Research On User Behavior Prediction Techniques Based On Smart Home

With the rapid development of the Internet of Things and sensor technology,the applications of smart home have been developed rapidly.However,current smart home systems is simply repeated according to predefined control procedures and rules,and cannot provide customized services that meet users individual needs based on their daily life habits.This thesis researches on the methods of predicting users behaviors based on data mining,in order to provide the relevant technical support to customized services in the smart home.This thesis discusses the state of the art of smart home and data mining,and carries out related technical analysis,including ZigBee technology for short-range wireless communication,data acquisition and previous data processing techniques,and behavior recognition methods such as support vector machine,naive Bayes classifier and Hidden Markov Model.The thesis researches on the behavior prediction methods in the application environment of smart home,and analyzes the requirements and technical issues from the aspects of data collection and previous data processing,data storage,and behavior prediction.The association rules mining used for behavior prediction in smart home application scenarios are determined,and the optimized and improved approach is proposed,in which the hashing or partitioning technique is adopted to improve the efficiency of the data mining process,basic neighbor sorting algorithm is used to clean up duplicated data,association rule mining algorithm is used to clean up missing data,and data transformation is performed by a function.A prototype of smart home behavior prediction is designed and implemented,in which the data acquisition and previous data processing module collects data in a periodic manner,deploy cameras and multiple sensors,and the data storage module uses the entity-relationship data model to store data,and a grid-based spatiotemporal indexing technology is proposed to improve data query efficiency.The behavior recognition and prediction module uses hidden Markov model for behavior recognition,and the association rule mining algorithm is used for behavior prediction.The experiments and tests results show the valid of the proposed optimized and improved approach in this thesis.