农业信息物理系统中不确定性复杂事件处理技术研究

发布时间：2018-04-24 19:17

本文选题：信息物理系统 + 复杂事件处理　；参考：《中国农业大学》2017年博士论文

【摘要】：信息物理系统(CPS)是通过计算(Computation)、通信(Communication)与控制(Control)技术的有机与深度融合,实现计算资源与物理资源的紧密结合与协调的下一代智能系统。CPS已被运用于医疗、能源、交通等多个重要发展领域,具有广阔的应用前景。CPS技术和农业的有机结合被称为农业CPS,农业CPS可能包含类型繁多的、相互连接的设备,这些设备不断产生大量的原始数据,同时目前的CPS系统大都是分布式部署,不同层次数据源的数据具有海量、异构和分散等特征,而传统的数据处理技术难以对其进行有效处理,可以引入基于事件的处理方法。农业CPS采集的事件流中,单个事件表达含义有限,用户更加关心的是反映农业智能控制逻辑的符合特定模式的事件序列,如何从事件流中识别这些有意义的事件序列是农业CPS数据处理中的一个难点,复杂事件处理(CEP)技术作为CPS的核心技术可以有效解决这个问题。而噪声、传感器误差、时钟不同步、网络负载和延迟以及其它原因,都会造成农业数据的不确定性,因此需要对不确定性的复杂事件处理技术开展研究。事件的不确定性可以量化为概率,因此不确定事件流的处理转化为对概率事件流的处理。概率复杂事件是由多个概率原子事件按照特定的模式复合后生成,在复合的过程中复杂事件的概率如何由原子事件概率计算生成以及如何按特定模式进行事件的匹配检测是需要研究的两个主要问题。本文在对目前CPS中CEP技术的研究现状和面临的挑战进行深入分析的基础上,以农业CPS为背景,用温室大棚作为实例,针对这两个问题,从复杂事件的概率计算、基于树的属性不确定时的复杂事件检测算法、以及基于概率时间Petri网的时间不确定的复杂事件检测模型几个方面开展了深入的研究。本文的工作主要包括以下方面:(1)研究了农业CPS中复杂事件的概率计算问题,提出了一种基于近似世系的概率计算方法。不确定复杂事件处理是要检测出满足概率阈值的事件序列,因此如何计算不确定原子事件流组成的复杂事件的概率是进行复杂事件处理时面对的主要挑战之一,本文加入数据世系管理理论,为了应对不断产生的海量的概率原始事件,引入充分近似世系计算算法,提出离散多项式近似世系计算算法,对两种算法进行分析和比较,实验证明两种算法均可进行数据压缩后再计算,从而减少计算量,提高计算效率,后者相较于前者压缩比例更高,更有优势。(2)研究了基于树的农业CPS中不确定复杂事件检测算法的问题,提出了针对一般属性不确定的PUCEP算法和时间属性不确定的编码ESI-tree解决方案。农业CPS系统中因为传感器误差、采集精度、网络通讯技术等等原因会造成事件属性的不确定性,本文提出一种解决一般属性不确定事件流的复杂事件查询检测方法--PUCEP算法,它在二叉树基础上融合了NFA方法,并将概率阈值引入其中进行优化,通过实验对比证明优化后的算法在多项性能上都有所改进;时间是事件众多属性中比较特殊的属性,针对其特殊性本文提出树形解决方案ESI-tree,再根据树的特点对树进行压缩,然后采用编码方法对其进一步优化,通过实验对比证明编码后的匹配算法在执行效率和内存占用等方面都具有优势。(3)研究了基于Petri网的农业CPS中不确定复杂事件检测建模的问题,提出了一种基于概率时间Petri网的复杂事件检测模型的方法。农业CPS中复杂事件由原子事件组合形成,原子事件的产生与CPS各个部分的采集、传输、转换等过程紧密相关,在这个过程中,由于系统误差或随机误差导致原子事件的时间不确定,本文针对时间的不确定性,以温室大棚为实例,利用概率时间Petri网对系统的原子事件建模,形成原子事件的模型,再将原子事件概率时间Petri网模块组合形成复杂事件模型,利用这个模型可以准确分析复杂事件语义,检测匹配过程中对复杂事件的误判、漏判,最后通过实例分析证明了该方法的正确性,并通过编程测试证明延时偏移概率越高,复杂事件检测的误判可能性越大,系统开销也越大。
[Abstract]:The information physical system (CPS) is an organic and deep integration of Computation, communication (Communication) and control (Control) technology, and the next generation intelligent system,.CPS, which realizes the close combination and coordination of computing resources and physical resources, has been applied to many important fields of development, such as medical treatment, energy source, traffic and so on, and has a broad application prospect,.C The organic combination of PS technology and agriculture is called agricultural CPS. Agricultural CPS may contain a wide range of interconnected devices. These devices produce a large number of original data. At the same time, most of the current CPS systems are distributed, and the data of different levels of data are characterized by sea, isomerism and dispersion and traditional data processing. It is difficult to deal with it effectively. The event based processing method can be introduced. In the event stream collected by agricultural CPS, the meaning of individual events is limited, and the user is more concerned about the event sequence that reflects the specific pattern of the agricultural intelligent control logic. How to identify these meaningful event sequences from the event flow is the agricultural CPS A difficult point in data processing, complex event processing (CEP) technology as the core technology of CPS can effectively solve this problem. Noise, sensor error, clock synchronization, network load and delay, and other reasons will cause uncertainty in agricultural data, so it is necessary to carry out complex event processing techniques for uncertainty. The uncertainty of the event can be quantified as a probability, so the processing of the uncertain event flow is converted to the processing of the probability event flow. The probability complex event is generated by multiple probability atomic events combined with a specific pattern, and how the probability of a complex event is generated by the probability calculation of the atomic event in a complex process and how The matching detection of events in a particular pattern is the two main problem that needs to be studied. On the basis of the current research status and challenges of CEP technology in CPS, this paper takes agricultural CPS as the background, uses greenhouse as an example, and aims at these two problems, from the probability calculation of complex events, based on the tree attributes. The complex event detection algorithm in uncertain time and the time uncertainty complex event detection model based on the probability time Petri net have been studied in several aspects. The work of this paper mainly includes the following aspects: (1) the probability calculation problem of complex events in agricultural CPS is studied, and a probability meter based on the approximate lineage is proposed. Calculation method. Undetermined complex event processing is to detect the event sequence that satisfies the probability threshold, so how to calculate the probability of the complex event which is not determined by the flow of the atomic event flow is one of the main challenges in the process of complex event processing. Starting events, introducing a full approximate lineage calculation algorithm, a discrete polynomial approximation algorithm is proposed, and the two algorithms are analyzed and compared. The experiment proves that the two algorithms can be compacted after data compression, thus reducing the amount of calculation and improving the computational efficiency. The latter has a higher compression ratio and a better advantage. (2) study Based on the problem of uncertain complex event detection algorithm in tree based agricultural CPS, a ESI-tree solution for PUCEP algorithm and uncertain time attribute is proposed. In agricultural CPS system, the uncertainty of event attributes will be caused by sensor error, acquisition precision, network communication technology and so on. A complex event query detection method --PUCEP algorithm is proposed to solve the general attribute uncertain event flow. It combines the NFA method on the basis of the two fork tree, and introduces the probability threshold into the optimization. The experimental comparison shows that the optimized algorithm is improved in many properties, and the time is more special among the many attributes of the event. According to its particularity, this paper puts forward the tree solution ESI-tree, then compresses the tree according to the characteristics of the tree, and then uses the coding method to further optimize it. Through the experiment comparison, it is proved that the matching algorithm after the coding has advantages in the execution efficiency and memory occupation. (3) the agricultural CPS based on Petri net is studied. The problem of complex event detection modeling is not determined. A method of complex event detection model based on probability time Petri net is proposed. The complex events in agricultural CPS are formed by the combination of atomic events. The generation of atomic events is closely related to the acquisition, transmission and conversion of each part of the CPS. In this process, the system error or the system error or the system error is closely related. The random error leads to the uncertainty of the time of the atomic event. In this paper, aiming at the uncertainty of the time, this paper takes the greenhouse as an example to model the atomic events of the system by using the probability time Petri net to form the model of the atomic event, and then combine the Petri net module of the atomic event probability time to form a complex event model, and use this model to be accurate. The complex event semantics is analyzed, and the misjudgments and misjudgments of complex events are detected in the matching process. Finally, the correctness of the method is proved by the case analysis. The higher the delay offset probability is, the higher the probability of the complex event detection is, the greater the system overhead is.

【学位授予单位】：中国农业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：S126

【参考文献】