Node Fault Diagnosis Of Wireless Sensor Networks Based On BN

Wireless Sensor Networks?WSN?as a new platform for obtaining information,enables people to obtain comprehensive and effective information during human-computer interaction,and has been successfully applied in many important fields.However,in practice work,due to the complexity and uncertainty of WSN,and when the deployment environment is relatively harsh,the node is easy damaged.Therefore,it is particularly important to effectively monitor the current operating state of the WSN node.Because the advantages of Bayesian Network?BN?,in dealing with uncertain information,it received many experts'and scholars'extensive attention.Aiming at the fault diagnosis for the uncertain complex system of WSN,this paper proposed a fault diagnosis method based on BN for WSN nodes.Based on the analysis and research on the operating mechanism and fault symptom attributes of WSN,this paper proposed a WSN node fault feature extraction method to obtain fault feature samples.However,during the BN modeling,the actual number of fault feature samples is often limited to obtain,aiming at the problem that the classical BN parameter learning algorithm learns the poor precision of the model parameters under the small data sets,a Constrained Data Maximum Entropy?CDME?algorithm for BN parameter learning is proposed,and applied CDME algorithm to WSN node fault diagnosis reasoning,verified the effectiveness and practicability of the algorithm in the actual WSN node fault diagnosis.The main research work of this paper is as follows:?1?Aiming at the problem of WSN node fault feature sample extraction,a WSN node fault feature extraction method is proposed.Firstly,analyzed the WSN operating mechanism and different WSN node fault states;Then simulate the S₁^S₈ fault symptom attributes such as?whether the measured node has a return value?generated by the WSN node under different fault conditions;Finally,used the node state resolver to collect the fault symptom attribute data of the WSN node under different fault conditions,and the WSN node fault feature sample is obtained.WSN node fault feature samples are obtained by out-of-order and clean processing,and are used as WSN node fault BN modeling and diagnostic reasoning input.?2?Aiming at the uncertainty problem in the fault diagnosis of WSN nodes,this paper proposed a fault diagnosis method based on BN for WSN nodes.Firstly,obtained the fault feature samples of different states of the WSN node;Secondly,under the sufficient fault feature samples,the BN parameter modeling is performed by the maximum likelihood estimation method;finally,input the fault symptom sample to be diagnosed into the established BN model,and used the Junction Tree inference algorithm to implement WSN node fault diagnosis reasoning.At the same time,under the same experimental conditions,the WSN node fault diagnosis experiment was also carried out through the Radical Basis Function?RBF?neural network.The comparison shows the WSN node fault diagnosis method proposed in this paper has higher diagnostic accuracy,classification effect and timeliness.?3?Aiming at the problem that the classical BN parameter learning algorithm has poor learning accuracy under small data sets,this paper proposed a CDME parameter learning algorithm.Firstly used a small data set to estimate a set of BN initial parameters;Secondly,the qualitative expert experience knowledge is transformed into a series of equality and inequality constraint sets;Then,under these constraint sets of BN parameter,used the Bootstrap algorithm to generate the parameter candidate set satisfying the constraint;finally,the BN parameter is calculated by weighting according to the maximum entropy of the information.Multiple sets of experiments were performed under virtual data and real data.The experimental results show that when the amount of data is sufficient,the learning precision of CDME parameter learning algorithm is similar to that of classical MLE algorithm,which indicates the correctness of the algorithm.Under the condition of small data sets,the CDME algorithm can be used to model the parameters of BN.Learning accuracy is better than Maximum Likelihood Estimation?MLE?algorithm and Qualitative Maximum Posterior?QMAP?algorithm.?4?Apply the CDME algorithm to WSN node fault diagnosis,verify the validity and practicability of using CDME algorithm for WSN node fault diagnosis under small data sets.Firstly,under the small data set and the sufficient data set,used the CDME algorithm to model the WSN node fault BN parameters;then,used the Junction Tree algorithm to compare the two sets of established BN models.The experimental results show that under the condition of smaller data sets,the diagnostic accuracy of WSN node fault diagnosis method is close to the diagnostic accuracy of WSN node fault diagnosis method under sufficient data sets.The diagnostic reasoning results also confirm the effectiveness of the proposed algorithm,it provides a new way for parameter modeling and WSN node fault diagnosis under small data sets.