Mechanism and suppression of flow characteristic deviation in control valves under fluid–structure interaction

The ANN was developed using the built-in neural-network application in MATLAB. No custom ANN training source code was developed. The network consisted of five inputs, one hidden layer with 10 neurons, and four outputs. The Levenberg–Marquardt algorithm and sigmoid activation were used, with a maximum of 1000 epochs and an MSE target of \(10^{-5}\). The corresponding fixed 30/10/10 partition and five-fold cross-validation assignments are provided in this workbook.