A TDOA joint optimization model (TDOA-PSO-NI-GD) that integrates Particle Swarm Optimization (PSO) and Newton's Iteration-Gradient Descent (NI-GD) method is proposed in response to the problem of insufficient positioning accuracy of multi-stage rocket debris, which improves the localization performance through the synergistic mechanism of global search and local optimization and constructs the multi-debris signal separation constraint model and the environmental interference compensation model. Experiments show that the model reduces the positioning error from 1~10 km to less than 0.5 km in the traditional single-stage optimization algorithm in the rocket debris recovery mission, and the multi-debris signal separation rate reaches 96.2%, and maintains the sub-kilometer accuracy under the mountainous terrain and strong wind disturbances; the validation of the Chan-Taylor algorithm combined with the least-squares method shows that its anti-jamming and positioning reliability are significantly better than that of the existing methods. The algorithm can be extended to mobile communications, unmanned vehicles and other fields, with both theoretical innovation and engineering application value.