Alternative Title拡張隠れセミマルコフモデルによる複数系列データモデリングとデータ収集・管理手法
Note (General)In recent years, with the development of devices and the development of data aggregation methods, data to be analyzed and aggregating methods have been changed. Regarding the environment of Internet of Things (IoT), sensors or devices are connected to the communication terminal as access point or mobile phone and the terminal aggregate the sensing data and upload them to the cloud server. From the viewpoint of analysis, the aggregated data are sequential data and the grouped sequence is a meaningful set of sequences because the group represents the owner's information. However, most of the researches for sequential data analysis are specialized for the target data, and not focusing on the "grouped" sequences. In addition from the viewpoint of aggregation, it needs to prepare the special terminals as an access point. The preparation of the equipment takes labor and cost. To analyze the "grouped" sequence and aggregate them without any preparation, this paper aims to realize the analysis method for grouped sequences and to realize the aggregation environment virtually. For analysis of grouped sequential data, we firstly analyze the grouped sequential data focusing on the event sequences and extract the requirements for their modeling. The requirements are (1) the order of events, (2) the duration of the event, (3) the interval between two events, and (4) the overlap of the event. To satisfy all requirements, this paper focuses on the Hidden Semi Markov Model (HSMM) as a base model because it can model the order of events and the duration of event. Then, we consider how to model these sequences with HSMM and propose its extensions. For the former consideration, we propose two models; duration and interval hidden semi-Markov model and interval state hidden-semi Markov model to satisfy both the duration of event and the interval between events simultaneously. For the latter consideration, we consider how to satisfy all requirements including the overlap of the events and propose a new modeling methodology, over-lapped state hidden semi-Markov model. The performance of each method are shown compared with HSMM from the view point of the training and recognition time, the decoding performance, and the recognition performance in the simulation experiment. In the evaluation, practical application data are also used in the simulation and it shows the effectiveness. For the data aggregation, most of conventional approaches for aggregating the grouped data are limited using pre-allocated access points or terminals. It can obtain the grouped data stably, but it needs to additional cost to allocate such terminals in order to aggregate a new group of sequences. Therefore, this paper focus on "area based information" as a target of the grouped sequences, and propose an extraordinary method to store such information using the storage of the terminals that exist in the area. It realize the temporary area based storage virtually by relaying the information with existing terminals in the area. In this approach, it is necessary to restrict the labor of terminals and also store the information as long as possible. To control optimally while the trade-off, we propose methods to control the relay timing and the size of the target storage area in ad hoc dynamically. Simulators are established as practical environment to evaluate the performance of both controlling method. The results show the effectiveness of our method compared with flooding based relay control. As a result of above proposal and evaluation, methods for the grouped sequential data modeling and its aggregation are appeared. Finally, we summarize the research with applicable examples.
2017
Collection (particular)国立国会図書館デジタルコレクション > デジタル化資料 > 博士論文
Date Accepted (W3CDTF)2018-02-03T13:25:44+09:00
Data Provider (Database)国立国会図書館 : 国立国会図書館デジタルコレクション