.png){kind=icon}
.png){kind=icon}
Return
EAS—An adaptive lossless time series data compression algorithm
.png){kind=icon}
PRE
AI
D
W
J
L
X
DOI：10.1016/j.future.2026.108742.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In response to the core challenges in the current floating-point data compression field, such as the rigid trade-off between speed and compression ratio, disregard for data sources, unfriendliness towards vectorization, and high cost of parameter optimization. This paper proposes EAS, an efficient adaptive sampling lossless compression framework. It adopts a hierarchical architecture with three layers to meet various requirements from real-time processing to data archiving. The fast execution layer uses fixed parameters (e=14,f=10) to achieve O(n) linear time complexity, suitable for high-throughput scenarios. The intelligent decision layer reduces the parameter search space to 3 to 5 candidate values based on data characteristics, thereby achieving near-optimal performance within a limited computational budget. The deep optimization layer pursues the final compression ratio through simulated annealing and parallel encoding competition. Key innovations include vector-to-amplitude parameter sharing, a three-level anomaly detection mechanism, and a dynamic selector for ALP/XOR encoding. Through optimization using vectorization and cache-aware design, EAS fully leverages the parallelism of modern CPUs. Experiments on 30 datasets show that EAS outperforms Gorilla, Chimp128, and PDE in terms of average compression ratio, and outperforms Zstd in time series data, while maintaining faster speed and random access compatibility. It performs particularly well in low-precision and low-repetition value datasets, providing a flexible solution for scientific computing and big data processing.
Journal
F
IF:
6.1
Papers:
6.8K
Citations:
2.3W
Organization
Cited Papers