基于模糊递归小波神经网络的葡萄酒品质预测

Quality Prediction of Wine Based on Fuzzy Recurrent Wavelet Neural Network

摘要:

针对葡萄酒品质预测模型难以建立的问题,提出一种基于模糊递归小波神经网络的葡萄酒品质预测模型。利用葡萄酒物理化学指标和品酒师打分作为模型的输入输出,采用梯度下降算法在线学习隶属函数层中心、宽度和小波函数平移因子、伸缩因子、自反馈权重因子以及输出层权值。仿真实验时,首先利用Mackey-Glass混沌时间序列进行了性能测试,然后利用UCI数据集葡萄酒品质数据对所建立的品质预测模型进行了验证。结果显示,与多层感知器、径向基函数神经网络等传统前馈神经网络相比,构建的模糊递归小波神经网络品质预测模型具有更高的预测精度,更加适合于葡萄酒的品质预测。

Abstract:

Due to the difficulty in establishing wine quality prediction model, the modeling method based on fuzzy recurrent wavelet neural network is developed in this paper. Physical and chemical indicators and taster scoring are used as the input and output of the model. The parameters of network, such as centers and widths of membership layer, translation and dilation factors of wavelet function, self-feedback weight factors, and weights of output layer, are trained online by gradient descent algorithm. In simulation experiments, Mackey-Glass chaotic time series is tested firstly, and then the wine quality data of UCI data set is used to verify the quality prediction model. The results show that prediction model based on fuzzy recurrent wavelet neural network has higher prediction accuracy, compared with traditional feedforward neural network.