Energy and Buildings 41 (2009) 445451

Contents lists available at ScienceDirect

Energy and Buildings

journal homepage: www.elsevier.com/locate/enbuild

Cost-optimized real-time operation of CHP systems

Heejin Cho a,*, Rogelio Luck a, Sandra D. Eksioglu b, Louay M. Chamra a
a b

Department of Mechanical Engineering, Mississippi State University, MS 39762, USA Department of Industrial and System Engineering, Mississippi State University, MS 39762, USA

A R T I C L E I N F O

A B S T R A C T

Article history: Received 3 September 2008 Received in revised form 22 October 2008 Accepted 10 November 2008 Keywords: Optimization CHP Micro-CHP Cogeneration system Energy dispatch

The Cooling, Heating, and Power (CHP) systems have been widely recognized as a key alternative for thermal and electric energy generation because of the outstanding energy efciency, reduced environmental emissions, and relative independence from centralized power grids. Nevertheless, the total energy cost of CHP systems can be highly dependent on the operation of individual components. This paper presents an energy dispatch algorithm that minimizes the cost of energy (e.g., cost of electricity from the grid and cost of natural gas into the engine and boiler) based on energy efciency constrains for each component. A deterministic network ow model of a typical CHP system is developed as part of the algorithm. The advantage of using a network ow model is that the electric and thermal energy ows through the CHP equipment can be readily visualized allowing for easier interpretation of the results. This algorithm has been used in simulations of a case study on the operation of an existing micro-CHP system. The results from the simulation are presented in the paper to demonstrate the economical advantages resulting from optimal operation. 2008 Elsevier B.V. All rights reserved.

1. Introduction The rising demand for electrical power as well as energy for cooling and heating of buildings is a growing worldwide concern. Economical, environmental, political, and logistical problems associated with raising centralized electrical power production are becoming more difcult to overcome. The Cooling, Heating, and Power (CHP) systems have been widely recognized as a key alternative for thermal and electric energy generation because of the outstanding energy efciency, reduced environmental emissions, and relative independence from centralized power grids. The schematic in Fig. 1 shows a typical CHP system for a building. A CHP system consists of a power generation unit interacting with thermally-activated components. A variety of power generation units can be used in a CHP system: micro-turbines, internal combustion engines, fuel cells, etc. Topping cycle CHP systems produce electric energy and recover thermal energy from the normally wasted stream to produce useful heating and/or cooling as byproduct. The electric energy produced by power generation units (PGUs) is used to power appliances and lights in the building and operate cooling and heating equipments. If the PGUs do not produce enough electric energy, the difference can be imported

* Corresponding author at: Department of Mechanical Engineering, Mississippi State University, P.O. Box ME, MS 39762, USA. Tel.: +1 662 617 1582; fax: +1 662 325 7223. E-mail address: hc86@msstate.edu (H. Cho). 0378-7788/$ see front matter 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.enbuild.2008.11.011

from the electric grid (EG). If there is excess electricity, it can be exported or sold to the EG. The waste heat can be recovered in the CHP system and converted to cooling and/or heating energy for the building. If the CHP components cannot recover enough thermal energy for space cooling and heating, supplementary heating devices (SHD) provide extra heat into the system. Refs. [1,2] provide a detailed technical description and analysis of the operation of several PGUs and thermally activated components that can be used in CHP systems. Optimal operation of a CHP system is not easy to achieve because of varying operational conditions such as the costs of fuel and electricity for the energy generation systems, and electric, cooling, and heating load demand from a building. Note that the price of electricity can vary seasonally, by the time of day, and may also exhibit sharp peak prices for very short periods of time. A proper optimization scheme is required to operate the CHP systems at the minimum cost and the maximum efciency. In practice, CHP systems are often operated manually. For instance, the optimal costs of fuels used in PGUs are often pre-calculated, and PGUs are activated if the costs of fuels are below certain values. This type of strategy does not provide a cost-optimized operation of the systems. In this paper, an energy dispatch algorithm is proposed to provide control signals for the cost-optimized operation of the equipment. The objective of this algorithm is to minimize the cost of energy, i.e., the cost of electricity from the grid and the cost of fuel into the PGUs and SHDs, based on energy ow in the system and energy efciency constraints for each component. This paper implements a linear program that identies

446

H. Cho et al. / Energy and Buildings 41 (2009) 445451

Fig. 1. Flow diagram of a typical CHP system.

the amount of cooling, heating, and electric energy to be produced in a CHP system in a time period for a given load forecast. The model captures the ow of energy in the system. For example, the model performs energy balances at each time step and calculates the loss of energy in the system. Previously, a dynamic computer simulation of a micro-CHP system at Mississippi State University (MSU) has been performed using a transient simulation program, TRNSYS [35]. It has been demonstrated that a dynamic micro-CHP model can be successfully simulated and an economical analysis can be readily incorporated as part of the dynamic simulation. It has also been shown that an optimization scheme on the operation of the system is necessary to operate it at the optimal condition. The algorithm proposed in this paper can be readily integrated into the dynamic simulation to provide the optimal operation at the minimum energy cost. Furthermore, the algorithm can be implemented physically in the test facility at MSU to provide control signals for the optimal operation of the CHP equipments. A number of papers in the literature have demonstrated the effectiveness of using linear programming to identify cost-efcient operation of CHP systems. Lahdelma and Hakonen [6] proposes a linear programming model that optimizes the overall production of heat and power on a number of power generation facilities for a given load forecast. Their model identies the amount of electricity and heat produced by each facility in a time period. A facility could be a power plant, hydropower, heat plant, etc. Rong and Lahdelma [7] extends this model to consider trigeneration systems. In Ref. [8], a tool for long-term optimization of cogeneration systems using mixed integer linear-programming and Lagrangian relaxation is developed, and it has been tested on an existing CHP system. The above studies focus on long-term unit-commitment of several generation facilities and neither use specic details of the operation of individual internal CHP components for a single facility nor do they provide information on the instantaneous control signals for the CHP components. 2. Optimal energy dispatch algorithm The goal of this algorithm is to provide the operational signal to the CHP components to operate the system in the most economical manner. A ow chart of the optimal energy dispatch algorithm is shown in Fig. 2. The costs of fuel and electricity are fed to the algorithm as inputs. In developing the Linear Programming (LP) algorithm, the following assumptions are used:  The electric, cooling, and heating energy demand from the building are exactly fullled by combining all the available sources of energy.  The CHP system satises conservation of energy, i.e., [total energy demand] = [total energy supply] [total energy loss].

Detailed descriptions of the network ow model and associated optimization scheme are presented in the following sections. 2.1. Description of the network ow model Network ow models are commonly used to help setting up linear programs [9]. A network ow model for a typical CHP system has been developed based on the ow diagram depicted in Fig. 1. The advantage of using the network ow model in the energy system problems is to effectively illustrate the energy ow, energy supply and demand, and efciency constraints. In addition, the network ow model facilitates setting up the objective function and constraints for the LP problem. The network ow model for a typical CHP system is shown in Fig. 3. The nodes in this network represent sources of energy and energy demand points. Arcs of this network demonstrate how the energy ows. Nodes 1 and 13 are conceptual nodes introduced to represent the total energy required for the system. Total energy required for the system is

Fig. 2. Flow chart of the optimal energy dispatch algorithm.

H. Cho et al. / Energy and Buildings 41 (2009) 445451

447

Fig. 3. Network ow model of a typical CHP system.

equal to the sum of the demands for electric, cooling, and heating energy of the building. Node 12 is also a conceptual node used to represent the energy loss in the system. 2.2. Optimization scheme The objective function and the network ow formulation are described next. The overall objective of this model is to minimize the total cost of running the CHP system over a time horizon that consists of T time periods to satisfy the total energy demand. The objective function can be expressed as Minimize fzxg
T X fc1 tx1 t c2 tx2 t c3 tx3 t t1

Eqs. (2)(14) represent constraints on the energy ow. These constraints ensure that the ow of energy coming into a node is equal to the demand plus the ow leaving the node: x1 t x2 t x3 t x13 t x21 t bt x4 t x1 t 0 x5 t x6 t x17 t x2 t 0 x7 t x18 t x3 t 0 x8 t x13 t x4 t x5 t 0 x9 t x10 t x6 t x7 t 0 (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

c4 tx13 tg

(1)

x11 t x19 t x9 t 0 x12 t x20 t x10 t 0 x14 t x8 t 0 x15 t x11 t 0 x16 t x12 t 0 x21 t x17 t x18 t x19 t x20 t 0 x14 t x15 t x16 t bt

Variables x1(t), x2(t) and x3(t) represent the portion of the energy desired from the electric grid (EG), power generation unit (PGU), and supplementary heating device (SHD) in time period t (t = 1, 2, . . ., T). c1(t) represent the cost of purchasing 1 kWh of electricity, c2(t) represents the cost of fuels that is used to produce 1 kWh of energy in the PGUs, and c3(t) represents the cost of fuels that is used to produce 1 kWh of energy in the SHDs. Variable x13(t) represents the amount of energy sold back to the EG in period t, and c4(t) represents the selling price per kWh of electricity. It is assumed here that any excess energy cannot be stored in the system for use in the next period. Instead, it is assumed that excess energy is sold back to the grid during the same period. Therefore, there is no interaction between variables at different time periods and the problem can be decomposed by examining each time period independently of the other time periods. The model will identify the amount of energy to be produced by each energy source so as to optimize the cost of operation.

As shown in Eq. (2), the total energy required for the system (b(t)) is equal to the total energy purchased from the EG (x1(t)) and supplied to PGU (x2(t)) and SHD (x3(t)) minus the energy exported to EG (x13(t)) and the energy lost (x21(t)). Note that b(t) is given as the expected electric, cooling, and heating loads at a given time step. Eqs. (15) and (16) show that, of the total energy produced by

448

H. Cho et al. / Energy and Buildings 41 (2009) 445451

the PGU (x2(t)), a1 100% is electricity, a2 100% is thermal energy and the rest (1 a1 a2) 100% of the energy is lost: x5 t a1 x2 t 0 x6 t a2 x2 t 0 (15) (16)

3.1. Micro-CHP facility at MSU The MSU micro-CHP demonstration facility is located in Starkville, MS, U.S.A. (latitude 33.48N and longitude 88.88W). This building is divided into two zones, a wood workshop area and an ofce area. The total area and height of the building is approximately 279 m2 (3000 ft2) and 3 m (10 ft), respectively. The micro-CHP system provides the electrical power and space conditioning to the building. The exterior and interior views of the mechanical room in the test facility are shown in Fig. 4. The detailed building specications including weather data can be found in Ref. [3]. In the test facility, a four-cylinder internal combustion (IC) engine with a 15 kW generator is used as a PGU and is operated to generate electric and thermal energy, and a 10ton absorption chiller is used for space cooling. A natural gas (NG) IC engine produces the shaft work. A generator converts the shaft work into the electric energy. The engine is operated at constant speed (i.e., 1800 rpm) to provide 60 Hz of electric power frequency which is synchronized to the U.S. grid. The thermal energy rejected from the engine through the engine coolant and exhaust gas is recovered using two heat exchangers attached to the coolant jacket and exhaust gas pipes. This recovered thermal energy is used for space cooling using a water red single-effect absorption chiller and space heating using an air handling unit. A boiler is used as a SHD in this study to provide extra thermal energy to the system when there is not enough heat recovery from the engine. The detailed information on the micro-CHP technology at MSU can be found in Refs. [3,4,10]. 3.2. Electric, cooling, and heating load demand The electric load demand of the micro-CHP system at MSU is estimated based on scheduled usage of appliances, lights, and CHP components in a day. While the shop and ofce appliances and the lights in the building require 2.8 kW, the micro-CHP cooling and heating components require 7.03 and 3.45 kW, respectively. The micro-CHP components are operated to offset the cooling and heating loads imposed on the building. The power consumptions of appliances, lights, and cooling and heating components are presented in Tables 13. Heating and cooling load proles cannot be accurately estimated without using actual information of weather, inltration, or heat generation due to human occupancy and appliances. In this paper, TRNSYS which is a transient simulation program is

Eq. (17) show that, of the total energy produced by the SHDs (x3(t)), a3 100% is thermal energy and the rest (1 a3) 100% of the energy is lost: x7 t a3 x3 t 0 (17)

Eq. (18) shows that, for the cooling component of the CHP system, (1 a4) 100% of the energy is lost: x11 t a4 x9 t 0 (18)

Eq. (19) shows that, for the heating component of the CHP system, (1 a5) 100% of the energy is lost: x12 t a5 x10 t 0 (19)

There is an upper bound on the values that the decision variables x2(t) and x3(t) can take. These bounds are equal to the maximum amount of energy that can be produced in a time period by the PGUs and SHDs, respectively, due to their production capacity. There is no such limit on the amount of electricity that can be purchased from the grid. Variables x14(t), x15(t), and x16(t) have lower bounds that are equal to electric, cooling, and heating loads at time t, respectively. Both upper and lower bounds for variable x14(t) are set to zero when its corresponding demand for electric energy is zero because no energy is needed in this case. The same logic applies to the upper and lower bounds for x15(t) and x16(t). The rest of the lower and upper bounds are free in that they are only required to be positive. 3. Implementation of the algorithm: a case study The optimal energy dispatch algorithm described above has been applied to a micro-CHP facility at MSU. A scenario for the operation of the facility on a particular day in summer has been performed to examine the performance of the optimal energy dispatch algorithm. The description of the micro-CHP facility at MSU, load demand proles, input and constraint information for the algorithm, and the test results of the algorithm are presented in the following sections.

Fig. 4. (a) Side view of building (engine-generator package and mechanical room) and (b) internal view of mechanical room.

H. Cho et al. / Energy and Buildings 41 (2009) 445451 Table 1 Building electric energy usage. Component Shop lights Ofce lights Additional equipment Total Electrical draw (kW) 1.20 0.60 1.00 2.80

449

Table 2 Electric energy usage of micro-CHP cooling components. Component Hot water pump Chilled water pump Cooling tower pump Heat recovery pump Absorption chiller Cooling tower fan Ofce fan-coil unit Shop fan-coil unit Total Electrical draw (kW) 1.50 0.75 1.12 0.75 0.21 1.50 0.20 1.00 7.03

Fig. 5. Estimated cooling load prole on a day in summer.

Table 3 Electric energy usage of micro-CHP heating components. Component Hot water pump Heat recovery pump Ofce fan-coil unit Shop fan-coil unit Total Electrical draw (kW) 1.50 0.75 0.20 1.00 3.45 Fig. 6. Estimated heating load prole on a day in winter.

used to estimate the cooling and heating loads of the test facility. TRNBuild, a building visual interface in TRNSYS, is used to characterize the multi-zone building model using the specications of the test facility. The multi-zone building model includes geometric and material information (e.g., size and type of walls and windows), internal gains (following the equipment and occupancy schedules), local weather information (e.g., ambient temperature, relative humidity, and solar radiation) in TMY21 data, and estimated inltration of the test facility. More detailed information on TRNSYS can be found in Ref. [5]. Particular days in summer, winter, and mid-season are selected to examine the cooling and heating load proles. The load proles are depicted in Figs. 57. Sudden increases and drops of cooling and heating energy are examined in Figs. 5 and 6 due to the internal heat gains (e.g., from occupants, lighting, and ofce equipment) during the ofce hours between 7 a.m. and 6 p.m. 3.3. Inputs and energy efciency constraints for the algorithm The costs of fuel and electricity are used as inputs to the algorithm. Electricity price of about 10 cents per kWh, which is about the national average in the U.S., is used in the simulation as well as a residential natural gas price of about 0.494 dollars per cubic meter (14 dollars per thousand cubic feet). The NG price is converted to 0.0463 dollars per kWh. More detailed information regarding the energy prices in the U.S. can be found in Ref. [12]. In order to be realistic, the fuel-to-electric-energy conversion efciency must correspond to the power output at each time step of simulation rather than remain constant for all power levels. The
1 The typical meteorological year 2 (TMY2) data is collected by the National Renewable Energy Laboratory (NREL) and is readily available to be used in a computer simulation. For more information, see Ref. [11].

Fig. 7. Estimated cooling and heating load prole on a day in mid-season.

fuel-to-electric-energy conversion efciency of the engine-generator package has been evaluated based on the experimental measurements of fuel consumption of the unit on site. A quadratic regression is used to determine the efciency function for the data. A function of the fuel-to-electric-energy conversion efciency (a1) can be determined as a1 0:0011x2 0:0287x5 0:0699 5 (20)

where x5 is the power output of the engine-generator package. The efciency data and curve t are illustrated in Fig. 8. This nonlinear constraint has been handled in the optimization using xed-point iterations. Basically, an initial fuel-to-electricenergy conversion efciency is assumed. Then the resulting power output (x5) from the LP through Eq. (20) is used to obtain a corrected fuel-to-electric-energy conversion efciency that in turn is used in a new LP simulation. This process is continued until the

450

H. Cho et al. / Energy and Buildings 41 (2009) 445451 Table 5 Operational conditions for each baseline case. Condition Importing electricity from the grid Possibly on Off On Engine operation On On Off Boiler operation Off Possibly on On

A B C

assess the performance of the optimal energy dispatch algorithm. The operational conditions for the baseline cases are as follows:  Condition A. The engine provides satisfactory amount of cooling energy required for the facility, and if the engine cannot provide enough electric energy demand, the difference is imported from the grid.  Condition B. The engine provides satisfactory amount of electric energy required for the facility, and if the engine cannot provide enough heat required by the absorption chiller to satisfy the cooling energy demand, the difference is covered by energy from the boiler.  Condition C. Electric and cooling energy demands are satised by using energy from the grid and boiler, respectively. In this case, the engine is not used. The cost of each condition can be easily evaluated using the energy ow relations and the efciency of each component for the

Fig. 8. Fuel-to-electric-energy conversion efciency of the engine-generator package.

efciency converges to a x value within a predetermined tolerance level. The entire process is repeated for each time step. Other nonlinear programming techniques can also be used but their implementation may be cumbersome due to the large number of variables used (i.e., 21 variables are used in the model). The ASHRAE handbook [13] provides the typical energy distribution for internal combustion engines. It suggests that 30% of the fuel energy is converted to heat energy rejected through the coolant and another 30% of the fuel energy is rejected as heat through the exhaust gas. The total efciency of heat exchangers for the coolant and exhaust gas is estimated to be 0.85, and the total fuel-to-thermal-energy conversion efciency (a2) (i.e., total heat recovered from the engine) is then calculated to be (30% + 30%) (0.85) = 51%. The boiler efciency (a3) is given as 90%. The total efciency of the cooling components (a4) can be estimated by considering the COP of an absorption chiller and the efciency of an air handling unit. A COP of 0.7 is used for the absorption chiller and an efciency of 0.85 is used for the air handling unit. The total efciency of cooling components is then calculated to be (0.7) (0.85) 100 = 60%. The total efciency of the heating components (a5) is estimated 85% which is an efciency of the air handling unit. The thermal energy losses due to energy transport/transmission in the network are neglected in this simulation because the pipes are well insulated in the facility. In case these losses are not negligible, they can be added to the total efciencies of the cooling and heating components (a4 and a5). The summary of input and efciency constraint information for the algorithm used in the scenario is listed in Table 4. 3.4. Baseline operation of the CHP system Results from a few baseline cases outlining different possible operating conditions for the CHP system can be considered to
Table 4 Summary of input and efciency constraint information for the algorithm. Cost ($/kWh) Input Cost of electricity imported from EG, c1 Cost of NG, c2 and c3 Cost of electricity exported to EG, c4 Efciency constraint Fuel-to-electric-energy conversion efciency of NG engine, a1 Fuel-to-thermal-energy conversion efciency of NG engine, a2 Efciency of the boiler, a3 Total efciency of the cooling components, a4 Total efciency of the heating components, a5

0.10 0.0463 0.02

Eq. (20) 0.51 0.90 0.60 0.85

Fig. 9. Partitioned energy supply based on the cost optimization.

H. Cho et al. / Energy and Buildings 41 (2009) 445451

451

A is evaluated based on fuel energy consumed by the engine and extra electricity imported from the grid, and Cost B is evaluated based on fuel energy consumed by the engine and boiler. Of course, these results are expected from the results of the energy partition shown in Fig. 9. 4. Conclusions An optimal energy dispatch algorithm has been developed for a CHP system. The algorithm provides the operational signals to the CHP system that results in minimum energy cost for the scenario considered. The LP model is formulated based on an intuitive network model of a CHP system. The energy dispatch algorithm has been used in simulations on a model of the micro-CHP system at MSU. The results from the simulations are compared to the results from a few baseline cases outlining different possible operation conditions. It shows that the algorithm provides the optimal cost throughout the simulation period. It is also seen that the fuel-toelectric-energy conversion efciency can be determined within the algorithm using a xed-point iteration procedure through the nonlinear efciency function. Finally the results from the case study illustrate the potential economical gain resulting from optimizing the operation of the CHP system. References
[1] Mississippi Micro-CHP and Bio-fuel Center, Cooling, Heating, and Power for Buildings (CHP-B) Instructional Module, Mississippi State, MS, USA, 2005, http://microchp.msstate.edu/pdf/chp-b_instructional_module.pdf. [2] Mississippi Micro-CHP and Bio-fuel Center, Micro-Cooling, Heating, and Power Instructional Module, Mississippi State, MS, USA, 2005, http://microchp.msstate.edu/pdf/m-CHP%20Instructional%20Module.pdf. [3] H. Cho, R. Luck, L.M. Chamra, Dynamic Simulation of a Micro-CHP Facility: A Case Study, ASME Paper No. ES2007-36156, Proceedings of 2007 Energy Sustainability Conference, Long Beach, CA, USA, 2007. [4] H. Cho, R. Luck, L.M. Chamra, Technical and Economical Analysis of a Micro-CHP Facility Based on Dynamic Simulation: A Case Study, ASME Paper No. IMECE200741885, Proceedings of 2007 ASME International Mechanical Engineering Congress and Exposition, Seattle, WA, USA, 2007. [5] Solar Energy Laboratory, TRNSYS 16 Manual, 2006. [6] R. Lahdelma, H. Hakonen, An efcient Linear Programming algorithm for combined heat and power production, European Journal of Operational Research 148 (2003) 141151. [7] A. Rong, R. Lahdelma, An efcient linear programming model and optimization algorithm for trigeneration, Applied Energy 82 (2005) 4063. [8] E. Thorin, H. Brand, C. Weber, Long-term optimization of cogeneration systems in a competitive market environment, Applied Energy 81 (2005) 152169. [9] R.K. Ahuja, T.L. Magnanti, J.B. Orlin, Network Flows: Theory, Algorithms, and Applications, Prentice Hall, Upper Saddle River, NJ, 1993. [10] N.A. Stone, Installation and Instrumentation of a Micro-CHP Demonstration Facility, Master Thesis, Mississippi State University, Mississippi State, MS, 2006. [11] Renewable Resource Data Center (RReDC) at the National Renewable Energy Laboratory (NREL), TMY2 Users Manual, 1996, http://rredc.nrel.gov/solar/old_data/nsrdb/tmy2/. [12] Energy Information Administration, U.S. Department of Energy, http://www.eia.doe.gov/. [13] 2008 ASHRAE HandbookHVAC Systems and Equipment, ASHRAE, Atlanta, GA, pp. 33 (Chap. 7).

Fig. 10. Optimal cost compared with the cost of baseline operational conditions.

energy requirements of the facility. The operational conditions are summarized in Table 5. 3.5. Test results of the algorithm The LP problem discussed in Section 2.2 has been implemented using the cooling energy demand shown in Fig. 5 throughout a particular day in summer. The corrected values of the fuel-toelectric-energy conversion efciency have been found through the xed-point iteration procedure discussed above. The simulation lasts 24 h with time steps of 1 min. The test results of the optimization with the nonlinear efciency constraint are presented in Figs. 9 and 10. The optimal input energy distribution, i.e., the electric energy imported from the grid, fuel energy supplied into the engine, and fuel energy supplied into the boiler, are presented in Fig. 9. It is seen from Fig. 9 that the engine is operated between 7 a.m. and 6 p.m., and only imported electricity and fuel into the boiler are used in the remaining time. This is expected as the efciency of the engine would be very low if operated in the early morning and evening hours on a summer day when the energy demand is low. Based on this energy partition, the rate of total optimal energy cost is shown in Fig. 10. The optimal cost is compared to the cost of baseline operational conditions described in Section 3.4. Cost A, B, and C are calculated based on Conditions A, B, and C, respectively. The results in Fig. 10 show that the optimal energy cost is indeed lower than the baseline cases throughout the day. As expected the optimal cost follows the path of Cost C in the morning and evening. As described in Table 5, Cost C is evaluated based on electric energy imported from the grid and fuel energy consumed by the boiler. It is interesting to see that the optimal cost follows the path of Cost A after 7 am and switch to the path of Cost B between around 10:30 a.m. and 6 p.m. As described in Table 5, Cost