A Wi-Fi Fingerprint Localization Method Based on Self-Adaptive K-Value and Improved Sparrow Search Algorithm

In indoor environments where GNSS signals are attenuated or blocked, Wi-Fi-based fingerprint positioning technology can be used as a complementary indoor positioning solution with implementation advantages. The technique achieves meter level positioning accuracy by collecting signal strength characteristics from pre-deployed static access points (APs). The weighted K-nearest neighbors (WKNN) algorithm, as a commonly used Wi Fi fingerprint positioning method, has two inherent limitations: First, there is the noise-induced measurement problem, which can result in incorrect matching of reference points (RP). Secondly, there is the issue of the K-value configuration. The commonly used fixed K values can affect the number of matched reference points. To address this problem, this study advances a Wi-Fi fingerprint localization method based on self-adaptive K-value and improved sparrow search algorithm: self-adaptive optimal weighted K-nearest neighbor (SAOWKNN). The algorithm uses the pauta criterion and the adaptive Kalman filter to process the signal, obtains stable received signal strength indicator (RSSI) values and then uses the signal strength to dynamically adjust the Kvalue, solving the problem of setting a fixed K value. Finally, an optimization model is established based on the improved sparrow search algorithm (ISSA). This model looks for the optimal solution of the weights of the K RPs in fingerprint localization to achieve optimal estimation of the position of the target point. Performance comparison experiments of the algorithm proposed in this study with WKNN, self-adaptive WKNN (SAWKNN), CNN BILSTM-Attention and graph neural network (GNN) method were carried out in simulated environments, field collection and public fingerprint database. The experimental results show that, compared to the traditional WKNN algorithm, this algorithm improves the localization accuracy by 25.5%; At the same time, it has a significant real-time advantage compared to the CNN BILSTM Attention method and GNN method.