چكيده به لاتين
A combined cycle power plant uses two or more different energy production processes, such as gas and steam, simultaneously to produce electricity. This type of power plant uses gas and steam cycle, which increases efficiency compared to traditional power plants. In combustion power plants, combustion control is very critical because it directly affects energy efficiency, reduction of greenhouse gas and pollutant emissions, and can also reduce the overall costs of the power plant. In combined cycle and gas power plants, the combustion chamber is considered as the most sensitive part of the power plant in this regard. Accurate control on the inlets and outlets of the combustion chamber is necessary to provide optimal combustion and improve the performance of the power plant. Distributed control systems (DCS) are the most widely used control systems in power plants, where different controllers can be used. The most important of them include proportional-integral-derivative (PID) controller, model predictive control (MPC) and adaptive control. In the first two types, with the increase in energy demand and problem size, their power and efficiency decrease greatly. Conventional PID control systems with fixed parameters cannot meet the increasing control requirements in production. As combustion systems must meet air pollution standards, their design and operation become more complex. In this study, a gas cycle of a combined cycle power plant using reinforcement learning and DQN algorithm in order to optimize and control performance variables including thermal efficiency, total cycle cost, and the amount of emissions during operation and finding a suitable trade-off between them without having a mathematical model It was investigated by controlling and changing control variables including fuel mass flow rate, air mass flow rate and compressor pressure ratio. This work has been done for the first time and there is no study in the field of gas cycle control using reinforcement learning. The results of this study and the learning process of the artificial intelligence agent in the defined environment show the good performance of this proposed method. The optimal value for 3 performance variables, including the thermal efficiency of the cycle, the total cost of the cycle, and the emission rate of the cycle during operation are 32.18%, 0.95 $/s, and 6.32 kg/s, respectively, which are reasonable values for a gas cycle. Based on the obtained results, it can be acknowledged that the controller based on reinforcement learning methods can have a high potential in acquiring knowledge and automatic performance in controlling and optimizing the combustion process in combustion power plants. Also, system control and management using reinforcement learning algorithms can overcome the most important disadvantages of conventional control methods.
