Study On Several Hybrid Intelligent Computing Methods And Applications

Intelligent computing involves mathematics, physics, physiology, biology,neuroscience, computer science, smart technology and other related disciplines.Intelligent computing mainly includes artificial neural networks, fuzzy system, andevolutionary computation. Based on the study of the compositional mechanism andthe thinking modes of human brain, in order to simulate human intelligence, artificialneural networks simulate the structure and the operating modes of human brain toconstruct artificial neural network models. Fuzzy system attempts to describe andprocess the fuzzy concepts existing in human languages and thoughts, which alsoaims to stimulate the human intelligence. Evolutionary computation is aself-organization and a self-adaptive artificial intelligence technology which solvingproblems by simulating biological evolution and mechanisms. In this paper, artificialneural networks, genetic algorithm (GA), particle swarm optimization (PSO), supportvector machines (SVM) and other intelligent computing methods are discussed intheir theoretical and applicable fields. In connection with the noninvasive intelligentdiagnosis of coronary heart disease, geological hazard risk assessment, forest-fireprediction, starch price prediction and other practical problems, different hybridintelligent computing methods are established. Topics include:1. Study on the problem of the noninvasive coronary heart disease diagnosis.Coronary heart disease (CHD) is a coronary artery atherosclerotic lesion which occursand causes vascular stenosis or occlusion, myocardial ischemia, hypoxia or necrosisand finally leads to heart disease.Coronary angiography is the only direct way to observe the coronary arteryshape, and is called “Golden Standard” by medical world. However, it is a traumaticdiagnosis, which needs high-level medical conditions, and it can cause some serious complications and even death with careless operation, which limits the widelydevelopment of this diagnosis.In this paper, for the noninvasive diagnosis of coronary heart disease problem,the author adopts14parameters related to CHD, combines BP algorithm with GA,selects Levenberg-Marquardt (L-M) algorithm to iterate instead of Gradient Descent,proposes GA-BP hybrid intelligent algorithm and builds the simulation model ofintelligent diagnosis based on the coronary heart disease cases of Cleveland ClinicFoundation. According to the data testing, it demonstrates that the proposed methodcan acquire diagnostic results with high accuracy, which can effectively solve thenoninvasive diagnosis of coronary heart disease problem.2. Study on the problem of geological hazard risk assessment. Natural hazardrisk assessment has a long established history, but its related branch, geologicalhazard risk (GHR) assessment, is a newly-emerging research area. In this paper, inorder to solve the problem of the GHR assessment, the author extracts seven impactfactors which relate to the problem, proposes GA-BP-PSO hybrid intelligentalgorithm. Firstly the network is trained by using GA to determine a superior solution,and then the result will be trained as the network initial parameters of BP algorithm.This method can increase the classification capacity of network, so that it can avoidthe local optimal solution. In this process, PSO is adopted to iterate instead of gradientdecent algorithm in order to optimize connection weights and thresholds, which cansignificantly improve network convergence rate, thus, quickly obtain the globaloptimal solution. By using GA-BP-PSO algorithm, calculating geological hazardmonitoring data of Jilin Province in recent ten years, the simulation model of GHRassessment of Jilin Province is established, and the distribution of GHR assessment inJilin Province is also acquired. The data testing shows that this novel algorithm cansolve the problem of GHR assessment with higher accuracy.3. Study on the problem of forest-fire forecasting. Forest-fire refers to the forestburning which is out of control. Furthermore, forest-fire has a huge impact onbiodiversity conservation and ecosystem sustainability issues, and will bring greatharm to human survival and living because it is sudden and strong, seriously destructive and difficult to be succored. In terms of forest fire forecasting problem,two hybrid intelligent computing methods, namely GRA-BRLMBP algorithm andMIV-BPSVM algorithms are proposed. On one hand, GRA-BRLMBP adopts graycorrelation analysis as a pre-processor to eliminate the factor properties whosecorrelation is relatively small in the predication property, and then uses the LMalgorithm after Bayesian regularization to iterate BP neural network, therebyestablishing GRA-BRLMBP simulation prediction model. On the other hand,MIV-BPSVM adopts MIV-BP algorithm to screen factors, and then creates asimulation model by SVM algorithm, thus completing the forest-fire forecasting. Thedata testing shows that for the forest-fire prediction, GRA-BRLMBP algorithmgreatly improves the convergence speed and generalization ability of the network.MIV-BPSVM algorithm has higher accuracy, both of which are the effective hybridintelligent algorithms when solving the problem of forest-fire forecasting.4. Study on the problem of starch price forecasting. Starch is one of importantbiological resources for human survival. However, due to large and complex starchindustries, the trends of starch prices cannot be forecasted correctly and timely. Tosolve this problem, the author proposes a GA-SVR hybrid intelligent computingmethod. Firstly, the shortcomings of the regression type of SVM algorithm are studiedand improved, using GA to do parameter optimizations, thus establishing a GA-SVRstarch price forecasting model. Based on the data from2003to2011, the starch pricesimulation forecasts the validity of the algorithm.