Application of feedforward neural network in prediction of continuous discharge coefficient
It is challenging to construct a continuous discharge coefficient model by using hydrodynamics when a storage tank discharges from a leakage hole continuously due to the complex interplay of multiple factors,such as leakage hole geometric parameters,liquid level,liquid physical properties,and flow state.This paper introduces a novel model to predict atmospheric storage tanks'continuous discharge coefficient(Cs)when atmospheric storage tanks continuously discharged,which utilizes the Feedforward Neural Network(FNN)algorithm and incorporates six input variables as liquid level above the leakage orifice(h),decreased velocity of liquid level(Vh),discharge time(t),equivalent diameter of orifice(d0),dynamic viscosity(m)and tank diameter(dT).Data samples are obtained by a series of continuous leakage experiments to construct a dataset,eighty percent of the dataset is allocated to a training set,and the remaining is set automatically as a testing set.The FNN algorithm is employed to establish the nonlinear relationship between Cs and its corresponding input variables.The model's evaluation performance reveals that the model's mean absolute error(EMA),explained variance score(SEV)and determination coefficient(R2)are 0.015 4,0.949 2,and 0.948 2,respectively,indicating that the prediction performance of the model is quite outstanding.When comparing with the corresponding experimental values of continuous discharge,it was shown that the mean absolute deviation of C.ranged from 0.06％to 15.40％,mean deviation of mass flow rate and continuous discharge mass ranged from 3.20％to 9.87％and 0.16％to 4.40％for the small-hole scenario,respectively;for the medium-hole scenario,mean absolute deviation of Cs ranges from 1.81％to 7.10％,mean deviation of mass flow rate and continuous discharge mass range from 2.06％to 6.95％and 0.29％to 1.37％,respectively.Therefore,the results calculated by C.outperformed those obtained by empirical constant,indicating that the proposed model can effectively predict values and trends of C.during continuous leakage.
safety engineeringcontinuous discharge from storage tankdischarge coefficientdeep learningFeedforward Neural Network(FNN)prediction model
万方数据
维普
