Optimized LSTM Model Using Simulated Annealing for Autoignition Temperature (AIT) Prediction as a Hazard Indicator

Abstract

Purpose: Autoignition Temperature (AIT) is the lowest temperature at which a substance will spontaneously ignite in normal air without any external ignition source. AIT is an important safety parameter in industries that handles flammable materials. Measuring AIT with conventional method is unfortunately slow, costly, and dangerous. As an alternative, an AIT prediction model can be developed using in silico approaches, specifically based on machine learning.

Methods: One of the methods that can be used is Long Short-Term Memory (LSTM) since it is good at modeling the complex relationships that is involved, but unfortunately it is difficult to tune manually due to their numerous hyperparameters. Therefore, an automated strategy can be used to find the best hyperparameters for the architecture. This study aims to develop an AIT prediction model as a hazard indicator using an LSTM model optimized with Simulated Annealing (SA).

Result: The experiment showed that the SA-LSTM model which uses a cooling schedule of Delta T = 0.7 outperformed the unoptimized baseline model.

Novelty: The optimization raised the R² on test data from 0.5682 to 0.5939 while also lowering the RMSE from 74.35 K to 72.10 K and the MAPE from 9.29% to 8.87%. These results confirmed that optimizing LSTM with SA gave a more robust tool for hazard indicator.