The Study And Application On BP Neural Networks And ANFIS In Multi-sensors Grain Information Fusion System

Most of th e grain must be stored after its production．The uses of th egrain and its quality on the market h ave close relations with the grainstorage technique．The grain storage is with characteristics of centralized，large volu me and space， which ma kes moni toring the grain storageinfor mati on and its adjust me nt a co mp licated project． If the infor mationcould not be properly fused a nd took measures in time， it will certainlyresult in either quality or grain losses． Conseq uently， in order to ensur ethe stored grain safety， only coul d we ob tain the st ore situationinfor mati on and ma ke an acc urate judgment in ti me， the s afe storag e canbe confir med．By using existing grain situation monitoring syste m， we co uldcollect the multi-factor s of grain storage， which ma k e s up fo r theinaccuracy and inco mplete i nfor mation only tes ted by a s ingle sensor an dlower the error conclusion possibility caused by the data deviation an dinsufficient． Multi-sensors technology could perfectly cope with theseproble ms， by appl ied multi-sensors infor mation fusion t echnology int ograin storage; we c ould conc lude accurate safety assess ments for g rainstorage at any ti me，which lay a fou nd ation to elevate grain storage safety，meanw hile， also give out a w ay to grain quality evaluation and scie ntificstorage．During the grain storage procedure， the eleme nts influencing thegrain mos t are temp erature，moisture，insect pests and micro-o rg anis m，these elements ma kes grain store stability and safety vary，causing qualityloss． Thi s paper is mainly research an appropr iate algorithm to fu se thedi fferent infor mati on with different physical meanings measured b y multi-sensors, finally get the consistent description to the moment. Moisture condensation, stack fever, stack mildew and insects are the most conditions appearing in storage process, and these common conditions are resulted from several influence elements, among these elements correlated and interacted with each other, so multiple influences with different degrees. Because of the AI, ANN and FIS are provided characteristics with highly nonlinear, perfect error tolerance and calculation randomness, also could calculate out the mathematic models with great complexity by invoking learning algorithms; Meanwhile, these systems mostly parallel process data and distributed store the information with flexible architecture, which makes ANN and FIS meet the data fusion requirement.Extract the characteristics of the collected data, make process and transformation according to the fusion goal, input them to the fusion system mainly built up by BP and ANFIS respectively, obtain a precise grain store assessment to a certain moment, and simulate in MATLAB, then carry out multi-aspect surveys, testing, analysis and contract to the simulation process and the results. The key are data collection must fulfill the object’s stochasticity, that data must cover generality and specificity, and then finding out a proper encoding method to construct an economic fusion technique, finally select optimized fusion technique to fuse the data.In this paper, the recognition frame is established on basis of the Technical Criterion for Grain and Oil Storage of PRC, Evaluation System on Safe Grain-Storage Technology Indexes and the practical experience of the staff in the studied grain depot. The fusion result mostly fit the reality.