Intelligent Algorithms For Flowrate Determination And Flow Profile Identification In Multipath Ultrasonic Flow Meter

As a kind of clean energy,the exploitation and utilization of natural gas is of great significance in relieving the energy crisis,dealing with climate change,and promoting the sustainable development.The industry of natural gas,in recent years,is being vigorously developed in most countries all over the world and at the same time,its proportion in the total primary energy consumption is gradually rising.As China is a big energy-consuming country,the strategic readjustment of energy structure in China,in addition to the realization of various major projects,such as "the west-east gas pipeline project","the sichuan-east gas pipeline project" and "landing of gas from sea plan",has accelerated the construction of gas pipeline and promoted the development of gas industry.However,the development of high-performance flow meters which belong to the fundamental sensing parts in gas pipeline has fallen behind relatively.Up to now the field of natural gas flow meters in China has been monopolized for a long time,which not only has a big impact on the related industry of flow meters in our countries but severely endangers the national energy information security.Therefore,developing the gas flow meters of high accuracy with independent intellectual property rights is now extremely urgent.Taking the ultrasonic gas flow meter(UFM)which is widely applied in natural gas metering as the research objective,this paper focuses on the research of the algorithm for both flowrate determination,which is crutial in influencing the metering accuracy,and flow profile identification,which is helpful in increasing the accuracy in various complex conditions.As for the algorithm for flowrate determination in UFM,the methods of weighted summation based on Gauss orthogonal integration are traditionally used,among which Gauss-Jacobi and OWICS are mostly utilized.This kind of method is convinent in calculation with a wide range of application.But in complicated flow fields it shows low accuracy,has very high requirements for installation,and easily affected by the type of flow field.It is because that the Gaussian method is deduced based on an assumption of ideal flow profile,which results in its weight coefficients fixed and also its only application in give path positions.With this method applied,flow conditioners are required additively in most cases while straight pipes of enough length both upstream and downstream are also necessary.In recent years,some researchers proposesd to apply intelligent algorithms,such as artifitial neural network(ANN),to flowrate determination,where the weight coefficients could be adaptively adjusted according to different flow fields,and it was effective.Therefore,intelligent flow determination is thought of an important develpement trend of future multipath UFMs.However,there still exist three major problems when intelligent algorithms are used to UFMs.1)The accuracy still need to be increased.Although its accuracy is higher than traditional Gaussian methods,the generalization ability of ANN hasn't been taken full advantage of in most cases since too many parameters have to be configured,which have great influence on the accuracy.2)The design process is too complicated.Architecture and internal parameters of intelligent networks need to be determined by the designer,but mostly it can only depend on their experience or trail and error because of a lack of uniform standards.3)It is not applicable to unknown situations.A trained netword is only suitable in one given situation.But a variety of flow fields will be faced with in actual application,which makes it very difficult to determine which network should be used to the unknown situation even if plenty of networks have been trained in advance.In order to solve the above three major problems,the following researches are made in this paper.1)The genetic algorithm(GA)is proposed to optimize ANN via determining the architecture and parameters so that its generalization ability could be taken full advantage of while the accuracy of flowrate determination would be increased.2)The extreme learning machine(ELM)proposed in recent years is introduced to analytically compute the weight in output layer.The training time is greatly recuced,design process is simplized,and many restrictions influencing the application of intelligent algorithms in UFM are removed.3)A kind of algorithm for flow profile identification is proposed based on multi-class support vector machine(SVM).In addition to realize the flow field visualization indirectely,it is able to automatically recognize the flow field in unknown conditions and further make the application of intelligent algorithms more convenient.Finally,the three proposed algorithms are combined together and experimentally validated based on an experimental setup based on an ultrasonic gas flow meter.According to the results,the algorithms for flowrate determination and flow profile identification proposed in this paper have obvious advantage when compared with traditional methods.The main contents of this paper is as follows:1)The geneticl algorithm optimized neural network(GANN)for flowrate determination of high accuracy.Aimed to solve many problems existing in the application of traditional ANN in UFM,the powerful searching ability of GA is utilized to seek the optimal ANN architecture and training methods and also optimize the internal parameters.The optimized ANN is then trained using data under a specific flow condition.In this way the mapping relationship between multipath velocities and flowrate as well as the internal flow information are included in the trained network.Thus very high accuracy of flow determination will be obtained.In addition,many other problems of traditional ANN are solved meanwhile,which mainly include that it is easily trapped in local minimum,its architecture is hard to determine,and the accuracy depends on the designer's experience.2)The algorithm of extreme learning machine(ELM)for flworate determination in UFMs.Different from traditional error backpropagation(BP)alogirithm,ELM is mainly proposed for neural network with single hidden layer,in which weights and biases in input layer are randomly produced while weights in output layer are computed analytically.When used to multipath UFM,it possesses a lot of advantages.Firstly,it is helpful in solving the problem of traditional ANN,which is that the training process is too slow because of error backpropagation and trail and error.Secondly,it helps to solve the problem of determining many parameters in ANN and SVM in order to simplify the design process.Traditional ANN is easy to be affected by lots of parameters including learning rate,training functions and so on.There also lack uniform rules in choosing the kernel functions and parameters for SVM.In these situations,there is no other choices but the methods of experience,experimental contrast,and cross validation.But for ELM,only one parameter,the number of neurons in hidden layer,need to be determined,which greatly simplify the design process.Thirdly,the accuracy of flowrate determination could be increased.ELM is effective in avoiding the accuracy decrease resulted from incorrect parameter configuration and moreover is able to reduce the risk of being trapped into local minimum.Therefore,with ELM utilized,the effectiveness and accuracy of neural network could be raised so that the application of ANN in UFMs would be further promoted.3)The intelligent algorithm for flow profile identification in UFM.Considering that present ANN is not applicable in unknown flow conditions,this paper proposes an intelligent algorithm for flow profile identification in UFM based on multi-class SVM.With the measured multipath velocities,the algorithm can be used to recognize flow profile of different pipe configurations,and installation positions and angles,which will acquire the flow information and indirectly realize the flow field visualization.When combined with the optimization methods in 1)and 2)together,it will additively provide the functions of monitoring and realtime diagnosis of flow field in addition to extremely increase the accuracy of flowrate determination.4)The experimental validation of intelligent algorithms in UFM.Based on UFM and a flowrate standard facility of sonic nozzle,an experimental setup is established for flowrate metering and the above three algorithms are experimentally validated.An intelligent ultrasonic flow metering system possessing the functions of both flowrate determination and flow profile identification is finally established when combining the proposed three algorithms.