| Resumo: | According to data provided by PORDATA, there was a sharp increase in domestic energy consumption in the last decade. In the last 3 to 4 years, this energy consumption has reached a stagnation point or even a slight decrease. One possible factor for this could be related to the introduction of smart systems and low-power equipment in our houses. In a context of reducing the energy consumption and increasing the comfort in a domestic environment, urges the need to develop a smart system, fully autonomous, capable of control and monitoring all housing equipment. In this dissertation, it was developed an algorithm capable of automatically control multiple variables. These variables were: heating or cooling temperatures, either for the air inside the house or its sanitary water, the optimal speed of the ventilator fans and the water pumps, and the air renovations needed in the different house divisions. This control was made according to internal variables related to the house itself: ambient temperatures, exterior temperatures, geothermal temperatures, air flows, etc.. For such a development, it was done, in the first line of work, a study of all the existent algorithms and its theoretical foundations. After that, it was made a performance test, so it could be chosen which studied algorithm was the best in an overall perspective. The test consisted of a self-tuning process of the parameters of a classical PID algorithm, so it could be controlled or regulated the temperature of an oven resistance when applied a PWM signal to it. The criteria used when choosing the best algorithm was the comparison between the performance values of the solutions (e.g. overshoot, rise time and steady-state error), plus the difficulty in its implementation. After choosing the best algorithm, the last part of this dissertation was developing an algorithm that would adapt to the exposed problem, predicting the optimal values for the aforementioned variables. The results obtained were congruent to what were expected, reaching all the pretended objectives for this dissertation. It was concluded that, even though the algorithm needs a lot of training data to achieve good results, the neural network achieved and predicted good results, based on a small database with previously taken results.
|
|---|