Algorithmic Improvements to EVO for Optimizing BPNN in Solving Dynamic Inverse Problems

This research develops evolutionary optimization (EVO) algorithm to resolve the dynamic inverse problem and make back propagation neural networks (BPNN) work better. Standard BPNN faces difficulties such as getting stuck in local minima and taking too long to converge while adjusting poorly to changing conditions. To overcome these challenges, this study introduces three key improvements: Our approach features mechanisms to preserve population diversity, updates that adjust to changing conditions, and an optimization system that handles multiple objectives. Our innovations strengthen the algorithm's global search power plus its fast reaction time while making it work better. The effectiveness of the improved algorithm is demonstrated through three experimental scenarios: Our research uses tools like dynamic system control, signal reconstruction, and fluid mechanics inverse models. Our experimental findings demonstrate that EVO-optimized BPNN achieves better optimization results than standard EVO and competing methods across all performance indicators. The new algorithm reduces reconstruction errors to 0.012 and cuts optimization time by about 35%. This study presents a new way to enhance BPNN performance and flexibility in changing environments.