Compressing and reconstructing the voltage data for lithium-ion batteries using model migration and un-equidistant sampling techniques

The long-term storage of the batteries' operating data is critical to tracing and analysing their historical use but challenged by the Trillions of bytes of raw data generated per day. For battery pack applications such as electrified transportation, recording the single-cell voltage requires tens of times more space than other signals such as the pack current. Therefore, an efficient data compressor for the voltage is urgently required to save storage. We here propose to record the entire current trajectory but only partial voltage data in the data-compressing phase to save space. Understanding that the battery's load profiles are often non-stationary, determining an optimum voltage-recording strategy is critical to the reconstruction accuracy but, unfortunately, an NP-hard problem. In this case, a heuristic method is proposed to seek a near-optimum solution with reduced computation. In addition, a battery model is also identified in the compressing phase so that the voltage trajectory can be readily calculated from the recorded current when data reconstructing is required. To compensate for the potential mismatch of the identified model, we establish a migration network using the recorded (partial) data. A piece-wise linear corrector is further fused into the reconstruction algorithm to not only guarantee zero errors at the voltage-recording points but also simplify the design of the above-mentioned heuristic optimisation algorithm. Experimental results show that the root-mean-squared-error of the reconstructed data could be bounded by 5 mV when more than 95% of the voltage data are compressed, paving the way to more efficient storage of large-scale battery operating data.