Abstract: As a classical clustering algorithm, DBSCAN algorithm is widely used in various fields. However, due to the poor adaptability of its parameters, the application effect depends entirely on the setting of parameters. Based on this, an adaptive DBSCAN algorithm based on improved harmony search is proposed to improve the adaptability of DBSCAN algorithm. The algorithm first uses the K-means nearest neighbor algorithm to optimize the initial population, thereby improving the quality of the initial population and providing a high-quality solution for subsequent evolutionary calculations. Secondly, an update operator based on double difference is designed to improve the search ability of the algorithm. Thirdly, two update strategy structures are used to avoid premature convergence of the algorithm, improve the optimization ability of the harmony search algorithm, and comprehensively improve the adaptability of the DBSCAN algorithm. Finally, a variety of datasets are used and comparative experiments are designed to verify the proposed algorithm. Experimental results show that the proposed algorithm has better recognition ability and adaptability.

Key words: adaptive DBSCAN, harmony search, parameter optimization, update operator

摘要： DBSCAN算法作为一种经典的聚类算法被广泛地应用于各领域，但由于其参数的自适应性较差，应用效果完全取决于参数的设置。基于此，提出了基于改进和声搜索的自适应DBSCAN算法，以提高DBSCAN算法的自适应性。算法采用K-平均最近邻算法优化初始种群，从而改善初始种群质量，为后续的进化计算提供优质解；设计了基于双差分的更新算子，提升算法的搜索能力；采用两种更新策略结构避免算法过早收敛，提高和声搜索算法的寻优能力进而全面提升DBSCAN算法的自适应性。采用多种数据集并设计对比实验验证提出的算法。实验结果表明，提出的算法具有更佳的识别能力和自适应性。

关键词: 自适应DBSCAN, 和声搜索, 参数优化, 更新算子