Rocket engines thrust parameters prediction based on long short-term memory attention model with Bayesian optimization

The sensor measurement parameters of liquid rocket engine (LRE) ground test are key indicators reflecting the engine's working performance, and parameter prediction is of great significance for engine performance evaluation and fault diagnosis. In response to the difficulty in predicting parameters due to the large range of parameter fluctuations under changing operating conditions, the thrust parameter is taken as the prediction target, a prediction model combining long short-term memory network (LSTM) with self attention mechanism has been established, to explore the long-term and short-term variation patterns of multivariate sensor time series. To maximize the predictive ability of the network, Bayesian optimization algorithm is used to optimize the hyperparameters combination of the model. The experiment shows that this method can predict changes in thrust parameters, with a root mean square error (RMSE) of 0.93% in the test set, and has strong accuracy and application value.