Balancing exploration and exploitation in conventional inversion algorithms is challenging. We evaluated the effectiveness of tandem neural network architecture (TNNA) in addressing this challenge. Inversion problems with different heterogeneous scenarios were designed to compare TNNA with four metaheuristic algorithms. Results show that TNNA significantly improves computational efficiency and accuracy, offering a promising framework for developing robust inversion algorithms.