基于优化模糊C均值算法的锚泊船聚集特性

针对模糊C均值算法随机选择初始聚类中心导致聚类结果对噪声样本点敏感性的不足, 采用局部密度加权的方法, 将初始聚类中心的选择范围限制在局部密度较高样本点区域, 优化初始聚类中心的选择方法; 利用样本点的局部密度改进目标函数, 提高局部密度较高的样本点在目标函数迭代过程中的影响力, 从而提升模糊C均值算法的聚类性能, 并采用人造数据集和鸢尾花真实数据集验证优化的局部密度模糊C均值算法的聚类效果; 通过计算锚泊船位置数据的局部密度, 分析了船舶锚泊偏好。试验结果表明: 对比模糊C均值算法, 优化的局部密度模糊C均值算法聚类精准率提高了2.9%, 召回率提高了3.8%, F度量值提高了3.9%, 说明优化的局部密度模糊C均值算法的性能优于模糊C均值算法; 在锚泊船位置数据上的聚类结果正确反映了天津港锚泊船的聚集特点和锚泊偏好, 其结果与船舶的常规做法一致, 说明优化的局部密度模糊C均值聚类算法是一种分析锚泊船聚集特性和锚泊偏好的有效方法。 

Aggregation characteristics of anchored vessels based on optimized FCM algorithm

For the lack of sensitivity of clustering results to noise sample points due to the random selection of initial clustering centers by fuzzy C-means(FCM) algorithm, by using the method of local density weighting, the selection range of the initial clustering centers was limited to the region of sample points with high local density, and the selection method of the initial clustering centers was optimized. The local density of sample points was used to improve the objective function, and then improve the influence of sample points with higher local density in the iterative process of the objective function, so that the clustering performance of FCM algorithm was promoted. The clustering effect of improved local density FCM(LD-FCM) algorithm was verified by artificial dataset and iris real dataset. The anchoring preference was analyzed by calculating the local density of anchored vessel's position data. Experimental result shows that compared with the FCM algorithm, the clustering accuracy rate, recall rate, and F-measure of the optimized LD-FCM algorithm improve by 2.9%, 3.8%, and 3.9%, respectively, which shows that the performance of the optimized LD-FCM algorithm is better than that of the FCM algorithm. The clustering results on the anchored vessels location data correctly reflect the aggregation characteristics and anchoring preference in Tianjin Port, and are consistent with the general practice of the vessels, which shows that the optimized LD-FCM algorithm is an effective way to analyze the aggregation characteristics and anchoring preference.