Research On Intelligent Information Fusion Based On Support Vector Machines And Fuzzy Technology

It is well known that the fusion of support vector machines (SVM) and fuzzy reasoning(FR) is based on their comparability and complements. The advantages of SVM are its strongability on learning, and the advantages of FR are its ability on reasoning according to rules.Based on the relationship between SVM and FR, some new information process structures,patterns and algorithms can be constructed by using of their complements such asinterpenetration and promotion. Therefore, intelligent information fusion of SVM and FR isattracted more attentions recently than the other information fusion techniques.In this desseration, based on the current research status about multi-transducerinformation processes, we introduced the artificial intelligence and pattern recognitiontechniques into the fusion of SVM and FR, and proposed some new methods on fusion ofSVM and FR. The main contributions of this dessertation are summerized as follows:(1) In order to eliminate disturb effects caused by interference parameters, aninformation fusion techinque of SVM based on progressive transductive classificationlearning algorithm was proposed. Its application on correcting the nonlinear effects ofphotoelectric displacement sensors caused by interference parameters was carried out. Theexperimental results show that by using of the SVM information fusion technique, the errorsof photoelectric displacement sensor can be less than1.5%.(2) We also proposed a new FR fusion technique based on presenting information frommulti-sensors by fuzzy set and membership function and establishing fuzzy rules based onexpert knowledges. Experimental results on angle increment control show that highperformance on angle increment control can be achieved by using the proposed informationfusion technique. However, this technique still has a drawback that the subjection functions ofinput and output variables must be determined subjectively and its fuzzy rules and fuzzyreasoning strongly dependent on fuzzy reasoning process.(3) An adaptive mutative scale chaos genetic optimization algorithm (AMSCGOA) wasproposed by using one-dimensional iterative chaotic self-map x n+1=sin (2/xn)with infinitecollapses. The numerical examples with three testing functions reveal that the AMSCIOA hasboth high searching speed and precision. The average truncated generations, the distributionentropy of truncated generations and the ratio of average inertia generations are used toevaluate the efficiency of AMSCIOA quantificationally. Numerical examples show that theefficiency of AMSCIOA is higher than that of conventional genetic algorithm.(4) A new fuzzy least squares support vector machines classifier based on chaos geneticalgorithm (FLS-SVMBCGA) was developed, in which the fuzzy membership functions are setby using clear sets to construct a fuzzy set and its parameters are optimized by chaos genetic algorithm. The experiments were conducted on three benchmarking datasets for testing thegeneralization performance of FLS-SVMBCGA. The experimental results show thatFLS-SVMBCGA is valid for improving the prediction accuracy of the classification problemswith noises or outliers. Diagnosis results of TPD signals from oil and gas transmission pipelinerevealed that classifying performance of TPD signals from oil and gas transmission pipeline byusing FLS-SVMBCGA classifier is more accurate. It is useful for getting an accurate diagnosison TPD signals from oil and gas transmission pipeline.(5) Due to the fuzzy speciality of beforehand alarm analysis on region grain security, anew fuzzy least squares SVM beforehand alarm model of region grain security is developed,in which the fuzzy membership function is set by using clear sets to construct a fuzzy set andits parameters such as penalty factor and kernel parameter are optimized by using chaosgenetic algorithm. The application results revealed that beforehand alarm relative errors of thebeforehand alarm model were small, and the beforehand alarm effect of the beforehand alarmmodel is very suitable for practical applications.(6) Due to the characteristics of economic system, ecological system and socical system,the evaluation index system of rural cyclic economic system was constructed in this thesis.Furthermore, the evaluation model of the rural cyclic economical system was establishedbased on SVM fuzzy reasoning information fusion and some support vectors extracted fromtraining sample library in the index system of rural cyclic economic system as fuzzy rules byusing SVM. Application results reveal that the correctness and effectiveness of the evaluationmodel of the rural cyclic economical system were improved.