Abstract
We give strong formulations of ramping constraints—used to model the maximum change in production level for a generator or machine from one time period to the next—and production limits. For the two-period case, we give a complete description of the convex hull of the feasible solutions. The two-period inequalities can be readily used to strengthen ramping formulations without the need for separation. For the general case, we define exponential classes of multi-period variable upper bound and multi-period ramping inequalities, and give conditions under which these inequalities define facets of ramping polyhedra. Finally, we present exact polynomial separation algorithms for the inequalities and report computational experiments on using them in a branch-and-cut algorithm to solve unit commitment problems in power generation.
Similar content being viewed by others
References
Agra, A., Constantino, M.: Lotsizing with backlogging and start-ups: the case of Wagner–Whitin costs. Oper. Res. Lett. 25, 81–88 (1999)
Aktürk, M.S., Atamtürk, A., Gürel, S.: A strong conic quadratic reformulation for machine-job assignment with controllable processing times. Oper. Res. Lett. 37, 187–191 (2009)
Baptisella, L.F., Geromel, J.: A decomposition approach to problem of unit commitment schedule for hydrothermal systems. IEEE Proc. D Control Theory Appl. 127, 250–258 (1980)
Carrion, M., Arroyo, J.M.: A computationally efficient mixed integer linear formulation for the thermal unit commitment problem. IEEE Trans. Power Syst. 21, 1371–1378 (2006)
Cohen, A.I., Yoshimura, M.: A branch and bound algorithm for unit commitment. IEEE Trans. Power Appar. Syst. PAS–102, 444–451 (1983)
Constantino, M.: A cutting plane approach to capacitated lot-sizing with start-up costs. Math. Program. 75, 353–376 (1996)
Constantino, M.: Lower bounds in lot-sizing models: a polyhedral study. Math. Oper. Res. 23, 101–118 (1998)
Damcı-Kurt, P.: Mixed-integer programming methods for transportation and power generation problems. Ph.D. Thesis, The Ohio State University (2014)
Fan, W., Guan, X., Zhai, Q.: A new method for unit commitment with ramping constraints. Electr. Power Syst. Res. 62, 215–224 (2002)
Frangioni, A., Gentile, C.: Solving nonlinear single-unit commitment problems with ramping constraints. Oper. Res. 54, 767–775 (2006)
Frangioni, A., Gentile, C., Lacalandra, F.: Solving unit commitment problems with general ramp constraints. Int. J. Electr. Power Energy Syst. 30, 316–326 (2008)
Frangioni, A., Gentile, C., Lacalandra, F.: Tighter approximated MILP formulations for unit commitment problems. IEEE Trans. Power Syst. 24, 105–113 (2009)
Garver, L.L.: Power generation scheduling by integer programming-development of theory. Trans. Am. Inst. Electr. Eng. Part III Power App. Syst. 81, 730–734 (1962)
Havel, P., Šimovič, T.: Optimal planning of cogeneration production with provision of ancillary services. Electr. Power Syst. Res. 95, 47–55 (2013)
Hobbs, B.F., Rothkopf, M.H., O’Neill, R.P., Chao, H.-P.: The Next Generation of Electric Power Unit Commitment Models. Kluwer Academic Publishers, Norwell (2001)
Kazarlis, S.A., Bakirtzis, A.G., Petridis, V.: A genetic algorithm solution to the unit commitment problem. IEEE Trans. Power Syst. 11, 83–92 (1996)
Lee, J., Leung, J., Margot, F.: Min-up/min-down polytopes. Discrete Optim. 1, 77–85 (2004)
Lowery, P.G.: Generating unit commitment by dynamic programming. IEEE Trans. Power Appar. Syst. PAS–85, 422–426 (1966)
Mantawy, A.H., Abdel-Magid, Y.L., Selim, S.Z.: Unit commitment by tabu search. IEEE Proceed. Gener. Transm. Distrib. 145, 56–64 (1998)
Morales-España, G., Latorre, J.M., Ramos, A.: Tight and compact MILP formulation for the thermal unit commitment problem. IEEE Trans. Power Syst. 28(4), 4897–4908 (2013)
Morales-España, G., Latorre, J.M., Ramos, A.: Tight and compact MILP formulation of start-up and shut-down ramping in unit commitment. IEEE Trans. Power Syst. 28(2), 1288–1296 (2013)
Nemhauser, G.L., Wolsey, L.A.: Integer and Combinatorial Optimization. Wiley, New York (1988)
Orero, S.O., Irving, M.R.: A genetic algorithm modeling framework and solution technique for short term optimal hydrothermal scheduling. IEEE Trans. Power Syst. 13, 501–518 (1998)
Ostrowski, J., Anjos, M.F., Vannelli, A.: Tight mixed integer linear programming formulations for the unit commitment problem. IEEE Trans. Power Syst. 27, 39–46 (2012)
Pekelman, D.: Production smoothing with fluctuating price. Manag. Sci. 21(5), 576–590 (1975)
Pochet, Y.: Mathematical programming models and formulations for deterministic production planning problems. In: Jünger, M., Naddef, D. (eds.) Computational Combinatorial Optimization, Lecture Notes in Computer Science LCNS, vol. 2241, pp. 57–111. Springer, Berlin (2001)
Pochet, Y., Wolsey, L.: Algorithms and reformulations for lot sizing problems. In: Cook, W., Lovasz, L., Seymour, P. (eds.) Combinatorial Optimization, Volume 20 of DIMACS Series in Discrete Mathematics and Theoretical Computer Science, pp. 245–294. Applied Mathematical Society, Philadelphia (1995)
Pochet, Y., Wolsey, L.: Production Planning by Mixed Integer Programming. Springer, Berlin (2006)
Rajan, D., Takriti, S.: Minimum up/down polytopes of the unit commitment problem with start-up costs. IBM Research Report RC23628, IBM, Yorktown Heights, NY (2005)
Rong, A., Lahdelma, R.: An effective heuristic for combined heat-and-power production planning with power ramp constraints. Appl. Energy 84, 307–325 (2007)
Saravanan, B., Das, S., Sikri, S., Kothari, D.P.: A solution to the unit commitment problem—a review. Front. Energy 7, 223–236 (2013)
Schrijver, A.: Theory of Linear and Integer Programming. Wiley, New York (1998)
Sheble, G.B., Fahd, G.N.: Unit commitment literature synopsis. IEEE Trans. Power Syst. 9, 128–135 (1994)
Silver, E.A.: A tutorial on production smoothing and work force balancing. Oper. Res. 15(6), 985–1010 (1967)
Snyder, W.L., Powell, H.D., Rayburn, J.C.: Dynamic programming approach to unit commitment. IEEE Trans. Power Syst. 2, 339–350 (1987)
Tseng, C.-L., Li, C.A., Oren, S.S.: Solving the unit commitment problem by a unit decommitment method. J. Optim. Theory 105, 707–730 (2000)
Wang, Q., Guan, Y., Wang, J.: A chance-constrained two-stage stochastic program for unit commitment with uncertain wind power output. IEEE Trans. Power Syst. 27, 206–215 (2012)
Zhuang, F., Galiana, F.D.: Unit commitment by simulated annealing. IEEE Trans. Power Syst. 5, 311–318 (1990)
Acknowledgments
Pelin Damcı-Kurt and Simge Küçükyavuz are supported, in part, by the National Science Foundation Grant #1055668, and an allocation of computing time from the Ohio Supercomputer Center. Deepak Rajan’s work is performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under contract DE-AC52-07NA27344. Alper Atamtürk is supported, in part, by the Office of Assistant Secretary of Defense for Research and Engineering and the National Science Foundation Grant #0970180.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1: Proof of convex hull of two-period ramping polytope without start variables
Corollary 1
For \(\bar{u} \le \ell + \delta \), \(conv(\mathcal {UNS}_t^2) = \{(p,x) \in \mathbb {R}^{2n}: (7){-}(15)\}\), and for \(\bar{u} > \ell + \delta \), \(conv(\mathcal {UNS}_t^2) = \{(p,x) \in \mathbb {R}^{2n}: (7){-}(13), (16), (17)\}\).
Proof
We use Fourier–Motzkin elimination (see [32] and [22]) of variable \(s_{t+1}\) from the convex hull of the feasible solutions to the formulation with start-up variables given by inequalities (5a)–(5i). Inequalities (5d), (5e) and (5h) continue to be facets [given by inequalities (7)–(9)] because these inequalities do not include variable \(s_{t+1}\) in their description. Similarly, if \(\bar{u}=\ell +\delta \), then inequality (5f) is also a facet, because the coefficient of \(s_{t+1}\) is equal to 0. In this case, inequality (5f) is equivalent to inequalities (14) and (15). If \(u=\bar{u}\), then inequality (5g) is also a facet, because the coefficient of \(s_{t+1}\) is equal to 0. In this case, inequality (5g) reduces to (12).
We need to consider all possible cross products of inequalities (5b), (5c), and (5f) (if \(\bar{u} > \ell +\delta \)) that provide a lower bound on \(s_{t+1}\) with inequalities (5a), (5g) (if \(u>\bar{u}\)), (5i), and (5f) (if \(\bar{u} < \ell +\delta \)) that provide an upper bound on \(s_{t+1}\).
We first consider the cross product of lower-bounding inequalities (5b) and (5c) with upper-bounding inequalities defined by (5a), (5g) (if \(u >\bar{u} \)) and (5i).
-
The pair of inequalities (5b) and (5a) gives \(x_{t} \ge 0\), which is dominated by inequalities (5e) and (5h).
-
The pair of inequalities (5b) and (5g) gives inequality (10) (if \(u >\bar{u}\)).
-
The pair of inequalities (5b) and (5i) gives inequality (11).
-
The pair of inequalities (5c) and (5a) gives \(x_{t+1} \ge 0\), which is dominated by inequalities (5d) and (12).
-
The pair of inequalities (5c) and (5g) gives inequality (12) (if \(u >\bar{u}\)).
-
The pair of inequalities (5c) and (5i) gives inequality (13).
Note that depending on whether the coefficient of \(s_{t+1}\) in inequality (5f) is positive or negative, we get a lower-bounding or upper-bounding inequality for \(s_{t+1}\), respectively. Therefore, we consider the following two cases:
- Case 1 :
-
If \(\bar{u} < \ell +\delta \), then inequality (5f) is an upper-bounding inequality given by
$$\begin{aligned} s_{t+1} \le \frac{(\ell +\delta )x_{t+1} - \ell x_{t} - p_{t+1} + p_{t}}{(\ell + \delta - \bar{u})}. \end{aligned}$$(27)
-
The pair of inequalities (5b) and (27) gives inequality (14).
-
The pair of inequalities (5c) and (27) gives inequality (15).
- Case 2 :
-
If \(\bar{u} > \ell +\delta \), then inequality (5f) is a lower-bounding inequality given by
$$\begin{aligned} s_{t+1} \ge \frac{-(\ell +\delta )x_{t+1} + \ell x_{t} + p_{t+1} - p_{t}}{(\bar{u}- \ell - \delta )}. \end{aligned}$$(28)
-
The pair of inequalities (5a) and (28) gives
$$\begin{aligned} p_{t+1} - p_{t} \le \bar{u} x_{t+1} - \ell x_{t}, \end{aligned}$$(29)which is dominated by inequalities (17) and (8). To see this, note that multiplying inequality (8) by \(-(\bar{u}-\ell -\delta )\) and adding to inequality (17) gives inequality (29) multiplied by \((u-\ell -\delta )\).
-
The pair of inequalities (5g) (when \(u >\bar{u}\)) and (28) gives
$$\begin{aligned} \frac{u x_{t+1} - p_{t+1}}{(u - \bar{u})} \ge \frac{-(\ell +\delta )x_{t+1} + \ell x_{t} + p_{t+1} - p_{t}}{(\bar{u}- \ell - \delta )}. \end{aligned}$$Rearranging terms we get inequality (17).
-
The pair of inequalities (5i) and (28) gives inequality (16).
\(\square \)
Appendix 2: Facet-definition proof of type-I multi-period ramp-up inequality
Proposition 10
Type-I multi-period ramp-up inequality (20) defines a facet of \(conv(\mathcal {U})\) if and only if \(\ell + j\delta < u\).
Proof
Necessity: For contradiction assume that \(\ell + j\delta \ge u\). From validity of inequality (20) we have \(\ell + j\delta \le u\), so \(\ell + j\delta = u\). Then inequality (20) can be written as \(p_{t+j}-p_t\le (\ell +j\delta )x_{t+j} - \ell x_{t} = u x_{t+j} - \ell x_{t}\), and it is dominated by inequalities (1a) and (1b).
Sufficiency: We use the technique in Theorem 3.6 of §\(I.4.3\) in Nemhauser and Wolsey [22]. We show that inequality (20) is the only inequality that is satisfied at equality by all points \((p,x,s) \in \mathcal {U}\) that are tight at (20), i.e., we show that if all points of \(\mathcal {U}\) at which inequality (20) is tight satisfy
then
-
1.
\(\alpha _{0}=0\),
-
2.
\(\alpha _{k} = 0\), \(k \in [1,n] {\setminus } \{t,t+j\}\),
-
3.
\(\alpha _{t} = -\bar{\alpha }\), \(\alpha _{t+j} = \bar{\alpha }\),
-
4.
\(\beta _{k} = 0\), \(k \in [1,n] {\setminus } \{t,t+j\}\),
-
5.
\(\beta _{t} = \bar{\alpha }\ell \),
-
6.
\(\beta _{t+j} = -\bar{\alpha }(\ell + j\delta )\),
-
7.
\(\gamma _{k}=0\), \(k \in [2,t] \cup [t+j+1,n]\),
-
8.
\(\gamma _{t+i'} = -\bar{\alpha }\min \{(\bar{u}-\ell -i'\delta ), (u - \ell - j \delta )\}\), \(i' \in [1,j]\).
In order to establish the values of the coefficients \(\alpha _{k}\), \(\beta _{k}\), \(\gamma _{k}\) and \(\alpha _{0}\), we construct a feasible solution to \(\mathcal {U}\) on the face defined by (20). Then a small change in the solution is made to obtain another feasible solution which is on the face defined by inequality (20). Comparing the resulting expressions, the possible values of a set of coefficients are obtained. Also note that from the validity assumption and (A2), \(\bar{u}\ge \ell +\delta >\ell \). We start by describing several points feasible to \(\mathcal {U}\) that will be used throughout the facet proof. We assume that \(k\ge 2\) if we set the value of \(s_{k}\). In the following feasible solutions [except for the zero vector (S1)] if the value of a variable is not given, then its value is equal to zero. Let \(k_1, k_2 \in [2,n]\) be two periods such that \(k_1 < k_2\). Let \(\epsilon \) be a very small number greater than zero.
Note that points (S3), (S6), (S7), (S8), and (S9) are feasible because \(\bar{u} > \ell \) and \(\ell + j\delta < u\).
Next we show the values of the coefficients \(\alpha _{k}\), \(\beta _{k}\), \(k \in [1,n]\), \(\gamma _{k}\), \(k \in [2,n]\) and \(\alpha _{0}\).
-
1.
\(\alpha _{0}=0\).
Consider solution (S1). Clearly, this point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to zero. Hence, \(\alpha _{0}=0\).
-
2.
\(\alpha _{k} = 0\), \(k \in [1,n] {\setminus } \{t,t+j\}\).
Consider the following two cases:
-
(a)
\(k \in [1,t-1] \cup [t+j+1,n]\).
Consider solution (S2) with \(k_1=k\). Clearly, this point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to zero. Now, consider solution (S3) with \(k_1=k\). This point also satisfies inequality (20) at equality and is a valid point because \(\bar{u} > \ell \) by assumption. Then evaluating (30) at both solutions we get \(\alpha _{k} \ell =\alpha _{k} (\ell + \epsilon )\). Hence, \(\alpha _{k}=0\).
-
(b)
\(k \in [t+1,t+j-1]\).
Consider solution (S5) with \(k_1=t\) and \(k_2=k\). This point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to \(-\ell \). Now, consider solution (S6) with \(k_1=t\) and \(k_2=k\) (this point also satisfies inequality (20) at equality). Then evaluating (30) at both solutions we get \(\alpha _{k} \ell =\alpha _{k} (\ell + \epsilon )\). Hence, \(\alpha _{k}=0\).
-
(a)
-
3.
\(\alpha _{t} = -\bar{\alpha }\), \(\alpha _{t+j} = \bar{\alpha }\).
Consider solution (S7). This point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to \(j\delta \). Now, consider solution (S8). This point also satisfies inequality (20) at equality. Because we showed that \(\alpha _{k}=0\), \(k \in [1,n] {\setminus } \{t,t+j\}\) in part 2, evaluating (30) at both solutions we get \(\alpha _{t} \epsilon = -\alpha _{t+j}\epsilon \). Let \(\bar{\alpha }:=-\alpha _{t}=\alpha _{t+j} \).
-
4.
\(\beta _{k} = 0\), \(k \in [1,n] {\setminus } \{t,t+j\}\).
Consider the following two cases:
-
(a)
\(k > t+j\).
Consider a solution to \(\mathcal {U}\) with \(x_{r}=1\), \(r \in [t,k]\), \(s_{t}=1\), \(p_{t+i} = \ell + i\delta \), \(i \in [0,j]\), \(p_{r} = \ell +j\delta \), \(r \in [t+j+1,k]\) and all other variables are equal to zero. This point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to \(j\delta \). Now, consider the same solution except we set \(x_{k}=0=p_{k}\) (this solution is on the face defined by inequality (20)). Evaluating (30) at both solutions we get \(\alpha _{k}(\ell +j\delta ) +\beta _{k} = 0\). Because we showed that \(\alpha _{k}=0\) in part 2 we get \(\beta _{k}=0\).
-
(b)
\(t < k < t+j\).
Consider solution (S5) with \(k_1=t\) and \(k_2=k\). This point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to \(-\ell \). Now, consider solution (S5) with \(k_1=t\) and \(k_2=k-1\) if \(t<k-1\), and solution (S2) with \(k_1=t\) if \(t=k-1\). Both of the points satisfy inequality (20) at equality. Note that if \(t=k-1\) we use solution (S2) because both \(k_1=t\) and \(k_2=k-1=t\) and we define \(k_1 < k_2\) in solution (S5). Evaluating (30) at the described solutions we get \(\alpha _{k}\ell +\beta _{k} = 0\). Because we showed that \(\alpha _{k}=0\) in part 2 we get \(\beta _{k}=0\).
-
(c)
\(k \le t-1\) for \(t \ge 2\).
Consider the following two cases:
-
i.
\(k=1\).
Consider solution (S4). This point is on the face defined by inequality (20) because both the left- and the right-hand sides of the inequality are equal to zero. Evaluating (30) at this solution we get \(\alpha _{k}\ell +\beta _{k}=\alpha _{0}=0\) and since \(\alpha _{k}=0\) (from part 2) we get \(\beta _{k}=0\).
-
ii.
\(k\ge 2\).
Consider solution (S10) with \(k_1=k\). This point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to zero. Now, consider solution (S2) with \(k_1=k-1\). Then evaluating (30) at both solutions we get \(\alpha _{k}\ell +\beta _{k}=0\), and because \(\alpha _{k}=0\) (from part 2) we get \(\beta _{k}=0\).
-
i.
-
(a)
-
5.
\(\beta _{t} = \bar{\alpha }\ell \).
If \(t=1\), then we use solution (S4). Because \(\alpha _{t} =\alpha _{1} =-\bar{\alpha }\) (from part 3) evaluating this solution at equality (30) we get \(\alpha _{1}\ell + \beta _{1}=0\) so \(\beta _{1} = \bar{\alpha }\ell \). For \(t\ge 2\) consider solution (S5) with \(k_1=1\) and \(k_2=t\). This point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to \(-\ell \). Now consider solution (S5) with \(k_1=1\) and \(k_2=t-1\). This point is on the face defined by inequality (20) because both the left- and the right-hand sides of the inequality are equal to zero. Then evaluating (30) at both solutions we obtain \(\alpha _{t}\ell +\beta _{t}=0\). Because \(\alpha _{t}=-\bar{\alpha }\) (from part 3) we get \(\beta _{t}=\bar{\alpha }\ell \).
-
6.
\(\beta _{t+j} = -\bar{\alpha }(\ell + j\delta )\).
Consider solution (S9). This point satisfies inequality (20) at equality because both the left- and the right-hand sides of the inequality are equal to \(j\delta \). Consider the same solution except now we set \(x_{t+j}=p_{t+j}=0\). This point is on the face defined by inequality (20) because both the left- and the right-hand sides of the inequality are equal to \(-\ell \). Then evaluating (30) at both solutions we obtain \(\alpha _{t+j}(\ell + j\delta ) +\beta _{t+j}=0\). Because \(\alpha _{t+j}=\bar{\alpha }\) (from part 3) we get \(\beta _{t+j}=-\bar{\alpha }(\ell + j\delta )\).
-
7.
\(\gamma _{k}=0\), \(k \in [2,t] \cup [t+j+1,n]\).
Consider solution (S2) with \(k_1=k\). This point satisfies inequality (20) at equality because both the left-and the right-hand sides of the inequality are equal to zero unless \(k=t\). If \(k=t\), then both the left- and the right-hand sides of the inequality are equal to \(-\ell \). Evaluating (30) at this solution we obtain \(\alpha _{k}\ell +\beta _{k} + \gamma _{k}=0\). If \(k \not = t\), then we have \(\alpha _{k}=\beta _{k}=0\) (from parts 2 and 4) so we get \(\gamma _{k}=0\). If \(k=t\), then because \(\alpha _{t}\ell = -\bar{\alpha }\ell \) and \(\beta _{t}=\bar{\alpha }\ell \) (from parts 3 and 5), we get \(\gamma _{t}=0\).
-
8.
\(\gamma _{t+i'} = -\bar{\alpha }\min \{(\bar{u}-\ell -i'\delta ), (u - \ell - j \delta )\}\), \(i' \in [1,j]\).
Consider a solution to \(\mathcal {U}\) with \(x_{t+i'}=1\), \(p_{t+i'}=\bar{u}\), \(s_{t+i'}=1\), \(x_{t+i}=1\), \(p_{t+i}=\min \{\bar{u} + (i-i')\delta , u\}\), \(i \in [i'+1, j]\) and all other variables are equal to zero. This point satisfies inequality (20) at equality because either both the left- and the right-hand sides of the inequality are equal to \(\bar{u}+(j-i')\delta \) or u depending on the value of \(p_{t+j}\). Evaluating (30) at this solution we obtain,
$$\begin{aligned} \alpha _{t+i'}\bar{u} + \beta _{t+i'} + \gamma _{t+i'} + \sum _{i=i'+1}^{j-1}(\alpha _{t+i}p_{t+i} + \beta _{t+i}) + \alpha _{t+j}p_{t+j} +\beta _{t+j} = 0. \end{aligned}$$From parts 1–4 and 6 we obtain \(\gamma _{t+i'} = -\bar{\alpha }(p_{t+j} - \ell - j\delta )\). Furthermore, because \(p_{t+j}\) is either \(\bar{u}+(j-i')\delta \) or u, the proof is complete. \(\square \)
Rights and permissions
About this article
Cite this article
Damcı-Kurt, P., Küçükyavuz, S., Rajan, D. et al. A polyhedral study of production ramping. Math. Program. 158, 175–205 (2016). https://doi.org/10.1007/s10107-015-0919-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10107-015-0919-9
Keywords
- Ramping
- Unit commitment
- Co-generation
- Production smoothing
- Convex hull
- Polytope
- Valid inequalities
- Facets
- Computation