<p>Energy conservation is an increasingly critical concern in manufacturing. The extensive application of industrial robots poses new challenges to energy conservation. To achieve excellent quality and operational performance, focusing on energy efficiency and time optimization is crucial. To tackle the challenges of multi-robot welding task planning, which involves multiple constraints and strong coupling, the problem is decomposed into three sub-problems. These include welding task allocation, collision detection, and single-robot trajectory planning. A multi-robot time-energy optimization (MRTEO) model is proposed, considering constraints such as robot speed, acceleration, accessibility, collision risks, and operation time. To optimize the MRTEO model, a squirrel search algorithm integrated with neural networks (NSSA) is developed. Firstly, to address the impact of the randomness of initial population results on algorithm convergence, a topology structure-based self-organizing mapping algorithm is introduced to generate the initial population. Secondly, considering the diverse characteristics of welding gun feed angles, a robot trajectory and welding task allocation optimization method based on squirrel search algorithm was constructed by integrating feed angle search. Additionally, mechanisms are implemented to prevent the squirrel search algorithm from converging prematurely, thereby maintaining individual and population diversity. Finally, to improve the efficiency of trajectory planning, a collision detection method based on backpropagation neural network is proposed. The performance of the algorithm is evaluated through a case study of body-side welding, which is compared with benchmarks. Results indicate that the proposed algorithm achieves an 18.62% reduction in energy consumption and a 64.20% improvement in the consistency of robot operation time.</p>