Research And Application Development On Knowledge Acquisition For Activity Pattern In Smart Environment

With the raise of people's concern degree for living comfort and health monitoring, Smart Environment based on Wireless Sensor Network technology is getting in-depth study and extensive application, and Smart Environment makes it possible for elderly. Children and patients to get automatically care. Activity recognition is an important basis for the deployment and implementation of Smart Environment, and its achievement needs to be guided by the activity pattern knowledge. However, at present, activity pattern acquisition methods mainly face two problems, one is spatial adaptability problem of activity pattern, and another is time consumption problem of activity pattern acquisition. These two problems affect the efficiency of Smart Environment's deployment and implementation, what's more, they increase the deployment and implementation costs of Smart Environment. As a result, these problems make it difficult to popularize Smart Environment in a large scale. Therefore, obtaining activity pattern knowledge that adapts to current Smart Environment quickly and effectively is one of the most urgent problems for research of Smart Environment.The main contributions of this paper on activity pattern acquisition method are shown as follows.Firstly, we analyze the typical activity data learning methods and activity pattern acquisition methods, combined with the goal of obtaining activity pattern quickly and efficiently, we propose that transfer learning is an ideal method for activity pattern acquisition.Secondly, due to the time characteristics and spatial characteristics of activities, activity model that can support the effective transfer learning of activity pattern is proposed. Then, activity information is preprocessed to meet the needs of activity pattern transfer learning. Considering the different effects of different types of nodes on the activity, node influence degree is introduced to do weighted processing of activity trajectories. Considering the fuzziness of time information, triangular membership is adopted to do the fuzzy processing of trigger duration.Thirdly, due to the transfer learning of activity pattern, trajectory mapping algorithm(TM) and trigger duration algorithm(TDT) are proposed in this paper, and they constitute transfer learning framework. Moreover, simulation experiments are carried out to verify the feasibility and effectiveness of the proposed transfer learning framework.Moreover, considering the non-uniqueness of the representation of activity information, activity model with the form of assembly is defined, and transfer learning framework for activity pattern based on ensemble information is proposed.Finally, to solve the problem that it is not easy to carry out experiments about Smart Environment actually, a simulation platform named Smart Environment Simulator Tool is designed and developed. It realizes activity monitoring function, activity simulation function and activity pattern data acquisition function, so that it provides a wealth of experimental platform for Smart Environment.