Thermal noise presents a major obstacle for quantum computing to scale up, and its existence puts stricter requirements on the robustness and fidelity of quantum control process. This work adopts the stochastic dynamic structure decomposition method and applies the Kubo-Einstein fluctuation dissipation theorem to the optimization of quantum dynamics, that is, how to improve the fidelity of quantum control process under thermal noise environment. Based on the characteristic that the classical path on a two-dimensional sphere can completely describe the motion of a single qubit, the research proposes a variational optimization scheme of gradient descent algorithm to reduce the thermal effect, and demonstrates its applicability via numerical simulation. It is found that the main factor affecting the fidelity of quantum control process under classical limit is thermal fluctuation. The method is expected to be mutually validated with experiments, and thus can further guide and evaluate experimental schemes towards achieving high fidelity of quantum gates.