面向智能温室物联网的分布式实时复杂事件处理研究

In Internet of Things (IoT) in intelligent greenhouse, controls are triggered by complex scenarios. A data analysis and mining technology is needed, which can precisely detect complex scenarios and then trigger appropriate controls in real-time. Traditional threshold control methods that are widely used in intelligent greenhouses hardly meet the demand because of their week description of complex scenarios..This project is aim to explore developing complex event processing (CEP) technology for IoT in intelligent greenhouse. In CEP, data from sensors is viewed as atomic event stream, and then complex scenarios i.e. complex events are intelligently detected by matching patterns of atomic events, in order to assist decision. The technology has been widely used in IoT for industry, smart home, etc..In the project, based on analysis of operating data from several real running greenhouses, event algebra will be improved to model complex events occurred in greenhouses, considering time and space relating among data from greenhouses, and a special real-time CEP algorithm will be developed, especially, which will guarantee satisfaction of strong time restriction in greenhouse control with the help of real-time scheduling theory. A software toolkit for modeling complex events from IoT in greenhouses and processing them in real-time will be also developed..Possible achievements will enrich theories of CEP and real-time scheduling. More importantly, agricultural production in China urgently requires more advanced information technologies. The project is a new exploration that advanced intelligent computing technologies will make greenhouse “smarter”.

智能温室物联网中，控制措施往往由复杂情景触发。迫切需要一种数据处理技术，能精确检测复杂情景，更有针对性地实时触发控制措施。当前温室常用的阈值控制方式的复杂情景表意能力较弱，难以满足需求。.本项目探索为智能温室物联网发展复杂事件处理技术，将传感器数据视作原子事件流，通过匹配原子事件发生模式，智能检测复杂情景，即复杂事件，辅助决策控制。该技术已经在工业、家居物联网领域得到了广泛应用。.本项目将在分析智能温室实际运行数据的基础上，首先，针对数据时空关联性，研究改进事件代数，建模智能温室复杂事件；其次，将研究实时复杂事件处理算法，特别借助实时调度理论，确保满足温室控制较强时间约束要求；第三，将研发智能温室物联网复杂事件建模及实时处理引擎工具集。.预期成果将丰富复杂事件处理、实时调度两个理论。更重要的是，农业生产对先进信息技术具有迫切需求，本项目是一种新探索，利用先进智能计算技术提升温室智能化水平。

在智能温室中，各类传感器设备监控作物、环境、土壤信息，协助精确调控环境、作物生长。一个传感器所监控的是原子信息，而触发温室控制措施的往往是复杂情景，由数个或数十个状态量或变化量反映。需要引入专家系统对传感器数据进行实时综合分析，检测复杂情景，指导作业控制措施的实施。如何更好的利用物联网数据更精细地指导生产作业，是发展智能温室物联网过程中的一个紧迫问题。迫切需要一种数据处理技术，能够应对复杂情景，更有针对性地触发控制措施，且能够满足控制实时性的要求。.本项目主要目标是探索将复杂事件处理引入智能温室物联网，将传感器数据视作原子事件流，将需要处理的复杂情景视作复杂事件，通过分析挖掘物联网数据，智能检测复杂事件发生情况，实时辅助决策控制措施。.项目组三年内完成了如下工作：.1）改造了4座温室、1个大田，取得了2年~半年不等的数据集，为本项目的展开奠定了基础。.2）研究了传感器优化部署算法，对温室传感器部署进行了优化，确保了用尽量少的传感器，反映温室的现实准确状态。.3）采用两类深度学习技术，对温室短期温度进行了预测，预测效果达到作物生长需求，4小时短期预测准确度，误差小于0.5℃。.4）设计并实现了大田及温室两类复杂事件描述语言，研发了复杂事件处理工具。.5）研发了温室物联网智能控制系统和温室VR物联网演示系统。.6）在国内不同地方5座不同温室得到了初步应用，取得了较好的经济社会效益。