In this paper an EMD-based (Empirical Mode Decomposition) technique is proposed in order to improve the data compression and reconstructed signal accuracy trade-off for PQ (Power Quality) signals. We demonstrate that our technique saves computational cost and reduces the mode mixing problem when compared to the EEMD (Ensemble EMD) and traditional EMD, respectively. After simulations with recorded power line test-cases, we compare our results with DWT (Discrete Wavelet Transform), and both traditional and EEMD. Results have shown that our technique outperforms the DWT in both data compression ratio and reconstruction accuracy metrics. Moreover, we analyze the balance between computational costs and mode mixing reduction for the three EMD techniques.