Research And Application Of User Behavior Anomaly Detection Algorithm In Smart Home Environment Based On Streaming Data

With the wide application of smart home technology,it has brought more convenience to people’s lives.The previous research mainly focused on the status monitoring and remote control of home equipment.However,how to use these data to further improve the life intelligence and user safety in the home environment has become a new research hotspot.On the other hand,with the rapid development of big data and cloud computing technology,data mining technology for big data is also widely used,especially the mining technology for realtime streaming data,which has become a new research boom.In this context,this paper takes the device flow data as the research object in the smart home environment,and studies the abnormal behavior detection algorithm of users,and builds a big data analysis platform for abnormal user behavior detection on this basis,in order to discover the home furnishing in time.The abnormal behavior of users in the environment will give an alarm,provide users with a stable security guarantee,and further improve the quality of life of users.The main research work and results of this paper are as follows:1.Aiming at the home equipment flow data collected in the smart home environment,a kind of user behavior anomaly detection algorithm based on time span and cluster difference index is proposed.The algorithm firstly performs offline clustering learning on the collected historical data of the status of smart home devices,and obtains each cluster set as the normal behavior pattern of users;The performance index algorithm is used to detect the concept drift of the device flow data;finally,the local outlier factor LOF algorithm is used to detect the specific time point when the user has abnormal behavior.Experiments show that the algorithm proposed in this paper can not only accurately detect whether the user has abnormal behavior,but also capture the time point when the abnormal behavior occurs.However,this algorithm also has three deficiencies:（1）due to the accumulation of certain data,there is a problem of detection lag;（2）the relevant parameters in the model need to be adjusted according to the streaming data mode,and cannot learn adaptively;（3）the algorithm It can only detect that an abnormality has occurred,and cannot identify what kind of abnormality has occurred.2.In order to further improve the real-time performance of user behavior change detection and identify specific anomalies,this paper proposes a user behavior anomaly detection algorithm based on the convolutional neural network model.The algorithm first converts the device flow data for a period of time into a grid map of user behavior patterns,and then obtains the corresponding pattern grid map according to several common abnormal behavior patterns of users,and then uses the convolutional neural network model to learn users.behavior pattern.In order to improve the computational efficiency,this paper also uses the density tracking algorithm of the behavior pattern grid graph to detect whether the user behavior pattern changes.When the change occurs,the behavior pattern grid graph is detected by the learned model,so as to obtain whether the user behavior is abnormal and the specific abnormal behavior.It can be seen from the experimental results that the proposed tracking algorithm can detect 100% of the time points when the user behavior pattern changes,and the trained model has an accuracy rate of 95.41% in detecting abnormal behavior categories,indicating that the algorithm can effectively Detect specific user abnormal behaviors.3.Based on the above research results,this paper constructs a user behavior pattern anomaly detection platform system for smart home environment.The system mainly includes two core modules: offline learning module and online computing module.In the offline module,the data is analyzed and processed,and the two-dimensional data is converted into a grid graph of behavior patterns and put into the VGGNe T16 convolutional neural network model for learning.In the online mode,the user’s real-time streaming data is calculated and processed through the Spark Streaming framework,and combined with the VGGNe T16 convolutional neural network model trained in the offline mode,the user’s behavior mode is analyzed and detected in real time.Finally,a visual web system is built to display the detection process and results of the system in the form of web pages.