Due to concerns about energy use in production systems, energy-efficient processes have received much interest from the automotive industry recently. Remote laser welding is an innovative assembly process, but has a critical issue with the energy consumption. Robot companies provide only the average energy use in the technical specification, but process parameters such as robot movement, laser use, and welding path also affect the energy use. Existing literature focuses on measuring energy in standardized conditions in which the welding process is most frequently operated or on modularizing unified blocks in which energy can be estimated using simple calculations. In this paper, the authors propose an integrated approach considering both process variation and machine specification and multiple methods’ comparison. A deep learning approach is used for building the neural network integrated with the effects of process parameters and machine specification. The training dataset used is experimental data measured from a remote laser welding robot producing a car back door assembly. The proposed estimation model is compared with a linear regression approach and shows higher accuracy than other methods.