Software And Hardware Feature Based System Failure Prediction Model And Application

With the development of computer technology,the related system is developing towards complexity,diversification and integration,the functions of the system need both complex hardware and software to complete.Interdependencies exist among subsystems,and failure of any subsystem may lead to the failure of the whole system.At present,many traditional methods only consider system failure purely from the perspective of software or hardware.The complex interaction between software and hardware that affects system failure is ignored and the failure data presents the characteristics of "big data".Based on the potential and advantages of deep learning in feature extraction and pattern recognition,a software and hardware feature based system failure prediction model is proposed in this paper.Hereby the failures of software and hardware in the system are analyzed,and the complex interaction between them and fault propagation are considered comprehensively.The main contents are as follows:Firstly,the software defect leads to software fault,which leads to software failure.An adaptive center-weighted oversampling algorithm is proposed for the imbalance of data classification in software defect prediction,and it is used to balance defect data sets.Regarding adaptive center-weighted oversampling algorithm,the appropriate neighborhood range and its neighborhood sample set for each minority class sample are determined,and then the adaptive gravity center in its neighborhood range,its neighborhood sample and the sample itself are used to synthesize new minority class samples.Meanwhile,according to the distribution characteristics of each minority class sample,the corresponding sampling weight is given to each minority sample.Finally,the balanced data set is used as the input of stacked denoising auto-encode to extract the feature of defect data and predict the defect.The experimental results showed that the proposed model has better prediction performance than traditional software defect prediction methods.Secondly,hardware fault results in hardware failure.A deep belief network fault diagnosis model based on adaptive GBRBM is proposed due to global constant learning rate often used for traditional deep belief network models may increase the number of model iterations and even affect the accuracy of model diagnosis.Based on an adaptive learning strategy,the continuous signal data is processed by Gauss Bernoulli restricted Boltzmann machine,and the learning rate is adjusted according to the error of iteration reconstruction in the process of model iteration.Moreover,in order to prevent the model from over-fitting due to the adjustment of learning rate,dropout is added to the training process of GBRBM to randomly delete hidden layer neurons with a certain probability.The time domain feature of fault signal and the energy feature of wavelet packet are combined as the input of the improved deep belief network model for feature extraction and fault diagnosis.The experimental results showed that the proposed model has better diagnostic performance than traditional fault diagnosis methods.Finally,aiming at the failure analysis often considers failure purely from the perspective of software or hardware,the extended Petri nets are proposed.The software defect,hardware fault and the interaction between software and hardware are analyzed respectively.Based on traditional Petri net,different set of places,logical relations and token color rules of hardware and software in the system are redefined.The extended Petri net model is used to analyze the possible failure reasons and fault propagation paths of complex underwater combat systems,and the effectiveness of the proposed model is verified.