答辩人：符修文 博士研究生

学科专业：机械工程

论文题目：工业无线传感器网络抗毁性关键技术研究

摘要：

无线传感器网络作为由大量传感器节点所构成的分布式网络系统，能够有效采集与传递各种环境和目标信息。在工业场景中，由于受到规模巨大、网络异构、传递时延、有向传输等内在因素以及外部环境干扰因素的共同作用，无线传感器网络难以长时间稳定可靠运行。抗毁性问题已经成为制约工业无线传感器网络规模化应用的主要技术瓶颈。本论文依据网络构建流程从三个方面对工业无线传感器网络抗毁性能进行优化：①在网络初始化阶段，通过对拓扑与容量参数进行优化配置，提升网络抵御随机失效与级联失效的能力；②在网络运行阶段，通过路由选择优化，实现感知数据的安全可靠传输；③在网络维护阶段，通过引入故障检测与诊断机制，解决网络因节点故障状态信息缺失所导致的后期维护难题。本文主要研究工作与创新点如下：

1）设计了一种工业无线传感器网络分簇拓扑演化机制。针对无线传感器网络在复杂工业环境下的拓扑抗毁性难题，构建了一种分簇无标度局域世界演化模型，使所生成网络拓扑贴近真实工业情形且容错性能较优。基于平均场理论证明了拓扑度分布符合幂律分布。考虑数据传输有向性，构造了一种用于评估网络负载均衡程度的有效测度-有向介数网络结构熵，并基于小世界网络理论提出了一种长程连接布局策略，有效解决了因无标度拓扑度分布异质性所引发的能量空洞问题。

2）提出了一种面向级联失效的工业无线传感器网络容量优化策略。针对工业无线传感器网络因遭受数据流量冲击所导致的级联失效问题，通过分析真实分簇网络动态负载变化规律，引入感知负载与中继负载概念，构建了一种参数可调的负载-容量模型，并分别研究了分簇无标度网络与分簇随机网络应对级联失效的抗毁性能。基于容量扩充方式，分别给出了扩容对象选择策略与新增容量分配策略，用于提升网络级联失效抗毁性能。

3）设计了一种工业无线传感器网络容错路由算法。考虑工业场景中复杂环境因素（如温度、湿度等）对网络路由性能影响，算法将工业无线传感器网络抽象为人工势场，且势场受环境场、能量场与深度场共同作用。通过构建权重可调的目标场，确保路由在满足低能耗与低延时等关键性能指标的基础上，使所建立不相交多路径可动态规避危险环境区域，提升消息路由抗毁性能。

4）提出了工业无线传感器网络故障检测与诊断算法。为满足工业无线传感器网络对实时故障检测与低延时故障诊断的迫切需求，基于邻近传感器节点数据采集所表现出的趋势相关性，设计了一种分布式故障检测算法，以消除故障检测触发时刻对检测准确率影响。在此基础上，提出了一种基于人工免疫理论的故障诊断算法，通过抗原分类、抗体库训练、抗体-抗原匹配等一系列步骤完成故障辨识。所提算法在具有较高诊断准确率的同时，运算耗时明显缩短，满足工业场景对服务低延时要求。

5）搭建了仿真平台与实际系统用于测试所提理论方法。针对工业无线传感器网络抗毁性仿真平台匮乏现状，结合工业无线传感器网络性能明显受环境因素影响与抗毁性行为受事件驱动等特征，引入部署环境组件与事件生成器，构建了一个工业无线传感器网络抗毁性仿真平台。在该平台基础上，以典型工业场景-立体仓库为部署环境，搭建了实际工业无线传感器网络系统，并验证了所提理论方法的实际性能。实际测试结果表明：所提理论方法能够有效提升实际工业无线传感器网络系统抗毁性能。

综上所述，本论文以解决工业无线传感器网络抗毁性问题为目标，研究用于提升网络抗毁性能的相关理论与方法。经仿真与实际测试，所提理论方法性能得到了有效验证，为构建抗毁性能较优的实际工业无线传感器网络系统提供了理论与实践参考。

Abstract:

As the distributed network system consisted by lots of sensor nodes, wireless sensor networks are capable of sampling and delivering different kinds of environmental data and target information. But in industrial scenarios, due to the internal impacts of the networks (e.g., large-scale deployment, heterogeneous structure, delivery delay, directional transmission) and external environmental interference, the long-term reliable operation of industrial wireless sensor networks cannot be ensured. Network invulnerability has become the major technical bottleneck for wider application of industrial wireless sensor networks. To tackle with this issue, in this dissertation we divided the task of invulnerability optimization into three phases according to the process of network building: ①the first phase is network initialization phase in which the parameters of topology and capacity would be optimized. The goal of this stage is to enhance the network invulnerability against random failures and cascading failures; ②the second phase is running stage, which is to ensure reliable transmission through routing-selection optimization; ③the third phase is maintenance stage of which purpose is to solve the later maintenance issues brought by the lack of failure information. To achieve this, we introduced mechanisms of fault detection and fault diagnosis into this stage. The main research works are as follows:

1) A clustering topology evolving mechanism of industrial wireless sensor networks is provided. To tackle with the topology invulnerability issue of wireless sensor networks in complicated industrial environments, we designed a clustering scale-free local-world evolving model to create a highly error-tolerant network topology and proved the degree distribution of the proposed model consistent with the power-law distribution. Given the property of directional data transmission, we built directional betweeness network structure entropy to evaluate the balance level of network loads. On this basis, we proposed a deployment scheme of long links based on the small-world theory, of which purpose is to solve the issue of energy hole caused by heterogeneity of degree distribution in scale-free topology.

2) Oriented to cascading failures, a capacity-optimization scheme of industrial wireless sensor networks is proposed. According to the cascading failures caused by traffic blasts in industrial wireless sensor networks, we firstly analyzed the genuine changing discipline of traffic loads in clustering network and built a clustering cascading model by introducing the concepts of sensing loads and relaying loads. And then, we researched the invulnerability performance of clustering scale-free network and clustering random network against cascading failures. To prevent cascading failures, we proposed two schemes: capacity selection scheme and capacity distribution scheme based on capacity-extending method.

3) An error-tolerant routing algorithm of industrial wireless sensor networks is designed. Given the impacts of complex environmental factors (e.g., humidity and temperature) on network routing performance, we proposed a disjointed multi-path routing algorithm based on potential field. In this algorithm, the industrial wireless sensor networks are abstracted as an artificial potential field which is influenced by environmental field, energy field and depth field. By building weight-adjustable target field, the routing invulnerability can be improved significantly by making disjointed multipaths avoid crossing dangerous areas. And at the meanwhile, the routing can also meet the requirements of low energy-consuming level and low delivery delay.

4) Fault detection and diagonosis methods are proposed in this dissertation. According to the common fault types of wireless sensor networks in industrial scenarios, we proposed a distributed fault detection algorithm based on the tendency-similarity characteristics of data sampling presented by adjacent sensor nodes. The most evident advantage of this algorithm is removing the effects of triggering moment on detection accuracy. On this basis, we presented a fault diagnosis algorithm based on artificial immune system. Through a series of procedures (i.e., classification of antigens, training of antibodies library, antibody-antigen matching and updating of antibodies), network fault types can be identified accurately.

5) We built a simulation platform and a practical system to validate the performance of proposed theories and methods. Given the current situation of lack of invulnerability simulation platforms, with considering of features (e.g., network performance is strongly affected by environmental factors and invulnerability behaviors are event-driven), we designed a simulation platform for invulnerability of industrial wireless sensor networks by introducing environment component and event creator. And then, we built a practical industrial wireless sensor network system to test the actual performance of proposed theories and methods in a stereoscopic warehouse. The test results show that the proposed theories and methods are able to enhance the invulnerability performance of industrial wireless sensor networks effectively.

       In summary, this dissertation researched the related theories and methods of network invulnerability, of which purpose is to solve the invulnerability issues in industrial wireless sensor networks. Through simulations and practical tests, the performance of proposed theories and methods are verified, which can provide theoretical and practical support for constructing highly invulnerable industrial wireless sensor networks.