Abstract:In order to solve the technical problems such as the low numerical accuracy of the classical method to forecast the total electricity consumption of the whole society and the excessively complex structural parameters of the model, this paper proposes a research method combining the big data of electric power with the deep learning algorithm in the field of artificial intelligence. Using computer to build a deep neural network with hierarchical structure, According to bionics principle, Rectified Linear Unit is introduced to solve the problem of gradient disappearance and slow convergence rate of neural network, using gradient descent to optimize model, at the same time, by introducing exponential decay vector automatically set by the neural network model in order to improve the model prediction accuracy and reduce the number of iterations. From the gradient principle of quantity field and Taylor formula, the mathematical principle behind gradient descent method is deduced. In order to solve the overfitting problem, the early stopping algorithm is introduced to improve the training speed and generalization ability of the model. Values forecasted by deep learning algorithms compared with the values forecasted by linear regression algorithm, deep learning algorithm for forecasting of the total electricity consumption of the whole society is superior to linear regression algorithm on the precision and stability index, and the depth of the neural network model for the forecast of future demand for electricity in the whole society has a high degree of credibility.