Abstract Wi-Fi indoor positioning provides a simple, convenient, ubiquitous and cost-effective solution by matching a pre-established Wi-Fi Received Signal Strength Indication (RSSI) fingerprint database with the RSSI values received from mobile terminals. However, due to the influence of the complex indoor environment on the signal, its accuracy can only reach the meter scale, and the huge fingerprint database leads to inefficient positioning. To solve this problem, the Canopy algorithm is used for coarse clustering, and then the K-means algorithm is used for fine clustering to determine the number of clusters and the initial clustering center to form multiple clustering sub-bases, which improves the positioning efficiency by about 95.05%. In the real-time matching stage, the sub-banks with the highest similarity are selected for matching by the correlation coefficient method, and combined with the Weighted K-Nearest Neighbors (WKNN) algorithm, this paper proposes an improved Bayesian probabilistic optimization algorithm, and the final experimental results show that the average positioning accuracy is improved by about 38.64%, the average runtime is shrunk by about 93.51%, and the stability of the system is slightly improved, which effectively improves the positioning accuracy, real-time performance, and stability.