基于小波神经网络的热值软测量建模

Modeling of Soft Measurement for Calorific Values Based on Wavelet Neural Network

提出了一种基于小波神经网络的软测量建模方法和学习算法,这种方法被用来"测量"裂解炉燃料气热值。小波神经网络具有2层结构:小波降噪层和多层感知器层。小波降噪层主要用来对过程数据进行降噪变换,提高源信号的信噪比,多层感知器层用来辨识过程模型。小波神经网络不但具有多层感知器网络的自学习和逼近性能,而且可以利用小波降噪理论,克服源信号的噪声干扰。小波变换可以变换初始数据并进行特征提取,变换后的数据具有更高的信噪比,仿真结果显示小波神经网络具有良好的逼近能力和泛化性能。

A kind of modeling and study method for soft measurement based on wavelet neural network（WNN） is proposed.It is used to measure the calorific values of fuel gas in the cracker system.The WNN model consists of two layers： the wavelet de-noising layer and multilayer perceptron layer.The wavelet de-nosing layer is mainly used for data de-noising transform to increase the signal and noise ratio of the source signal.Multilayer perceptron layer is used for identifying the process model.WNN not only owns the property of self-learning and approximation which are held by multilayer perceptions network,but also can overcome the noise disturb to the source signal using the de-noising theory of wavelet.The wavelet can transform the original data and extract the features.The transformed data shows much higher signal noise ratio.The simulation results indicate that the WNN owns much better approximation and the generalizing performance.