Research On Neural Network Fault Diagnostic Method Based On BBO

Taking problem of the pump-jack fault diagnosis in the process of oil recoverytechnology as background, a new type optimization algorithm biogeography-basedoptimization was improved, and was applied to neural network training, the optimized neuralnetwork was applied to pump-jack fault diagnosis, the concrete content is as follows:First of all, the chaotic thoughts was fused into BBO (Biogeography-based Optimization)algorithm, to form a new hybrid algorithm CS-BBO: Chaotic sequence was used to initializethe habitat, the ergodicity of the initial population was improved; Hybrid migration strategywas added in the migration mechanism, a small disturbance was added into the selectedemigration habitat, the directivity of the migration mechanism was enhanced,and a range ofsearch accuracy was increased, the convergence rate was accelerated; Degradation of chaoticmutation operator was added in the mutation, the search scope was enhanced, some extent thelocal optimum was avoid.The each step of the CS-BBO algorithm was introduced in detail.The convergence of the CS-BBO algorithm was proved, through the benchmark function test,and comparing with traditional BBO algorithm performance.Second, CS-BBO algorithm was applied to train BP (Back Propagation) neural network,The SIV in the BBO is weights and thresholds in the neural network; The HIS in the BBO isSystem target error in the neural network; The neural network based CS-BBO algorithm wasbuilded. The defects of conventional BP algorithm, i.e. slow training speed and prematureresult, were improved by the algorithm which has search ability and the ability to make fulluse of existing informations, the method and optimization process were given, according tothe training sample experiment simulation, compared with the traditional BP algorithm andthe other group of intelligent algorithm optimized BP algorithm.Finally, the trained neural network was used in pump-jack fault diagnosis. Inrod-pumped well current data processing to extract the feature vector; The neural networkbased CS-BBO algorithm was builded, the diagnosis was carried out on the test sample, andthe diagnosis result comparing with other diagnosis methods of diagnosis.