Research On Sensor Data Complement Technology Based On Collaborative Filtering

The Internet of Things (IoT) is an important part of the new generation of information technology. According to the agreed protocol, it can also connect any items to the Internet for information exchange and communication by radio frequency identification, infrared sensors, laser scanners and other information sensing device. And it implements intelligent identification, positioning, tracking, monitoring and management. Compared with the traditional Internet, the IoT is a ubiquitous network which is established on the Internet. A variety of sensing technology is widely used in the IoT. What is more, the IoT has converged wired and wireless networks. To this end, people need to deploy a variety of types of sensors for networking structures. The IoT periodically collects environmental information to obtain real-time data, updates real time dynamic data in data repository to provide a unified information resource for all types of business support, and achieves the application of the IoT across all industries. However, because of the sensor physical environment, as well as the sensor itself energy, memory and other hardware constraints, sensor often suspends or even fail, and are unable to submit data. Therefore, because of the limitations of the sensors, real-time completion of the missing data to ensure the reliability of data transmission has aroused people’s attention, and has become one of the focuses of the current research.Aiming at the problem of real-time filling missing data in sensor networks, this thesis presents a complete framework based on sensor networks data collaborative filtering. This thesis uses the idea of collaborative filtering through the cooperation mechanism, and evaluates the sensor data to obtain the recommended values to complete the missing data. Firstly, in regard to the sensor with historical information, this thesis takes Pearson similarity as indicators to examine the match degree of the current monitoring information and historical information, and selects the best historical data as the completion data. Secondly, in regard to the missing sensor without historical information, this thesis uses cluster analysis to establish the sensor evaluation model, and clusters the sensor network according to the Euclidean distance and Mahalanobis distance, and then completes the missing data by the sensor monitoring data. In a number of environmental indicators that sensor network monitor, in regard to the indicators that physical and chemical relationship exists, this thesis presents a missing data completion method from a logical point of view, in order to effectively improve the to complement data quality and run time and to meet the real-time constraints premise, this thesis uses BP neural network algorithm based on the real-time data stream of sensor networks. BP neural network algorithm obtains the mathematical relationship between the amount of physical and chemical through training, and achieves efficient data completion function. Finally, the thesis designs three experimental programs, analyzes and evaluates the algorithm respectively from the efficiency of the algorithm, parameter setting, and the applicable conditions to verify the effectiveness and practicality of the method.