ART2 Neural Network And Chemical Patterns Clustering Analyse

Clustering analyse is one of the ways to know things. We can know the essential characters of things more clearly by clustering analyse. In chemistry and chemical engineering study, we often need clustering analyse to the data and patterns. There are mostly two means of clustering analyse now. One is statistical clustering, and the other is neural network. The statistical clustering includes systemic clustering, dynamic clustering, fuzzy clustering, best division, etc. Neural network is a new clustering way which is studied much and develop rapidly recent years. Neural networks used to cluster include BP-network, fuzzy network, self-orgnization map network, adaptive resonance theory network, etc. In this paper, adaptive resonance theory network(ART2) is chosed. ART2 neural network will not destory patterns already stored by system when studies new patterns. So it can quickly recognize stored patterns and cluster patterns. It also can harmonize stability and flexibility of network by modify vigilance parameter. This paper introduces the structure and theory of traditional ART2(T-ART2) network, and analyse the clustering mechanism of T-ART2 network. Becourse T-ART2 network uses only category template to verify similar degree between new pattern and stored patterns, it isn't sensitive to gradual altering patterns. So the clustering capability of T-ART2 network is not very good. This paper bringsforward a new ART2 network named bidirectional matching ART2(BM-ART2), which can cluster pattens more strictly by setting two bounds for every category. BM-ART2 network is more senstive to gradual altering patterns than T-ART2 network. The simulation experiment proves that the clustering capability of BM-ART2 network excels previous ART2 networks.Systemic clustering is another method often used. It has excellences and defects camparing with neural networks. It is discovered by experiment that the stability of systemic clustering is better than neural network. It is affected less by the sequence of patterns. Its precision is high. But it need too much time and space so its efficiency is low. Neural network can cluster quickly and has higher efficiency than systemic clustering. But it is affected by the sequence of patterns more and its precision is lower than systemic clustering. So, this paper explores combining BM-ART2 network and systemic clustering and using the twos to cluster chemistry patterns. It manifests through experiment that the combination of BM-ART2 and systemic clustering gets a balance between efficiency and precision. It can accommodate the relation of efficiency and presion by modifying vigilance parameter.This paper also introduces correlative knowledge of clustering analyse. It explores how to evaluate the result of clustering and the abnormal phenomena in clustering. In the last, it brings forward theprospect in later study.