A method and system for multi-objective energy efficiency optimization of a residue hydrocracking unit driven by automatic machine learning for low carbon operation

Abstract

The present application belongs to the technical field of petroleum chemical process optimization and energy saving and emission reduction, and relates to a kind of automatic machine learning driven low carbon operation multi-objective energy efficiency optimization method and system of residue hydrocracking unit, and solves the problems of insufficient consideration of carbon emissions, model dependence on experience and low solving efficiency of existing optimization methods.The present application builds a multi-objective operation optimization model that integrates carbon emission accounting, minimizes economic benefits, energy consumption and carbon emissions as co-optimization objectives, and uses an automatic machine learning framework based on TPOT to build a high-precision surrogate model to replace the complex and rigorous mechanism model, and then uses a particle swarm optimization algorithm for efficient global solution.The method and system can automatically output the optimal operating parameter combination while ensuring the safe and stable operation of the device, achieve economic benefit improvement per unit time, public engineering consumption reduction and carbon dioxide emission reduction, and provide a low-carbon and intelligent operation optimization solution for complex industrial processes such as residue hydrocracking.

Description

Technical Field

[0001] This invention belongs to the field of petrochemical process optimization and energy conservation and emission reduction technology. Specifically, it relates to an automatic machine learning-driven multi-objective energy efficiency optimization method and system for low-carbon operation of residue hydrocracking units, which is applicable to the low-carbon, high-efficiency and intelligent operation of heavy oil processing units in the oil refining industry. Background Technology

[0002] Residue hydrocracking units are crucial process units in oil refining, responsible for the deep processing of heavy oil products. They are significant for improving heavy oil resource conversion levels, optimizing refinery product structure, and enhancing overall processing flexibility. These units primarily process high-boiling-point heavy feedstocks such as residue oil. The feedstock composition is complex, and its properties vary widely. Operation involves multiple stages of reaction and the synergistic generation of various products, resulting in a long process flow, complex structure, and numerous operational variables. Furthermore, the intertwined and highly coupled material and energy flows within the unit make its operation highly sensitive to changes in operating conditions. Under long-term continuous operation and high-load production conditions, residue hydrocracking units typically consume significant amounts of energy and generate substantial carbon dioxide emissions, making them a typical example of high energy consumption and emission levels among oil refining units. With increasing global attention to climate change and the gradual advancement of low-carbon development goals, emission control in energy-intensive industrial processes is becoming increasingly prominent. As a vital component of the energy conversion and supply system, the carbon emission levels of the oil refining industry directly impact the achievement of regional and industry-wide emission reduction targets. Residue hydrocracking units account for a significant proportion of a refinery's overall energy consumption and emissions. The selection of their operating modes, parameters, and load levels directly impacts their energy efficiency and carbon emission intensity. Therefore, optimizing operating conditions to improve energy efficiency and reduce carbon emissions, while ensuring safe, stable, and continuous operation, is of practical significance for improving refinery environmental performance, enhancing enterprises' ability to meet carbon emission constraints, and promoting sustainable development in the industry. Existing research and engineering practice show that most optimization efforts for residue hydrocracking units primarily focus on improving economic efficiency. This typically involves adjusting operating conditions such as reaction temperature, pressure, feed schemes, and unit load to increase target product yield or reduce unit raw material processing costs. While these economic-indicator-centric optimization strategies can improve operational profitability to some extent, they often neglect the impact on energy efficiency and carbon emission levels in practical applications. Under complex operating conditions, simply pursuing economic improvements may lead to increased energy consumption or deviations from reasonable operating ranges, thus limiting the release of the unit's energy-saving potential and hindering long-term stable operation under low-carbon constraints. From a comprehensive energy and environmental perspective, carbon emissions from residue hydrocracking units are characterized by diverse sources and complex structures. Emission levels are not only affected by the process reaction but also closely related to fuel consumption and the use of utilities such as steam and electricity. The superposition of multiple emission sources makes the overall carbon emission characteristics of the unit more complex. Furthermore, due to frequent changes in operating conditions and significant nonlinear coupling relationships between various operating variables, the patterns of energy consumption and carbon emission changes under different operating strategies are difficult to accurately predict using empirical methods. In existing operational optimization methods, carbon emissions are usually not modeled as an independent or synergistic target, but rather as a byproduct of energy consumption changes. This weakens the relevance and practical effectiveness of optimization schemes in low-carbon operation to some extent. With the continuous development of industrial automation and information technology, mechanistic models and data-driven methods are gradually being applied to the operation analysis and optimization research of residue hydrocracking units. Mechanistic models can describe the operating behavior of the unit from the perspective of reaction kinetics and material conversion mechanisms, providing a theoretical basis for operation analysis. However, due to the complexity of the reaction network and the large number of model parameters, when participating in optimization calculations under multiple operating variables and constraints, they often face problems of large computational scale and low solution efficiency, making it difficult to meet the needs of rapid decision-making and frequent adjustments in actual production. At the same time, although traditional data-driven models have certain advantages in characterizing complex nonlinear relationships, their model construction process is heavily dependent on human experience and lacks adaptive capabilities in feature selection, model structure design, and parameter adjustment. When the operating conditions of the unit change or the load fluctuates, the model's prediction accuracy and stability are easily affected. In addition, the low-carbon operation problem of residue hydrocracking units essentially involves the synergy and trade-off between multiple objectives such as economic benefits, energy utilization efficiency, and carbon emission levels. These types of problems typically require multi-objective optimization methods. However, in practical engineering applications, multi-objective optimization often suffers from problems such as large solution set size, complex objective trade-offs, and low computational efficiency, making it difficult to promptly translate optimization results into executable operational plans. This problem is particularly prominent in production environments that require frequent adjustments to operating conditions, thus hindering the widespread application of related optimization methods in industrial settings. In summary, existing optimization methods for residual oil hydrocracking units still have shortcomings in areas such as explicit quantification and modeling of carbon emissions, synergistic optimization of economic benefits, energy utilization, and carbon emission reduction, and model adaptability. These limitations make it difficult to fully meet the needs of the refining industry's development towards low-carbon and intelligent operations. Therefore, it is necessary to further explore optimization methods and systems that can comprehensively consider economic benefits, energy efficiency, and carbon emission impacts at the unit operation level, and possess high modeling efficiency and optimization solution capabilities. This will provide reliable technical support for achieving low-carbon, high-efficiency, and sustainable operation of residual oil hydrocracking units. Summary of the Invention

[0003] (a) Purpose of the invention This invention aims to address the problems of insufficient carbon emission consideration, low optimization efficiency, and reliance on human experience and limited generalization ability in the operation optimization of existing residue hydrocracking units. It provides a method and system for low-carbon operation and energy efficiency optimization of residue hydrocracking units based on automated machine learning. This method constructs an operation optimization model that considers carbon emission factors, synergistically optimizing economic benefits, energy utilization, and carbon emission impacts. It utilizes automated machine learning technology to achieve adaptive construction and efficient solution of the surrogate model, thereby reducing computational complexity and improving optimization efficiency and stability. This invention can achieve a balance between improved operational economy and reduced carbon emissions while ensuring the safe and stable operation of the unit, providing effective technical support for the low-carbon and intelligent operation of residue hydrocracking units and similar industrial processes. (II) Technical Solution To achieve the above objectives, the technical solution of this invention is summarized as described in the claims, and mainly includes the following core contents: Construction of a multi-objective low-carbon operation optimization model: Three core optimization objectives are established: maximizing profit per unit time, minimizing energy consumption, and minimizing carbon emissions. The profit objective integrates product value with raw material and utility costs; the energy consumption objective quantifies the consumption of steam, electricity, and fuel; and the carbon emission objective calculates process emissions and indirect emissions using the emission factor method. The model also integrates key process constraints such as reaction temperature, pressure, space velocity, and product yield. Automated Machine Learning Proxy Modeling: To address the issue of slow computation of mechanistic models, a high-precision proxy model is constructed using the TPOT-based automated machine learning framework. TPOT automatically completes the entire optimization process from raw data to the final model, including feature preprocessing, construction, and selection, as well as selecting the optimal model structure and hyperparameters from various regression algorithms (such as random forest and gradient boosting), establishing a rapid mapping relationship from operational variables (such as reaction temperature, hydrogen-to-oil ratio, and feed rate) to optimization objectives (profit, energy consumption, and carbon emissions). Optimization based on intelligent algorithms: Carbon emissions are cost-based and integrated into economic objectives, transforming a multi-objective problem into a single-objective problem. A trained surrogate model is used as the objective function evaluator, and a particle swarm optimization algorithm is employed to perform an efficient global search within the constraint space of the operational variables, quickly obtaining the optimal or near-optimal combination of operational parameters. System Integration and Online Application: The above methods are encapsulated into an integrated system with functions such as data access, model self-learning, optimization calculation, and result push. The system can automatically trigger optimization processes periodically or based on changes in operating conditions, providing operators with clear parameter adjustment suggestions. (III) Beneficial Effects Compared with the prior art, the present invention has the following beneficial effects: Achieving multi-objective synergistic optimization: For the first time, carbon emission targets were explicitly and systematically quantified and synergistically optimized in the operation optimization of residue hydrocracking units, achieving a unified improvement in economic, energy consumption, and environmental benefits. Improved optimization efficiency and adaptability: The proxy model built through automatic machine learning reduces the optimization calculation time from hours to minutes for the mechanistic model while ensuring accuracy; and the model can automatically adapt to changes in working conditions, reducing human intervention. Enhanced engineering practicality: The optimization results are output in the form of specific operational parameter adjustment suggestions, with clear boundaries, making it easy to implement in engineering. Moreover, the optimization process is strictly carried out within safety constraints, ensuring operational safety. IV. Detailed Implementation Example 1: Method Implementation Process This embodiment details the implementation steps of the method of the present invention. Nearly one year of plant operation data was collected from the factory database, including reactor bed temperature, pressure, circulating hydrogen flow rate, feed properties, product yield, and steam, electricity, and fuel consumption, totaling approximately 200 variables. After data cleaning, outlier removal, missing value imputation, and standardization, a high-quality dataset was obtained. Based on process knowledge and correlation analysis, 15 key operating variables were selected as optimization variables, including inlet temperature of the reactor, temperature rise of each bed, system pressure, hydrogen-to-oil ratio, and feed flow rate. A multi-objective optimization model is constructed: Objective 1 is to maximize hourly profit; Objective 2 is to minimize total energy consumption (standard fuel oil equivalent); Objective 3 is to minimize total carbon dioxide emissions. Constraints include: upper limit of key temperature, range of liquid level fluctuation, lower limit of product diesel yield, etc. Carbon emission accounting: Using the methods recommended in the "Guidelines for Carbon Emission Accounting of Petrochemical Enterprises," carbon emissions from the reaction process, carbon emissions from fuel combustion in the heating furnace, and indirect carbon emissions from purchased steam and electricity are calculated separately and summed to obtain the total emissions. A carbon tax price of X yuan / ton CO2 is set, and the total emission cost is included in Target 1. Proxy model construction: The preprocessed data was divided into training and test sets in a 7:3 ratio. Using the TPOT framework, the optimization objective was set to minimize the mean squared error of prediction, and automatic search was run. Ultimately, TPOT recommended and trained a pipeline model consisting of a feature selector, multinomial feature construction, and gradient boosting regressor. On the test set, the predicted R² for profit, energy consumption, and emissions were all greater than 0.92. Optimization Solution: The economic objective integrating carbon costs is used as the fitness function of the particle swarm optimization algorithm, employing a surrogate model for rapid prediction. The particle swarm size is set to 50, and the algorithm iterates 200 times. Under the condition of satisfying all process constraints, the algorithm converges to a set of optimal operating parameters. 7. Application of Results: The optimized parameters (e.g., suggesting a reduction of the reactor inlet temperature by Y°C and an adjustment of the hydrogen-to-oil ratio to Z) should be submitted to the production department for evaluation and implementation. Figure 2 As shown, the parameter adjustments were all within the allowable range, and the implementation proceeded smoothly. Example 2: Effect Verification and Analysis This invention was continuously applied for three months in a 4.0 Mt / a residue hydrocracking unit. Analysis Figure 1 It can be seen that after adopting the method of the present invention, the process emissions of the device are reduced by about 0.5% compared with those before optimization, the utility emissions are reduced by about 0.3%, while the exhaust emissions remain basically stable. The results show that the method of the present invention mainly achieves carbon emission reduction by improving process emissions and utility emissions, and achieves targeted optimization of the carbon emission structure of the device without introducing additional exhaust emission risks. Figure 2 The trends of changes in the main operating parameters of the unit before and after optimization were compared. It can be seen that the adjustment range of each operating variable is within a reasonable engineering range and conforms to the process operation law of the residue hydrocracking unit, indicating that the optimization results obtained by the present invention have good engineering rationality and practical feasibility.

Claims

1. An automated machine learning-driven multi-objective energy efficiency optimization method for low-carbon operation of a residue hydrocracking unit, characterized in that, Includes the following steps: 1) Data acquisition and preprocessing: Collect historical and online operating data of the residue hydrocracking unit, and perform data cleaning, anomaly removal, missing value processing and standardization; 2) Screening of key variables and establishment of mapping relationships: Analyze and screen key operational variables and performance evaluation indicators that have a significant impact on the operating performance of the device, and establish the mapping relationship between operational variables and performance indicators; 3) Construction of multi-objective optimization model: Construct a multi-objective operation optimization model with the objectives of maximizing economic benefits, minimizing energy consumption and minimizing carbon emissions, and including process safety and operational constraints; 4) Carbon emission accounting and objective function integration: The carbon emissions generated by the plant process and utilities are uniformly accounted for, and the carbon emissions are cost-based based on the carbon tax mechanism and integrated into the economic objective function, thus transforming the multi-objective optimization problem into an equivalent single-objective optimization problem. 5) Construction of automated machine learning surrogate model: Using the TPOT-based automated machine learning framework, feature engineering, model selection and hyperparameter optimization are completed automatically, and a surrogate model that can accurately map the relationship between the operational variables and the optimization target is trained. 6) Intelligent optimization solution: Based on the constructed surrogate model, the particle swarm optimization algorithm is used to perform a global search in the operation variable space to solve for the optimal combination of operation parameters that satisfy the constraints. 7) Optimization result output and application: Output the optimal combination of operating parameters as operation optimization suggestions to guide the actual operation adjustment of the device. The method as described in claim 1 is characterized in that, in step 3), the multi-objective optimization model has the following objectives: economic benefit objective is calculated by subtracting raw material cost and utility cost from product revenue; energy consumption objective is quantified by steam and electricity consumption; and carbon emission objective is obtained by accounting for direct emissions from the reaction process and indirect emissions from the consumption of steam, electricity, and other utilities.

2. The method as described in claim 1, characterized in that, The construction process of the automatic machine learning surrogate model described in step 5) includes using the TPOT framework to automatically search and evaluate the complete machine learning process, including data preprocessing, feature construction, model algorithm and hyperparameters, and using the model prediction accuracy and generalization ability as evaluation criteria to determine the optimal surrogate model.

3. The method as described in claim 1, characterized in that, The carbon emission accounting described in step 4) specifically includes calculating process emissions, fuel combustion emissions, and indirect emissions corresponding to purchased steam and electricity based on the equipment material balance, energy balance, and emission factors, and converting the total carbon emissions into economic costs.

4. The method as described in claim 1, characterized in that, The search process of the particle swarm optimization algorithm described in step 6) is based on the objective function value predicted by the surrogate model. Under the process constraints of reaction temperature, pressure, space velocity, product distribution and upper and lower limits of equipment load, it iteratively searches for the optimal combination of operating variables that makes the equivalent single objective function optimal.

5. An automated machine learning-driven multi-objective energy efficiency optimization system for low-carbon operation of a residue hydrocracking unit, characterized in that, include: (1) Data acquisition and processing module: used to acquire the device's operating data in real time and complete data preprocessing; (2) Model Management Module: Used to store and manage mechanistic model knowledge, automated machine learning agent models, and multi-objective optimization models; (3) Automatic machine learning engine: Based on the TPOT framework, it automatically trains and updates agent models using historical and online data; (4) Optimization and solution module: integrates particle swarm optimization algorithm and calls surrogate model to perform fast optimization and solution; (5) Carbon emission accounting module: Calculates the carbon emission level and cost of the device in real time according to preset rules and emission factors; (6) Human-computer interaction and decision support module: used to display optimization results, key indicator comparisons, and provide operation guidance and suggestions.

6. The system as described in claim 5, characterized in that, The automatic machine learning engine can trigger automatic retraining and updating of the agent model to adapt to the new operating state when the device's operating conditions change significantly.

7. The system as described in claim 5, characterized in that, The optimization solution module supports performing optimization calculations at fixed intervals or manually triggered, and pushes the obtained optimal combination of operating parameters to the production control system or operator interface in real time.

Images