Void fraction estimation in vertical gas-liquid flow by multi-layer long short-term memory implemented in current-voltage system (mlLSTM-SM-CV)

抄録

This study presents the void fraction â estimation by multi-layer long short-term memory with sparse model implemented in multi-layer current-voltage system (mlLSTM-SM-CV) in a vertical gas-liquid flow. In mlLSTM-SM-CV, the voltage vector _lVⁿ is measured at measurement time number n, in two layers l which are upstream-layer u and downstream-layer d, for each measurement pair k, under condition numbers c. SM determines which k is indispensable to â estimation resulting in sparse voltage vector _lVⁿ in l. Here, the upstream-layer sparse voltage vector _u^cVⁿ and downstream-layer sparse voltage vector _d^cVⁿ are reflected by the spatial distribution of bubbles in gas-liquid flow. mlLSTM-SM-CV system consists of two LSTM layers which are 1^st LSTM layer and 2^nd LSTM layer. In the 1^st LSTM layer, both _lVⁿ are arranged based on the time series of each l. In the 2^nd LSTM layer, arranged _lVⁿ are used for â estimation. For train dataset, both _lVⁿ were experimentally measured under c = 30 of the temporal-mean true void fraction ă_true calculated by the drift flux model under all c. For the test dataset, both _lVⁿ were measured under c = 18 of ă_true. Model parameters are optimized resulting in the best parameters of Sid = 64, M1 = 10, M2 = 50 with RMSE = 0.0134 and MAPE = 5.3%, respectively.