1 Motivation
2 Demand response and the concept of gas response
2.1 Gas-to-power demand response
-
Price-based DR: The economic signals are electricity prices, and special electricity tariffs, such as real-time pricing (RTP), where the electricity price is adjusted hourly (or at shorter intervals), allowing end-users to adjust their electricity consumption to these dynamic prices.
-
Incentive-based programs: Customers are motivated to provide load flexibility via incentives or penalties. Under emergency DR, for example, grid operators instruct participants to reduce electricity consumption at very short notice during periods when the grid is jeopardized. Customers receive appropriate incentives—of a financial nature, for example—or penalties if they do not comply.
2.2 Two gas-to-power demand response programs
-
Factor-based G2P-DR: Real-time-factorization (RTF) This program follows the idea of RTP, but uses hourly (or more frequent) gas factors instead of electricity prices as economic signals. Figure 2 shows a corresponding gas factor trajectory for the gas-fired power generation shown in Fig. 1. Along real-time gas factors, companies, for example, can shift electricity-intensive processes to times when these factors are low. Such an approach is usually followed in the context of production planning under RTP when electricity costs are to be reduced; cf. Bänsch et al. (2021).
-
Incentive-based G2P-DR: Emergency G2P-DR (EDR) The idea is that participants will receive an external signal to reduce electricity consumption at very short notice during periods when more electricity is being generated from gas-fired power plants and fed into the power grid. A prespecified critical gas factor can be defined and a breaching of such critical value triggers a notice. In Fig. 3, a critical factor of 0.9 is selected, which results in 6 different time frames of different length. For example, a company can reduce its electricity consumption for production processes in response to such a notice. A justification for external signaling in production planning can be found in Scholz and Meisel (2022).
3 Application of G2P-DR in bicriteria production planning
3.1 Brief literature overview
Articles | \(\text {C}_{max}\) | EC | \(Price-/{\hbox {CO}}_{2}- factorized EC\) | Gas-factorized EC |
---|---|---|---|---|
Anghinolfi et al. (2021) | ✓ | ✓ | ✗ | ✗ |
Zhang et al. (2019) | ✓ | ✓ | ✗ | ✗ |
Zhou et al. (2021) | ✓ | ✓ | ✗ | ✗ |
Li et al. (2018) | ✓ | ✓ | ✗ | ✗ |
Mansouri et al. (2016) | ✓ | ✓ | ✗ | ✗ |
Dai et al. (2013) | ✓ | ✓ | ✗ | ✗ |
Chen et al. (2020) | ✓ | ✓ | ✗ | ✗ |
May et al. (2015) | ✓ | ✓ | ✗ | ✗ |
Lu et al. (2018) | ✓ | ✓ | ✗ | ✗ |
Dai et al. (2019) | ✓ | ✓ | ✗ | ✗ |
Wei et al. (2022) | ✓ | ✓ | ✗ | ✗ |
Cao et al. (2021) | ✓ | ✗ | ✓ | ✗ |
Schulz et al. (2019) | ✓ | ✗ | ✓ | ✗ |
Moon et al. (2013) | ✓ | ✗ | ✓ | ✗ |
Schulz et al. (2020) | ✓ | ✗ | ✓ | ✗ |
Heydar et al. (2022) | ✓ | ✗ | ✓ | ✗ |
Ho et al. (2021) | ✓ | ✗ | ✓ | ✗ |
Ding et al. (2016) | ✓ | ✗ | ✓ | ✗ |
3.2 Problem description
-
The planning horizon (18/07/2022–22/07/2022) is divided into 480 periods of one quarter hour each.
-
The two conflicting criteria, makespan and gas-fired electricity consumption, are minimized simultaneously.
-
The single-stage parallel machine environment consists of several machines whith non-identical electricity coefficients.
-
Different machine states and discrete production speed levels that allow for integer output in a period are taken into account to leverage gas-fired electricity consumption (five different speed levels in total).
-
Electricity coefficients (in kW) are randomly generated for each state of each machine: Off (0 kW); Ramp-up (drawn from \({\mathcal {U}}_{\lbrace 20,21,\ldots ,60\rbrace }\)); Standby (drawn from \({\mathcal {U}}_{\lbrace 4,5,\ldots ,8\rbrace }\)); Production (drawn from \({\mathcal {U}}_{\lbrace 145,146,\ldots ,210\rbrace }\)).
-
The change in electricity consumption with variation of the production speed is calculated using the conversion formula in Schulz et al. (2020).
-
All jobs are known at the beginning of the working week and the due dates are at the end of the working week. The quantities demanded for each job are randomly generated.
-
Preemption and lot-splitting are possible.
-
A machine can process at most one job in one period, and the selected production speed cannot change in one period. The latter also applies to a selected machine state.
-
For simplicity, warehousing, backlog and machine setups are neglected.
Indices | |
---|---|
m | Machine \(m \in {\mathcal {M}} = \{1,..., M\}\) |
j | Job \(j \in {\mathcal {J}} = \{1,..., J\}\) |
i, h | Machine states \(i, h \in {\mathcal {I}} = \{0,..., I\}\) with the following states: off (\(i=0\)), ramp-up (\(i=1\)), standby (\(i=2\)) and production (\(i=3\)) |
t | Period (a quarter hour) \(t \in {\mathcal {T}} = \{ 1,..., T\}\) |
\(\nu\) | Production speed level \(\nu \in {\mathcal {N}}=\lbrace 1,...,N \rbrace\) |
Parameters | |
---|---|
\(c^{RTF}_t\) | Gas factor in period t of the electricity mix considered |
\(c^{EDR}_t\) | \(c^{EDR}_t = \rho\) if a critical event occurs in period t, 0 otherwise. \(\rho > 0\) is a prespecified penalization factor |
\(a^{prod}_{\nu j}\) | Quarter hourly production rate of job j on each machine at speed level \(\nu\) |
\(d_j\) | Demanded quantities of job j |
\(\gamma ^{tran}_{ih}\) | Transition parameter from machine state i to state h (1 if possible, 0 otherwise) |
\(a^{elec}_{im}\) | Quarter hourly electricity consumption of machine m in state \(i \in {\mathcal {I}}\setminus \lbrace I \rbrace\) |
\({\widehat{a}}^{elec\_I}_{\nu m}\) | Quarter hourly electricity consumption of machine m in production state I at speed level \(\nu\) |
Decision variables | |
---|---|
\(x_{jmt \nu }\) | Equals 1 if job j will be processed on machine m in t at speed level \(\nu\), otherwise 0 |
\(\delta ^{state}_{imt}\) | Equals 1 if machine m has state i in period t, otherwise 0 |
\({\widehat{\delta }}^{state\_I}_{mt \nu }\) | Equals 1 if machine m is in production state I at speed level \(\nu\) in period t, otherwise 0 |
\(s^{buy}_t\) | Amount of electricity \(\left[ \text {in kWh} \right]\) to be purchased in period t |
\(\alpha _t\) | Equals 1 if a job is produced in period t and 0 otherwise |
\(\beta\) | Nonnegative auxiliary variable used for linearization |
\(\text {C}_{max}\) | Equals the makespan |
\(\text {G}_{RTF}\) | Equals the gas-factorized electricity consumption to be minimized |
\(\text {G}_{EDR}\) | Equals the penalization from electricity consumption in critical periods to be minimized |
4 Results and discussion
4.1 Results in the case of RTF
4.2 Results in the case of EDR
-
Step 2.1: When an event is triggered, we fix the decision variables related to the previous periods of the event to the corresponding values of the previously generated solution. We solve the MIP again, computing a Pareto front representation via the MWTM, but considering the penalization of the electricity consumption in the emergency time frame.
-
Step 2.2: Since a Pareto front representation is calculated in step 2.1, a single solution has to be selected by a decision-maker. This solution is saved. However, if a single solution is to be directly calculated, a no-preference method such as MGC can be applied instead.
-
Step 3: Step 2.1 and 2.2 are repeated until no further event occurs in the considered time horizon. The last saved solution is then the final schedule.