【摘要】 原始步态数据普遍存在时间长度异构与空间尺度异构问题，传统动态时间规整(DTW)等方法抗惯性测量单元(IMU)噪声能力弱，难以适配信息化场景，致使K-means聚类簇内混杂率偏高。为解决该问题，本文提出创新方案：设计“数据清洗-空间对齐-时间对齐”三步预处理流程，提取时间统计动态特征、空间几何形态特征及时空耦合特征，再从初始质心(K-means++)、K值选择(肘部法+轮廓系数+CH指数)、距离计算(PCA降维+加权欧氏距离)及异常值抑制(迭代加权)四方面优化K-means算法。实验验证表明，本文方法聚类纯度达87.8%、簇内混杂率仅8.3%,聚类耗时≤100 ms,满足康复医疗信息化平台实时传输需求；消融实验证实，时间对齐模块可使纯度提升≥8%,空间对齐可使纯度提升≥11%。该方法有效突破传统技术瓶颈，可为步态评估提供可靠技术支持。更多还原

【Abstract】 The original gait data generally have problems of heterogeneous time lengths and spatial scales. Traditional methods such as Dynamic Time Warping(DTW) have weak anti-IMU noise capabilities and are difficult to adapt to informationized scenarios, resulting in a high intra-cluster contamination rate in K-means clustering. To solve this problem, this paper proposes an innovative solution: designing a three-step preprocessing process of “data cleaning-spatial alignment-temporal alignment”, extracting temporal statistical dynamic features, spatial geometric shape features, and spatio-temporal coupling features, and then optimizing the K-means algorithm from four aspects: initial centroid(K-means++), K value selection(elbow method+silhouette coefficient+CH index), distance calculation(PCA dimension reduction+weighted Euclidean distance), and outlier suppression(iterative weighting). Experimental verification shows that the clustering purity of this method reaches 87.8%, the intra-cluster contamination rate is only 8.3%, and the clustering time is ≤100 ms, meeting the real-time transmission requirements of the rehabilitation medical information platform. Ablation experiments confirm that the temporal alignment module can increase the purity by ≥8%, and the spatial alignment can increase the purity by ≥11%. This method effectively breaks through the technical bottlenecks of traditional methods and can provide reliable technical support for gait assessment.更多还原