This paper presents a bidirectional ac-dc topology based on modular multilevel converter and employing the interleaving concept suitable for solid-state transformer applications. The main features of this topology are derived from the interleaving and the cascading converters concepts, as reduced conduction losses, passive elements operating with effective high frequency, reduced weight and volume, and reduced output high-frequency current ripple. The basic operation of the proposed topology as well as a pulse-width modulation scheme and proper control strategy are described. Simulation results of the proposed converter operating with switching frequency of 2.5 kHz, rated output power of 33.33 kW output voltage 800 V and fed from a 7.96 kV/ 60 Hz ac grid are presented.
This paper aims to present an application of a continuous states and actions actor-critic algorithm based on artificial neural networks function approximations, capable of tracking the maximum power point of a wind turbine system. The application does not use any mechanical sensor for wind. The proposed system is composed of a three phasic permanent magnet synchronous generator of 1 kW, a diode rectifier, a Boost converter and a resistive load. The algorithm consists of two Multi-layer Perceptron Neural Networks (NN) witch approximate the state value function and the DC bus voltage reference, the power signal feedback adjusts NN parameters through backpropagation algorithm. To validate the proposed algorithm, simulations using average models of the system are made on Simulink/Matlab software with different wind speed profiles. A wind turbine emulator test bench using a DC motor to emulate the torque made by the wind on the generator rotor provides experimental results Over more dynamic wind profiles tested, the proposed MPPT algorithm response overcomes the classic P&O.
This paper aims to present an algorithm based on artificial neural networks, with Multilayer Perceptron architecture, capable of tracking the maximum power point of a wind turbine system: (i) without the use of any mechanical wind speed sensors and (ii) with less oscillatory steady state. The proposed method it is based on the traditional Perturb & Observe algorithm. However, it uses the neural network to learn a strategy to define the best action (the update step size) for each operation point. The learning is done through backpropagation. To validate the proposed algorithm, it is shown simulations made on Simulink/Matlab software. The simulations consists of a generic wind turbine of 1kW. The average values of generated electric power are the performance index.