| Abstract [eng] |
When developing modern artificial intelligence based systems, it is important for them to be useful and able to adapt to the dynamic environment. Particular attention is paid to the development of smart services, improvement of their quality, monitoring behaviour of their users and integration of adaptive components. Therefore, the aim of this dissertation work is to modify the existing methods and propose new ones of affect recognition and adaptive control in order to develop a control system of the closed environment microclimate, implementing dynamic provision of services, based on the estimated affect state of humans. The dissertation work proposes the method designed to remove labelling error of the training sample that allows increasing the accuracy of classification, while solving a wide range of challenges when the algorithms of supervised learning are used. The integrated adaptive control method has been proposed for microclimate control system of closed environment. It is an improved method of microclimate control system, when a person participates in a system as a sensor of ambient comfort and the evaluation of affect enables the estimation of suitability of environmental parameters for that person. The results of experimental simulation has shown that using a reinforcement learning and the function of affect evaluation, it is possible to achieve the optimal light, temperature, and air quality parameters in accordance with the estimates of the affect of a human. |
