Scholarlink

<p>In this paper, we address a cooperative differential game for partially observed stochastic systems by virtue of a data-driven approach. First, leveraging the convexity property of cost functionals, the cooperative differential game problem is recast as a linear quadratic optimal control problem of equal validity using a weighted sum optimization method. Then, by integrating state decomposition techniques, variational methods and stochastic filtering theory, a feedback-form characterization of Pareto optimal strategy is rigorously derived. Furthermore, a data-driven algorithm is developed to learn the Pareto optimal strategy without using the information of system coefficients <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(A\)</EquationSource> </InlineEquation> and <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(B\)</EquationSource> </InlineEquation>. Subsequently, the convergence of this algorithm is established. Finally, the algorithm’s performance is verified through a numerical simulation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Partially observed stochastic linear quadratic cooperative differential game in finite horizon: a data-driven approach

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Guangchen Wang,
  • Yu Wang,
  • Heng Zhang

摘要

In this paper, we address a cooperative differential game for partially observed stochastic systems by virtue of a data-driven approach. First, leveraging the convexity property of cost functionals, the cooperative differential game problem is recast as a linear quadratic optimal control problem of equal validity using a weighted sum optimization method. Then, by integrating state decomposition techniques, variational methods and stochastic filtering theory, a feedback-form characterization of Pareto optimal strategy is rigorously derived. Furthermore, a data-driven algorithm is developed to learn the Pareto optimal strategy without using the information of system coefficients \(A\) and \(B\) . Subsequently, the convergence of this algorithm is established. Finally, the algorithm’s performance is verified through a numerical simulation.