The welding deformation problem of pump truck arms has gradually attracted widespread attention. The current welding process lacks an effective prediction model, resulting in unstable welding quality and low production efficiency. This paper aims to establish a pump truck arm welding deformation prediction model combining CNN and LSTM, and collect multiple parameter data during the welding process, including welding temperature, welding speed, material thickness and stress distribution. Then, CNN and LSTM are combined to design and train the model, in which CNN is used to extract spatial features and LSTM is used to capture time series features. Finally, the model performance is optimized through cross-validation and hyperparameter tuning. The accuracy of the model in welding deformation prediction reaches 92.5%. The prediction model based on deep learning can effectively improve the deformation prediction accuracy of pump truck arm welding and provide strong support for the optimization of welding process.