Simulation is a fundamental solution for investigating the capacity and efficiency of complex workshops in discrete manufacturing. With increasing custom production orders, it is more efficient to establish an extendable simulation model and make auto-configuration to analyze different kinds of reconfigurable production workspace, rather than construct a simulation model for each scenario. However, it is difficult to automatically configure a simulation model with long programming codes and complex coupled constraints among orders, materials, production machines, and production lines. To solve this problem, this paper establishes a key-value mapping rule to map key and repetitive information and decouple constraints among different parts of simulation model. Then, a chain-of-thought-based model auto-configuration method is proposed to configure a simulation model by using large-language model for programming code generation. To ensure correct and credible simulation model configuration, an automatic evaluation pipeline is also established. Experimental results on a typical extendable Anylogic simulation model for reconfigurable production workshop show that the proposed method can automatically generate appropriate simulation models for different scenarios with high precision greater than 90%.