Short-term load forecasting is an essential foundation for the optimization and economic operation of microgrids. In response to the impacts of seasonal, periodic, and random fluctuations in the original microgrid load data on load forecasting accuracy, this paper proposes a short-term load forecasting model for microgrids based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Long Short-Term Memory Neural Network (LSTM). First, the CEEMDAN method is employed to decompose the original load sequence into several sub-sequences of different frequencies in order to better capture features at various time scales. By calculating the fuzzy entropy value of these sub-sequences, components with similar fuzzy entropy values are superimposed to reduce data randomness and decrease prediction time. Finally, meteorological factors, the type of day, and the superimposed historical load sequences are used as inputs for the LSTM model to forecast and obtain the total predicted power. To validate the effectiveness of the method proposed in this paper, a simulation is conducted using a microgrid in a specific region as an example. The results are compared with other prediction models, demonstrating the effectiveness of the proposed method in short-term load forecasting for microgrids.