Predicting the conductivity of purified water for use in the pharmaceutical industry
Purified water, Pharmaceutical industry, Prediciton, Time series.
Purified water plays a crucial role in the pharmaceutical industry, directly impacting the quality of processes and final products. This study proposes a machine learning model based on LSTM networks to predict the conductivity of purified water in a treatment plant intended for industrial pharmaceutical use. Real data from a pharmaceutical laboratory's water treatment plant, NUPLAM, was used. Data pre-processing included data preparation and cleaning, as well as an attribute engineering stage. The LSTM architecture was chosen because it is widely used for time series prediction tasks, thanks to its ability to memorize short- and long-term data to predict the future. The previous values are called lags, while the number of steps you want to predict in the future is called the horizon. The investigation covered identifying the best performance between different lags for a horizon of 1 and determining the best results for various combinations of lags and horizons, seeking an acceptable error. The performance evaluation used the RMSE metric. The results indicated that although a greater number of lags improves performance, this benefit is attenuated by an increase in training time. In addition, it was noted that increasing the horizon is associated with an increase in error, but that with the use of more lags it is possible to obtain good results. However, after a certain point, the increase in lags began to reduce the effectiveness of the models. This work contributes to the improvement of monitoring and control processes in pharmaceutical industries, including NUPLAM, as well as making an academic contribution to the study of time series prediction for water treatment systems.