Research On Data Credibility Evaluation Method Based On Multi-source Heterogeneous Information Fusion

Safety supervision and monitoring system is an important application of industrial IOT(Internet of Things),with intelligent supervision,risk warning and other capabilities.Stable monitoring system has a great guarantee for enterprise property as well as people's life safety.However,the wireless sensor network data will become unreliable and untrustworthy because of some factors,such as sensor aging,instrument failure,environmental and human interference,resulting in supervision and monitoring system false alarms and missed alarms,substantially affect the efficiency of the system in intelligent supervision and system performance.Therefore,in order to improve the effectiveness and stability of the safety monitoring system in practical applications,the system should perform quality assessment of the input data.In this thesis,the data reliability assessment algorithm is the research topic.which is based on the information fusion technology to achieve data reliability assessment,so as to lay the technical foundation for ensuring the high quality of data.The main contents of this thesis include two parts shown as follows.(1)A data reliability assessment method based on multi-source homogeneous information fusion is proposed.Firstly,the characteristics of wireless sensor network data are analyzed in the spatio-temporal domains.Secondly,A data similarity metric based on the extend Jaccard function is proposed using multi-source data,which is a method to judge data similarity from a global perspective.Thirdly,in response to the drawbacks of similarity metric with inaccurate evaluation in some cases,the inter-data matching metric from a local perspective is proposed.The data reliability evaluation algorithm is composed of a weighted combination of this two metrics.Through testing on different data sets,it is verified that the method is effective in evaluating the reliability of different types of data.Finally,the assessment method based on single-source data and other assessment methods based on multi-source data are selected.Compared with other methods under multiple experimental scenarios,algorithm in this thesis are more accurate and can effectively evaluate the reliability of the data.(2)A data reliability assessment method based on multi-source heterogeneous information fusion is proposed.Firstly,the heterogeneous data of different magnitudes are standardized,and the spatio-temporal correlation between the data is analyzed by Pearson coefficients.Secondly,The multi-node multi-feature prediction model of this thesis is constructed based on the long and short-term memory network,and the output of the model is used as the estimation of the measurement data.Then the reliability measure module is used to calculate the data credibility of the actual measured values.Finally,different network models are selected for performance comparison,and the experimental results show that the multi-node multi-feature prediction model had higher prediction accuracy and better performance.Then two more models are constructed based on long and short-term memory networks,and compared with these two models,the prediction accuracy of the model in this thesis is improved by 28% and 85%,respectively.The results of the assessment methods based on different prediction models are analyzed,and the proposed reliability method is more accurate for multiple types of data,which verifies the accuracy and effectiveness of the method in this thesis.