based on the nonlinear residual sequence and then new deviations are attained as a result of operating with neural network, so that the richer infor-mation can be obtained under uncertain conditions and the prediction accuracy is improved. Some common goodness-of-fit criteria are employed to examine the performance of the proposed adap-tive fuzzy semiparametric time series model. A simulation example is given to illustrate the effec-tiveness of this method. Last, statistical analyses of obtained results indicate that the proposed model is reliable and effective for predicting fuzzy time series data and superior to other fuzzy time series forecasting models.