Onsite condensing power is cheaper than the grid but carries four times its carbon

CheAI Research Inc. | Research article | May 26, 2026

Keywords: steam systems; combined heat and power; dispatch optimization; Scope 2 emissions; grid losses; chemical manufacturing

Abstract

1. Introduction

A chemical site's utility system converts fuel into steam at high pressure, lets that steam down through turbines and valves to the pressures its processes need, and buys or generates the electricity the site consumes [1]. The equipment is usually fixed for years. What changes from hour to hour is the dispatch: how hard each boiler fires, how much steam passes through each turbine rather than a letdown valve, and how much electricity is imported. Because turbines convert the pressure drop between headers into power, dispatch couples the steam balance and the electricity balance, and a decision taken for one appears in the other.

Dispatch studies are difficult to compare. Published results differ in equipment, in prices, in whether emissions are counted at the stack only or across purchased electricity, and in whether the baseline they improve on is stated at all. A saving reported against an unstated baseline cannot be checked, and an emissions reduction counted at the stack alone can conceal an increase at the power plants supplying the grid.

Here everything except dispatch is held fixed. The site, its equipment limits, demands, prices and emission factors were specified before any optimization was run, and the baseline is a written heuristic recorded before the first optimization. Emissions are counted gate to gate, Scope 1 for fuel burned on site and, for imported electricity, grid emissions that include Scope 2 and the associated grid losses [2], and exported electricity is never netted against either. The study asks three questions: how much optimal dispatch saves against a documented baseline and where the saving comes from; what limits the optimum; and how far apart the cost and emissions optima are, and why.

2. Methods

2.1 Modeled site

The model, the Steam and Cogeneration Dispatch Testbed, represents a single site. The site has a high-pressure header at 6.0 MPa and 480 degrees C, a medium-pressure header at 1.8 MPa and a low-pressure header at 0.45 MPa. Two natural-gas boilers rated 120 and 80 t/h feed the high-pressure header. A backpressure turbine passes up to 60 t/h from high to medium pressure, and an extraction-condensing turbine admits up to 90 t/h, extracts to medium pressure and condenses a tail of at least 10 t/h. Turbine sections run at an isentropic efficiency of 0.78. The processes take 55 t/h of medium-pressure and 70 t/h of low-pressure steam, return a fraction 0.65 of it as condensate, and consume 18 MW of electricity. Gas costs 4.50 USD/GJ on a higher heating value (HHV) basis; grid electricity costs 0.075 USD/kWh to buy and earns 0.042 USD/kWh when exported. Grid emissions use the eGRID2023 CAMX total-output factor of 195.04 kg CO2e/MWh [3], grossed up for 4.1 percent transmission and distribution loss. They therefore comprise Scope 2 and the emissions of grid losses, which the GHG Protocol assigns to an end user's Scope 3, and are 4.3 percent higher than a strict Scope 2 figure.

These values describe a plausible mid-size chemical site consistent with published industrial steam-system practice. They are not measured data from any facility.

2.2 Baseline

The baseline is a conventional operating heuristic, recorded before any optimization: route process steam through the turbines rather than the letdown valves, run the backpressure turbine at its maximum, hold the condensing tail at its minimum, take the greatest extraction the network accepts, load both boilers to the same fraction of their rating, and balance electricity with the grid. The extraction at which the high-to-medium letdown reaches zero was found by bisection on the network model.

2.3 Optimization

The dispatch problem was formulated as an IDAES flowsheet [4] in Pyomo [5] and solved with Ipopt [6]. Steam and water properties follow IAPWS-IF97 [7, 8], implemented in a separately verified property layer. Boiler fuel follows a published part-load efficiency form in which efficiency rises with load. Because this form makes fuel use strictly concave in the split of steam between the boilers, the minimum at fixed total steam lies at an endpoint of the split, and each feasible endpoint was solved as a separate problem. Boiler commitment was enumerated over all 4 on-off assignments. Each optimization was started from 20 seeded random points, and every solver termination was checked; a non-optimal termination was treated as a failure, never as a result.

2.4 Cost-emissions frontier

The frontier minimizes cost subject to an emissions cap, set at 21 equally spaced values between the two single-objective optima. Emissions are the sum of Scope 1 CO2 and grid emissions CO2e. Each frontier point is the best of its eligible starts; failed starts are counted and not retried.

2.5 Verification

Every reported optimum was replayed through the unchanged network, turbine and boiler models and re-accounted by an independent script built on separately written IF97 equations. Mass, energy and second-law balances were evaluated at each optimum from an independently assembled stream table rather than from the solver's residuals.

3. Results

3.1 Optimal dispatch saves 114.77 USD per hour

The baseline costs 2,921.72 USD/h and the cost optimum 2,806.94 USD/h, a reduction of 3.93 percent. Because neither dispatch exports electricity, the difference separates exactly into 28.67 USD/h of boiler fuel and 86.11 USD/h of grid import (Figure 1). The import term dominates: the optimum raises the extraction-condensing turbine's inlet from 85.3 t/h to its 90 t/h limit, which increases onsite generation by 1.15 MW and reduces import from 5.47 to 4.33 MW.

The smaller fuel term comes from the boilers. The baseline runs both at the same fraction of rating, 87.2 and 58.1 t/h, at an efficiency of 0.777. The optimum runs the larger boiler at its full 120 t/h, where efficiency reaches 0.84, and reduces the smaller one to 30.0 t/h, where its efficiency is 0.715 (Figure 2). With efficiency rising in load, concentrating steam in one boiler uses less fuel than sharing it equally.

3.2 Every optimum lies on an equipment limit

No reported optimum is interior. At the cost optimum the backpressure turbine is at its 60 t/h maximum, the condensing turbine's inlet at its 90 t/h maximum, the high-to-medium letdown closed, export at zero and the larger boiler at its rating. The emissions optimum and every frontier point also lie on equipment limits, though not the same ones: at the emissions optimum the larger boiler remains at its rating and the condensing tail is at its 10 t/h minimum, while both turbines are below their maximum flows. An active-set inventory across all 599 reported results confirms that each lies on at least one limit. The optima are therefore set by equipment limits rather than by a thermodynamic balance within them; how far relaxing a particular limit would move them was not quantified.

3.3 The cost and emissions optima move the condensing turbine in opposite directions

The emissions optimum costs 2,810.67 USD/h, 3.73 USD/h more than the cost optimum, and emits 27,937 kg/h against 28,624 kg/h, a reduction of 2.4 percent. It reverses the cost optimum: the condensing tail falls to 10.0 t/h, the condensing inlet to 87.7 t/h and the backpressure flow to 57.6 t/h, and import rises (Figure 3, Table 1).

The explanation is the carbon intensity of the power each source supplies. Moving from the cost optimum to the emissions optimum gives up 1.15 MW of onsite generation, removes 921 kg of Scope 1 CO2 per hour and adds 233 kg of CO2e in grid emissions. Per megawatt-hour, averaged over the move, the onsite power given up carried 802 kg CO2, while the replacing grid power carries 203 kg CO2e at the loss-inclusive CAMX factor for delivered electricity. Condensing generation rejects most of its fuel energy in the condenser, so on a low-carbon grid it is the most carbon-intensive electricity available to the site. On cost it still prevails, because gas at 4.50 USD/GJ makes condensing power cheaper than grid power at 0.075 USD/kWh. The implied abatement cost along the frontier is 5.43 USD per tonne.

Table 1. Baseline, cost-optimal and emissions-optimal dispatch.

3.4 Numerical reliability

All 20 starts of the cost optimization agree to a relative spread of 4.9 × 10^-16. In the first formulation, which left the boiler split to the solver, the spread was 1.8 percent, above the half-percent acceptance criterion; this result led to the endpoint treatment of the boiler split described in Section 2.3. The study made 1,420 solver calls, all checked; 401 returned checked failures, almost all boiler configurations that were locally infeasible under a frontier cap, and none was counted as a result. The largest balance residual at the optimum is 1.3 × 10^-15, and the independent re-accounting agrees with the optimizer's cost to a relative 3.2 × 10^-9.

4. Discussion

Three findings generalize as method, although their magnitudes do not. First, a documented baseline changes what a saving means: against this one the advantage of optimal dispatch can be attributed term by term, and most of it is electricity rather than fuel. Second, dispatch optima of this kind tend to lie on equipment limits, so the binding limits are as informative as the set-points. Third, the direction of the emissions optimum depends on the grid. On a grid as clean as the CAMX average, onsite condensing generation is the most carbon-intensive electricity available and the emissions optimum reduces it; on a coal-heavy grid the comparison could reverse. A cogeneration study reporting Scope 1 alone would omit the emissions of the replacement grid electricity and so overstate the net benefit of reducing onsite generation.

Separate methodological studies, on simpler or different systems, examine the main simplifications: how robust a commitment decision is when demands and prices vary, what a steady-state model cannot capture about pressure transients, and how far an annual inventory factor and an hourly marginal factor diverge [9]. They inform, but do not validate, the configuration studied here.

5. Limitations

• The site is a specified scenario, not a measured plant, and is not statistically representative of US chemical manufacturing.
• Prices are point estimates, and the grid factor is an annual average rather than the hourly marginal factor that governs what an additional megawatt-hour displaces.
• Optima are the best found by checked local optimization from 20 starts and are not certified global optima.
• The frontier is resolved only at its 21 computed points, and the intensities and abatement cost are averages between the two optima, not local marginal values; the abatement cost applies to no real site or grid.
• Counting grid electricity without the loss gross-up would raise the onsite-to-grid intensity ratio from 3.9 to about 4.1.

6. Conclusions

Against a documented baseline, optimal dispatch of the modeled site saves 114.77 USD per hour, mostly through additional onsite generation. Every optimum lies on equipment limits. On the CAMX grid, condensing generation carries about 3.9 times the carbon of delivered grid power, so the cost and emissions optima pull the condensing turbine in opposite directions, and an emissions assessment of cogeneration dispatch requires accounting for purchased electricity.

References

1. Smith R. Chemical Process Design and Integration, 2nd edition. Wiley, Chichester, 2016.
2. World Resources Institute and World Business Council for Sustainable Development. GHG Protocol Scope 2 Guidance. Washington, DC, 2015.
3. U.S. Environmental Protection Agency. Emissions and Generation Resource Integrated Database, eGRID2023 Revision 2. Washington, DC.
4. Lee A, Ghouse JH, Eslick JC, et al. The IDAES process modeling framework and model library: flexibility for process simulation and optimization. Journal of Advanced Manufacturing and Processing, 3(3):e10095, 2021.
5. Bynum ML, Hackebeil GA, Hart WE, et al. Pyomo: Optimization Modeling in Python, 3rd edition. Springer, Cham, 2021.
6. Wachter A, Biegler LT. On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming. Mathematical Programming, 106(1):25-57, 2006.
7. International Association for the Properties of Water and Steam. Revised Release on the IAPWS Industrial Formulation 1997 for the Thermodynamic Properties of Water and Steam, IAPWS R7-97(2012). 2012.
8. Wagner W, Cooper JR, Dittmann A, et al. The IAPWS Industrial Formulation 1997 for the Thermodynamic Properties of Water and Steam. Journal of Engineering for Gas Turbines and Power, 122(1):150-182, 2000.
9. Siler-Evans K, Azevedo IL, Morgan MG. Marginal emissions factors for the U.S. electricity system. Environmental Science and Technology, 46(9):4742-4748, 2012.

Related studies

• Verification of the modeled steam network: property layer, fuel accounting and network balances
• Electrification and heat recovery scenarios: configuration changes on the same site
• Price sensitivity of the optimal dispatch: how the optimum moves with gas and power prices
• Robust cogeneration commitment: worst-case commitment tested out of sample
• Steam header transients: pressure excursions a steady-state model omits
• Inventory and avoided emissions: annual and hourly grid factors compared