Cascade cooling systems containing different cooling methods (e.g., air cooling, water cooling, refrigerating) are used to satisfy the cooling process of hot streams with large temperature spans. An effective cooling system can significantly save energy and costs. In a cascade cooling system, the heat load distribution between different cooling methods has great impacts on the capital cost and operation cost of the system, but the relative optimization method is not well established. In this work, a cascade cooling system containing waste heat recovery, air cooling, water cooling, absorption refrigeration, and compression refrigeration is proposed. The objective is to find the optimal heat load distribution between different cooling methods with the minimum total annual cost. Aspen Plus and MATLAB were combined to solve the established mathematical optimization model, and the genetic algorithm (GA) in MATLAB was adopted to solve the model. A case study in a polysilicon enterprise was used to illustrate the feasibility and economy of the cascade cooling system. Compared to the base case, which only includes air cooling, water cooling, and compression refrigeration, the cascade cooling system can reduce the total annual cost by USD 931,025·y−1 and save 7,800,820 kWh of electricity per year. It also can recover 3139 kW of low-grade waste heat, and generate and replace a cooling capacity of 2404 kW.