Prediction of gasoline flash point based on deep learning

A soft sensing method based on Variational Mode Decomposition and GRO(Gold Rush Optimizer) algorithm to optimize TCN convolutional neural network was proposed for flash point prediction of gasoline on top of atmospheric pressure tower. The proposed network model is divided into three modules: the first data preprocessing module, the input data using VMD to decompose multiple sub-signals; The second module is parameter optimization module. GRO algorithm is used to optimize the hyperparameters that need to be set manually in TCN to find the optimal prediction model parameters. The third module is the prediction module, which obtains the prediction model through training and predicts several sub-signals, and finally superposes the prediction results. Through the actual data of gasoline flash point on the top of a normal pressure tower in a refinery, the validity of the model for flash point prediction is verified.