| Streszczenie: | Plural long short-term memory (pLSTM) with sparse model (SM) has been implemented in a developed multiple current-voltage (MCV) system (called pLSTM-SM-MCV) for the a estimation in upward vertical gas-liquid flow. The new MCV system injects constant current to measure the voltage vnk at multiple electrode pairs k at measurement time n with lower noise than the conventional multiple voltage-current (MVC) system. The measured voltage vector Vn is processed to the extracted voltage vector Ven by SM under the assumption in Vn. The Ven is trained for the flow regime identification by the 1st LSTM and the a estimation by the 2nd LSTM in pLSTM-SM-MCV. Experiments were conducted in a vertical gas-liquid experimental setup under the conditions of liquid single-phase flow, bubbly flow, and slug flow. As a result, the Ven was successfully extracted by determining the regularization parameter value λ in the sparse model as λ = 0.01 and λ = 0.042 in bubbly and slug flows, respectively. Under the λ, pLSTM-SM-MCV identifies the flow regime with zero percentage error and estimates the void fraction with a total mean of mean relative error hεi = 0.0275 which is reduced by 42 compared to the former pLSTM without SM (pLSTM-MCV) and by 51 compared to the former pLSTM implemented in conventional MVC system (pLSTM-MVC).
|
|---|