Industrial customers are the primary participants in incentive-based demand response programs. However, centralized optimization of their response strategies struggles to balance the interests of multiple stakeholders and is prone to privacy leakage risks. To address these issues, this paper proposes a hierarchical distributed demand response strategy for industrial parks utilizing the federated learning alternating direction method of multipliers (FedADMM). This approach simultaneously considers the interests of both aggregators and customers while effectively protecting sensitive data. First, a hierarchical response model is established, accounting for the benefits of multiple stakeholders. The aggregator layer optimizes customer-level response strategies to maximize demand response profits, while the customer layer optimizes device-level strategies to maximize additional revenues. Second, an improved FedADMM (IFedADMM) is proposed. By improving global parameter updating, information transmission encryption, and client-side iteration rules, the proposed method achieves privacy protection during the iterative process and enhances computational efficiency. The optimization model is then solved in a distributed manner, where both parties dynamically update compensation prices until convergence to the optimal solution. Finally, a case study of an industrial park is conducted to verify the effectiveness of the proposed method in improving privacy protection and computational efficiency.