Adaptive K-Means Clustering for Indoor Localization Models

The patent implements a dynamic K-means clustering approach where the number of clusters K is not fixed but adaptively determined through a two-phase process: a split phase that increases K when clusters are too large, and a merge phase that decreases K when clusters are too small. This dynamic adjustment resolves the contradiction by optimizing location differentiation accuracy while managing computational complexity through adaptive rather than exhaustive computation.

The patent performs preliminary actions by establishing initial clusters and then iteratively refining them through split and merge phases. The algorithm pre-processes the data by creating initial cluster assignments and then systematically improves them through controlled iterations, reducing the overall computational burden compared to attempting to determine the optimal K value directly from the beginning.

The patent dynamically adjusts K during the clustering process rather than using a fixed small value. The split phase increases K when clusters contain too many elements, and the merge phase decreases K when clusters are too fine-grained. This dynamic approach maintains computational efficiency by starting with a reasonable initial K while improving area differentiation accuracy through adaptive refinement based on actual data characteristics.

The algorithm incorporates feedback mechanisms where cluster characteristics (size, density, spatial distribution) are continuously evaluated during the split and merge phases. This feedback loop allows the system to automatically adjust K to achieve optimal differentiation accuracy without requiring manual intervention or exhaustive computation, thus maintaining efficiency while improving precision.

The patent segments the clustering process into distinct phases: an initial clustering phase, a split phase for increasing granularity, and a merge phase for reducing complexity. This segmentation allows the algorithm to achieve high cluster granularity when needed while managing computational difficulty by processing adjustments in controlled steps rather than attempting to determine the final optimal structure all at once.

The patent implements dynamic adjustment of cluster granularity through the split and merge phases. Rather than committing to a high K value from the start (which would increase computational difficulty), the algorithm starts with a moderate K and dynamically increases granularity only where and when needed based on data characteristics, thus achieving high precision without proportional increases in computational difficulty.

The patent performs preliminary clustering to establish initial cluster assignments before executing the split and merge refinement phases. This preliminary action provides a reasonable starting point that reduces the number of iterations needed to achieve optimal localization accuracy, thereby reducing time consumption compared to starting from scratch or using more exhaustive optimization methods.

The algorithm dynamically determines when to stop iterating through the split and merge phases based on convergence criteria or maximum iteration limits. This dynamic termination strategy ensures that the model achieves sufficient localization accuracy without unnecessary additional iterations that would only increase time consumption, thus optimizing the trade-off between accuracy and time.