Artificial Nerve Network Integration Technology And Its Mineralsforecast Applied Research

The artificial nerve network(ANN)is the non-linear complex networksystem which the massive simple neurons joint becomes, by paralleldistribution processing, from the organization, auto-adapted, from the study, fault-tolerant and so on the good distinctive quality obtains people's attention. ANN in the solution such as pattern recognition, the distinction classification, the forecast appraisal, the examination diagnosis and the automatic control, the plan and the decision-making, the pattern recognition and the processing of the same kind question,has the remarkable superiority. But the above question in the placestudy information processing is the universal existence, thereforecaused generally has majored in engineering author's research to bewarm The nearly more than 20 years very many study recent years, along with the ANN technology unceasing consummation, solved each kindof place study question case using the ANN model success massively toappear And study the ANN model development presents the many kinds oftechnical synthesis integration the situation, the heredity algorithm,the wavelet transforms, method and the ANNs model fusion and so onsimulation algorithm as well as fuzzy logic, displays the solution study to analyze the Central Africa linear question thesuperiority.Although the nerve network already obtained the successful applicationin very many domains, it exists the traditional question certainlynot very good solution like partial minimum, the study, has exudedquestion and so on ability. In practical application, because lacksthe question experience, often needs to pass through massively takesthe trouble to consume when the experiment to try to find out candetermine the appropriate nerve network model, the algorithm as wellas the parameter establishment, its application effect completely isdecided by user's experience Even if uses the similar method solutionsimilar question, because the operator different its result very isalso possible to be widely divergent In the practical application, the operator often is the ordinary engineers and technicians whichlacks the nerve computation experience, if does not have the projectnerve computational method which easy to use, the nerve networktechnology application effect very will be difficult to obtain theguarantee.In 1990, Hansen and the Salamon foundation proposed the nerve networkinte gration (neuralnetworkensemble) the method They proved that, maysimply through train many nerve network and carry on its result thesynthesis, remarkably enhances the nerve network system to exude theability Afterwards, Sollich and Krogh were the nerve networkintegration has given a definition, namely"the nerve networkintegration was carries on the study with the limited nerve network tothe identical question, integrated the output decides together undersome input demonstration by the constitution integration various nervenetwork under this demonstration output"Because realized containsthe huge potential and the application prospect to the nerve networkintegration, after Hansen and Salamon, very many researchers have allconducted this aspect research How but did then research work mainlyconcentrate in uses in the nerve network integration technology theconcrete application domain Starts from the 1990s in termediatestages, the related nerve network integration fundamental research hasreceived the enormous value, the massive researchers well up into thisdomain, the theory and the application achievement unceasingly emerge,cause the nerve network to integrate into at present internationalmachine learning and a nerve computation quite active research hotspot.This article studies the heredity algorithm (Genetic Algorithm, GA)with the radial direction base nerve network (Radial Basis Function,RBF) integrated, simultaneously, attempted the heredity algorithm REFintegration applies to the minerals forecast appraisal research. This article's sresearch results mainly has:1. studied nerve network integration algorithm Bagging and Boosting, RBF and the GA algorithm, has analyzed the algorithm goodand bad points, discussed several methods fusion superiority. 2. The heredity algorithm to optimized the question to have theoverall situation to seek the superior ability, this article proposedbased on heredity algorithm GARBaggRBF and the GARBoostRBF sorterchoice algorithm. The attempt carries on the optimization using theheredity algorithm to the RBF integration synthesis weight, thenchooses most superior partial members RBF to participation integratesincreases the classified precision. With the aid of the heredityalgorithm auto-adapted heuristic overall situation search, final willrestrain separately corresponds the nerve network integration eachmonomer, the realization integration simultaneously the trainingprocess The simulation result indicated this algorithm is effective,both maintained the heuristic optimized method parallelism, enhancedthe nerve network integration study efficiency, and may guaranteebetween the monomer network is mutually independent, has biggerdifference, further strengthens the nerve network integration to exudethe performance Potential.3. proposed one kind based on partial precision dynamic integratedalgorithm DERB agg RBF and DERBoostRBF, according to waits theclassified test sample in its periphery to be close to spatial thepartial precision to choose part of members sorters to participate inthe integration, obtains the classified result after the voting or theweighting voting. The dynamic RBF nerve network integration method, inview of certain nerve networks, uses the weighting most close neighbortechnology to collect them to exude the error message constitutionper formance matrix, the dynamic choice exudes the error smaller nervenetwork in this foundation, forms integrated after the consecutivemean the finally output result. The experiment indicated that,compares with the other means, this method has the satisfyingperformance.4. based on Matlab nerve network toolbox improvement heredityalgorithm radial direction base network integration simulationtraining 5. It uses inherit algorithm's radial direction base networkintegration to apply in the mineralization quantitative evaluation,and with many Yuan statistical, the mathematical quantification theoryand other nerve network methods will carry on the comparison. Hasdemonstrated this method in the minerals forecast appraisalapplication prospect.