May 2022

Digital Technology

Save energy and reduce CO2 emissions with closed-loop optimization of utilities networks

Oil and gas, petrochemical and chemical companies face the difficult challenge of maximizing profitability while achieving aggressive decarbonization objectives set for 2030 and beyond.

Lodolo, S., Aspen Technology

Oil and gas, petrochemical and chemical companies face the difficult challenge of maximizing profitability while achieving aggressive decarbonization objectives set for 2030 and beyond. The transition to a digitally powered business model is generally accepted across industries as companies adopt new processes, pursue new feedstocks and expand automation. Advanced process control (APC) and optimization technologies have long been used to help automate and optimize operations. More recent innovations have tackled strong process nonlinearities and multi-unit dynamic optimization of large process envelopes to better meet evolving, volatile market demands. Now, artificial intelligence (AI) and deep learning have further enhanced APC and optimization capabilities to help upstream and downstream companies achieve energy and carbon dioxide (CO2) reduction goals without significant capital investments. When applied to fuel gas, steam and hydrogen utility networks, APC and dynamic optimization technologies can have a tremendous impact in meeting an organization’s operational and sustainability targets. This article explores why this is so important and how to achieve beneficial results with a utility network.

Transitioning to a new business model

Oil and gas, petrochemical and chemical companies are focused on tackling the dual challenge to meet the growing demand for resources from a growing population with increasing standards of living, while also addressing sustainability goals. To do so, these companies are transitioning to a new sustainable business model. This typically involves rebalancing their production portfolio and investment areas, focusing on new feedstocks, new products, new energy sources and new process technologies. A digitalization journey framework leveraging new technologies like AI and deep learning can help with this transition. This journey is different from one organization to the next, with each being at a different digital maturity level. However, sustainability is always at the forefront of the transition and the new business model.

In fact, most companies have established aggressive sustainability targets for 2030, and particularly for CO2 emissions reduction, with the European Union (EU) leading the way, setting a goal of 55% reduction from 1990 levels with full decarbonization to be achieved by 2050.

The global pandemic added more challenges and squeezed corporate margins, accelerating this transition process. Last year we witnessed high volatility in supply/demand and in energy costs, which has recently tripled in some regions. Extremely high CO2 costs are prevalent in areas where a CO2 trading and permit system is in place, such as the EU’s Emissions Trading System (ETS).

A well-performing, complex 200,000-bpd European refinery that emits 1.5 MMtpy of CO2 may need to buy around 200,000 tpy of CO2 permits now worth around $18 MM/yr, given current CO2 prices of $90/t—carbon pricing is based on cost in November/December 2021. Free allowances are forecast to be reduced each year and permit costs are likely to increase even more. This creates a sense of urgency to reduce energy and CO2 costs to stay competitive, achieve sustainability goals or simply survive, as CO2 costs may even move margins from positive to negative.

Owners and operators are planning significant capital expenditure (CAPEX) investments to achieve decarbonization targets set for 2030 and beyond. Examples of key investment areas include carbon capture, electrification, green/blue hydrogen, cogeneration and fuels-to-chemicals. These CAPEX investments are strategic, mid-to-long term and will require significant upfront investment with a high CO2 abatement cost to achieve the targets. It is expected that these investments could help reduce carbon emissions by approximately 30%–35%.

An additional 20%–25% in carbon reduction is possible with more tactical investments to increase process efficiency. These would generally be less costly, low hanging fruit improvements that would not require executive level approval. There would be a negative CO2 abatement cost since CO2 reduction is typically accompanied by energy consumption reduction, making this secondary approach an attractive complement to longer-term, more strategic CAPEX investments.

Process efficiency improvements

APC and optimization technologies have been utilized in the process industries for several decades to enable improved process efficiency. Focusing first on large complex units and then secondary units, and more recently with large multi-unit envelopes in scope, like the entire middle distillates circuit in a refinery or a full ethylene plant in petrochemicals, optimized dynamically minute by minute. Optimization objectives have generally been in relation to increased capacity or yields, greater quality control or reduced giveaway, and—to a lesser extent—energy savings. Priorities for most providers have shifted dramatically during the past 2 yr, with energy reduction, production cost reduction and increased agility becoming the top focus areas.

APC and optimization technologies are the most cost-effective option to improve process efficiency and support the decarbonization process. It is worth noting that with carbon prices continuing to increase, the value from CO2 reduction becomes as significant a benefit as energy reduction savings.

There are a multitude of efficiency improvement opportunities available to refiners. These include:

  • Increasing furnace efficiency
  • Minimizing distillation columns’ pressure against actual unit constraints
  • Minimizing load and optimizing the purification level of distillation columns
  • Minimizing quality giveaways to avoid excess energy consumption
  • Minimizing circulation, reprocessing and recycling
  • Making the right product recovery/energy consumption tradeoffs
  • Maximizing heat/cold recovery
  • Integrating multiple units and managing heat balances
  • Balancing demand/supply to avoid losses (e.g., hydrogen, steam)
  • Minimizing downgrades (e.g., steam, fuels, hydrogen, products)
  • Maximizing equipment efficiency
  • Optimizing utilities networks: fuels, hydrogen and steam.

APC is now capable of solving difficult nonlinear problems. In addition, it enables an adaptive workflow to maximize model accuracy, lets users make changes to economic objectives on the fly, and embeds AI and deep learning capabilities to leverage a wealth of historical data to obtain models and solve complex problems—for instance, inferring highly nonlinear qualities that depend on the type of run, feedstock qualities or grade to be produced.

With dynamic, self-adapting optimization capabilities that automatically align models and the process, a true overall optimum can be achieved across a large process scope.

These capabilities and technology enhancements enable users to solve complex problems, such as utilities networks, which were previously difficult to solve.

The rest of this article focuses on optimization strategies, which represent tremendous energy- and CO2-saving opportunities in the three primary site networks: fuel gas, hydrogen and steam.

Fuel gas network optimization

For many facilities, fuel gas network optimization can be a significant issue that must be addressed. The following are some of the most common challenges:

  • Fuel gas network pressure (volume) and quality interact heavily
  • Volume balance models are integrating processes as pressure builds up as a ramp
  • Quality models are nonlinear, and gains may even change from positive to negative or vice versa
  • Various makeup gases are typically available, and their quality may change over time
  • Additional makeup gas reduces the effect on overall network quality due to nonlinearities from mixing
  • Inherent instability due to disturbances continuously entering the system.

The quality of gas streams entering the network varies continuously. If enrichment gas with a high heating value is added to control the pressure, the heating value of the fuel gas increases. The furnaces will then have to cut their consumption, which will upset the volume balance, and the enrichment gases will also be cut back.

This behavior keeps repeating, leading to a self-propagating cycle impacting the performance of fuel gas users across the whole site. Fuel gas instability can result in scenarios such as units cutting back load, increasing load to blowdown networks and gas recovery compressors, and flaring events.

This behavior, illustrated in FIG. 1, can cause excessive energy consumption and CO2 emissions. With integrating processes like volume balance models, which exhibit strong nonlinearity, model gains can flip their sign (i.e., go from positive to negative or vice versa) as the quality effect of the enrichment gas depends on the current network’s calorific value.

FIG. 1. Fuel gas network pressure and quality interaction.
FIG. 1. Fuel gas network pressure and quality interaction.

Multiple enrichment gases may be available to control pressure, and each gas can increase or decrease the network calorific value. The impact may change over time depending on current calorific values of both gas and the network. Adding more makeup gas changes the effect as the gap between the two qualities changes.

FIG. 2 illustrates the potential effects of these nonlinearities. For instance, the same changes to the same enrichment gas may have very different effects on the network’s calorific value. Addressing these challenges is possible by adopting a nonlinear adaptive process control solutiona to solve highly nonlinear and interactive problems and reach desired outcomes. The adaptive process control solutiona lets users easily revise economic objectives to continuously minimize the overall cost of the streams entering the fuel gas network.

FIG. 2. Fuel gas networks’ nonlinearities.
FIG. 2. Fuel gas networks’ nonlinearities.

The following are some other key benefits of using such a nonlinear adaptive process control solution:

  • Simultaneously control both fuel gas pressure and quality (calorific value), stabilizing operations throughout the site
  • Minimize/avoid flaring
  • Minimize blowdown discharge and subsequently gas recovery compressor power consumption
  • Improve furnace efficiency through consistent and stable firing, which enables further reduction in excess air
  • Tackle every furnace throughout the refinery to ensure proper control over combustion and oxygen concentration (e.g., for a furnace with 80% efficiency, a 5% reduction in oxygen from 3.5% to 3.33% can result in roughly a 0.3% increase in efficiency)
  • Control emissions in closed loop within the APC applications (not simply monitoring emissions) wherever it makes sense to do so
  • Reduce the overall cost of fuels entering the fuel gas network or being directly fired, whenever flexibility to use different types of fuel exists, and improve overall profitability
  • Significantly reduce CO2 emissions, considering that around 2.78 t of CO2 are emitted for each ton of fuel gas.

The 200,000-bpd refinery in our example can reduce fuel gas flaring and blowdown by more than 50%, improving overall combustion efficiency and process unit performance, reducing CO2 emissions by 19,000 tpy and achieving savings of $4.6 MM/yr.

Typical available control handles include:

  • Natural gas entry points
  • LPG vaporizers
  • Fuel oil/fuel gas balancing in double firing boilers/furnaces
  • Streams manipulated to support fuel gas network pressure (e.g., propane, LPG)
  • Fuel gas exchange streams to/from various headers and sections.

Pre-requirements are usually minimal, and a calorific value analyzer may be required only if accurate and sufficiently robust inferentials cannot be built based on available density analyzers and lab analysis. Nonlinear inferencing capabilities of proprietary softwareb can be leveraged to build such inferentials.

FIGS. 3 and 4 show fuel gas header pressure and calorific value before and after deploying the APC solution.

FIG. 3. Fuel gas pressure before and after deploying the APC solution.
FIG. 3. Fuel gas pressure before and after deploying the APC solution.
FIG. 4. Fuel gas calorific value before and after deploying the APC solution.
FIG. 4. Fuel gas calorific value before and after deploying the APC solution.

Hydrogen network optimization

Hydrogen network optimization is another significant opportunity to improve overall process efficiency and support decarbonization objectives.

Most refineries are either hydrogen or desulphurization capacity limited (some are both). In all cases, hydrogen must be produced and used optimally. A refiner’s planning/scheduling department typically provides guidance on hydrogen producers’/consumers’ load and what to do in case of excess or the lack of hydrogen, but coordination remains infrequent, manual and sub-optimal.

Due to the high cost of natural gas, the cost of producing hydrogen is expensive (presently topping $3,000/t in some sites for pure hydrogen, depending on how hydrogen is produced). Moreover, producing 1 t of hydrogen in a steam reformer results in approximately 10 t of CO2, having significant impact on site emissions.

It is imperative to avoid wasting hydrogen to fuel gas—every kilogram counts since a reduction of just 200 kg/hr of hydrogen results in overall savings of $2 MM/yr in energy and CO2 benefits.

Besides the economic loss, hydrogen discharges to the fuel gas network disturb overall operations given that the hydrogen calorific value is 2 times–3 times that of a typical fuel gas. There is certainly the need to dynamically coordinate hydrogen producing and consuming units, which are often under different operators’ responsibilities. Some challenges and peculiarities must be considered, including:

  • The hydrogen network can be very complex, with multiple headers at different pressures and purities, up to 4–6 suppliers (reformers, steam reformers, partial oxidation units, gasifiers, etc.) and up to 10–15 consumers (hydrocrackers, hydrotreaters, desulphurization units, tail gas treaters, isomerization units, etc.) and compressors, pressure swing adsorption (PSA)/membranes, or other purification units.
  • Hydrogen headers usually have different pressures and qualities and are interconnected through valves or compressors.
  • Disturbances in one of the suppliers or consumers can impact the entire system with variations in hydrogen quality and pressure.
  • Optimum dynamic operational changes. Constraints may show up in different units and scenarios may change significantly based on costs and the availability of feeds, conversion/desulphurization units’ margins and market conditions.
  • There are nonlinearities to be addressed and tradeoffs to be managed like purities, catalyst lifecycles, losses and recoveries based on required hydrogen purities and associated consumptions.
  • There is sometimes a lack of ownership of coordination of multiple producers and consumers.

Hydrogen consumption is forecast to increase in the short- to mid-term due to multiple reasons, such as tighter desulphurization targets being enforced and the need to co-process more renewable feedstocks in hydrotreaters due to local regulations. For example, a 3% vegetable oil coprocessing could require 20% more hydrogen.

Hydrogen networks interact with both gasoline and middle distillates’ circuits. Within a gasoline circuit, tradeoffs may exist between hydrogen production and gasoline production requirements. Within a middle distillates circuit, the tradeoffs are between unit loads, feed blending (e.g., vegetable oils, cracked feeds or vacuum gasoils) and hydrogen availability.

If such tradeoffs are identified, the most practical way to address them is to integrate the hydrogen network optimizer with the production circuit optimizer(s).

With a scope encompassing production circuits (targeting reduction of quality giveaways), additional benefits in terms of energy and CO2 savings can be achieved, as any giveaway results in additional energy consumption.

A hydrogen network can be effectively optimized by leveraging dynamic optimization technologyc in combination with an adaptive process control solution.a This addresses nonlinearities and coordinates multiple units dynamically by relying on an ever-green adaptive APC layer to reach desired objectives.

Such an application typically enables users to:

  • Balance production of and demand for hydrogen, ensuring it is optimally distributed
  • Optimize steam reformer operations and conversion
  • Minimize hydrogen discharge to fuel gas network and blowdown
  • Reduce power consumption in gas recovery compressor(s), if any
  • Minimize hydrogen production costs whenever this flexibility exists
  • Increase profitability of conversion and desulfurization units
  • Stabilize hydrogen header pressures and purities
  • Optimize hydrogen network purities
  • Significantly reduce CO2 emissions.

Assuming the 200,000-bpd European refinery discharges 500 kg/hr of hydrogen to fuel gas, it can reduce the loss by at least 50%, reduce CO2 emissions by 18,000 tpy and achieve savings of approximately $4 MM/yr. Additional benefits can be realized due to reduced hydrogen production costs and more profits coming from hydrogen users.

It is not uncommon for sites to discharge approximately 1 t/hr of hydrogen or more on average to the fuel gas network, as shown in FIG. 5.

FIG. 5. Approximately 1 tph of hydrogen discharged to fuel gas over 2 mos for a 200,000-bpd refinery.
FIG. 5. Approximately 1 tph of hydrogen discharged to fuel gas over 2 mos for a 200,000-bpd refinery.

The following are some typical available control handles:

  • Steam reformer(s) hydrogen production
  • Reformer(s) severities (when possible and not fixed on gasoline balance)
  • Bypasses between headers of different purities
  • Purge rates from consuming units
  • PSA and membrane operation
  • Hydrocracker (or another selected unit) load to balance hydrogen consumption.

Pre-requirements are usually minimal. Unlike the applications on production circuits, a hydrogen network optimizer does not require an APC layer on most of the units in the envelope. If the load (throughput) of a process unit is to be optimized, then an APC layer is required on the unit.

FIG. 6 illustrates four consecutive years of hydrogen losses to flare, two years before and after implementation of dynamic optimization technologyc, with a minimum 70% reduction in losses.

FIG. 6. Hydrogen discharged to the flare 2 yr before and after the dynamic optimization technologyc was implemented in a 220,000-bpd refinery. Source: A. Porcel and K. Kahlgren, “Dynamic optimization moves into mainstream,” ERTC, November 2018.
FIG. 6. Hydrogen discharged to the flare 2 yr before and after the dynamic optimization technologyc was implemented in a 220,000-bpd refinery. Source: A. Porcel and K. Kahlgren, “Dynamic optimization moves into mainstream,” ERTC, November 2018.

Steam network optimization

Steam production systems must satisfy the demand coming from process units at the lowest possible cost. These networks can be very complex with multiple headers, boilers, gas turbines and heat recovery steam generators (HRSGs), as well as different types of turbines and turbogenerators.

FIG. 7 shows a complex network in an ethylene plant with seven headers, two boilers, six turbines, numerous turbomachines, hundreds of users and more than 10 steam pressure controllers. Of course, optimizing this network is quite challenging. Header pressures for high-, medium- and low-pressure steam must be controlled tightly, and the overall efficiency must be maximized by properly addressing nonlinearity, which is dependent on load distribution and equipment availability.

FIG. 7. Ethylene cracker steam network.
FIG. 7. Ethylene cracker steam network.

Another issue to consider is that steam networks are continuously affected by disturbances from process units. Within the network, header pressures interact between themselves. Moreover, there are constraints on power import/export and a need to balance steam/power production. Cost variability may impact the optimal operating point dynamically, as often as daily or even more frequently.

Steam letdowns and venting must be minimized or avoided, as they can result in excessive fuels consumption and higher CO2 emissions/t of steam produced. Consumptions and related emissions strongly depend on the steam production infrastructure, efficiencies and fuels used. A reasonable estimate is that 12-t of high-pressure steam consumes 1 t of fuel gas, and each ton of fuel gas emits 2.78 t of CO2. Therefore, 0.23 t of CO2 are emitted for each ton of high-pressure steam produced.

Any amount of steam lost or downgraded to a lower pressure level without producing energy impacts the site’s energy bill and CO2 balance. It is necessary to dynamically coordinate energy producers and consumers, often under different operators’ responsibility or even operated in separate control rooms.

A steam network can be effectively optimized by leveraging the nonlinear adaptive process control solutiona, typically in combination with an optimizer that sets targets for the APC layer considering the overall utilities system. This must be considered, given that the steam and fuel networks interact, and a single overall optimum exists. Such an application typically enables users to:

  • Stabilize high-, medium- and low-pressure steam header pressures by properly managing boiler/turbine loads and turbine spills
  • Maximize boiler efficiency, while guaranteeing proper high-pressure control
  • Optimize boiler(s) combustion
  • Maximize overall efficiency by optimally distributing loads across equipment.
  • Continuously minimize direct letdowns (and vents if any) if this is the most economical solution
  • Significantly reduce CO2 emissions.

In addition, every steam user must be evaluated to ensure that energy targets are actively pushed by local APC applications. For example, columns must be operated at their minimum pressure against process constraints and not at pressure limits set by an operator. Also, live steam constraints of columns or fractionators must be active (e.g., flash point or hydrogen limit instead of with steam/flow ratios).

Every APC application objective should be revised with the energy targets in mind. The following are several examples of units to consider for APC applications:

  • Distillation columns
  • Amine regenerators
  • Sour water strippers
  • Fractionators live steam
  • Side strippers
  • Compressors/steam turbine drives.

Multiple opportunities exist in using the steam network. For example, FIG. 8 shows that more than 10 t/hr of high-pressure steam is downgraded to medium-pressure header during normal operations. Some typical available control handles include the following:

FIG. 8. Letdown flowrate and valve position.
FIG. 8. Letdown flowrate and valve position.
  • Boiler/turbine/HRSG loads
  • Turbine spills
  • Carbon monoxide (CO) boiler load (if present)
  • Other local steam production equipment and medium-/low-pressure streams to the network
  • Boiler combustion handles
  • Flexibilities in header pressure control
  • Steam exchange streams.

Significant benefits can be achieved (FIGS. 9 and 10). Both figures show a reduction in fuel gas consumption and letdown before and after optimization. Using the same refinery example, savings of $2.9 MM/yr can be achieved and up to 12,000 tpy of CO2 saved through steam network optimization.

FIG. 9. Fuel gas consumption before and after optimization. Source: G. Ruggeri, S. Chillemi, J. C. Duarte and S. Dhaliwal, “Balancing steam network production and demand improves power generation,” Aspentech OPTIMIZE 2015.
FIG. 9. Fuel gas consumption before and after optimization. Source: G. Ruggeri, S. Chillemi, J. C. Duarte and S. Dhaliwal, “Balancing steam network production and demand improves power generation,” Aspentech OPTIMIZE 2015.
FIG. 10. Letdown before and after optimization. Source: G. Ruggeri, S. Chillemi, J. C. Duarte and S. Dhaliwal, “Balancing steam network production and demand improves power generation,” Aspentech OPTIMIZE 2015.
FIG. 10. Letdown before and after optimization. Source: G. Ruggeri, S. Chillemi, J. C. Duarte and S. Dhaliwal, “Balancing steam network production and demand improves power generation,” Aspentech OPTIMIZE 2015.


In these volatile times, recent advancements in APC and optimization technologies provide strong options for oil and gas and chemical companies faced with the dual challenge of achieving energy efficiency and reducing carbon emissions. With no CAPEX investment needed, these solutions can quickly deliver significant process efficiency improvements and a negative CO2 abatement cost (i.e., a net profit).

Utilities networks in refineries, petrochemical and chemical sites present clear optimization opportunities to save energy and reduce CO2 emissions, supporting widely adopted decarbonization programs across the process industry.

Considering energy and CO2 reduction benefits for the three networks, TABLE 1 shows optimization achievements for the 200,000-bpd European refinery example used throughout this article. CO2 emissions reduction in this case represents 3.3% of the overall refinery emissions, with a value of more than $4 MM/yr at the current price for CO2 in Europe of $90/t. Note that the 49,000 tpy of CO2 in column three of TABLE 1 is approximately 25% of the CO2 permits that the 200,000-bpd European refinery example would need to buy every year. Additional benefits can be obtained from process units, primarily from capacity and yield increases. An increase in yields generally results in additional energy savings due to a reduction in costly quality giveaway.

These results can be achieved by leveraging recent APC and optimization technology enhancements that enable users to efficiently address problems that were previously difficult to solve. These enhancements include:

  • Adaptive process control to maintain ever-green models
  • Full nonlinear control
  • AI to identify models and build inferred properties to cope with different run modes and nonlinearities
  • Flexibility to change optimization strategy rapidly
  • Self-adapting dynamic optimization for multi-unit optimization of large envelopes. HP


   a  Aspen DMC3™

   b  Aspen Deep Learning IQ

   c  Aspen GDOT™

The Author

Related Articles

From the Archive



{{ error }}
{{ comment.comment.Name }} • {{ comment.timeAgo }}
{{ comment.comment.Text }}