随着新型工业化、信息化战略的推进，以智能电网为代表的信息物理系统的发展在国民经济领域中具有战略性和基础性地位。由于其重要性和开放性，信息物理系统易遭受物理攻击和网络攻击，导致系统组件失效，甚至引发级联反应和造成系统瘫痪，从而产生巨大的生命财产损失。因此，在信息物理系统中配置防御资源抵御物理攻击和网络攻击至关重要。 攻防博弈模型能够通过刻画攻击者与防御者的交互行为，优化防御资源配置效率。然而，现有基于攻防博弈模型的防御资源分配优化研究对于攻击者行为的有限理性，信息安全漏洞广泛关联及其导致的攻防策略空间动态性，以及多主体非合作防御引发的无政府状态现象的认识存在不足，使得信息物理系统的防御资源配置效率低下。为提升防御资源实际配置的有效性，针对上述三个场景，本文开展了如下研究。 首先，本论文研究了考虑攻击者有限理性行为的防御资源分配优化问题。考虑到攻击者行为的有限理性，建立了基于随机最优反应均衡的攻击者有限理性行为模型。构建了防御资源分配双层优化框架并设计了二乘二分搜索算法，求解最优的防御资源总预算和相应的最优资源分配策略。针对IEEE-14节点系统和IEEE-118节点系统的案例实验证明了模型在平衡防御资源成本和防御效果的有效性。案例实验也证明了正确评估攻击者有限理性的重要性。 其次，本论文研究了考虑攻击路径关联和攻击策略可变的动态攻防博弈图及资源配置优化问题。建立攻防图描述了信息物理系统信息安全漏洞的相关关系和系统安全状态，并基于攻击图构建多阶段攻防图博弈模型。同时结合多阶段交互分析中攻击者的有限理性行为和理性进化现象，以防御者贴现累计收益为目标函数，设计粒子群优化算法求解了最优多阶段事前防御资源和事后修复资源配置。基于乌克兰电网攻击事件构造的攻防博弈图案例验证了本研究的有效性。 最后，本论文研究了考虑子系统防御者不合作的风险共担机制设计问题。建立了三层结构的非合作多防御者斯塔克尔伯格博弈模型，分析系统参数与因子系统防御者不合作导致的无政府状态代价的相关关系。设计了一种风险分担机制，在非合作多防御者系统中协调执行全局最优防御策略而非个体最优防御策略。将该模型应用于广西电网计量系统防御资源配置问题，表明了风险共担机制的优越性，并对大型信息物理系统的管理者提供了配置系统的建议。

With the advance of new industrialization and informatization strategy, the development of the cyber-physical system, e.g., smart grid, plays a strategic and fundamental role in the national economy. Because of its importance and openness, the cyber-physical system is vulnerable to physical and cyber attacks, which will lead to the failure of system components, and even trigger cascade reactions and system paralysis, thus causing huge economic losses. Therefore, it is vital to defend against physical and cyber attacks by allocating defense resources within the cyber-physical system. The attack-defense game model can improve the efficiency of defense resource allocation by analyzing the interaction between attackers and defenders. However, existing researches on defense resource allocation optimization based on the attack-defense game model have insufficient understanding of the bounded rationality attacking behavior, the wide correlation of cyber security vulnerabilities and the space dynamics of attack and defense strategies, and the anarchy phenomenon caused by the multi-agent non-cooperative defense, which makes the efficiency of defense resource allocation of the cyber-physical system. In this paper, we first study the defense resource allocation optimization problem considering the bounded rational behavior of the attacker. Considering the bounded rationality of attacker behavior, a model of bounded rationality of attacker behavior based on Quantal Response Equilibrium is established. A two-layer optimization framework is constructed and a combined power-of-two and dichotomy search algorithm is designed to determine the optimal total defense resource budget and the corresponding optimal resource allocation method. The case analysis of the IEEE-14 bus system and IEEE-118 bus system shows that the proposed model is beneficial for power companies to better balance defense resource cost and defense effect. Secondly, this paper studies the multi-stage attack-defense graph game and resource allocation optimization considering the attack path correlation and variable attack strategy. An attack graph is established to describe the relationship between information security vulnerabilities and the system security state of the cyber-physical system. Based on the attack graph, a multi-stage attack and defense game model is constructed. At the same time, based on the bounded rational behavior and rational evolution of attackers in the multi-stage interactive analysis, a nonlinear optimization model with the discounted cumulative rewards of defenders is designed. Then, the particle swarm optimization algorithm was designed to obtain the optimal allocation of pre-defense resources and post-repair resources among multiple stages. The validity of this study is verified by the case of attack and defense game diagram constructed from Ukraine power grid attack events. Finally, considering the increasing size of the cyber-physical system, this paper studies the design of the risk-sharing mechanism considering the non-cooperation of subsystem defenders. A three-layer non-cooperative multi-defender Starkelberg game model is established, and the relationship between system parameters and the price of anarchy caused by the non-cooperation of subsystem defenders is analyzed. A risk-sharing mechanism is designed to coordinate the individual optimal defense strategy with the global optimal defense strategy in a non-cooperative multi-defender system. This model is applied to the defense resource allocation of the Guangxi power grid metering system, which shows the superiority of the risk-sharing mechanism, and provides some suggestions for the administrator of a large-scale cyber-physical system to build the system.