A chlor-alkali cell house coupled to a cogeneration system draws every extra megawatt from the grid

A membrane chlor-alkali cell house modeled from published electrochemistry and coupled through its electricity demand to a modeled steam and cogeneration system.

CheAI Research Inc. | Research article | July 14, 2026

Keywords: chlor-alkali electrolysis; membrane cell; cogeneration; process integration; model validation

Abstract

Chlor-alkali electrolysis is among the largest electrical loads in chemical manufacturing and is often co-located with steam and cogeneration systems. We modeled a membrane cell house producing 40 tonnes of chlorine per day from published electrochemistry and coupled it, through its electricity demand, to a modeled multi-pressure steam and cogeneration system. The cell operates at 5.034 V, of which 56 percent of the DC power is rejected as heat; the cathode activation term, extrapolated from laboratory data on plain nickel, accounts for 42 percent of the voltage. At constant prices the coupled optimum imports the cell's entire 6.99 MW AC demand from the grid and leaves every steam and turbine set-point unchanged: at reference prices, additional onsite generation would cost more than grid import. All 59 sensitivity cases of the cell model fall inside the DOE energy-intensity band, so that comparison cannot discriminate between model variants. The specified twenty-start optimization found three distinct local optima; the result was settled by a vertex certificate based on interval arithmetic rather than by additional starts.

1. Introduction

Membrane chlor-alkali cells convert brine into chlorine, caustic soda and hydrogen at a cost dominated by electricity [1]. At integrated chemical sites the cell house shares a utility system with steam users and onsite generation, and coupling the two could in principle shift electricity supply between the grid and the turbines. Whether it does depends on the relative cost of onsite generation and grid power at the margin.

This study builds a cell model from published electrochemistry, examines how its plausibility can and cannot be assessed against published energy intensities, and determines how the coupled optimum differs from the steam system optimized alone.

2. Methods

2.1 Cell model

A membrane cell is solved at 90 degrees C and 6,000 A/m² with a chlorine current efficiency of 0.95. The cell voltage is the sum of a reversible voltage, corrected from the standard value of 2.186 V for temperature and for brine and caustic activities; anode and cathode activation overpotentials; and the ohmic drop across the membrane. Each term is also computed by a separately written route for comparison. Activation overpotentials follow forward-branch Tafel laws with Arrhenius temperature extrapolation: the cathode from polarization data on plain nickel [2, 3] and the anode from temperature-dependent kinetics on RuO2(110) model electrodes [4]. The membrane drop is the area-specific resistance, thickness divided by conductivity, with conductivity extrapolated in temperature from measurements on a commercial perfluorinated membrane [2]. These laboratory data were obtained at lower temperatures and current densities than the modeled operating point and were not tuned toward industrial cell voltages.

2.2 Sensitivity analysis and benchmark

Thermodynamic data, activities, gas pressures, cathode and anode kinetics and membrane conductivity were varied one family at a time, giving 59 cases. DC electricity per pound of chlorine was compared with the DOE bandwidth range [5] as a sanity check, not a fitting target.

2.3 Coupling and optimization

A rectifier at 95.5 percent efficiency supplies the cell from the site's electrical system; only electricity crosses between the cell house and the utility system. The coupled dispatch was optimized hour by hour over a 168-hour horizon with IDAES, Pyomo and Ipopt [6, 7, 8], from 20 random starts, with an acceptance criterion of 0.5 percent on the spread of results.

3. Results

3.1 Cell voltage and heat rejection

The cell voltage of 5.034 V comprises a reversible voltage of 2.365 V, anode activation of 0.179 V, cathode activation of 2.138 V and a membrane ohmic drop of 0.352 V (Figure 1). The cathode term, 42 percent of the total, is extrapolated from low-current-density data on plain nickel. The whole-cell reaction power, including the oxygen side reaction implied by the 0.95 current efficiency, corresponds to an effective zero-heat voltage of 2.230 V at 1,326 kA (the main chlorine reaction alone has a thermoneutral voltage of 2.270 V). Electrical work above this voltage is rejected as heat: 3.72 MW from 6.68 MW of DC power. The energy balance closes to a relative 1.4 × 10^-16.

Stacked bar of cell voltage: reversible 2.36 V, anode 0.18 V, cathode 2.14 V, membrane 0.35 V, with the whole-cell zero-heat voltage marked near 2.23 V — **Figure 1. Decomposition of the cell voltage.** The four voltage terms. Voltage above the dashed whole-cell zero-heat line leaves the cell as heat. The cathode term is extrapolated from a laboratory plain-nickel electrode.

3.2 The DOE band does not discriminate between model variants

The central case uses 6,200 Btu of DC electricity per pound of chlorine, inside the DOE range of 5,116 to 6,578 Btu/lb. Across the sensitivity cases the intensity ranges from 6,058 to 6,370 Btu/lb, and all 59 cases lie inside the band (Figure 2). At this current efficiency the band admits any DC cell voltage from 4.15 to 5.34 V. The comparison is also not like for like: the DOE intensities describe chlorine production as a whole, whereas the model counts DC cell electricity only, excluding rectifier, auxiliary, brine-treatment and evaporation energy.

Ticks for each sensitivity case, grouped by family, all falling inside a shaded DOE band from 5,116 to 6,578 Btu per pound — **Figure 2. Energy intensity of each sensitivity case and the DOE range.** Each tick is one sensitivity case; the shaded region is the DOE bandwidth range.

3.3 Coupled optimum

Table 1. Cost-optimal dispatch of the steam system with and without the cell house.

Quantity	Steam system alone	Coupled with the cell house
Operating cost, USD/h	2,806.94	3,331.28
Onsite generation, MW	13.67	13.67
Grid import, MW	4.33	11.32
Grid emissions, kg CO2e/h	880	2,301
Scope 1, kg CO2/h	27,745	27,745

Onsite generation changes by 3.6 × 10^-7 MW (Table 1). Import rises by 6.99 MW, matching the cell's AC demand to 3.6 × 10^-7 MW, and the additional cost of 524.34 USD/h is that import at 0.075 USD/kWh. Grid emissions rise by 1,422 kg CO2e per hour and Scope 1 is unchanged. The steam system could physically generate more: at a lower gas price the same equipment generates 22.37 MW by raising boiler output and condensing more steam. At reference prices that additional condensing generation costs more than grid import, so the optimum buys the cell's electricity instead. Solving the two systems separately agrees with the coupled cost to within 2.3 × 10^-10 USD over the horizon. Under constant prices and grid factor every hour is identical, with 1 distinct hourly cost, so the horizon reduces to one representative hour.

3.4 Multiple local optima and their resolution

The specified multistart did not meet its criterion. Of the 20 starts, 15 reached 3,331.28 USD/h, 3 stopped at 3,382.69 and 2 at 3,395.44: three strict local optima on different boiler configurations, a spread of 1.93 percent (Figure 3). Tightening solver tolerances returned the same three points. The worst is 1.93 percent more expensive than the best.

Bar chart of how many of twenty starts ended at each of three local optima: fifteen at the best, three at the second, two at the third — **Figure 3. Distribution of multistart results across three local optima.** Number of starts ending at each local optimum. Labels give the hourly cost of each optimum.

The result was settled by proof rather than additional starts. Because both turbines exhaust at the same state and do the same specific work, cost depends only on their combined flow, and along the feasible boundary it is first decreasing, then strictly concave, then increasing. A certificate establishes each premise with outward-rounded interval arithmetic, including negative controls that must fail, and concludes that each configuration's minimum lies at one of its two vertices. Vertex enumeration then gives a spread of 2.1 × 10^-16. The certificate holds only at the inputs where it is evaluated; a related price-sensitivity analysis finds it failing at three of six price extremes.

4. Discussion

At reference prices the modeled site supplies a new electrical load most cheaply from the grid, so coupling adds cost without changing the steam-side dispatch. Coupling would matter if prices or emission factors varied by hour, or if the cell's rejected heat were recovered into the steam system; neither is included here.

The DOE comparison illustrates a general point about validation. A plausibility check that every reasonable model variant passes cannot distinguish a good model from a poor one; it should be reported as a consistency check, not as validation.

5. Limitations

The site, cell house and capacity are modeled, not measured.
Electrode kinetics and membrane conductivity are extrapolated from laboratory sources at other temperatures, concentrations and current densities; film, bubble, contact and electrode resistances are omitted, so the voltage is a lower bound with respect to those losses.
Prices are constant, so the multi-hour horizon is one hour repeated.
Coupling is through electricity only: no heat recovery from the cell and no steam to evaporation.
The heat balance uses standard-state species enthalpies and omits the concentration and mixing enthalpies of the actual brine and caustic streams, so the 3.72 MW of rejected heat is a modeled estimate with unquantified uncertainty, and its exact closure does not validate it.
Grid emissions include the emissions of grid losses, which the GHG Protocol assigns to Scope 3 rather than Scope 2 [9].
Hourly solutions are the best found by local optimization; vertex enumeration is exact only where the certificate holds.

6. Conclusions

Coupled at constant prices, the modeled cell house's 6.99 MW demand is met entirely by grid import, because additional onsite generation costs more than grid power at reference prices. All 59 cell-model variants fall inside the DOE energy band, so that benchmark cannot validate the model.

References

O'Brien TF, Bommaraju TV, Hine F. Handbook of Chlor-Alkali Technology. Springer, New York, 2005.
Dias AC. Chlor-Alkali Membrane Cell Process: Study and Characterization. Doctoral dissertation, University of Porto, Portugal, 2010.
Franceschini EA, Lacconi GI, Corti HR. Kinetics of the hydrogen evolution on nickel in alkaline solution: new insight from rotating disk electrode and impedance spectroscopy analysis. Electrochimica Acta, 159:210-218, 2015.
Sohrabnejad-Eskan I, Goryachev A, Exner KS, Kibler LA, Hensen EJM, Hofmann JP, Over H. Temperature-dependent kinetic studies of the chlorine evolution reaction over RuO2(110) model electrodes. ACS Catalysis, 7(4):2403-2411, 2017.
U.S. Department of Energy. Bandwidth Study on Energy Use and Potential Energy Saving Opportunities in U.S. Chemical Manufacturing. Washington, DC, 2015.
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.
Bynum ML, Hackebeil GA, Hart WE, et al. Pyomo: Optimization Modeling in Python, 3rd edition. Springer, Cham, 2021.
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.
World Resources Institute and World Business Council for Sustainable Development. GHG Protocol Scope 2 Guidance. Washington, DC, 2015.