Based on the combustion process, CC technologies can be classified into oxy-fuel combustion, pre-combustion capture, and post-combustion capture. Based on the operating modes, they can be divided into slipstream mode, storage mode, and integrated flexible operation mode.

Considering the commercial and technological maturity, the post-combustion capture technology with integrated flexible operation mode is adopted by FCCGT, which is realized by installing a flue gas bypass system and a solvent storage tank on the gas turbine (Song et al., 2024). The operation process of FCCGT is shown in Figure 1.

FCCGT controls the flue gas bypass system to feed a certain proportion of the flue gas generated by the gas turbine into the absorption tower. In this tower, the flue gas reacts with an ethanolamine (MEA) solution to obtain a CO₂-rich solution, which is stored in the rich solution storage tank. The rich solution pump is then adjusted to control the rich solution fed to the regeneration tower, where the rich solution is heated with extracted steam to cause counter-reaction to separate CO₂ from the MEA. The separated CO₂ is partially used as feedstock of a power-to-gas facility to produce methane for the natural gas network, while the remaining CO₂ is compressed for storage. The regenerated lean MEA solution is pumped back into the lean solution storage tank and recirculated to the absorption tower for reuse. This integrated flexible operation allows FCCGT to efficiently capture and utilize the CO₂ generated by the gas turbine.

Compared to gas turbines equipped with traditional CC device, the unique features of FCCGT are:

1. Flexible and controllable flue gas diversion ratio: By adjusting the flue gas bypass system, the proportion of flue gas entering the CC device can be controlled, allowing a balance between operating costs and carbon emissions.

Due to these distinctive features, a targeted model needs to be established for FCCGT.

According to Figure 1, the model reflecting the power output and real-time CC amount of FCCGT is expressed as follows: P t FCC = P e , t FCC + χ M t TC + P t Y M t TC = M t UC + E t S − E t A M t UC = θ max CC γ FCC δ t P t FCC M t AC = γ FCC P t FCC M t GC = M t AC − M t UC δ min ≤ δ t ≤ δ max , P min FCC ≤ P t FCC ≤ P max FCC 0 ≤ M t TC ≤ η θ max CC γ FCC P max FCC (1) where t is the index of time. P t FCC is the total power output of FCCGT. P e , t FCC is the net power delivered to the distribution network by FCCGT. P t Y is the CC power consumption of the CC device. χ is the consumed power for capturing unit mass of CO₂ (MWh/t). M t TC is the total mass of the captured CO₂, M t UC is the mass of CO₂ entering the CC device after flue gas diversion (t). E t S is the mass of CO₂ supplied from the rich solution storage tank (t). E t A is the mass of CO₂ absorbed by the rich solution storage tank (t). δ t is the flue gas diversion ratio. θ max CC is the maximum CC level. M t AC is the total mass of CO₂ produced by FCCGT (t). γ FCC is the CO₂ emission intensity of FCCGT (t/MWh). M t GC is the mass of CO₂ emitted into the atmosphere from FCCGT after flue gas diversion (t). P max FCC and P min FCC are the upper and lower power limits of FCCGT.

According to (Equation 1), by controlling the inflow and outflow of the rich solution storage tank, the amount of CO₂ flowing into the absorption tower M t UC can be different from that processed by regeneration tower M t AC at the same time, which realizes the decoupling of CC amount and power consumption mentioned in the Introduction.

Since the CO₂ is stored in the solution in the state of MEA compound, the variables related to the mass of CO₂ in (Equation 1) should be converted the volume of MEA solution, which is then used to model the volumetric constraints of the solution storage tanks. The conversion between CO₂ mass and MEA volume is shown as: V t in = E t A M MEA ψ S t MEA ρ t MEA M CO 2 V t out = E t S M MEA ψ S t MEA ρ t MEA M CO 2 (2) where V t in and V t out are the MEA volume flowing into/out of the rich solution storage tank (m³). M MEA and M CO 2 are the molar masses of MEA and CO₂ (g/mol). ψ is the mass of CO₂ that can be absorbed by 1 mol of MEA (mol/mol). S t MEA is the solution concentration. ρ t MEA is the solution density (t/m³).

Then the volumetric constraints of the solution storage tanks can be modeled as follows: V t F = V t − 1 F − V t out + V t in V t W = V t − 1 W + V t out − V t in V 0 F = V T F 0 ≤ V t F ≤ V L , 0 ≤ V t W ≤ V L (3) where V t FL and V t WL are the MEA volumes in the rich and lean solution storage tanks at time t, respectively. V t − 1 FL and V t − 1 FL are the MEA volumes in the rich and lean solution storage tanks at time t-1, respectively. V 0 FL and V T FL are the MEA volumes in the rich solution storage tank at time 0 and T, respectively. T is the total time slots in a dispatching cycle. V L is the volume of the solution storage tanks (m³).

Besides, the solution storage tanks should not flow in and out MEA simultaneously, and the inflow and outflow have upper limits, which are expressed as follows. 0 ≤ E t S ≤ η θ max CC γ FCC P max FCC U t W 0 ≤ E t A ≤ η θ max CC γ FCC P max FCC U t F 0 ≤ E t A ≤ M t TC U t W + U t F = 1 (4) where η is the maximum operational efficiency of the regenerator tower and the compressor. U t W and U t F are the 0-1 indicator variables reflecting the MEA volume changes in the lean and rich solution storage tanks, respectively.

The bilinear term δ t P t FCC in (Equation 1) leads to a non-convex FCCGT model. To convexify this bilinear term, McCormick envelopes are used and shown as below (Deng et al., 2021). ω t = δ t P t FCC (5) ω t ≥ δ min P t FCC + δ t P t , ⁡ min FCC − δ min P t , ⁡ min FCC ω t ≥ δ max P t FCC + δ t P t , ⁡ max FCC − δ max P t , ⁡ max FCC ω t ≤ δ max P t FCC + δ t P t , ⁡ min FCC − δ max P t , ⁡ min FCC ω t ≤ δ t P t , ⁡ max FCC + δ min P t FCC − δ min P t , ⁡ max FCC (6) where ω _t is an auxiliary variable used to replace δ t P t FCC in (Equation 1). Therefore, By combining the replaced (Equations 1, 6), the proposed FCCGT model is convexified.

Based on the production principles and operating modes of the MLEAF, its DR model for day-ahead optimal dispatch is constructed, which consists of the following constraints.

(1) Constraints of MLEAF regulatable capacity

During a complete production cycle, the regulation times for each MLEAF should not exceed the predetermined upper limits to avoid adverse impact on the product purity. The relevant constraints are shown as follows: y m , t − z m , t = s u m , t − s u m , t − 1 y m , t + z m , t ≤ 1 ∑ y m , t + ∑ z m , t ≤ M (8) where y m , t and z m , t are the 0-1 variables indicating that MLEAF changes to the upward and downward regulation states, respectively. M is the maximum regulation times of the MLEAF.

(3) The constraints of the MLEAF production output

To ensure that the MLEAF production output still meets requirements after participating in the DR project, the following constraints are constructed: ∑ t = 1 T λ P m , t ⋅ P m , t ≥ O m (9) where O m is the target of MLEAF production output. λ P m , t is the piecewise linear function of the MLEAF yield to its power, which is shown as λ P m , t = λ 1 ,   P m base − P max d ≤ P m , t ≤ P m r , d λ 2 ,   P m r , d ≤ P m , t ≤ P m r , u λ 3 , P m r , u ≤ P m , t ≤ P m base + P max u (10) where λ ₁、λ ₂ and λ ₃ are the MLEAF yield under the power decrease state, nominal state, and power increase state. P m r , d and P m r , u are the lower and upper limits of the rated state of MLEAF.

The MLEAF production output constraints given by Equations 9, 10 are conditional constraints, which cannot be solved by common commercial solvers directly. To address this problem, auxiliary variables are first used to perform piecewise linearization of (Equation 10): s m , t d + s m , t r + s m , t u = 1 λ m , t = λ 1 s m , t d + λ 2 s m , t r + λ 3 s m , t u P m , t ≤ P m r , d s m , t d + P m r , u s m , t r + P m base + P max u s m , t u P m , t ≥ P m base − P max d s m , t d + P m r , d s m , t r + P m r , u s m , t u (11) where λ _m,t is the output yield of the m-th MLEAF at time t. s m , t d , s m , t r and s m , t u are 0–1 variables to indicate the three intervals of P _m,t divided in (Equations 10).

Based on (Equations 9, 11) can be rewritten as: ∑ t λ m , t P m , t ≥ O m (12)

Similarly with (Equation 1), the bilinear term in (Equation 12) can also be convexified using McCormick envelopes: ∑ t o m , t ≥ O m (13) where o _m,t is an auxiliary variable used to replace the bilinear term λ _m,t P _m,t in (Equation 12).

Although λ _m,t is discrete, it still has minimum and maximum values, which are λ ₁ and λ ₃, respectively. The minimum and maximum values P _m,t of are P m base − P max d and P m base − P max d , respectively. Using the maximum and minimum values of λ _m,t and P _m,t , the linearized limits of o _m,t is constructed as: o m , t ≥ λ 1 P m , t + λ m , t P m base − P max d − λ 1 P m base − P max d o m , t ≥ λ 3 P m , t + λ m , t P m base + P max u − λ 3 P m base + P max u o m , t ≤ λ 3 P m , t + λ m , t P m base − P max d − λ 3 P m base − P max d o m , t ≤ λ m , t P m base + P max u + λ 1 P m , t − λ 1 P m base + P max u (14)

Finally, the MLEAF production output constraints are composed of (Equations 11, 13, 14).

The scenario generation method in (Zhang et al., 2021) is adopted here, whose procedures are given below.

Step 1: Organize the historical forecast error data of RES and load power as a matrix.

The objective function of the proposed dispatch model is to minimize the total operation cost of the distribution network.

(1) The start-up and shut-down cost of the FCCGT

This paper considers the implementation of a carbon tax policy in the distribution network and calculates the carbon tax cost associated with CO₂ emissions as follows: C carb = c CO 2 ∑ k = 1 N sce ∑ t = 1 T p k ∑ i = 1 N FCC M t , k , i GC + λ e P e , t , k buy (19) where c CO 2 is the carbon tax price. λ e is the average carbon emission factor corresponding to purchased electricity.

(6) The penalty cost for wind and solar power curtailment

Based on (Equations 15–21), the total cost of the distribution network is calculated by Equation 22 as follows: min   C FCC + C R + C Buy + C Mg + C carb + C Curt + C LD (22)

(1) Constraints of FCCGT

The relevant constraints have been given by Equations 1–6.

(2) Constraints of MLEAF

The relevant constraints have been given by Equations 7–14.

(3) Power balance constraints

A modified IEEE 33-bus standard system is designed as the test distribution system, whose network structure is shown in Figure 2. This designed system includes three identical FCCGTs, i.e., GT1-GT3, which are connected to buses 22, 25, and 33, respectively. The technical parameters of GT1-GT3 are shown in Table 1. The three FCCGTs have the same type of CC devices, with the parameters listed in Table 2.

The wind farm is connected to Bus 18, with its forecasted power output shown in Figure 3. The penalty price for curtailing wind power is 100 $/MW·h. Apart from MLEAF, the forecasted power of all the other loads in the distribution network is also shown in Figure 3. The ratio of load at each bus to the total load is the same with that in the original IEEE 33-bus standard system, which can be found in (Yang et al., 2021) and not elaborated here.

The MLEAF is connected to Bus 18. When not participating in DR project, the MLEAF daily power curve is a horizontal line with its value equal to the rated power. The relevant parameters are shown in Table 3.

The electricity price for purchasing power from the superior power grid adopts time-of-use pricing, where the peak, flat and valley periods are divided according to the load curve shape shown in Figure 3. The corresponding prices for each period are shown in Table 4. The average carbon emission factor for purchased electricity is 0.65 t/MW h.

Based on the parameters above, three different scenarios are designed for comparative analysis to verify that the incorporation of FCCGT and DR of MLEAF can improve the economic efficiency, low-carbon attribute and flexibility of the distribution network. These scenarios are as follows.

Scenario 1: GT1-GT3 are equipped with CC devices without solution storage tanks and with a fixed flue gas diversion ratio. The MLEAF does not participate in the DR project.

Scenario 2: GT1-GT3 are FCCGTs with adjustable flue gas diversion ratios. The MLEAF does not participate in the DR project.

Scenario 3: GT1-GT3 are FCCGTs with adjustable flue gas diversion ratios. The MLEAF takes part in the DR project. This scenario corresponds to the flexible low-carbon optimal dispatch model proposed in Section 4.

All three scenarios are modeled with the YALMIP toolbox and solved by the CPLEX solver on the MATLAB platform.

The dispatch costs are shown in Table 5.

According to Table 5, it can be found that.

1) In Scenario 2, the carbon emission is reduced by 18 t compared to Scenario 1. The wind power curtailment penalty is decreased by 4,340 $. The total cost is decreased by 8,000 $. This indicates that the FCCGT has higher flexibility than gas turbines equipped with CC devices without solution storage tanks, which improves CC efficiency and promotes the utilization of renewable energy.

In summary, the effectiveness of the proposed flexible low-carbon optimal dispatch model is validated by comparing the total costs, carbon emissions and wind curtailment of the three scenarios.

The dispatch schemes of the three scenarios are shown in Figures 4–6.

According to Figures 4–7, the outputs of all the power sources during the peak load periods are compared between the three scenarios. It can be found that the proposed model can optimize the output of FCCGTs and the wind farm, so the distribution network can effectively interact with the superior power grid to achieve the “peak shaving and valley filling” effect. More specifically, during peak periods, the external power purchase is decreased by utilizing MLEAF to reduce its own load. During off-peak periods, the consumption of wind power is increased through reasonable dispatching, which also reduces the external power purchase. Since the carbon emission factor of purchased electricity is higher than that of gas turbines and wind power, the proposed model results in a decrease in the total daily carbon emissions, which proves its low-carbon property.

According to Figures 8, 9, the fixed gas diversion ratio in Scenario 1 results in a nearly constant high CC power during peak periods, which further leads to a shortage in the feed-in power of the FCCGTs and exacerbates the power generation deficit. In Scenarios 2 and 3, CO₂ that cannot be processed during peak periods is stored in the solution tanks and processed later in the off-peak periods. In this way, the CC power consumption during peak periods is reduced to zero, which allows FCCGTs to provide more feed-in power during peak periods and more CO₂ processing load during off-peak periods. Such flexibility promotes RES power consumption and achieves peak-shaving and valley-filling effects.

According to Figure 10, during the peak period from 18:00 to 20:00, the gas diversion ratio in Scenario 3 is higher than that in Scenario 2. This is because the DR of the MLEAF alleviates the power output pressure of FCCGTs during peak periods. More specifically, MLEAF actively responds to the distribution network’s peak shaving instructions and reduces its load demand, so the required power output and the corresponding CO₂ generated by FCCGTs are both lowered. Therefore, FCCGTs can provide more CC power to process the smaller amount of carbon emission, which results in the higher flue gas diversion ratio in Scenario 3.

According to Figure 11, the solution volumes in the rich and lean solution tanks vary with time to react with the operation condition of the distribution network. During the peak periods, the MEA solution volume in the rich solution tank increases, which indicates that CC power is reduced in these periods and CO₂ is stored temporarily for later processing. During the valley periods, MEA flows out of the rich solution tank to release CO₂ and then flows into the lean solution tank, which indicates that the CC power increases to process the CO₂ temporarily stored in peak periods.

Through the coordinated operation of the rich and lean solution tanks, the CC amount and the CC power consumption are decoupled, contributing to the flexible low-carbon economic dispatch of the distribution network.

From the comparison of Figures 11, 12, it can be observed that compared to Scenario 2, Scenario 3 additionally considers the flexibility brought by the DR of MLEAF. Therefore, the CC devices can operate more efficiently across different load periods, which better balances the operating efficiency and energy consumption of the FCCGTs. With the flexibility from both the generation and load sides, the overall efficiency and stability of the distribution network is improved, which means lower operation costs and smaller environmental impact.