صفورا حسینی

  roduction and consumption of energy carriers from fossil fuels (coal, oil and gas) is the biggest source of global warming and greenhouse gas emissions. Therefore, there is a need to develop and implement strategies to achieve the goal of the Paris Agreement and limit global warming. In this context, energy-intensive industries, such as oil and gas refineries, play a vital role. The first step in the energy management of a refinery is to identify the potential for saving energy. In this way, in the first step, a comprehensive framework for data processing and development of prediction models with statistical methods and machine learning was carried out to predict energy consumption in the gasoline unit of Shazand Oil Refining Company. The control variables of the gasoline making process were evaluated and the most important parameters affecting energy consumption in this unit were identified. Historical data was collected daily from the beginning of ۱۳۹۷ to the end of ۱۴۰۱ and to determine the minimum number of variables required for modeling, a multi-stage processing plan including multi-linear analysis and selection of important features based on the model was used. In the modeling stage, to achieve the best and most accurate model, five common algorithms reported in the articles have been reviewed and compared. Among all the algorithms, gradient boosting regression model and random forest showed better performance than other methods and have the lowest validation error. The results show that these models, with a determination coefficient of ۰.۹۱ in the test data set, predicted the energy consumption in the gasoline unit and other variables very accurately. In addition, the evaluation of the test data showed that by adjusting the variables in accordance with the modeling results, there is a ۴.۴۳% energy saving potential, equivalent to a reduction in the consumption of ۵۶۴۸.۸۳ kilo normal cubic meters of refinery gas fuel in one year. In this research, a new framework for operational control based on machine learning with the aim of improving energy performance is presented. In the operational control strategy, the operational modes of the gasoline production unit were first categorized with the K-means clustering algorithm, then the lowest amount of energy consumption in each operational mode was selected as the operational control reference of the proce   In the next step, controlled parameters and output variables were determined using predictive models. The analysis of the test data proved that by adjusting the control parameters of the mentioned unit in accordance with the best historical performances, there is a ۷% energy saving potential, equivalent to reducing the consumption of ۸۸۹۴.۴۷ cubic meters of refinery gas fuel in one year. Keywords: energy saving and performance management, Naphtha hydrogen refining unit, energy baseline ?