Decomposition-Based Approximation of Time Series Data with Max-Error Guarantees

With the growing popularity of IoT nowadays, tremendous amount of time series data at high resolution is being generated, transmitted, stored, and processed by modern sensor networks in different application domains, which naturally incurs extensive storage and computation cost in practice. Data compression is the key to resolve such challenge, and various compression techniques, either lossless or lossy, have been proposed and widely adopted in industry and academia. Although existing approaches are generally successful, we observe a unique characteristic in certain time series data, i.e., significant periodicity and strong randomness, which leads to poor compression performance using existing methods and hence calls for a specifically designed compression mechanism that can utilise the periodic and stochastic patterns at the same time. To this end, we propose a decomposition-based compression algorithm which divides the original time series into several components reflecting periodicity and randomness respectively, and then approximates each component accordingly to guarantee overall compression ratio and maximum error. We conduct extensive evaluation on a real world dataset, and the experimental results verify the superiority of our proposals compared with current state-of-the-art methods.