本项目针对连续状态空间的"维数灾"问题，提出利用支持向量核方法对连续状态空间强化学习问题进行建模并逼近所建模型。主要思想是结合机器学习中最具发展前景的核方法来解决强化学习问题，结合了支持向量核方法的强化学习系统，学习过程与人类更加的相似，以自学习为主导、监督学习为辅助，相互配合从而更好的解决复杂问题。首先利用核方法对原输入空间复杂问题进行非线性建模，然后利用核函数机制将输入空间非线性问题转化到高维特征空间中的线性问题，并巧妙避开升维所带来的计算复杂性，这样就可以对原输入空间复杂问题进行精确的模型构建，同时在高维特征空间可以高效的求解线性模型，线性模型具有很优良的泛化特性，而泛化能力正是解决连续空间强化学习问题的核心需求。因此，基于支持向量核方法的连续空间强化学习模型构建与逼近研究极富理论价值和应用前景。

For the "Curse of Dimensionality" of the continuous state space, this research proposes a modeling mathod for reinforcement learning in continuous state space, which using support vector kernel methods and approach the model. The main idea of this reaearch is to solve reinforcement learning problems integrating with support vector kernel method. The learning process of the reinforcement learning system is more similar to human, in which self-learning as the leading, supervised learning for auxiliary, cooperate with each other and thus a better solution to the complex problem. Firstly, it uses kernel methods nonlinear modeling complex problems of the original input space. Then it transforms nonlinear problem into linear problem in high dimensional feature space, which takes advantage of the kernel function mechanism, and avoids the computational complexity caused by rising dimension. So we can build an accurate model of the original input space of complex problems, and can solve the linear model efficiently in high dimensional feature space. The linear model has excellent generalization performance, which is the core needs of solving the continuous space reinforcement learning problems. Therefore, the research of the continuous space reinforcement learning model construction and approximation has a highly theoretical value and application prospect.