Distributed sensor systems enmesh almost every part of our lives. Hence it is an important challenge to provide an effective solution for the controlling task of these systems especially because nowadays resource scarcity is a serious constrain. In my thesis I describe a possible controller implementation for such a sensor system – a model greenhouse constructed in order to provide support for analyzing various agricultural and environment protection tasks.
First, I review the indispensable fundamentals of control engineering, and then I provide an overview about the different neural network architectures as I choose them to perform the forecasting task for the intensity of an infrared lamp system, which simulates the heat of the Sun for the model greenhouse. The importance of this prediction task derives from that the heat provided by these lamps is a good substitution for the built-in heating system. Hereby it is possible to spare energy using the lamps for heating purposes instead of the heating system.
It is a prevalent method when similar areas are being analyzed to study the problem in a simulated environment before it is implemented in the eventual system. This is why I create a model then a simulator for the greenhouse using previous measurement data and my knowledge of thermophysics.
Next, I create a controller for the system. Utilizing the simulator, it is a simpler task to tune the parameters of the controller and examine the influence of the implemented predictor on the system.
I integrate the realized controller into a graphical user interface, which was created as a previous assignment, then I verify the functional controller and its ability to operate on the actual target system and then I evaluate its results.
Finally I show possible ways of improvement for the created application.