Existing sound source localization algorithms based on time delay estimation (TDE) predominantly assume the sound source to be near-field. However, in scenarios involving small-scale microphone arrays, such as handheld localization devices like smartphones, the sound sources are primarily far-field. Traditional sound source localization algorithms based on time delay estimation perform poorly in dealing with far-field sources. In order to achieve rapid and accurate localization in such scenarios, a localization algorithm suitable for far-field sources was proposed. Additionally, a criteria was proposed to determine the reliability of time delay estimates. This reliability was used to estimate the time difference of arrival (TDOA) covariance matrix and select appropriate microphone pairs. The microphone was regarded as a node and the distance between nodes was expressed by the reliability of time delay estimates. The challenge of microphone pair selection as a path planning problem in graph theory, i. e. was transformed, finding the longest path passing through all nodes. The results of MATLAB simulation experiments show that considering the sound source as a far-field source, the algorithm proposed exhibits significant improvements in accuracy and precision compared to the traditional Chan algorithm. It is concluded that, after the incorporation of the path planning-based microphone pair selection algorithm, the far-field localization algorithm exhibits exceptional anti-interference capabilities. This implies that even under adverse acoustic conditions, such as low signal-to-noise ratios or high reverberation times, the algorithm proposed can achieve accurate sound source localization.