Accurate load forecasting can not only help microgrids improve the utilization efficiency of energy resources, but also ensure the stability and reliability of power supply. In this paper, a deep learning-based load forecasting model for microgrids is proposed, and its application effect in different microgrids is verified through experiments. First, deep learning algorithms such as LSTM, BiGRU and CNN are used to construct a hybrid prediction model, and TVFEMD technique is introduced to signal decompose the load data to reduce the influence of noise. Through comparative experiments, the results show that on microgrid 1, the proposed model has a higher prediction accuracy with a minimum MAPE value of 2.0485% compared with the traditional methods, while on other microgrids, the model still maintains a more stable performance. In microgrid 3, the prediction results are more reliable in general, although there is a large error. Based on the experimental results of the model, this paper also discusses the interpretability of the model and its potential application in real microgrid scheduling. Ultimately, the proposed deep learning model can effectively improve the accuracy of microgrid load prediction with strong adaptability and stability.
