The mechanical performance of the negative bending moment zone of steel-concrete composite continuous beams has always been a focus of attention in bridge engineering. Existing research mainly proposes improvement measures for the tension of concrete bridge decks, but rarely considers the common phenomenon of steel beams bearing pressure. The steel beams in the steel-concrete composite beam, especially the steel bottom plate in the negative bending moment area of the pier top, bear almost the maximum compressive stress of the entire bridge. Therefore, taking a steel-concrete composite beam bridge in Zhejiang as the engineering background, it is proposed to fill a part of the cast-in-place concrete inside the negative bending moment area of the pier top box girder during the construction phase to reduce the compressive stress borne by the steel bottom plate. Firstly, an Ansys finite element model of the composite beam bridge is established for stress analysis. Then, the length and vertical thickness of the filled concrete along the longitudinal direction of the bridge are used as variable parameters, and the minimum peak compressive stress of the steel beam bottom plate is taken as the optimization objective. IGRNN (Improved Generalized Regression Neural Network) is used to optimize its size. Finally, the predicted optimal size result is substituted into the finite element model to verify the accuracy of the prediction results. The research results indicate that pouring and filling a certain amount of concrete blocks into the negative bending moment area of the pier top inside the box beam can significantly reduce the compressive stress on the bottom plate of the steel beam; At the same time, IGRNN can greatly improve the efficiency of size optimization. For the case of composite beam bridges, the predicted optimal size and corresponding steel beam bottom plate compressive stress value are within 5% of the compressive stress value calculated by the finite element model, and the optimized steel beam bottom plate compressive stress is reduced by 74.9% compared to the original structure, with good results. This research achievement and method can provide reference for reducing the compressive stress of the steel beam bottom plate at the pier top and related issues for similar bridges.