Guntupalli Radhika, Sudhakaran M, Raj P Ajay-D-Vimal
Department of EEE, Pondicherry technological university, Puducherry, India.
Heliyon. 2022 Jun 6;8(6):e09669. doi: 10.1016/j.heliyon.2022.e09669. eCollection 2022 Jun.
Renewable Energy Resources (RERs) are widely used on the concern of global environment protection. Solar energy systems play an important role in the generation of electrical energy, remarkably minimize the utilization of nonrenewable fuel sources. Solar energy can be extracted and transformed into electrical energy via solar photovoltaic process. Several traditional, soft computing, heuristic, and meta-heuristic maximum power point tracking (MPPT) techniques have been developed to extract Maximum Energy Point (MEP) from the solar photovoltaic modules under different atmospheric conditions. In this manuscript, the combination of reinforcement learning algorithm (RLA) and deep learning algorithm (DLA) called deep Reinforcement Learning Algorithm based MPPT (DRLAMPPT) is proposed under partial shading conditions (PSC) of the solar system. DRLAMPPT can deal with continuous state spaces, in contrast to RL it can be operated only with discrete action state spaces. In this proposed DRLAMPPT, deep deterministic policy gradient (DDPG) solves the problem of continuous state spaces are involved to reach the GMEP in photovoltaic systems especially under PSC. In DRLAMPPT, the representative's strategy is parameterized by an artificial neural network (ANN), which uses sensory information as input and directly sends out control signals. This work develops a 2 kW solar photovoltaic power plant comprises of a photovoltaic array, DC/DC step-up converter, 3-Φ Pulse Width Modulated Voltage Source Inverter (PWM-VSI) integrated with conventional power grid using Constant Current Controller (CCC Effectiveness of the proposed DRLAMPPT with CCC can be validated through an experimental setup and with MATLAB. Simulation and tested at different input conditions of solar irradiance. Experimental results prove that, in comparison to existing MPPTs, suggested DRLAMPPT not only attains the best efficiency and also adopts the change in environmental conditions of the photovoltaic system at a much faster rate and able to reach the GMEP within 0.8 s under PSC. Experimental and simulation results also prove that suggested CCC with LC filter makes the inverter output voltage and the grid voltage are in phase at the lower value of THD i.e. 1.1% and 0.98% respectively.
可再生能源（RERs）出于全球环境保护的考虑而被广泛使用。太阳能系统在电能产生中发挥着重要作用，显著减少了不可再生燃料源的使用。太阳能可以通过太阳能光伏过程被提取并转化为电能。已经开发了几种传统的、软计算的、启发式的和元启发式的最大功率点跟踪（MPPT）技术，以在不同大气条件下从太阳能光伏模块中提取最大能量点（MEP）。在本论文中，提出了一种基于强化学习算法（RLA）和深度学习算法（DLA）的名为基于深度强化学习算法的MPPT（DRLAMPPT）的组合，用于太阳能系统的部分阴影条件（PSC）。DRLAMPPT可以处理连续状态空间，相比之下，RL只能在离散动作状态空间下运行。在这个提出的DRLAMPPT中，深度确定性策略梯度（DDPG）解决了在光伏系统中尤其是在PSC下达到全局最大功率点（GMEP）所涉及的连续状态空间问题。在DRLAMPPT中，智能体的策略由人工神经网络（ANN）参数化，该网络将感官信息作为输入并直接发出控制信号。这项工作开发了一个2千瓦的太阳能光伏电站，它由一个光伏阵列、DC/DC升压转换器、与传统电网集成的三相脉宽调制电压源逆变器（PWM-VSI）组成，使用恒流控制器（CCC）。所提出的带有CCC的DRLAMPPT的有效性可以通过实验装置和MATLAB进行验证。在不同的太阳辐照度输入条件下进行了仿真和测试。实验结果证明，与现有的MPPT相比，所建议的DRLAMPPT不仅获得了最佳效率，而且能够以更快的速度适应光伏系统环境条件的变化，并且在PSC下能够在0.8秒内达到GMEP。实验和仿真结果还证明，所建议的带有LC滤波器的CCC使逆变器输出电压和电网电压在较低的总谐波失真（THD）值下即分别为1.1%和0.98%时同相。
