随着移动设备数量的不断增长和移动数据流量的急速攀升，无线网络架构已经演化成为日益复杂的异构无线网络，从而使得依据传统的人工或半人工的管理方式，无线网络运维管理和优化将变得非常昂贵和脆弱，并且非常耗时。因此，期望未来无线网络可以实现智能、高效的自主运维管理功能，自主解决无线网络遇到的网络问题(网络中断，性能下降)，提升网络服务质量，减少网络能量开销。具备多种SON功能的自组织网络被认为是实现网络智能化的有效工具，通过密集部署的SON功能，在网络中引入自组织能力，从而实现无线网络的自配置、自优化以及自愈合功能，实现网络的自主规划运维和优化管理。然而，面对如今高度动态的无线网络场景，尤其是超密无线网络，如何设计一种综合高效的自愈合机制，去识别密集网络中的网络问题从而减少网络运维成本和管理开销，是现在面临的巨大挑战。其次，随着SON功能的密集部署，在同一网络中，多个不同的SON功能同时运行时，容易导致网络的参数调整冲突和目标优化冲突，从而引发网络性能的震荡。因此，如何设计多个SON功能间的协同管理模型和协同机制是无线网络智能管理面对的关键问题。此外，考虑到有限的无线网络资源(可获得的带宽和传输功率)，以及数据流量到达过程和无线信道的高度动态特性，如何设计有效的网络资源管理机制，减少网络能量开销，提升网络性能，从而实现可持续的网络自主管理是未来网络亟需解决的另一关键难题。因此，在本论文中，主要工作如下： 1.研究了关键性能指标信息部分缺失的超密无线网络中的自愈合问题。提出了一种在部分关键性能指标信息缺失场景下的综合高效的网络自愈机制，包含小区中断检测机制和中断补偿机制。具体而言，网络自愈机制主要包含网络中断检测阶段和中断补偿阶段。首先进行中断检测，当检测到中断发生后，进入中断补偿阶段，对中断小区中的用户进行补偿，确保用户的正常服务。在中断检测阶段，基于部分关键性能指标信息和用户位置信息，提出了基于支持向量机描述学习的高效且低复杂度的检测算法。在中断补偿阶段，为补偿中断小区的用户，提出了分布式的资源分配算法，描述成混合整数优化问题，通过拉格朗日对偶理论求解，从而保证网络中的负载均衡和用户的服务质量需求。 2.研究了无线网络中的多个SON功能间的多时间尺度协同管理问题。提出了多时间尺度的SON功能协同管理机制。具体而言，把SON功能中的自组织机制刻画成马尔科夫控制环，探讨了不同时间尺度下的马尔科夫决策过程。本文把多时间尺度的SON功能协同管理问题转化为多时间尺度的马尔科夫决策过程，不同时间尺度的SON功能对应不同时间尺度的马尔科夫决策过程，考虑到网络环境的动态变化，本文提出了Q-学习算法求解提出的决策过程。此外，为了提升网络能量效率，本文设计了多时间尺度的MLB-ESM功能用例，基于设计的网络效用函数，对无线网络进行自主优化，从而提升无线网络能量效率，并印证了所提出的多时间尺度协同管理机制的有效性。最后通过仿真验证了理论分析和所提算法的有效性。 3.研究了基于电网和收集的能量同时供电的网络中的资源分配问题。通过定义网络效用函数，考虑无线网络吞吐量和电网能量开销等影响，设计了有效的资源分配算法。为了充分利用可再生能源，提出了一个基于能量感知的资源分配算法，从而最大化网络效用函数，提升网络性能。本文提出的基于能量感知的资源分配算法不需要任何有关于信道分布、可再生能源到达和数据到达的先验分布知识，而仅仅需要网络当前的队列状态信息。因此，本文提出的基于能量感知的资源分配算法，在高度动态的无线网络中具有实际的应用价值。此外，本文对所提出的算法进行了性能分析，并提出了一种实现架构，从算法复杂度和信令开销等方面进行了分析。最后通过仿真验证了理论分析和所提算法的有效性。 4.研究了基于混合供电的多种无线接入网络中的按需资源分配问题。考虑到能量收集过程，数据流量到达过程以及时变信道等过程的动态特性，本文把灵活的能量调度问题描述成最小化电网能量开销的随机优化问题。具体而言，基于收集能量的稀疏性和不连续性，本文提出了一种动态的网络能量队列过程模型。基于李雅普诺夫优化模型，随机优化问题转换化为一系列的子优化问题，包括网络数据流量控制子问题，网络能量管理子问题和网络资源分配子问题。本文提出了一种灵活的能量调度机制，在此机制中多种无线接入网络收集的能量可以进行共享，按需进行调度资源，从而有效地减少电网能量开销，提升收集的能量的利用率。此方法不需要无线信道的先验知识分布和数据达到状态信息，适用于实际的无线系统。此外，本文还探讨了电网能量消耗与网络时延之间的均衡问题，通过理论分析可知，网络时延与控制参数V存在线性增长的关系，电网能量消耗与控制参数V存在1/V的关系。该研究内容的理论分析可以为基于混合供电的多种无线接入网络的资源分配提供有效的理论指导。 关键词：自组织网络，网络自愈，资源分配，能量收集，网络管理

 With the exponential growth of mobile traffic and the increasing number of mobile termi-nals, the wireless network infrastructure has evolved as a heterogeneous and increasingly complex network, making the network operation management and optimization very costly, more vulnerable and time-consuming with the traditionally manual and semi-automatical approaches. Hence, the future wireless network is expected to realize intelligent and effi-cient self-management in managing and controlling the network with an autonomic manner, dealing with network problems (outage, performance degradation), increasing the quality of service (QoS) and reducing energy consumption cost. The self-organizing network (SON) with different SON functions which is envisioned as a powerful tool, is densely-deployed to realize self-configuration, self-healing and self-optimization in the wireless network. How-ever, in such highly dynamic network scenarios, especially for dense small cell networks, a comprehensive self-healing solution becomes necessary to identify performance issues (Key Parameter Indicators (KPI)) and to reduce the Operation Expenditure (OPEX) and manage-ment cost with the increasing cell density. In addition, different SON functions can lead to parametric and objective conflicts resulting in network performance oscillation when oper-ate simultaneously in the same area, so a theoretical scheme of a SON coordination problem and of the solution should be addressed. Furthermore, considering the highly dynamics of the traffic arrival and wireless channel states with the limited network resources (available bandwidth and transmission powers), efficient network management is also urgently needed to reduce the energy consumption cost and improve the network efficiency to realize fully sustainable network management for future wireless networks. Hence, in this thesis, we mainly focus the works as follows, 1.we investigate the self-healing (SH) problem in SON-based ultra dense small cell net-works, where KPI information of some small cells may not be available. We propose a com-prehensive self-healing scheme including both small cell outage detection mechanism and small cell outage compensation mechanism, and is capable of dealing with partial KPI situ-ations. Specifically, the proposed SH scheme includes small cell outage detection (SCOD) stage and small cell outage compensation (SCOC) stage, where in SCOD stage the outage is detected, followed by the SCOC stage to compensate the users in the outage zones. For the SCOD stage, we propose a SCOD algorithm through machine learning approach based on partial KPI statistics. An effective Support Vector Data Description (SVDD) algorithm with low computational complexity is developed to detect and locate the outaged small cells with the context of KPIs and user position information. For the SCOC stage, to compensate the users in detected outage zone, we propose a novel distributed resource allocation algorithm that aims to optimize load balancing of the detected outage small cells area. The SCOC scheme is designed to allocate resources of neighboring small cells to the outage users, con-sidering the dynamics and density of small cell environments. Specifically, we formulate resource allocation as a mixed integer optimization problem, solved by Lagrangian dual theory to guarantee both load balancing and QoS requirements of users. 2.we investigate a novel multiple time-scale coordination scheme (MTCS) for SON func-tions to ensure steady wireless network operation. Specifically, the self-organizing mecha-nisms in SON functions are modeled as Markov control loops and the stability conditions of control loops with different time scales are derived. Furthermore, we devise an ana-lytical model, multiple time-scale Markov decision process (MMDP) for hierarchically M SON coordination decision making processes, where SON coordination decisions in each level (each time-scale) are made in a specific time-scale, and the Q-leaming algorithm is developed to solve the proposed model, considering the dynamics of wireless network en-vironments. Then, in order to improve energy efficiency of the wireless networks, we study a MLB-ESM use case with two typical SON functions (MLB (mobility load balancing) and ESM (energy saving management)) with designed network utility in ultra-dense small cell networks. At last, the coordination capabilities of the proposed MTCS scheme is evaluated by the simulation results. 3.we investigate the resource allocation for hybrid energy supplied OFDMA cellular net-works where the base station is powered by renewable energy and electric grid. With the aim of exploiting the harvested energy, we propose an energy-aware resource allocation (EARA) algorithm for maximizing the network utility, which represents the tradeoff between the net-work throughput and the grid power consumption. In particular, the EARA algorithm only needs the current network states without a relevant priori distribution knowledge. Hence, it is applicable with unpredictable dynamics of wireless channel states, renewable energy arrivals and stochastic traffics in practical OFDMA cellular networks. In addition, we char-acter the performance of EARA algorithm theoretically considering the complexity and the overhead of communications. Furthermore, we propose an implementation architecture for the EARA algorithm, and analyze the low implementation costs such as the low computa-tional complexity. The simulation results verify the theoretical analysis and show that the proposed algorithm is effective in the wireless networks. 4.we investigate the resource-on-demand (RoD) allocation strategy for multiple radio ac-cess technologies (multi-RAT) wireless networks, which are simultaneously powered by both harvested energy and grid power. By considering the dynamics of randomly varying harvested energy, dynamic traffic arrival and time-varying wireless channels, the flexible energy scheduling is formulated as stochastic optimization problem to minimize grid pow-er consumption cost. Specifically, under the sporadic availability and discontinuity of the harvested energy, a dynamic network energy queue model is proposed to provide the en-during operation for resource optimization in multi-RAT wireless networks with renewable energy. Following the Lyapunov optimization framework, the stochastic optimization prob-lem is transformed into a sequence of optimization subproblems, including the network flow control subproblem (NFCS), the network energy management subproblem (NEMS) and the network resource allocation subproblem (NRAS). By solving these subproblems, an optimal resource-on-demand algorithm for multi-RAT wireless networks is developed in the dynam-ic environment. We propose a RoD strategy for flexible energy scheduling in multi-RATs with heterogeneous energy sources, where the harvested energy is effectively shared be-tween RATs to satisfy varying energy demand dynamically. The grid power consumption can be effectively reduced with efficient utilization of harvested energy, where a priori dis-tribution knowledge of the wireless channel and data arrival state is not required, which is very practical in real systems. The tradeoff between grid power consumption and network delay is derived, where the increase of delay is approximately linear in V and the decrease of grid power consumption is at the speed of 1/V with the control parameter V. It can pro-vide guidelines for dynamic resource allocation in multi-RAT networks with heterogenous energy sources. Keywords: self-organizing network, self-healing, resource allocation, energy harvesting, network management