Dolphin: an adaptive lossless compression algorithm for oscillating floating-point time series

The exponential growth of floating-point time series data in IoT and Industry 4.0 applications has made efficient compression critical for alleviating storage and transmission bottlenecks. Oscillating time series, characterized by sign alternation and amplitude fluctuations between adjacent values, typically yield XOR results with few leading zeros. Although this limits leading-zero-based encoding effectiveness, the prefix region exhibits exploitable regular bit distribution patterns. Existing XOR-based streaming algorithms capture temporal correlations through differential encoding but are optimized for moderate leading zero counts, failing to exploit consecutive zero bits in prefix regions. To address this limitation, we propose Dolphin, an adaptive lossless compression algorithm for oscillating floating-point time series. Dolphin introduces a skip encoding mechanism that bypasses redundant prefix intervals by detecting specific bit patterns and consecutive zero sequences, significantly reducing processing overhead. Building upon this, a multi-branch adaptive framework dynamically selects optimal encoding strategies based on data characteristics. Evaluation on 18 real-world datasets demonstrates that Dolphin achieves 1.6%–1.8% compression ratio improvement over state-of-the-art algorithms, with a maximum gain of 3.73%. Compression speed improves by 9.3%–18.3% and decompression by 8.5%–9.7%, confirming significant improvements in both compression effectiveness and throughput.