Neural Network-Based Household Electricity Load Prediction and Optimal Dispatch in Smart Grid

With the advancement of smart grid technology, the efficient management of household electricity consumption has become a pivotal aspect of energy optimization. In this paper, a Transformer-based model for household electricity consumption prediction is proposed and compared with the LSTM model, demonstrating the superior prediction accuracy and response capability of the Transformer. We first construct a dataset containing multi-dimensional features such as weather and time, and perform feature selection using Pearson correlation coefficient and Random Forest to screen out the factors that have significant impact on household electricity consumption. Then, this paper employs the Transformer model for electricity consumption prediction. In order to further optimize the household's electricity expenditure, we introduce a time-of-use tariff mechanism, which integrates the scenario of electric vehicles supplying electricity to household loads through V2H (Vehicle-to-Home) technology. On this basis, an optimal scheduling strategy is developed with the objective of minimizing the electricity bill. Simulation results indicate that the proposed strategy not only achieves high prediction accuracy for household electricity demand, but also reduces the cost of electricity by more than 40% through a reasonable scheduling strategy, which significantly alleviates the burden of household electricity consumption.