Izvorni znanstveni članak
https://doi.org/10.7307/ptt.v38i4.1222
An Improved K-means Clustering Algorithm Based on EIQ Analysis for Order Batching of Shuttle-Based Storage/Retrieval Systems
Chuanjun CHEN
; Department of Automation, Tsinghua University, Beijing, China
*
Hongqiang FAN
; Department of Intelligent Engineering and Automation, Beijing University of Posts and Telecommunications, Beijing, China
Junjie LIU
; BZS (Beijing) Technology Development Co., Ltd., Beijing, China
Shun LI
; BZS (Beijing) Technology Development Co., Ltd., Beijing, China
* Dopisni autor.
Sažetak
Shuttle-based storage/retrieval systems (SBS/RS) require efficient order batching to optimise split-case picking. Original K-means clustering, which groups orders based on overlapping SKUs to minimise bin presentations, struggles with high-dimensional, sparse pharmaceutical data due to computational inefficiency, unsuitable distance metrics and unstable initialisation. We propose an enhanced K-means algorithm based on EIQ analysis. High-frequency SKUs are selected using IK frequency filtering, while Pearson correlation is applied to remove redundant features and reduce dimensionality. Cluster centre initialisation is improved using a roulette-based strategy, and cosine distance replaces Euclidean distance to better capture SKU similarity. Case studies using real data from Company A show that the proposed method outperforms both first-come-first-serve (FCFS) and standard K-means in reducing bin presentations and enhancing processing stability. The algorithm remains robust regardless of SKU popularity shifts. Sensitivity analysis confirms strong performance within appropriate thresholds for feature selection (n: 20–25) and correlation filtering (Pearson correlation: 0.8–0.9). Furthermore, as the number of item-lines per order increases, the improved algorithm yields greater efficiency gains. This algorithm can also be well applied to other industries.
Ključne riječi
order batching; SBS/RS; pharmaceutical order; EIQ; improved K-means clustering
Hrčak ID:
346673
URI
Datum izdavanja:
28.4.2026.
Posjeta: 19 *