Smart meters and other advanced metering infrastructures have the capability to collect data at ultra-high frequencies, which helps to more accurately depict the characteristics of electricity usage. However, this process also poses a risk of user privacy leakage. To address this issue, a decomposition-reconstruction technique is proposed, which generates privacy-preserving load curves from the original data while maintaining the basic trend and peak features. Through discrete wavelet transformation, the high-frequency load curve is decomposed into low-frequency basic load components and high-frequency variation components. Subsequently, load curves from different users are adjusted, shifted, and recombined in a random load curve generator according to certain rules to obtain a high-fidelity new load curve. Experiments on real data show that the new load curves generated by this method can fully retain the statistical characteristics of the original data, ensuring the reliability of downstream data mining tasks, while avoiding the privacy leakage risks from high-frequency metering data.