Abstract: To address the issues of slow convergence and susceptibility to local optima in traditional differential evolution algorithms, as well as the poor optimization stability caused by the randomness in individual selection, a multi-restart strategy is introduced in this paper. The algorithm is executed multiple times with different random seeds, increasing the algorithm’s spatial exploratory capability and, to a certain extent, resolving the problem of easily falling into local optima. Through the incorporation of a new mutation strategy, the optimization stability is improved by approximately 10%. Additionally, a parameter self-adaptive tuning mechanism is introduced, dynamically adjusting the algorithm’s parameter values, resulting in an approximately 10% increase in convergence speed and enhancing the algorithm’s robustness.