Abstract: Non-invasive load monitoring (NILM) technology has great significance in advancing the intelligence of power system management and guiding users towards more rational electricity usage plans. However, the accuracy of its monitoring results is limited by the scale and authenticity of the electrical load datasets. In addition, the existing public datasets lack sufficient sample quantity and variety, and the comprehensiveness and authenticity of self-constructed datasets also need improvement. To address these issues, this paper proposes a method for simulating residential load data applicable to NILM. First, limited original electrical load data obtained from sampling are subjected to standardize preprocessing and fast Fourier transform (FFT) calculations to derive their dynamic harmonic admittance parameters. Then, an admittance transformation method is introduced to process the dynamic admittance parameters, constraining them within the first/fourth quadrant for easier simulation verification. After that, the processed admittance is utilized at various harmonic orders along with source parameters to deduce a mathematical model of electrical load harmonic admittance. Next, a simulation model is established to generate standard current waveforms for this electrical load. Compared with other methods, the proposed approach demonstrates satisfying fitting effects across multiple scenarios, ranging from simple on/off loads to multi-stage continuously varying loads, from microsecond-level periodic currents to hour-long duration processes, and from single-type load simulations to multi-type load usage scenario constructions. It shows significant merits over existing data generation methods in terms of fidelity, comprehensiveness, and scope of application. Finally, random variables following probability distributions are further introduced into the dynamic parameters to simulate the random errors of actual loads, enabling the generation of electrical load interval currents that account for practical errors, and effectively enhancing the scientific validity and richness of the generated data. This can serve as a reliable source of datasets for NILM.