Study On Content Of Soil Component Based On Neural Network And Genetic Algorithm

The Near Infrared Spectroscopy(NIR) analysis method has the characteristics:high speed of analyzing,no sample pretreatment,no devastating analysis,analyzing at long bowls and real time processing,good measuring repeatabilityand simple operating. And now,NIR has been the most developing andremarkable spectroscopy analytic technique. NIR combines with computertechnique and chemistry metrology, so it has the representative and bright epochcharacteristic with the ceaseless improving of digital of instrument measuringsignal and greenness in the course of analyzing.Our country is a farming country,majority of farmers. In the new century,China is confronted with the circumstance of dense people and exiguous soil,being short of resource,exasperate environment. The key of solving with 1.3billion people's food supply is driving the technology of agriculture. And all thesehave the large relation of soil in which the crop grows. The component of soil isone of the most influencing factors of crop gowning. And inspecting thecomponent of soil in good time and accurately is significant in agricultureproduction.The paper states the development of NIR and the research status in theanalysis of component of soil firstly,afterwards, introduces the theory of NIR,thestrategy of establishing model and establishes model of Principal ComponentRegression (PCR). And following,to the question of non-linearity in NIRanalyzing, the paper studies the application of annual neural network (ANN) andgenetic algorithm (GA) in analysis of component of soil and establishing modelwith BP algorithm and its improved grads descending and adapting learning ratealgorithm (BPX) and Levenberg-Marquart algorithm (LMBP). At the last,due tothe good ability of genetic algorithm' external searching,the author establishesmodel based on the combination of BP and GA.The paper primarily studies the analysis of soil with ANN,including threesegments:Segment 1:It introduces the theory of NIR,the strategy of establishing modeland establishes model of PCR. The way is based on Principal ComponentAnalysis (PCA) that compresses data,i.e. reduces dimensions. But it hasdisadvantages of evident non-linearity and weak anti-jamming ability.Segment 2:It studies the analysis of component of soil with BP algorithm,BPX and LMBP. The paper selects 50 samples in test field of the engineering ofmechanization of farming,Jilin university,yushu,Changchun. It selects most ofthe samples as training set. The author chooses 35 for training and the other 15 fortesting. Then gather the map of near infrared spectrum and reduce dimensionswith DCT to improve accuracy and rapidity. The simulation shows that BP cannearly describe the analysis of soil. BPX imports momentum to improveconvergent speed and error precision,but the momentum is conformed viaexperiment. The speed is much faster than BP algorithm,only several seconds toachieve the goal. But it needs more memory.In a word,BP network is nearly successful for analysis of non-linearitysystem,as a result of simple structure and algorithm,accomplishing training in arequired time. Moreover,its improved algorithm has good effect on analysis andcan predict the content of soil nearly correctly.Segment 3:It adopts GA to optimize weights and LMBP to optimizesubsequently as a result of GA's external searching and BP's local searchingability. The author designs and realizes the model based on combination of BP andGA. Its has high accuracy and the result shows the average relative error,averageabsolute error and determinative coefficient is better than the other models.Combining with the research content,the structure of paper is arranged asfollows:Chapter 1:It introduces the development of NIR and the research status inthe analysis of component of soil. It analyzes the need and feasibility of the paper.Make the aim and significance of the paper definite. At last,it confirms thecontent.Chapter 2:It introduces the theory of NIR,the strategy of establishing modeland establishes model of PCR.Chapter 3:It introduces the math model,network and common algorithm ofANN.Chapter 4:It introduces the theory,flow,basic operation and the method ofpreferences of GA.Chapter 5:It studies the theory and material realization of combination of BPand GA.Chapter 6:It studies realization of model of analysis of soil component andcompare the result.Chapter 7:It summarizes the full text and put forward the improved suggest-ions in the future.Meanwhile,the paper also has the shortcomings to be improved:1.Select more representative samples and process data more effective.2.As for the shortcomings of standard GA(SGA),adopt improved GA tooptimize wholly.3.The analysis of soil component is not precise because of less samples. Incomparison,BP and GA need more samples. So we consider other means,such asSupport Vector Machine (SVM) for finite samples searching optimizing duringcomplexity and learning ability of model. So it has unique predominance in theproblem of less samples , non-linearity and complicated many dimensionsrecognizing.