AI-driven refinery optimisation

A successful refinery must overcome numerous hurdles to meet these dual targets. Foremost is producing clean products while minimising energy consumption, meaning that any digitalisation project must be viewed through the lens of environmental responsibility and emissions reduction. This can sometimes call for a delicate balancing act between commercial competitiveness and environmental responsibility.

The energy sector faces increasing challenges as complexity grows, pushing plants to embrace digital transformation and sustainability. Traditional optimisation methods must be revised, necessitating a shift to AI-driven solutions. However, with numerous options available, it is crucial to prioritise actionable strategies that enhance operational excellence.

Cutting through the noise, leaders must focus on practical, AI-enabled solutions that address the industry's evolving demands and drive meaningful progress. Success hinges on adopting innovative approaches that effectively meet these complex challenges as the landscape changes.

The market faces constant variations in demand, but most long-term predictions point towards reducing demand for fossil liquid fuels. Despite the EU's pledge to ban the registration of new combustion engine cars by 2035, there will still be decades before the existing fleet is replaced. With that in mind, the role of an operator is crucial in improving the performance of refinery assets that can effectively adapt to the changing landscape of the energy transition.

When initiating strategies to improve environmental performance, there are two routes to travel. The first involves using renewables as an energy source or a feedstock. The second pathway is improving efficiency to produce more output from less energy input. To successfully deliver on these two strands, a multipronged approach is crucial. This approach opens a world of innovation, and fundamental to this is making the best use of the myriads of data available from the plant.

The growing role of AI

The ability to leverage advanced data analytics and machine learning algorithms delivers many benefits for the industry. The strategic adoption of AI can fine-tune operational processes, proactively manage risks, and drive the industry towards higher levels of sustainability. However, it is important to remember that not all AI is created equal. The buzz around AI should not sway operators; instead, they should carefully consider what they want to achieve through its deployment in their operations.

The rapid rise of AI is reshaping the landscape for refinery operators. AI applications are not just altering but revolutionising refinery operations, offering a multitude of benefits. However, it is essential to note that the deployment of AI has its challenges.

The imperative of true AI

For decades, plant optimisation has relied on linear and first principles models. While recent hybrid approaches integrating AI with these models have gained momentum, they remain constrained by the underlying assumptions of linear and first principles methodologies. This results in assumption-driven rather than data-driven collaborations. Since different organisational domains (operations, controls, planning and economics) typically have their own set of assumptions, hybrid AI approaches are limited in their ability to appropriate reality across the different domains. This limitation hinders the ability to achieve fully optimised, end-to-end collaboration across all domains.

In contrast, a true AI approach is based first and foremost on data, allowing domain experts to guide learning using only the assumptions and first principles that are widely agreed upon. This shifts the focus to evidence, while integrating insights from diverse domains into a unified model accessible to the entire organisation. A true AI model is typically based on a deep and large neural network trained on months to years of historical data.

Unlike advanced process control (APC) and hybrid models, which struggle with nonlinearity and dynamic changes, true AI models are designed for such complexities. They effectively manage the interdependencies of refinery processes, optimising operations in real time from hours to minutes. This approach not only maximises yields and reduces emissions, but also enhances human collaboration, driving plants toward operational excellence.

Harnessing AI for process control

To harness the power of true AI, Imubit has developed the Optimizing BrainTM that seamlessly integrates data from all operational aspects, driving a level of refinery optimisation previously unattainable through traditional methods. With its ability to adapt to dynamic, real-time conditions and proactively address operational inefficiencies, this AI platform offers European refineries a pathway to not only meet regulatory demands but also thrive in a future where sustainability and profitability are inseparable.

This solution mimics the human brain that can integrate sensory data to create a unified perception of reality, guided by natural laws like gravity and momentum. It learns through experience and adapts to gaps and noise in historical data. This concept can be applied to industrial plants by integrating data from various areas, such as planning, economics, engineering, and operations, to build a comprehensive model of plant operations.

This approach optimises processes and enhances collaboration across teams. Transparency is essential, allowing engineers and operators to understand and trust the system's decisions, leading to more effective and informed decision-making.

Consider a plant with units like a hydrocracker (HCU), fluid catalytic cracker (FCC), gas oil hydrotreater (GOHT), and cat gasoline hydrotreater (CGHT). These units face constant disruption from fluctuating feed compositions, degrading catalyst activity, and varying ambient conditions. To further complicate matters, external factors such as volatile product prices, inventory limitations, and downstream capacity constraints add layers of complexity to plant operations.

A small, cross-functional team of engineers and operators must navigate these challenges daily, making high-stakes decisions – such as determining vacuum gas oil (VGO) cuts and adjusting hydrocracker conversions in response to unseen shifts in feed composition. They must also manage the intricate feed and conversion processes of the FCC, all while trying to maintain efficiency and profitability in an unpredictable environment.

The company’s technology replaces disconnected, individual models with its Foundation Process Model™, a holistic, plant-wide optimisation strategy built on advanced deep and large neural networks. This model integrates real-time data from every aspect of the refinery, providing unparalleled clarity into plant operations. Rather than relying on fragmented insights or human judgement alone, the system processes factual data, optimising decisions across the entire plant in a consistent and collaborative way.

By harnessing AI for process control, this approach directly supports the sustainability goals of modern refineries. The Optimizing Brain™ drives profitability by improving operational efficiency, and reduces energy consumption and emissions. This is particularly important as refineries increasingly face regulatory pressures to reduce their carbon footprint. With AI-enabled precision, plants can optimise energy-intensive processes like hydrocracking and fluid catalytic cracking, achieving up to 30% energy efficiency gains. These improvements are not just beneficial for the bottom line – they play a crucial role in helping refineries meet stringent environmental regulations and transition toward a more sustainable future.

By enabling plants to process more challenging, lower-carbon feedstocks, such as biofuels and alternative energies, the technology empowers operators to align operational excellence with sustainability goals. This ensures that refineries can thrive in a future where the balance between profitability and environmental responsibility is no longer a choice but a necessity.

Training the models

When training a model on historical data, Imubit integrates first principles, such as mass balance. This ensures the model aligns with mass balance principles even with imperfect historical data, enabling us to develop highly dynamic nonlinear models.

A developed reinforcement learning model that simulates the plant's behaviour, defining process constraints and allowing the model to run through thousands of simulated years. This rigorous training is about optimising the refinery’s core processes, driving efficiency and precision.

Once trained and verified by engineers, the model is deployed in closed-loop control, effectively managing an actual plant's noisy conditions. This approach enhances plant performance and fosters better collaboration and synergy among the various teams involved.

Before transitioning to closed-loop control, users must grasp how the technology makes decisions. The company streamlines this understanding with its industrial AI platform that allows engineers, operators, and economists to explore variable relationships using real-site data. The approach is straightforward: there is no complex control theory or abstract equations, just a clear focus on process, constraints, objectives, and economic factors.

This transparency empowers users to actively engage with the model, analyse its behaviour, adjust, and observe how it responds. The result is a deep, intuitive understanding of the optimal production moves, driving more innovative, informed decision-making.

As AI advances within the energy sector, the Optimizing Brain leverages historical data and experience to refine plant optimisation strategies, turning ambitious goals into achievable outcomes. Companies like Marathon Petroleum Corp., Citgo, and Chevron are at the forefront of this digital transformation.

By-lined by Dennis Rohe, Business Consulting Leader, Imubit.

Read the article online at: https://www.hydrocarbonengineering.com/special-reports/01112024/ai-driven-refinery-optimisation/