We analyze the mutation characteristics of multiple differential evolution methods and their respective appropriate search state. In order to enable individual's efficient searching with selflearning and selfregulation, we establish a feedback loop, which can help the population choose mutation strategies dynamically according to its own search conditions. Moreover, we also use the conjugate gradient method to find out the optimal search routes in the neighborhood of the best individuals of each generation, and it is useful for searching locally in the neighborhood of optimal solution. Results of the next generation theoretically prove that the proposed hybrid algorithm converges to the global optimal solution at a probability of 1. Experiments on benchmark functions show that the proposed algorithm can improve the efficiency and precision in searching the optimum value. Applied to tension string design problem, the proposed algorithm is able to achieve good structure parameters.