A COMPARATIVE STUDY OF LARGE LANGUAGE MODEL FOR CONCENTRATION PREDICTION OF OIL SLUDGE WITH NON-STATIONAL HEAT TRANSFER

Abstract

As data accumulates and computational power increase, the performance of large language models (LLMs) have been significantly improved, which promoted it entered a stage with rapid development in various research field. In order to explore the application capability of LLMs in complex physical problem, we took oil sludge as the research object to predicted the concentration based on the temperature at the corresponding location, using the dataset with dynamic viscosity uf equal 2.5 and 5.0 for training and testing. We selected six LLMs for experiments, and found that four of them had hallucination problems which was the outputs were inconsistent with the actual program. Then, we built a random forest (RF) and compared it to the RF model predicted by LLM in five-fold cross validation, to verify whether the parameters were potential optimized or not. Results shows that the artificial model was superior to the LLMs solution in terms of generalization and accuracy, whose RMSE and R2 are 0.02158, 0.9690.