基于新阈值函数去噪及SVM的建模方法研究

Study of Modelling Based on a New Kind of Thresholding Function's De-Noising and SVM

研究谷氨酸发酵生产过程中采集到的数据信号通常具有随机性和非平稳性,中间附加了各种噪声,以至于影响数据建模的拟合效果和泛化性能。利用小波分析的特点,在研究D.L.Donoho和I.M.Johnstone提出的小波阈值去噪方法的基础上,采用了一种新的阈值函数去噪方法。新阈值函数在整个定义域内统一定义,表达式简单易于计算,同软阈值函数一样具有连续性,而且是高阶可导的,便于进行各种数学处理。将方法与支持向量机(SVM)相结合对谷氨酸发酵过程进行建模。仿真结果表明,上述方法能有效恢复数据的真实性,所建SVM模型具有较高的拟合能力,且预测误差小,稳健性好。

The data gathered from the fermentation production process have the nonstationary and random nature. The data contain different noises inevitably, which will affect the fitting and generalization capability in data modelling. To resolve the problem, a novel thresholding function is presented based on the wavelet shrinkage put forward by D. L. Donoho and I. M. Johnstone and has many advantages over traditional thresholding functions. Comparing with the traditional thresholding functions, it is simpler in expression, as continuous as the soft - thresholding function, and has a higher order derivative which makes some kinds of mathematical disposals convenient. Combined the method with Support Vector Machines and applied to a modelling of glutamate fermentation, the experimental results indicate that it is effective to regain the factuality of the data. The model based on SVM has higher fitting abilities and less prediction errors.