LSTM recurrent neural network for identification of a thermal system

The focus of this graduation thesis is the application of a Long Short-Term Memory (LSTM) recurrent neural network (RNN) in the context of identifying and modeling a complex thermal system consisting of two heaters and two sensors. The proposed approach is based on the remarkable capability of LSTM cells within an RNN to capture and model long-term dependencies in sequential data, thus enabling precise characterization of the thermal system’s behavior. The effectiveness of this methodology is evaluated through the analysis of results obtained from detailed simulations and experimental tests conducted on the thermal system in question. Based on these results, it becomes evident that the application of LSTM recurrent neural networks in thermal system identification holds great potential for advancing the field of control and automation. This technique emerges as a current and effective tool for solving complex problems, offering new possibilities in the analysis and modeling of thermal systems and, by extension, in various practical applications in the field of electrical engineering and industrial automation.