Process parameter optimization model for tablet compression based on random forest and proximal policy optimization...

In tablet manufacturing, optimizing process parameters is critical for improving tablet quality. Aiming at the problem of quality fluctuations in tablet compression caused by limited precision in automatic adjustment of process parameters and continued reliance on manual intervention, this study proposed a process parameter optimization model (RF-PPO) that integrated Random Forest (RF) and Proximal Policy Optimization (PPO) algorithms, using a Lactobacillus tablet compression line from a pharmaceutical company as the research case. The model first employed feature selection to identify critical process parameters (CPPs) for constructing the RF quality prediction model. Subsequently, the PPO algorithm was applied to iteratively optimize parameters by minimizing the deviation between the RF-predicted and nominal tablet weight (0.8 g), and the optimized parameters were then fed into the RF model to predict the optimized tablet weight. Finally, the effectiveness of the RF-PPO model was evaluated using two approaches: offline validation based on historical production data and a tablet-weight closed-loop control simulation in MATLAB/Simulink. The results indicated that the RF prediction model achieved high accuracy, with root mean square error (RMSE) values of 0.0035, 0.0117, and 0.0118 in the training set, test set, and 5-fold cross-validation, respectively, and all corresponding coefficients of determination (R²) exceeding 0.92. Meanwhile, the offline validation and the Simulink simulation results consistently show that the RF-PPO model can output optimized process parameters within a millisecond-level response time (0.0030 s), reducing manual intervention and maintaining tablet weight within the premium product range, closely approaching the nominal weight (0.8 g), providing a promising solution for intelligent pharmaceutical manufacturing.