Research On Dynamic Model Of CO₂ Compressor Based On Feedback Neural Network

With the development of society and the advancement of technology, CO₂ centrifugal compressor can meet industry on a variety of gas compression requirements, thus the use of the field is very wide. Therefore, in-depth understanding of the working characteristics of the compressor, with the most appropriate algorithm to control the compressor to work in the best condition is necessary. Because many factors affect the performance of the compressor, the traditional method of modeling is often not accurately reflect the performance curve, and neural networks, an analog behavioral characteristic of animal nervous systems, information Distributed parallel process in algorithm model, has a good ability to learn and close any non-linear function characteristics. The main research contents are as follows:?1? The determination of model of centrifugal compressor. First, work process CO₂ compressor system is analyzed, initially identified consists of two parts:the turbine section and the compressor section. Then get the response of the test data, and the data filtering, smoothing. Then, using the processed data to analyze the correlation between the different state variables, excluding the irrelevant state variables or the correlation is relatively small. Finally, the turbine speed model and the input and output of the mufti-input and mufti-output model are determined respectively.?2? Prediction research based on neural network. Construction of compressor state quantity prediction model were based on BP and Elman network, and the prediction results are compared. Then the input nodes of BP network and Elman network to increase self feedback by adding a feedback output to the input, and establish the model based on the neural network. The transfer function of the decoupling control system is obtained, and the model is established by the step response method. Finally, the overall prediction results of the above models are compared.?3? When the control variables are changed, corresponding to the other variables, and the variation pattern is similar to the one with the delay of the first order inertia step response curve, the model is established by the step response method. On the basis of feedback from BP and external feedback Elman neural network is also modeled, and the model of the step response method were compared.?4? Established based on mufti-variable decoupling PID method and fuzzy control method based on mufti-variable decoupling, and the decoupling effect of decoupling of the two methods were compared, the latter in the model control and decoupling are better than the former. The PID manipulation rules and neural network hidden layer unified together to form a new network to achieve decoupling, the decoupling control has a good performance relative to conventional PID control and fuzzy decoupling, and quickly stable state.In summary, CO₂ compressor model established in this paper can play a good state quantity prediction, at the same time can be implemented according to requirements of the compressor decoupling requirements.