The Welded Hollow Sphere Joint (WHSJ) is frequently employed as a connection form in spatial grid structures. Structures don't retire immediately upon damage, necessitating reinforcement. However, effective solutions for reinforcing WHSJs still have shortcomings in practical engineering applications. Given this research context, a reinforcement scheme for WHSJs is proposed. Feasibility is confirmed through numerical simulation, showing that this reinforcement method is applicable to WHSJs with initial loads, enhancing their mechanical performance without being influenced by the initial loads. The study also initially uncovers the patterns of load-bearing capacity variation with geometric dimensions. Subsequently, an Artificial Neural Network (ANN) is used to predict compressive load-bearing capacity. It utilizes input variables such as sphere diameter (Ds), steel pipe diameter (Dp), reinforcement cone base diameter (Dt), sphere thickness (Ts), steel pipe thickness (Tp), cone thickness (Tt), and cone base angle (θ), with the output being the compressive load-bearing capacity of the welded hollow sphere joint post-reinforcement. Prediction errors are confined to within 10%. Using the Garson algorithm, the sensitivity analysis reveals that steel pipe thickness (Tp) and sphere thickness (Ts) play relatively significant roles in the prediction process.