To address the coordination issue between the fluctuating output of wind and photovoltaic (PV) power and the demand for continuous and stable hydrogen supply of hydrogen-based shaft furnaces, a bi-level capacity allocation optimization model for the wind-PV hydrogen production system of hydrogen-based shaft furnaces is established, which takes into account the characteristics of weakly flexible continuous loads. The system consists of wind turbines, PV arrays, electrolyzers, hydrogen storage tanks, power grids and hydrogen-based shaft furnaces. The upper layer conducts capacity planning with the objective of minimizing the total system cost, while the lower layer performs electricity-hydrogen coordinated dispatch to minimize the operating cost. YALMIP and CPLEX are adopted to solve the operational optimization problem.Aiming at the deficiencies of traditional dung beetle optimizer (DBO) in convergence accuracy and stability when solving capacity allocation problems with complex constraints, an improved LSDBO-WOA algorithm is proposed. This algorithm integrates Latin hypercube sampling, opposition-based learning, spiral search strategy of whale optimization algorithm (WOA), engineering boundary candidate solution injection and elitist preservation strategy.Numerical examples show that the proposed LSDBO-WOA algorithm can obtain a superior capacity allocation scheme. Its average objective function value is 1.03% lower than that of the grey wolf optimizer (GWO), demonstrating strong optimization capability and favorable convergence stability. Under the optimal allocation scheme, the hydrogen supply reliability of the system reaches 100%, which effectively ensures the continuous and stable operation of hydrogen-based shaft furnaces. The research results verify the effectiveness of the established model and the improved algorithm in the capacity optimization of wind-PV hydrogen production systems for hydrogen-based shaft furnaces.