Research On Entity Similarity Analysis Technology For The Internet Of Things

"Interconnection of all thingsl" forms the Internet of things(IoT).The core of the IoT is "things",that is,entities.An entity may have several features,such as room's temperature and humidity.A type of entity feature is sensed by a type of sensor,so the relationship between the entity and the sensor is one-to-many.A large number of sensors are sensing the state information of various entities in the IoT,forming massive heterogeneous dynamic real time data information.People focus on the entity state information,rather than the sensor data.Therefore,how to deal with the massive sensor data to effectively mine its potential information and knowledge has become an important issue in the IoT technology.The similarity analysis technology has been widely used on the Internet and will be more widely used in the IoT.The similarity analysis technology can be applied not only to traditional similarity search services and recommendation services,but also to industrial IoT,intelligent agriculture,intelligent city and other fields.The complexity of IoT sensor data makes traditional similarity analysis technology no longer meet the user needs.Therefore,the IoT entity similarity analysis technology has become an important research issue.The entity similarity analysis technology of IoT is just emerging and the research based on quantitative sensor data is relatively few.Previous researches on sensor data fitting have relatively large errors,so the accuracy of similarity calculation is not high.In the case of small difference in entity features,the accuracy of single attribute entity similarity analysis technology is not good.In the complex environment of the IoT,few users put forward the requirement of single attribute similarity analysis.The single feature entity similarity analysis technology cannot meet the needs of users.Furthermore,the entity similarity analysis technology calculate similarity based on distance;redundant and irrelevant data not only increase the amount of calculation,but also affect the accuracy of calculation.In this thesis,the existing problems and challenges in the researches are studied from three aspects:similarity calculation method of entity state data piecewise,multi-feature entity similarity analysis method for IoT,features selection mechanism for the IoT entity similarity analysis technologies.(1)In order to solve the problems of large fitting error of sensor data and low accuracy of similarity calculation in previous studies,this thesis proposes a method of entity similarity calculation based on entity feature data piecewise-linear fitting method.Firstly,the sensor data piecewise points calculation method is proposed,and the sensor data piecewise-linear fitting method based on the data points can solve the problems of the priori polynomial obtained difficult for the least squares polynomial algorithm and the large fitting errors for the sensor data of simple linear fitting.And then,according to the piecewise point fitting functions,we proposed the sensor data similarity models construction method and the similarity calculation method.Finally,this method is applied for similarity search.Compared with the existing algorithms,the similarity search accuracy and speed are improved.Compared with the original data of the sensors,the data storage costs are reduced by at least two orders of magnitude.(2)The single feature entity similarity analysis technologies cannot meet user needs,which brings challenges to the IoT similarity analysis.To solve these challenges,this thesis proposes a multi-feature entity similarity analysis method for the IoT.Firstly,a new unsupervised aspheric clustering algorithm is applied to sensor data clustering,which can groups of inflection points of feature data and different attributes of IoT entities.And then,a multi-feature weighting method is proposed,according to the different degree of discrimination between entity attributes.Finally,combined with sensor data clustering algorithm and feature weighting method,we propose a multi-feature entity similarity analysis method using group fitting correlation calculation algorithm(GFC method).The GFC method is used for similarity search.The experimental results show that the GFC method compared with the single feature similarity search methods can improve the average search accuracy and increase the search speed,as well as compared with the original data of the sensor reduce the data transmission and storage costs.(3)To solve the problems that complex redundant data increasing the computational costs and reducing the computational accuracy of similarity analysis technologies,this thesis proposes the feature data selection method for IoT entity similarity analysis technologies.In this thesis,the typical feature selection algorithm(Relief algorithm)is improved for sensor data selection.In addition,the entity feature matrix between entities and features is defined to filter the common entity features.A three-component storage relation table of the entities,models and features for the dynamic data models creation is proposed.Finally,a feature selection method for IoT entity similarity analysis technologies is proposed(SMEF method).The SMEF method is applied to the similarity analysis technologies of single feature and multi-feature for similarity search.The experimental results show that similarity search methods based on SMEF method can improve the average search accuracy by more than 10%,as well as increase the search speed and reduce the data storage costs.