Energy Cost Optimization in Paper Tissue Industry Using Predictive Modeling: A Machine Learning Based Management Approach

Abstract

However, due to the high energy requirements of the production process, along with the instability of the power purchase agreement prices, managing energy costs becomes complicated, receiving little attention in scientific research. In this regard, the current study aims to forecast and minimize the expenses on electricity and natural gas at a tissue paper mill in Peru using machine learning algorithms. It fills five gaps: no integrated energy cost model, no model based on process variables, poor integration with ISO 50001, an excessive number of energy intensity-oriented studies, and a lack of empirical evidence for Latin America. The total number of data points was 445 daily operational records (January 2023–June 2024). Two-stage structure: the first stage is to estimate electricity and natural gas consumption (MWh/day) through Ordinary Least Squares (OLS), Ridge, Lasso, and Random Forest (RF) regressions based on selected operational factors; the second stage is to estimate the specific cost (USD/ton) by calculating the sum of specific energy consumption (SEC) and PPA rates. The significance of the factors was evaluated through Lasso and RF regressions, verified via one-at-a-time sensitivity analysis. The baseline has energy consumption at 3,935 MWh/ton (15.1% higher compared to the BAT/BREF benchmark) and cost at $154.37/ton. Energy costs account for 62.9% of the total cost although electricity only consumes 30.2%, being relatively expensive. In this case, the PPA price is the key variable in the cost equation (±19.37 USD/ton), being significantly higher compared to the operation ones. The EA efficiency ratio is the critical operating factor. Moreover, a reduction of 38% in the boiler CD3 efficiency was discovered, affecting energy consumption by +5.3%, or approximately $500,000/year. The Standard Operating Conditions scenario saves approximately USD 227,000/year. The optimized scenario, which includes renegotiation of the PPA price, is worth approximately USD 2.6 million/year. In other words, the effect of contract management on energy costs is much higher than the effect of the operational optimization. In general, the model used works as a dynamic baseline according to ISO 50001 standards, and it can be used for benchmarking purposes.