Motor optimization design
In order to answer this project, the description of the problem can be found, which consists of 3 main points. Firstly, the aim is to find the 5 best engine designs given a multi-objective problem using as few simulations as possible. Then, by means of reinforcement learning techniques, the stabilization of the active power of the motor at the minimum point is achieved. Finally, the signals of 7 motors are treated trying to detect possible anomalies. Optimisation of more than 1 target is difficult to visualise, the generated visualisation is shown below. The MOEAD, NSGA II, NSGA III & SPEA2 algorithms have been used for this propose.
Reinforcement learning is powerful in many cases due to its constant iteration to explore the unfamiliar space. When it comes to that iteration, it is important to mention that there is no supervisor, but a reward signal and the feedback is not instantaneous. Besides, as data is sequential, time really matters as well as the chosen actions affect the subsequent received data. In this problem, there are 143647 states and 4 actions (+X, -X, +Y, -Y), and there are 2 objetives to optimise.
