Soft sensing model of flow rate for independent metering valves based on LSTM

Flowmeters and mechanism models are common in hydraulic system for estimating flow rates. However, the flowmeters are costly and the establishment of high-precision mechanism models is difficult. To address these challenges, a soft sensing model of flow rate based on a long short term memory (LSTM) neural network is proposed and applied to four-spool independent metering (IM) valves. In this study, the data set was produced by experiments, and gray correlation analysis was utilized for feature filtering. Then the model was trained and tested. The results indicate that the proposed LSTM model has a highly flow rate soft sensing accuracy, with a root mean square error of 1.93 and an average absolute percentage prediction error of only 3.16%.