The invention discloses a power load prediction method based on phase space reconstruction and data driving, and the method comprises the steps: firstly calculating the delay time and embedded dimension of historical load data, dividing a data set into a training set and a test set, carrying out the phase space reconstruction of the training set and the test set according to the delay time and embedded dimension, and carrying out the normalization processing; secondly, comparing the prediction precision of the twelve data driving models under the same data set, wherein the obtained XGBoost isthe optimal model in the statistical learning model, tand he LSTM and the extreme learning machine are the optimal models in the neural network; carrying out model, day, week, half month and month load prediction on the three models respectively, and keeping high prediction precision; and finally, performing unequal weight combination on the XGBoost, the LSTM and the extreme learning machine by using a grey relational degree method, and constructing a combined data driving model. The method can improve the prediction precision.