The invention discloses a PCA and quantile regression forest-based short-term load probability density prediction method. The method comprises the following steps of 1) collecting basic data requiredby the short-term load prediction of a power system, such as load historical data, temperature, humidity and other meteorological information, date type and the like; 2) selecting input variables relatively strong in correlation from influence factors, and constructing a proper training sample set; 3) carrying out dimensionality reduction treatment on an input variable set by adopting the principal component analysis; 4) establishing a quantile regression forest prediction model for data after being subjected to dimensionality reduction treatment, and obtaining a regression prediction result under any quantile condition; 5) obtaining the short-term load probability density prediction through the kernel density estimation. According to the invention, by adopting the above method, the short-term load prediction precision is effectively improved. According to the method, the prediction result of the load probability density at any moment can be obtained. The short-term load prediction problem of the power system can be well solved.