Implementation of temperature and humidity control in air-conditioning test cabinets with deep learning methods

This thesis investigates the performance differences between traditional PID (Proportional-Integral-Derivative) control systems and modern Deep Learning (DL) methods, in particular Long Short-Term Memory (LSTM) networks-based approaches, in implementing temperature and humidity control in climate test chambers. Climatic test chambers play critical roles in a variety of industrial and research applications, and accurately controlling the temperature and humidity levels of these environments is vital for the reliability and repeatability of experiments and products under test. In this study, PID controllers and LSTM-based models are comparatively tested in real-time temperature control scenarios. PID controllers are a long-standing, well-documented and widely used method in control systems design. However, this traditional method has limitations and difficulties of adaptation to complex conditions. The main weakness, which led to the topic of this thesis, is its inability to provide fast adaptation to the variable thermal loads of test objects and environmental variable conditions in climate test chambers. On the other hand, LSTM is an emerging technology in the field of deep learning and has attracted attention thanks to its superior capabilities, especially when working on time series data. Within the scope of the thesis, how effectively and precisely both systems can regulate temperature levels is evaluated on various parameters such as response times, energy efficiency and adaptation capabilities. The experiments are carried out in cabinets simulating air conditioning test chamber conditions and the results obtained are supported by detailed analyses. This study reveals the advantages and limitations of PID and LSTM based systems for temperature control of air conditioning test chambers and makes recommendations for future use in these areas.