FPGA-based on-line Neural Network in Energy Storage System for Power Regulation of Wind-turbine Generator

Stochastic characteristics of wind power generation may lead to reliability and power quality problems in power systems. This paper presents two control schemes considering state-of-charge (SOC) in the energy storage system for the wind-turbine-generator: constant power mode and power balance mode. Because the response of energy storage system is not fast enough, estimated coming wind power is gained by wind speed forecasting for the charging/discharging control applied to the energy storage system. An on-line learning algorithm is developed in an Elman-based recurrent neural network implemented in the power balance mode to conduct the wind speed forecasting several minutes ahead. This paper uses a Real-time Digital Simulator (OPAL-Lab OP5600) to investigate the performance of the proposed method. The on-line learning Elman-based recurrent neural network is implemented by a Xilinx FPGA. The wind-turbine-generator, power converters, energy storage, resistance load and the power grid are modeled in the Real-time Digital Simulator using SIMULINK/ARTEMIS. Thus, a co-simulation (i.e., FPGA-in-the-loop simulation) is developed in this work. The ratio of CPU times required to study a scenario by a regular PC, Windows Target, and OPAL-Lab on the average is about 14.2: 1.87: 1. The studied results obtained from this FPGA-in-the-loop simulation verify the applicability of the proposed method.