Partial discharges (PD) in power transformers are major latent hazards that can lead to insulation deterioration and equipment failure, directly threatening the safe and stable operation of power systems. Traditional ultrasonic localization methods mainly optimize the localization results based on a single set of observation data, which is highly susceptible to noise and abnormal measurements, thereby limiting their reliability. In contrast, combining multiple sets of observation data can significantly improve the robustness and reliability of localization results. Therefore, this paper presents an ultrasonic localization of PD in power transformers using the Artificial Bee Colony (ABC) algorithm and a novel Intelligent Adaptive Localization (IAL) strategy. Firstly, the ABC algorithm is utilized to independently localize each set of sensor data. Secondly, the IAL strategy is proposed to dynamically identify abnormal conditions and adaptively obtain the optimal results of sample selection and combination of localization methods. IAL achieves a balance between localization accuracy and robustness by analyzing the distribution and consistency of localization results of sets. Experimental results show that the proposed ABC-IAL method attains a localization error of less than 5 cm in noisy environments, providing a new solution for ultrasonic localization in power transformers.