在过去的二十年，随着电子与传感技术的发展，无线传感器网络(Wireless Sensor Networks，WSNs)已经成为许多监测和感知应用中的重要支撑技术，包括室内／室外环境检测、工业监控、农业生产、医疗保健、电网监控、军事侦察等。与传统的无线传感器网络相比，当前及未来的无线传感器网络应用需要同时优化多个网络目标，包括网络可靠性、网络连通性、网络时延、能量效率等。以无线传感器网络的节点能量为基础，结合网络覆盖度、连通性、可靠性等特性，同时考虑到网络节点的能量消耗、通信距离、部署密度等关键因素，本文优化无线传感器网络路由调度从集中式和分布式选择控制策略开始，逐步实现对WSNs路由调度的启发式智能优化；同时也从网络生命周期的单一目标优化扩展到网络连通性和可靠性的多目标优化，实现对无线传感器网络由简单到复杂的不断研究，主要贡献总结如下： 对于大多数的无线传感器网络应用，网络生命周期和算法复杂度是网络研究的关键问题，吸引许多研究者的深入分析。对无线传感器网络中传感器随机分布的场景，以网络生命周期为优化目标，我们提出DHCO(Dynamic Hierarchical Energy-Efficient Method Based on Combinatorial Optimization for Wireless Sensor Networks)算法，旨在通过均衡节点能量消耗延长网络生命周期。DHCO算法是集中式控制算法，将网络路由调度问题转化为组合优化问题。具体地，DHCO算法根据传感器节点的通信距离建立分层网络架构，并以此为基础构造每个传感器节点的可行路由集合，利用节点剩余能量和最大最小原则选择最优通信路由，实现传输感知数据。仿真实验表明，DHCO算法比其它经典算法延长无线传感器网络生命周期，同时也降低计算复杂度。 在无线传感器网络应用中，节点能量和计算资源是严格受限的，这使得管理更简单的分布式网络调度方式更有研究意义。对无线传感器网络中传感器随机分布的场景，我们提出DORAHP(Distributed Joint Optimization Routing Algorithm Based on the Analytic Hierarchy Process)算法，旨在通过均衡无线传感器网络的通信负载延长网络生命周期。在DORAHP算法中，如果传感器节点具有较大剩余能量、到基站的通信距离较短并且邻居节点数较小，该节点对应的汇聚权重会较大且容易被选为转发节点。同时，DORAHP算法给出基于统计的量纲消除方法处理不同网络因素的量纲，并利用层次化分析法评价它们的重要性。仿真实验表明，DORAHP算法不仅是可行的，而且相比于其它经典算法延长网络生命周期。 对于像火山监测等危险的场景，部署高密度的传感器有助于延长网络生命周期，但给无线传感器网络路由优化带来巨大挑战。针对高密度传感器随机部署且被检测目标也是随机部署的应用场景，我们提出ULGAT(Joint Unsupervised Learning and Genetic Algorithm Approach for Topology Control)算法，旨在延长高密度无线传感器网络的生命周期。在ULGAT算法中，机器学习的K-均值分簇技术被用于构造二级分簇网络拓扑，然后通过遗传算法的多次迭代识别最佳染色体用于构造近似最优分簇网络拓扑，实现感知数据的传输。同时，基于网络覆盖度和ULGAT算法我们还提出ULGATGAF(ULGAT with Geographical Adaptive Fidelity)算法，调度部分传感器睡眠实现更好地节省网络能量。仿真实验表明，算法ULGAT和ULGATGAF不仅延长高密度无线传感器网络的生命周期，同时算法计算复杂度是可以接受的。 随着无线传感器网络的发展，对网络可靠性和网络时延等特性提出更高要求，为此我们改进ULGAT算法后提出MLPGA(Machine-Learning-Based Parallel Genetic Algorithms for Multi-objective Optimization)算法，旨在同时延长网络生命周期、增强网络连通性和可靠性。针对传感器和被检测目标均是随机部署的应用场景，MLPGA算法采用机器学习的K-均值分簇技术构造二级网络拓扑架构，并设计能量转化分簇方法避免产生通信负载过重的网络簇头节点，实现均衡网络通信负载。MLPGA算法运用网络的多个关键因素建立优化模型，定义染色体种群的最小模式集合并利用遗传算法求解该模型。在优化模型中，机器学习的主成分分析法被用于消除不同优化目标之间的依赖关系和评价它们的重要性，实现将多目标优化函数转化为评估不同染色体的适应值函数。通过分析基于染色体概率特性的局部收敛和基于基因概率特性的全局收敛两子收敛过程，MLPGA算法被证明收敛到最优染色体。仿真结果表明，MLPGA算法延长无线传感器网络生命周期、增强网络连通性和可靠性，同时计算复杂度是可以接受的。 本文研究无线传感器网络的路由问题，从简单的网络生命周期优化，到基于网络覆盖度网络生命周期优化，再到网络生命周期、网络连通性和可靠性的多目标优化，对无线传感器网络的研究不断加深。 关键词：无线传感器网络，拓扑控制，路由调度，能量效率，网络性能优化

 In the past twenty years, with increasing advances in electronics and sensing technologies, Wireless Sensor Networks (WSNs) have been the foundation of monitoring and sensing applications, such as indoor/outdoor detections, industry monitoring, agricultural production, health care, power-grid monitoring, military reconnaissance, and so on. Different from conventional WSNs, current and future WSNs are subject to simultaneously satisfying multiple network objectives including the network reliability and connectivity, network latency, energy efficiency, and so on. Based on sensors ’ energy, this thesis carries on the thorough analysis and research by both optimizing the coverage, connectivity, and reliability of WSNs and using energy consumption, communication distance, and deployment density of sensors. Meanwhile, this thesis not only improves WSNs ’ routing schedules from simple selection strategyies based on centralized control and distributed control to intelligent optimization based on heuristic strategy, but also achieves an expansion from the single-objective optimization with the network lifetime to the multi-objective optimization with additional network connectivity and reliability. This thesis has the main advantages as follows: Being vital metrics of the WSN applications, the network lifetime and computational complexity are key issues, deserving more investigations. For random deployment scenarios of sensors, DHCO (dynamic hierarchical energy-efficient method based on combinatorial optimization for wireless sensor networks) was proposed to prolong the network lifetime. DHCO is a centralized algorithm while transformsing the WSN routing problem into the combinatorial optimization problem. DHCO constructs the hierarchical network structure according to neighbor communication links, being the foundation of the feasible routing set for each sensor. Using both the residual energy of sensors and the maximum-minimum criterion, DHCO obtains the optimal route to transmit the sensed data. Simulation results demonstrate the viability of DHCO compared to state-of-the-art algorithms at a good computational complexity. Traditionally, WSNs are more stringent resource-limited and computation-limited, which makes the simple distributed algorithm more promising. For random deployment scenarios of sensors, DORAHP (distributed joint optimization routing algorithm based on the analytic hierarchy process) was proposed to prolong the WSN lifetime by balancing communication load of sensors. In DORAHP, sensors, that are with more residual energy, shorter distance to the base station (BS), and less neighbors, are prone to possess more aggregated weight and to be the forwarding hop. The data dimensional elimination method and the analytic hierarchy process algorithm are used to eliminate different dimensions of metrics and to rank their importance levels, respectively. Simulation results demonstrate that DORAHP is feasible while having superior performance on extending WSNs’ lifetime. For some hazardous scenarios such as volcano monitoring, although it is beneficial to deploy high-density sensors, many challenges come forth. For random deployment scenarios of both high-density sensors and sensed objectives, ULGAT (joint unsupervised learning and genetic algorithm approach for topology control) was proposed to prolong the lifetime of high-density WSNs. ULGAT utilizes the K-means clustering algorithm of machine learning to design the 2-tier clustering network topology and adopts the genetic algorithm to identify the optimum chromosome, being the foundation of designing the near-optimal clustering network topology to transmit the sensed data. Based on both ULGAT and the network coverage, ULGATGAF (ULGAT with Geographical Adaptive Fidelity) was developed to schedule partial sensors to sleep for further energy conservation. Simulation results demonstrate that ULGAT and ULGATGAF not only prolong the lifetime of high-density WSNs, but also have an acceptable computational complexity. With developments of WSNs, many stringent requirements including network connectivity and reliability arise. Hence, by improving the ULGAT’ s single-objective optimization, the MLPGA (machine- learning-based parallel genetic algorithms for multiobjective optimization) algorithm was proposed to simultaneously satisfy multiple optimization objectives including prolonging network lifetime and enhancing network connectivity and reliability. For random deployment scenarios of both sensors and sensed objectives, the MLPGA algorithm adopts the K-means clustering algorithm of machine learning to design the 2-tier clustering network topology, and develops the clustering method of energy conversion to prevent overloaded CHs and to balance network load. A fair optimization model is formulated based on multiple objectives, and is solved by using both the minimal schema and the genetic algorithm. Using the principal component analysis (PCA) algorithm of machine learning, the proposed algorithm eliminates dependencies between multiple optimization objectives and ranks their importance levels, which constructs the fitness function to evaluate different chromosomes. Additionally, the convergence property of the proposed algorithm is proved both locally and globally based on the chromosome probability and the gene probability. Simulation results demonstrate that MLPGA has an acceptable complexity while outperforming state-ofthe-art algorithms in terms of the network lifetime, connectivity, and reliability. This thesis studies different optimization problems of WSNs, from the simple network lifetime optimization to the coverage-based lifetime optimization, to the network lifetime optimization with additional network connectivity and reliablity. The WSN scheduling problem is deepened to meet the complex requirements. Keywords: Wireless sensor networks, topology management, routing scheduling, energy efficiency, network performance optimization