Study And Application Of Self-organization Structure On The Extreme Learning Machine

With the development of the artificial intelligence technology,the ANN(artificial neural network)algorithm is widely used in the fields of data mining,pattern recognition,image detection,face recognition.And the ANN algorithm has achieved excellent results in the modeling and the analysis.The ELM is a typical single hidden layer feed-forward neural network.The initial weight and the threshold value are randomly determined by the extreme learning machine.As long as the number of the hidden nodes is reasonably set,it can have faster running speed and lower computational complexity as well as good generalization performance.The number of the hidden layer nodes of the extreme learning machine needs to be determined according to the experience.And the number of the hidden nodes has a great influence on the modeling results.The random determination of the input layer to the hidden layer weight and threshold will result in the instability of the model performance.Therefore,the new self-organizing neural network SO-CSLN(self-organizing cosine similarity learning network)is proposed in the paper,which can automatically determine the number of the hidden layer nodes an d get a stable structure.The research contents of the paper are as follows:1.Because the determination of the hidden layer nodes number for extreme learning machine depends on the experience.The different number of the hidden layer nodes has a great influence on the result.In this paper,based on the research results of the central limit theorem,the number of the hidden layer nodes of the extreme learning machine is determined according to the self-organization of the sample.The experimental results show that the propose method can accurately determine the number of the optimal hidden nodes.2.Because the extreme learning machine randomly generates the number of the nodes between the input layer and the hidden layer,the performance of the extreme learning machine is unstable.Based on the information entropy and the cosine similarity theorem,the weights between the input layer to the hidden layer and the hidden layer to the output layer are optimized.Optimize the network structure of the extreme learning machine.Verify the correctness and the effectiveness of the SO-CSLN on the different UCI datasets.Compared with the traditional ELM and BP,the SO-CSLN has the higher accuracy and the generalization ability.3.Finally,the SO-CSLN is applied to the construction of the production forecast model of ethylene production system in petrochemical industry.The experimental results show the effectiveness and practicability of the method.At the same time,it can guide the production of ethylene and improve the utilization of energy.4.The modeling and analysis system of chemical production based on B/S architecture is designed and developed.And the system can visualize,predict and classify the chemical data.