Time Series Analysis On Differential Pressure Signals Of Gas-liquid Two-phase Flow Through Slotted Orifice

Based on a self developed prototype of wet gas meter and experimental data set, a series of works have been carried out. In this thesis, time series analysis techniques were used to process the differential pressure of slotted orifice for gas-liquid two-phase flow. The correlations between the parameters of time series model and gas and liquid flow rates were investigated and mapped by neural network. The main works are listed as follows:(1) A novel signal processing method was put forward. Firstly, the differential pressure signal was de-noised by wavelet transform using the modulus maximum method; and then de-noised signal was decomposed into a finite number of intrinsic mode functions (IMF) by empirical mode decomposition. The proposed method is a powerful tool for turning the non-stationary original differential pressure into a set of stationary time series, which satisfy the conditions of time series analysis.(2) Autoregressive (AR) models were built with different IMFs. Correlations among AR model parameters, flow regimes and individual flow rates were analyzed in detail. The model parameters can not only distinguish flow regimes but also have close correlation with the variation of gas and liquid flow rates. The results are fundamentally important for the individual flow rates measurement by neural network.(3) The neural network was used to map the relationship between the AR model parameters and gas liquid flow rate. The genetic algorithm (GA) was employed to optimize BP network's initial weights, and then performances of the BP and GA-BP network were compared under different model parameter combinations. The results show that the GA-BP network is more accurate than the BP network with the same inputs. Relative errors of the GA-BP network are within 10% and 15% for gas and liquid flow rates respectively at 90% confidence level when the liquid flow rate is more than 0.75m3/h.