Photovoltaic maximum power point tracking is an important way of improving photovoltaic power generation efficiency. In partial shading conditions, the characteristic curve of a photovoltaic array has multiple peaks, and it is easy for the conventional algorithm to fall into a local optimum. How to find the global maximum power point (GMPP) in partial shading conditions is crucial. In this paper, a locate and shrink algorithm (LSA) is proposed. This applies the concept of boundary shrinkage to gradually shrink the two boundary points to the GMPP. The first stage of the LSA proposes a peak location method that the duty cycle range of the main peaks can be located by adaptive sampling combined with the I-V characteristic curve. The peak location method can be combined with other unimodal algorithms and has strong ability to expand. In the second stage, a shrinkage method based on a three-point criterion is proposed. This can quickly find the peak point in the range of one single peak via shrinking boundaries and has strong environmental adaptability. Both simulations and experiments are performed for the LSA and several other algorithms. The results show that the LSA has obvious advantages in tracking speed, tracking efficiency and steady-state oscillation.