Centrifugal Compressor Performance Based On Fuzzy Neural Network Research

Centrifugal compressor is important high speed equipment widely used in industry.Its operation charaterictics is related with production,whose actual operating state usually differs with the designal operation.The performance curve based on actual situation should be recalculated and estimated.Traditional theoretical arithmetic models are often dependent on the similarity hypothesis or empirical formula,or complicated fluid dynamics theory. Since centrifugal compressor has many influencing factors,the traditional arithmetic models have low precision and nonuniversalization.Performance model of centrifugal compressor based on fuzzy neural network have been established in this paper.Its feature of fuzzy neural network is possible to settle down the above problem.The related problems as to adopting fuzzy neural network to predicate the performance of centrifugal compressor have been discussed,which provides better model for performance prediction and control of centrifugal compressor.Performance model of centrifugal compressor based on Elman dynamic recursion artificial neural net has been established.The calculation results based on the model indicate that when output is 2-D vector,testing error obtain least numerical value if node number of the hidden layer and structure layers is 4.Training error is 3.12201,testing error is 0.5906,and model is capable to reflect systemic running states,but its presicion is not high enough.Performance model of centrifugal compressor based on Takagi-Sugeno fuzzy reasoning has been established by applying fuzzy neural network with self-adapting function.Fuzzy clustering arithmetic is MMC.Results indicate that the net has better performance when r_a is 0.6 and r_b is 1.1,training error is 0.52844,testing error is 0.59579.The model is less fuzzy rules,higher precision,less calculation.Main performance model of centrifugal compressor based on fuzzy C mean clustering to reflect essential running state of centrifugal compressor has been established.Fuzzy neural network will be trained by output of main model and another input 3-D vector.Net initiatory weight is not random,resulted in commendably reducing the net training time and improving the net converges to disqualification local minima.Result indicate that the model is capable of reflect running states of centrifugal compressor.The training error is 0.2618 and the testing error is 0.3622.It is turn up trumps.The turbine and centrifugal compressor have been looked upon a unit which boils down to a single order seismic system.The fuzzy control model has been established with the part,which combines the unit with the model of centrifugal compressor based on fuzzy neural network.Result indication:(1)only adjusting deviation quantizing gene or deviation change quantizing gene is not ideal enough for system response capability.The system response time and overshoot have been gradually improved when adjust deviation quantizing gene in company with adjust deviation change quantizing gene in a certain range.The model obtains better capability.(2)The enlargeing control increment proportion gene or increaseing control rule table sensitive degree to deviation,resulted in reducing responses time and enlarging responses overshoots,increasing convergance time.(3) Making variable membership function approach to midst,resulted in increasing response time,minishing overshoot,and increasing convergence time.Integrating factor self-regulating mechanism has been joined esential fuzzy controller. The calculation results based on the model show that the controlled model is capable of rapid respond to the variety of desired value,reaction time not exceeds five second,and the response have rapid stabilized after achieved desired value.