Research On Algorithm And Model Of Artificial Immune System In Gas Outburst Prediction

As a massively parallel intelligent message process system, artificial immune system (AIS) has offered the new ideas for real-time project problem settlement. Immune algorithm is a new information processing technology, which collects unique evolution learning mechanism of biological immune system, such as adaptability, self-organization, diversity, immune memory, evolve, etc. And AIS has been a hot topic in computational intelligence after artificial neural network (ANN) and evolutionary computation nowadays. According to the demands of coal and gas outburst forecasting, the research in this dissertation is oriented on immune algorithm and immune network model and their application to coal and gas outburst prediction.Meeting gas monitoring large-scale database, some aspects of association rules mining algorithm need to be improved mining efficiency, availability, accuracy and so on. A concentration restraining clonal selection algorithm (CRCSA) is proposed. CRCSA obtains inspiration from concentration restraining principle and secondary response of immune system, which introduces immune memory to divide the antibody into the memory unit and ordinary unit and adopt different variation tactics to the memory unit and ordinary antibody. Through joining random antibody and mutation antibody, CRCSA can keep antibody diversity and ensure the quick convergence and the optimal solution. CRCSA association rules mining method is put forward and is applied to coal and gas outburst forecasting. The results show that CRCSA association rules mining method is effective in practice.Aimed at the sensitive restraining threshold shortcoming of artificial immune network (aiNET), distance concentration adaptive artificial immune network (DCAaiNET) is given in this dissertation, and network learning algorithm convergence of DCAaiNET is proved. DCAaiNET model adopts antibody selection and restraining mechanism based on distance concentration, which can limit the antibody of higher concentration and strengthen antibody diversity and reduce some parameters influence to DCAaiNET especially restraining threshold. Then cut down calculation cost and accelerate convergence speed and reduce subjective factors influence of DCAaiNET. DCAaiNET is applied to clustering analysis, and average running time of clustering algorithm is the shortest and iteration steps is the least and clustering accuracy is the highest. Using DCAaiNET clustering method in coal and gas emission prediction, satisfied results are achieved and the DCAaiNET clustering method is good.Based on analysis of the basic principle and features of immune algorithm, this dissertation integrates respectively the space search advantages of immune algorithm and chaos optimization algorithm and presents a chaotic immune optimization algorithm (CIOA). By applying chaos mutation operator to producing new antibody and applying immune selection operator to realizing"the survival of the fittest", CIOA is able to maintain a good diversity. At the same time, CIOA has higher convergence speed and it can avoid falling into local optima effectively. CIOA is used to optimize radial basis function neural network (RBFNN) which be effective for nonlinear time series prediction and prediction accuracy than conventional forecasting models have some improvement. Finally, the CIOA RBFNN is applied to coal and gas outburst forecasting and the result is satisfied.At last, gas outburst forecasting prototype system based on AIS is developed. In real coal and gas outburst forecasting application, through clustering analysis and association rules mining of underground work face gas gushing, seam's depth, coal construction thickness, gas content and other factors on the impact of gas outburst is found. By constructing RBFNN for gas emission quantity time series prediction, gas gushing prediction is realized, and the prediction result is identical with the fact.