Prediction of Ladle Furnace Refining Endpoint Temperature Based on Particle Swarm Optimization Algorithm and Long Short-Term Memory Neural Network

Abstract

The temperature control of molten steel is crucial in ladle furnace (LF) refining. In this paper, the factors influencing endpoint temperature were obtained by analyzing the LF refining process, and high-quality data sets were obtained by combining data cleaning and processing. Then, a black box model based on particle swarm optimization (PSO) and long short-term memory (LSTM) neural network machine learning algorithm was proposed to predict the endpoint temperature of LF, avoiding the inadequacy of manual parameter adjustment. The constructed PSO–LSTM model yielded the following results: The coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE) were 0.924, 4.064, and 5.109, and the prediction hit rates within the ranges of [−3,3], [−5,5], [−8,8], and [−10,10] were 43.16%, 71.12%, 88.15%, and 94.83%, respectively. The accuracy of multiple linear regression (MLR), backpropagation (BP) neural network, genetic algorithm back propagation (GA-BP) neural network, particle swarm optimization back propagation (PSO-BP) neural network, LSTM neural network, and the present model was compared from different dimensions, and the results demonstrated that the consistency and correlation between the PSO–LSTM model prediction values and the measured values were higher, the model prediction accuracy was superior, and the model effectively predicted molten steel temperature in the LF furnace.