Probabilistic forecasting of power grid project cost control index based on a variational Bayesian deep learning model

Aiming at the high-dimensional, multi- variable, non-linear and complex uncertainty problem of power grid project cost control index prediction, this paper proposes a power grid project cost control index prediction method based on variational Bayesian deep learning theory. First, the Bayesian LSTM network is used to construct a probabilistic prediction model of control indicators. This model can not only use the learning ability of long- and short-term recurrent neural networks to mine the complex nonlinear relationship between time series cost data, but also capture the influence of uncertainty on the model structure and Influence of potential parameters. Then, the KLqp variational Bayesian inference method is used to train the probability prediction model to obtain the posterior approximate distribution of the parameters of the probability prediction model. Finally, the SGHMC-KDE method is used to obtain the probability density function of the predicted conditional probability distribution, so as to realize the probability prediction of the power grid engineering cost control index. In this paper, the actual case of a provincial power engineering cost database is used for verification, and the simulation results show that the method proposed is significantly better in terms of forecasting skills and forecasting reliability. The model can provide effective uncertainty estimation and prediction results.