Research And Application Of Semantic Pattern Mining Technology For Time Series In Internet Of Things

In recent years,due to the constant popularization of Internet of Things(IoT),sensor devices are widespread.In order to achieve the purpose of controlling network,managers and researchers need to find and analyze the implicit state in time series data.Thus,it is very common to apply data mining technology to time series analysis in recent researches.The time series mining method based on modeling can reveal the inner working rules of the sequence,so it is widely used in IoT.In order to reveal the inherent relationship of time series data with modeling method,this paper chooses the Hidden Markov Model(HMM)as foundation.Combining with the characteristics of time series and diagnosis and prognosis requirement,this paper proposes A cluster-based HMM for high-level state discovery(CHMM).In this model,the original time series is extracted by clustering method.And then the result of clustering is used as the initial observation sequence.Furthermore,this paper uses extended Viterbi algorithm to learn the observation sequence and hidden state sequence simultaneously and proposes an iterative method to acquire the optimal sequences.For purpose of analyzing the system fault and remaining useful life,this paper proposes a state-based HMM for Fault Diagnostics and Prognostics(SHMM-DP).According to the obtained hidden states with CHMM,this model can be used in two applications.One is about diagnostics.The diagnosis rules are established according to the sequence states information and a combination rule set with predefined ontology is used as the knowledge base.So that the possible faults in the current sequence can be reasoned,and the root cause of the fault can be obtained.Another is for prognostics,in this application,the HMM modeling of the obtained states through the Baum-Welch algorithm.By comparing the current sequence in each state HMM model on log-likelihood probabilities,we can get the most possible state of current sequence.On this basis,this paper uses regression model to establish the relationship between log-likelihood probability sequence and remaining useful life sequence.The remaining useful life predict results can obtained through this regression model.The experiment results show that:the CHMM model is effective in finding the hidden states of a system.The SHMM-DP model can predict the remaining useful life of a system with high precision.