A large scale mobile Cyber Physical System (CPS), which consists of a large number of mobile devices interacting with each other and the physical environment, is an integrated system of computation, networking and physical processes. In recent years, CPSs have gradually transformed how people interact with and control the physical world in many domains: agriculture, transportation, health-care, manufacturing, energy, defense, aerospace, buildings, etc.;A large scale mobile CPS understands the physical world by sensing data to estimate the status of physical fields. This thesis focuses on two major tasks of large scale mobile CPSs: field estimation and route planning. The task of field estimation is to use sensing data of physical fields to estimate two statuses: 1) physical field: a physical quantity, represented by a number or tensor, that has a value for each point in space and time, such as air pollution, temperature, moisture, noise, traffic, etc; 2) system status: the conditions of the system's mobile devices such as location, mobility, sensing accuracy, etc. The task of route planning is to design the routes for mobile devices in the system for data collection, which guarantees field estimation to achieve application specific accuracy.;However, the real system faces two main challenges: lacking dense coverage and lacking even distribution of data collection. A dense coverage requires that the percentage of the overall space and time period being sensed by the mobile devices in the system should exceed a minimum number. An even distribution requires the information entropy of data distribution over space and time should exceed a minimum number. To improve the coverage and evenness of the data distribution, route planning designs routes for mobile devices to make sure that they sense data at designated locations and times. Since route planning relies on field estimation, especially system status estimation (e.g. locations of mobile devices), inaccuracy from field estimation deteriorates route planning performance. In addition, many real-world systems are semi-controllable. Only a fraction of total mobile devices follow the suggested routes from the system. This leads to two challenging problems: how to select mobile devices for route planning and how to design routes for the selected mobile devices.;The thesis presents a spatiotemporal relationship aided framework for large scale mobile CPSs, which incorporates a new spatiotemporal relationship analysis layer to address the challenges of lacking dense coverage and lacking even distribution of data collection. By utilizing the spatiotemporal relationships of physical field and system status in the spatiotemporal relationship analysis layer, which are discussed in Section 2, models and algorithms are designed to improve the performance of major system tasks: field estimation (physical field and system status) and route planning. I deploy real testbed experiments and extensive simulations with real world collected data to validate the system design. As a part of the evaluation for uncontrolled to controlled motion aspects of our system, air pollution sensors are deployed on the taxi-based testbed to collect data in the city of Shenzhen for 2 years in collaboration with Tsinghua University. In addition, a swarm of 8 micro aerial vehicles are deployed in an indoor environment for autonomous navigation. The results show incorporating the spatiotemporal relationship analysis layer can achieve 2.1x and 6x error reduction on physical field and system status estimation and 3x improvement on route planning. This illustrates the potential of the spatiotemporal relationship analysis layer to improve the performance of field estimation and route planning in large scale mobile CPSs.
展开▼
