Study Of Active Learning In The Automatic Gauge Control System

Active learning algorithm is an effective methoed that can be used to reduce the demand of machine learning algorithms mark data, which uses unlabeled sample in the form of aid marked sample for classifier training, in order to build the high accuracy classifier. The application of active learning in the thickness control system is an important issue in the field of machine learning and intelligent control.In view of the intelligence and the adaptability of the control system are weak,establishing a active learning development model from the simulation of biological selflearning. Based on this model, thickness control system can make corresponding adjustment according to the feedback value, and accumulate experience to improve the intelligent degree of adjustment.Firstly, for the sample collection time-consuming problem in BP neural network training, introducing a BP neural network control algorithms based on active learning.simulation results show that by using active learning effectively reduces training time.Secondly, aiming at the problem that traditional sampling strategy algorithms is easy to collect duplicate samples, proposing a modified sampling strategy to achieve the training sample collection, and using the rapid decline for network self-learning,proposing a modified multilayer perceptron neural network nontrol algorithm.Finally, in view of the ultimate network performance affected by its own fixed structure, a dynamic neural network is put forward. In order to build a active learning development model, adjusting the dynamic network structure according to the global sensitivity analysis to achieve PID controller tuning parameters online, then the strip thickness intelligent control systems are established.Because of the application of active learning algorithms, the intelligence and adaptability of the thickness control system can be improved. The simulation results prove the validity of the study. Due to the limitation of time and conditions, there are still many aspects waiting for improvement and further discussions.