Combined Heat and Power Production ManagementIntroduction
The future development of energy production and distribution relies on the combined exploitation of different, renewable energy sources, each one with its own costs and profits, sometimes dependent on weather and/or daylight hours (e.g., thermal solar). In this contest Combined Heat and Power (hereafter CH&P) assumes a increasing relevance due to its high efficiency when compared to conventional condensing power plants.
We consider CH&P Energy production optimization a tactical and operational issue, because the management cycle has a yearly horizon and subsequent short term planning, with the aim of maximization of EBITDA (Earnings Before Interest, Taxes, Depreciation, and Amortization).
Mathematical models
The Mixed Integer Linear Programming (MILP) problems arising in this business are particular cases of the wellknown Unit Commitment (UC) problem (see, e.g. \cite{padhy2004unit, wood2012power}), in which the goal is to determine how to dispatch the committed units in order to meet the demands and other constraints costefficiently. As presented in \cite{bettinelli2016}, a plant manager has to serve one or more demands (heat demand mainly, but also chill demand and/or electricity demand if the plant is connected to a building or to an industrial facility) and these demands can be satisfied by producing energy using different machines, depending on the plant configuration and the various economic drivers of input and output commodities. Also, if the electricity produced by the plant is sold to the National Electricity Network, there is an additional variable (the amount of energy sold) and an additional constraint (the amount of electricity committed to the market the day before for the following day).
Manual management relies on several elements to manage the typical business processes:
 budgeting, to define the overall energy production for the year to come;
 revised budgeting, to modify the budget objectives on the basis of the results of (at least) the first semester;
 weekly and daily operations, most of the times based on the experience of the plant manager (in respect of the previous years).
The goal of a plant manager is to satisfy the customers’ demands by choosing the best energy production mix to maximize the EBITDA. Consequently, the decisionmaking process must take into account the following factors:
 costs, profits and fiscal advantages (if any) of each energy source;
 technical constraints of the plant itself and of the machines;
 regulatory constraints;
 ordinary and extraordinary maintenance requirements.
For each machines, the workload has to be optimized for each time window. Possible nonlinear efficiency curves of the machines are can be approximated in nonconvex piecewise linear functions and linearized in the model. Time is discretized in a finite set of time intervals, having a duration defined by the user. Smaller time windows yield to more detailed production plans but also increase the size of the associated problems and the corresponding computational effort required for their resolution. Fiscal advantages, as well as power production regulatory constraints are defined by the legislator over indicators that measure the performances of the plant over a solar year and play a fundamental role in economic terms, since strict respect of certain energy/heat rations is the condition to leverage on incentives that are often fundamental to economic sustainability of the plant. This explains why even for operational dayahead optimization, an extended time horizon including the entire month or, more often, the remaining part of the solar year should be considered. The presence of thermal energy collectors, as well as plantspecific period constraints (maximum number of startup over period) are other examples that binds together what otherwise may be considered as single timewindows problems.
Optimization methods
UC problems, even when CH&P is not considered, are complex in practice. The solution methods used to solve UC problems span from Lagrangian relaxation, as proposed in (Borghetti, 2003) and (Li, 2005) to genetic (Kazarlis,1996) or Tabu search (Mantawy,1998) heuristics. Attention has been put as well on obtaining efficient MILP formulations (see e.g. (Carrion, 2006) and (Ostrowski, 2012)). Interdependencies between power and heat productions make realistic CH&P power units even more difficult to be optimized (Song, 1999). In some cases general purpose MILP techniques are applied, such as the Branch and Bound (see (Illerhaus, 1999)). Resolution methods may otherwise rely on timebased decomposition, as in (Lahdelma, 2003), dynamic programming (Rong, 2009) or again Lagrangian relaxation (Thorin, 2005). In the business case considered in (Bettinelli, 2016) a Matheuristics based on time and machines decomposition have been implemented to reach near optimal solutions in a reasonable amount of time. The key idea of the heuristic strategies consists in the resolution of several easier subproblems rather than directly addressing the original large MILP formulation.
Data and Software
Anchor 

 CHPreferences 

 CHPreferences 


References
1. Borghetti, A., Frangioni, A., Lacalandra, F., Nucci, C.A.: Lagrangian heuristics based on disaggregated bundle methods for hydrothermal unit commitment. Power Systems, IEEE Transactions on 18(1), 313–323 (2003)
2. Bettinelli A., Gordini A., Laghi A., Parriani T., Pozzi M., Vigo D.: Decision support systems for energy production optimization and network design in district heating applications. Technical Report OR165, DEI –University of Bologna, under review by Integrated Series in Information Management
3. Carrion, M., Arroyo, J.M.: A computationally efficient mixedinteger linear formulation for the thermal unit commitment problem. Power Systems, IEEE Transactions on 21(3), 1371–1378 (2006)
4. Illerhaus, S., Verstege, J.: Optimal operation of industrial chpbased power systems in liberalized energy markets. In: Electric Power Engineering, 1999. PowerTech Budapest 99. International Conference on, p. 210. IEEE (1999)
5. Kazarlis, S.A., Bakirtzis, A., Petridis, V.: A genetic algorithm solution to the unit commitment problem. Power Systems, IEEE Transactions on 11(1), 83–92 (1996)
6. Lahdelma, R., Hakonen, H.: An efficient linear programming algorithm for combined heat and power production. European Journal of Operational Research 148(1), 141–151 (2003)
7. Li, T., Shahidehpour, M.: Pricebased unit commitment: a case of lagrangian relaxation versus mixed integer programming. Power Systems, IEEE Transactions on 20(4), 2015–2025 (2005)
8. Mantawy, A., AbdelMagid, Y.L., Selim, S.Z.: Unit commitment by tabu search. In: Generation, Transmission and Distribution, IEE Proceedings, vol. 145, pp. 56–64. IET (1998)
9. Ostrowski, J., Anjos, M.F., Vannelli, A.: Tight mixed integer linear programming formulations for the unit commitment problem. IEEE Transactions on Power Systems 1(27), 39–46 (2012)
10. Padhy, N.P.: Unit commitmenta bibliographical survey. Power Systems, IEEE Transactions on 19(2), 1196–1205 (2004)
11. Rong, A., Hakonen, H., Lahdelma, R.: A dynamic regrouping based sequential dynamic programming algorithm for unit commitment of combined heat and power systems. Energy Conversion and Management 50(4), 1108–1115 (2009)
12. Song, Y., Chou, C., Stonham, T.: Combined heat and power economic dispatch by improved ant colony search algorithm. Electric Power Systems Research 52(2), 115–121 (1999)
13. Thorin, E., Brand, H., Weber, C.: Longterm optimization of cogeneration systems in a competitive market environment. Applied Energy 81(2), 152–169 (2005)
14. Wood, A.J., Wollenberg, B.F.: Power generation, operation, and control. John Wiley & Sons (2012)