Considering the small base of consumer load time series, strong volatility and randomness, along with the difficulty of obtaining high forecasting accuracy, a hybrid model based on Empirical Mode Decomposition (EMD) and Quantile Regression Forest (QRF) is proposed for consumer load probability density forecasting, which is aimed at enhancing prediction precision. Firstly, the signal processing algorithm of EMD is applied to decompose the original consumer load time series, where the sample entropy of each decomposed mode function is calculated. Based on the values of sample entropy, the mode functions can be reconstructed. Then, each reconstructed component is modeled separately using QRF for consumer load forecasting, where the conditional distribution of predicted values can be obtained by superimposing prediction results of different components. Finally, the Kernel Density Estimation (KDE) is used to output the consumer load probability density forecasting results at any time. Compared with deterministic point prediction methods, the proposed probability density forecasting model has advantages of describing the possible fluctuation range and uncertainty of the consumer load in the future, where the case study has also verified its validity. This work is supported by National Natural Science Foundation of China (No. 51507052).