Abstract: In the arc welding manufacturing process, accurate prediction of weld seam size is important for controlling the quality of weld formation. In this study, a prediction model fusing whale optimization algorithm (WOA) and deep belief network (DBN), referred to as WOA-DBN, is proposed, which constructs a prediction model of weld size for arc welding manufacturing, using current, frequency, duty cycle, and welding speed as the input parameters. Welding seam size prediction model. In order to improve the search efficiency of the algorithm, enhance the convergence performance and avoid falling into the local optimal solution, this paper introduces the strategies of chaotic inverse learning initialization population, nonlinear convergence factor, as well as the simulated annealing operation and adaptive mutation perturbation, etc., and establishes a deep belief network model optimized by the chaotic whale optimization algorithm, i.e., AAMCWOA-DBN. Through experimental comparison, the AAMCWOA-DBN model outperforms the traditional WOA-DBN model in terms of prediction accuracy and performance metrics, the MAPE of the fused wide forecast is only 1.85% and the MAPE of the remaining high forecast is only 0.47%. This study utilizes artificial intelligence algorithms to predict the weld seam size of arc welding manufacturing, which provides new research perspectives and methods for the intelligent control of weld shaping and weld quality, and is expected to be applied in related fields.