An indoor positioning optimization method based on dimensionality reduction and clustering fusion

Indoor positioning technology can effectively compensate for the shortcomings of traditional positioning technology, and has broad development prospects in location-based services, innovative business models, emergency rescue, and other areas. The Wi-Fi fingerprint location often requires large database and high location time cost. To obtain a stable and efficient offline fingerprint database and achieve indoor positioning process, this paper proposes a novel accurate and fast indoor positioning optimization method based on dimensionality reduction and clustering fusion joint algorithm. In the data pre-processing phase, the missing rate for access points (APs) screening is introduced to ensure that the APs involved in localization are stable and reliable. For the parameter selection problem of kernel principal component analysis (KPCA), the fitness function of particle swarm optimization (PSO) is redesigned by the mean positioning error to get the principal components which are more suitable for positioning. Moreover, a clustering fusion method using agglomerative hierarchical clustering and clustering evaluation metrics to improve K-means is presented. Improved clustering algorithm can determine the number of clusters and the initial cluster centers, and reduce the positioning error caused by the instability of K-means clustering significantly. The experiments show that the proposed method achieves a mean positioning error of 1.045 m and a positioning time of 0.02372 s compared to the original fingerprint database under the same environment. The positioning accuracy and efficiency are improved by 15.2% and 73.8% respectively. Compared with the unimproved K-means, the proposed method can obtain unique and stable clustering results and the localization accuracy is improved by 6.5%. The robustness and generalizability of the method is verified by multiple evaluation metrics.