Research On Flatness Recognition And Control Based On Improved Cloud Network

Plate and strip steel is the main composition of the steel products, which is widelyused for cars, bridges, buildings, instrumentation and other industries. With thedevelopment of the national economy, higher requirement for steel quality was made.Flatness is an important quality indexe of strip steel, and flatness controlling technique isthe hot topic in the rolling area. For the last few decades, artificial intelligence has beenused widely in flatness pattern recognition and controlling, and satisfactory results areobtained in practical applications. This paper choose the cold strip mill flatness patternrecognition and control based on improved cloud network as a research object. Onanalyzing the existing intelligent control approach, improved cloud network was studied,which was applied to flatness control system and was in-depth research.First of all, aiming at the weakness of the existing cloud neural network on trainingand practicality, a new improved structure of cloud neural network is designed. Theimproved network overcomes the deficiencies of the original network; Through theresearch on the similarities and differences between the cloud model and the Gaussianfunction and the combination of T-S fuzzy neural network, T-S cloud inference network isdesigned. The network not only has all the advantages of T-S fuzzy neural network, butalso has the ability of the cloud model to deal with data uncertainty, thus it is a reasonableand effective network.Secondly, the improved cloud network is used in flatness pattern recognition.Flatness pattern recognition model is designed based on improved cloud neural networkand T-S cloud inference network. Simulation experiment demonstrates that the recognitionaccuracy of the improved cloud neural network is slightly higher than the original neuralnetwork. The improved network overcomes the deficiencies of the original network;Antijamming capability of flatness pattern recognition model based on T-S cloud inferencenetwork is significantly stronger than that based on T-S fuzzy neural network. Therecognition accuracy is higher,and it is a new method of flatness pattern recognition.Finally, T-S cloud inference network is applied to the the900HC reversible cold roll. The flatness predictive mode is established. And the mill flatness control system isdesigned. Also a simple controller is developed, and is compared with T-S fuzzy controller.The initial parameters of controller are firstly determined through offline trainingaccording to measured data, then they are adjusted online. The adjust method uses theerror back propagation algorithm. Simulation experiment demonstrates that the system iseffective.