In response to the lack of physical consistency in the load model of household electrical equipment, a modeling method for household electrical load is proposed based on the working and composition principles of household appliances. Firstly, the electrical model is equivalent to different combinations of differential equations. Secondly, based on the Physical Information Neural Network (PINN), the physical equations are used as part of the neural network loss function to build three types of household electricity load models. Finally, the simulation data waveform was compared with the measured data waveform to verify the accuracy of the model. The research results indicate that the proposed method for modeling household electricity load based on physical information neural networks exhibits significant advantages under data constraints. In the validation experiment, the fitting error of the model for typical loads was less than 3.5%, which is about 40% lower than the error of traditional LSTM models. It can generate household load waveforms that are close to the actual situation, providing a new physically interpretable model support for non-invasive load monitoring.