Electronic Nose Dynamic Characteristics Of Information Processing Method

With the development of interrelated technology such as sensors technology, data processing technology, computer technology, artificial intelligence technology, software and hardware technology of parallel computers and the development of industry, the dynamic data fusion technique of electronic nose will be the important means for the development of electronic nose in the future.In this thesis, the principle and essence means of dynamic data fusion technique are introduced. Some common used methods of dynamic data pretreatment and pattern recognition algorithm are summarized. On the basis of this, the Multi-BP sub-Network and the model of Least-Square algorithm for dynamic data fusion in electronic nose are presented. In addition, RBF Neural Network is applied for dynamic data fusion and the method of multi-group forecast in electronic nose system is proposed. In order to compare the discriminating effect of different pattern recognition algorithm in dynamic data fusion technique, a set of detection system, which is combined gas sensor array with pattern recognition technology, is designed and constructed. By this system, experiments are carried out with drink samples in exoteric circumstance. The dynamic data are obtained in the condition of changeless and changed temperature, respectively.Qualitative recognition is processed from the dynamic data and results show that the capability of BP network is weak relatively because of long learning time and difficulty of convergence. It is usually difficult to identify the complicated odor successfully using general BP network in electronic nose. Ratio of recognition is 100% when we combine Multi-BP sub-Network with dynamic data in changeless temperature. When RBF Neural Network is applied in two kinds of dynamic data fusion, ratio of recognition can be enhanced from 89.7% and 70.4% to 100% and 95.3%, respectively. The model of least-square algorithm is simple and the efficiency is relative high. The preciseness ratio of recognition reached 100%. Therefore, the data of dynamic response can increase dimensions of analyzed sample and supplies sufficient information for gas recognition, which will benefits to enhance analysis precision of electronic nose system.