DDQ‐FKAN Assisted Dynamic Software Cache for Computing Task Offloading in Mobile Edge Computing

ABSTRACT In Mobile Edge Computing (MEC), task offloading and software caching face the challenges of high computational complexity and limited storage capacity. Existing methods are prone to local optima and suffer from low caching efficiency. To address these issues, we propose a joint optimization framework. Specifically, we design an Improved Crested Porcupine Optimizer (ICPO) to obtain the Nash equilibrium solution, where improved population initialization and mutation operations are introduced to avoid local optima. In addition, we develop a Fast Kolmogorov‐Arnold Network (FastKAN)‐enhanced Double Deep Q‐Network (DDQ‐FKAN) to determine the optimal cache vector, leveraging learnable activation functions and Gaussian Radial Basis Functions (GRBF) to accelerate algorithm convergence and reduce energy consumption. Simulation results show that, compared with baseline algorithms, ICPO reduces average energy consumption by 9.78%. Meanwhile, DDQ‐FKAN significantly accelerates convergence and achieves up to a 26% reduction in system energy consumption in multi‐user MEC scenarios, thereby effectively improving offloading efficiency and caching performance.