Monocrystalline silicon rods (MSR) are key materials in the semiconductor industry and are crucial for the development of information technology. Therefore, this paper addresses the integrated problem of MSR’s batch planning and hybrid flow shop scheduling problem with batch sequence dependent setup times (MSR_BPHFSP_BSDST), which is commonly encountered in monocrystalline silicon production activities. A mixed-integer programming model is established with the objective of minimizing the makespan. Since this problem is NP-hard, a novel Q-learning based variable neighborhood search (QVNS) algorithm is designed for solving it. Firstly, to tackle two highly coupled subproblems (i.e., the batch planning subproblem and the hybrid flow shop scheduling subproblem with batch sequence dependent setup times) within the MSR_BPHFSP_BSDST, the new encoding and decoding strategies are developed to decouple them. Subsequently, five efficient neighborhood search operators are designed for this problem. The Q-learning algorithm is employed to dynamically determine the execution order of search operations during the search process. This intelligent search mode can reduce ineffective search caused by randomly selecting neighborhood search operators in traditional VNS, and can execute deeper search within a limited time. Finally, the effectiveness of the proposed algorithm is validated through simulation experiments and comparisons with other algorithms.