Self-Tuning Neural Network PID With Dynamic Response Control

PID controllers are widely used and adaptable to various types of systems. However, for the response to be adequate under different conditions, the PID gains must be adjusted. The tuning is made according to the difference between the reference value and the real value (error). This work presents a self-adjusting PID controller based on a backpropagation artificial neural network. The network calculates the appropriate gains according to the desired output, that is, the dynamic response desired which is composed of the transient part and the stationary part of the step response of a system. The contribution of the work is that in addition to using the error for network training, the maximum desired values of overshoots, settling times, and stationary errors were used as input data for the network. An offline training database was created using genetic algorithms to obtain the dynamic response data associated with PID gains. The genetic algorithm allows getting data in different operating ranges and allows using only stable gains combinations. The database was used for training. Subsequently, the neural network estimates an appropriate gain combination, adapting to the error and the desired response. The method performance is evaluated by controlling the speed of a direct current motor. The results indicate an average error of 4% for the database between the requested and system response. On the other hand, the gains estimated by the network in the test dataset (1544 combinations) did not cause instability and complying with the expected dynamic response in 86% of the dataset.