Design of a multi-level deep residual neural network for short-term prediction of electrical loads in power systems
To maintain supply-demand balance, it is crucial to design a method to provide prior knowledge on load consumption in look-ahead time with high level of accuracy and reliability. The load prediction problem is becoming more and more challenging due to emerging new concepts in the electrical grids and reconstruction of the power networks. This paper develops a residual neural network to predict the electrical loads with high level of accuracy. In the designed network with combining two powerful deep residual network, a new residual deep network is proposed to improve the learning ability as well as prevent problems like overfitting and gradient reduction/explosion. Furthermore, to fully understand the spatial-temporal features, convolutional neural network (CNN) and gated recurrent unit (GRU) are combined and integrated into the designed multi-level deep network. The seasonal analysis and investigating several cases using actual electrical load consumption in Shiraz, Iran verifies the effectiveness of the proposed method and higher accuracy of the proposed deep network in comparison with other methods demonstrate the superiority of the proposed method.
-
Electrical Load Parameter Identification using Multi-Variant Structure Based on Deep Learning
Omid Izadi Ghaforkhi, *, Ahmad Forouzantabar
Journal of Intelligent Procedures in Electrical Technology, -
Short-term Load Forecasting Using a Graph-based Deep Learning Structure
Mahtab Ganjouri, *, Ahmad Forouzantabar, Mohammad Azadi
Journal of Novel Researches on Smart Power Systems,