The Application And Study Of Sensor Modeling Based Fuzzy Analysis And Multi-Model

Soft-sensing technology, also known software technique, it select a set of variables easy to detect and auxiliary variables which are closely with the dominant variables in the industry under the optimal certain criteria, by constructing the mathematical relationship between the dominant variables and auxiliary variables to establish soft-sensor model. Soft sensor technology, it is popular with industrial sector for its significant diversification benefits and modeling diversity. Neural networks, support vector machines, they have been applied to the petroleum, chemical industry, because they can approximate linear and nonlinear function of any capacity with any precision. However, due to the existence of the actual production process and nonlinear characteristics, a single model often can not meet the technical precision.Since adding several models to improve prediction accuracy and robustness, it has made considerable progress that the prediction method based on multi-model research, and has been successfully applied in industry. To solve the problem that a single model can not fully describe precisely the non-linear characteristic of the wood moisture content in the full scale. Wood moisture measurement based fuzzy of multi-modeling method is presented in this paper, firstly, the soft measurement technique and wood moisture measurement methods are described, and sub-modeling algorithm and multi-model theory are studied. Finally, multi-modeling method and its application in the measurement of wood moisture content are given in the paper.The first step of the method is to classify the output of the training set samples by similarity criteria by FCM algorithm, then correspond the input samples to output samples after classification, and select the typical value of each input samples as the center of the mean vector; the second step use BP or SVM method to model for a subset of each type of samples, select which method based on the number of the subsets after FCM clustering,to adopt the SVM algorithm or BP network based on characteristics of the different application of choice.To select the different modeling method is drawn from the experience. We discover that SVM can solve the small sample, nonlinear and high dimensional problems and when sufficient training samples, BP can better describe the complex nonlinear relationship in the use of the modeling process. Therefore, if the number of subset samples is more than 100, use the BP algorithm; if the number of subset samples is less than, or equal to 100, then use the SVM algorithm model. Finally, we get the average root mean square error. Simulation experiments prove that the precision of multi-modeling is better than single models for wood water content measuring.