Transient Current Maintenance using Controller Based on Recurrent Neural Network in High Voltage Direct Current Approximation Model

A detailed high voltage direct current (HVDC) is developed to simulate transmission system behavior in transferring electric energy from the sending to the receiving end of the power system simulation. This model is the interaction between the pulse signal control and current/voltage AC/DC on semiconductor devices, including switching-level systems. However, significant computational resources are needed to simulate the detailed HVDC model. An approximation model or average value model (AVM) is considered to replace the exact HVDC model for a light computation burden. High peak overshoot in transient current is a problem in this HVDC system, especially during the start-up mode. So controller based on the recurrent neural network (RNN) is proposed to maintain the direct current transient response at starting time. The RNN is built with three inputs and one output structure. The training processes are conducted in the offline stage. The training data are obtained from the HVDC equipped with a proportional integral controller and various parameters and conditions. The RNN control can maintain the transient current response with lower peak overshoot and ITSE than the conventional control. The RNN control settling time is also slightly smaller than the competing control.