Abstract: The port is an important logistics and transportation hub in our country, the high-precision port personnel positioning system plays an important role in the process of safety production and management. By combining radial basis function (RBF) and extreme learning machine (ELM), an RBF-ELM port personnel location method based on channel state information (CSI) is proposed to realize personnel location by a single access point (AP). In the offline training stage, the positioning area is first gridded and the CSI data of each grid position is measured; position fingerprints are constructed by fusing the amplitude difference and linearly reconstructed phase information, and dimensionality reduction was performed by principal component analysis (PCA); subsequently, simulated annealing (SA) is embedded into quantum particle swarm optimization (QPSO) to improve the global search ability, find the optimal parameters of the model, and thereby improve the accuracy and generalization ability of positioning model. In the online prediction stage, the prediction results of the model are matched with labels and coordinates in the location library by weighted K-nearest neighbors (WKNN) matching algorithm to obtain more accurate positioning coordinates. At the same time, the weight update method of dynamic forgetting factor is adopted, and only part of calibration data can be collected to complete the update of positioning model weight, which effectively alleviates the problem of positioning accuracy decline caused by changes in time and environment. In the simulated port scenario, the probability of the positioning accuracy of the system reaching 1m in line-of-sight (LOS) environment is more than 92%, the static average error is 0.58 m, and the dynamic average error is 0.86 m. The static average positioning error of 0.67 m can also be obtained in non-line-of-sight (NLOS) environment. Compared with other positioning methods, the proposed algorithm has improved positioning accuracy, timeliness and stability.