The guide wall structure in hydraulic engineering is subject to long-term complicated loads, such as alternative water pressure and wind pressure, which may lead to the damage of structures. However, damage detection is difficult to implement in large hydraulic structures under ambient excitation because of the uncertainty of ambient excitation and the limitation of the test condition and precision. Therefore, a new damage detection method, which employs a real encoding hybrid algorithm of clonal selection and particle swarm optimization to optimize the modal frequency index, is proposed for guide wall structures. The proposed method only requires low modal frequency, thus making the method suitable for nondestructive dynamic damage detection of large hydraulic structures under ambient excitation. Taking a guide wall structure as an example, results show that this method has advantages in the global searching performance and identification accuracy. The proposed method is effective and can be applied in many types of large hydraulic structures.