A Classified Identification Deep-Belief Network for Predicting Electric-Power Load

A classified identification deep-belief network (CIDBN) was proposed to improve accuracy and effectiveness in forecasting electric-power loads. The CIDBN is composed of a multilayer Restricted Boltzmann Machine (RBM); the pre-weight of the RBM was trained by means of contrastive divergence. Two classifier models, a constant step classifier and a probability sampling classifier, were employed to build the CIDBN. In the proposed method, the historical load input data were normalized. Then, the layered pre-training of the DBM was implemented by the hierarchical unsupervised greedy pretraining method, and the results were used as the initial value of the supervised learning training model. The simulation results of the power network data in a region of east China from 2015–2017 show that the CIDBN algorithm with 3 network model layers predicted the results with an average absolute error of only 0.802% based on 200 load and temperature training runs. Compared with the self-organizing fuzzy neural network (SOFNN) and the back propagation (BP) neural network, the CIDBN forecast was more accurate.