基于支持向量机和再生核理论的随机微分方程求解OA
A novel approach based on support vector machines and reproducing kernel theory for solving stochastic differential equations
结合支持向量机、再生核理论和正则化等方法对随机微分方程问题展开研究.将随机微分方程问题视为算子方程问题,利用再生核空间的性质得到方程解与已知条件的关系式,将方程问题转化为逼近问题,采用Tikhonov 正则化将逼近问题转为二次规划问题,然后利用支持向量机进行求解,最终得到一个由支持向量构成的稀疏解.这种稀疏性意味着最终模型仅与支持向量有关,从而减少了模型的复杂性并提高了计算效率.以随机热方程问题为例,仅有离散值数据集作为初始条件展开数值求解,数值实验表明,所提方法与随机预正交自适应 Fourier 分解(SPOAFD)方法相比精度更高.
This paper investigates the solution of stochastic differential equations(SDEs)by integrating Support Vector Machines(SVMs),reproducing kernel theory,and regularization techniques.Specifically,the stochastic differential equation is formulated as an operator equation,and properties of reproducing kernel spaces are leveraged to derive the relationship between the equation's solution and given conditions.The equation problem is transformed into an approximation problem,which is then converted into a quadratic programming problem using Tikhonov regularization.The problem is solved using SVMs,resulting in a sparse solution composed of support vectors.This sparsity means that the final model is only related to the support vectors,thereby reducing model complexity and improving computational efficiency.Taking the stochastic heat equation as a case study,numerical solutions were performed using only discrete-valued datasets as initial conditions.Taking the stochastic heat equation problem as an example,numerical solution is performed with only discrete-value datasets as the initial condition,and numerical experiments show that the proposed method achieves superior accuracy when compared with the Stochastic Pre-Orthogonal Adaptive Fourier Decomposition(SPOAFD)method.
莫艳;黄健业
广东工业大学 数学与统计学院,广东 广州 510000广东工业大学 数学与统计学院,广东 广州 510000
数理科学
随机微分方程正则化再生核理论支持向量机
stochastic differential equationsregularizationreproducing kernel theorysupport vector machine
《佛山大学学报(自然科学版)》 2026 (5)
58-65,8
中山大学广东省计算科学重点实验室2024开放基金(2024003)
评论