大型场景海量传感信息弹性感知机制及验证

Large-scale sensor networks (LSSNs) are usually characterized by a huge amount of sensor data, complicated connectivity, wide coverage areas and increasingly stringent response-time requirements. These requirements present a significant challenge to sensor data processing and perception. This is especially so when the computational resources available are limited. This project aims to tackle this challenge by using attention-like mechanisms, inspired by the human-sensory system. This attention process will be based on the integration of both user-defined (top-down) preference and sensory (bottom-up) information. By approaching the large scale and complexity problem through Bayesian attention modeling, we will provide an elastic framework for sensor data management in LSSNs. The theoretical and practical value lies on the facts that (1) it provides a unified theoretical system where the design of several major components including data representation, resource allocation and scheduling and human-machine interface control can be guided by the attention mechanism. (2) A decentralized framework will be developed based on a mechanism that processes sensor data locally in both spatial and temporal domains. (3) The events to be detected are divided into two groups: user pre-specified events and unpredictable abnormal events. The processes of detecting the two groups of events are firstly separated and then integrated by using the proposed Bayesian attention mechanism. By doing so, the system is able to pay attention to user pre-specified events and, in the meantime, process unpredictable abnormal events. (4) Our work will also include developing a functional prototype equipment. It will be validated in a large-scale practical industrial production monitoring project.

大型传感系统通常具有多源海量数据、复杂连接、大空间尺度等特点，在处理资源受限的条件下实现高效场景感知是一个严峻挑战。本项目探索学习人类感知智能，通过融合自底向上传感信息和自顶向下用户信息，研究面向大型场景感知的贝叶斯注意力选择方法，建立海量传感数据条件下弹性感知机制。主要研究内容包括：（1）利用注意力选择机制指导数据表达、资源分配、人机交互等方法的设计与实现，使大型复杂场景感知建立在一个统一的理论体系之上；（2）研究空间域和时间域中信息局部处理机制，建立分散式信息管理框架，提高海量信息处理效率；（3）将事件检测分解为面向已知确定事件和面向未知异常突发事件检测过程，利用贝叶斯注意力选择机制实现两者的融合，使系统能够在重点完成预定的检测任务的同时，兼顾对未知异常突发事件的处理；（4）研制实验设备，在实际大规模的工业生产监控项目中，验证理论成果和关键技术。

大型传感系统通常具有多源海量数据、复杂连接、大空间尺度等特点，在处理资源受限的条件下实现高效场景感知是一个严峻挑战。本项目取得的成果主要包含“理论和关键技术”和“实验设备研制和系统验证”两个方面。在“理论和关键技术”方面：本项目探索学习人类感知智能，通过融合自底向上传感信息和自顶向下用户信息，研究大型场景海量传感数据条件下高效感知机制。本项目组发表学术文章23篇，申请发明专利4项，软件著作权1项。在“实验设备研制和系统验证”方面：本项目强调在大规模工业生产监控实际项目中，验证理论成果和关键技术，保证技术的研究价值和成果的真实性。我们结合煤层气工业生产大规模、海量传感数据传输和处理的实际需求，研制了“智能节点”实验设备，完成了实验室测试和现场实验。实验结果显示了本项目研究成果的理论价值和实际应用价值，具有较广的推广应用前景。