A Multi-Objective Design Methodology for Hybrid Renewable Energy Systems R. Chedid, S. Karaki and A. Rifai Abstract — This paper describes a methodology to design a hybrid renewable energy system over a certain planning horizon. Traditionally a system plan was developed to achieve a minimum cost objective (MCO) while satisfying the energy demand, reliability, stability and battery constraints. The minimum emissions objective (MEO) is now an important target to achieve subject to the above mentioned constraints. Each of the above problems may be solved using linear programming, but minimizing the two preceding objectives at the same time forms a multi-objective problem which is solved by the ε-constraint and the goal attainment methods. The ε-constraint method minimizes the total cost while the emissions are less than a certain value ε determined by the linear programming when minimizing emissions only or by the designer. The goal attainment method tries to balance all the objectives and make them as close as possible to the initial goals determined by MCO and MEO. A case study is presented to illustrate the applicability and the usefulness of the proposed method. Index Terms-- Renewable energy, Hybrid systems, Multiobjective optimization. I. INTRODUCTION S olar and wind energy systems are among the most developed renewable energy systems (RES), and have been widely used in both autonomous and grid connected applications. Much research has been carried out to optimize their size and evaluate their performance. Chedid and Rahman [1] developed a linear programming model to optimize the size of a hybrid system with battery storage and diesel sets. However, the solution provided did not consider system’s expansion over a future horizon. Kellogg et al. [2] presented a numerical algorithm to determine the optimum size of system’s components for three different configurations: wind alone, photovoltaic (PV) alone and hybrid wind/PV. Karaki et al. [3] presented a probabilistic model of a stand-alone wind/PV power system. The model takes into consideration system stability, outages due to the primary energy fluctuations and hardware failure. Gavanidou and Bakirtzis [4] developed a multi-objective planning technique to design a R. B. Chedid is with the Faculty of Engineering and Architecture, American University of Beirut, PO Box 11-0236, Beirut, Lebanon, (e-mail: rchedid@layla.aub.edu.lb). S. Karaki is with the Faculty of Engineering and Architecture, American University of Beirut, PO Box 11-0236, Beirut, Lebanon, (e-mail: skaraki@layla.aub.edu.lb). A. Rifai is a graduate student with the Faculty of Engineering and Architecture, American University of Beirut, PO Box 11-0236, Beirut, Lebanon. hybrid system based on minimization of both capital investment and loss of load probability (LOLP). They applied the trade off/risk method which rejected inferior plans and gave a set of robust scenarios to the designer. Protogeropoulos et al. [5] tried to determine the optimum size of a hybrid system and to assess its economical and technical merits against single PV and wind standalone systems. Morgan et al. [6] described the development of a simulation program that enables the designer to find the reliable level of a renewable energy system. When both load and meteorological data are known, the program calculates the system autonomy level and predicts the battery voltage. II. PROBLEM DEFINITION AND MATHEMATICAL FORMULATION The problem is to develop a multi-objective model to design a hybrid PV-wind system with battery storage and diesel generators taking into consideration future system expansion. The design objectives are cost and gas emission minimization, which include minimization of carbon dioxide (CO2), sulfur dioxide (SO2) and nitrogen oxides (NOx) from diesel units over the life-time of the project. The problem constraints are:(1) reliability constraint which dictates that a certain percentage of the peak demand must be secured as a reserve, (2) energy balance constraint, (3) stability constraint to limit the renewable capacity to a certain percentage of the peak demand, and (4) battery constraint, which determines the upper size of the batteries installed each year. As mentioned above, the problem has 2 categories of objectives: cost and emission minimization, and is solved subject to a set of constraints: A. Cost function: The cost objective can be written as follows: min[T wlife +t −1] X wt (Cw − S wt )Wt + ∑ α w Ow LFn H Wn n =t min[T slife +t −1] + X st (Cs − S st )Wt + (1) ∑ α s Os LFn H Wn n =t T min[T dlife +t −1] ∑ t =1 + X dt (Cd − Sdt )Wt + ∑ α d Od LFn H Wn n =t min[T blife +t −1] η − 1 + X bt (Cb − Sbt ) Wt + dod α b Ob LFn H Wn ∑ 2 n =t + X rt (Cr − Srt )Wt Where: - X’s represent ratings of system’s units installed in year t. Authorized licensed use limited to: IEEE Xplore. Downloaded on March 19, 2009 at 11:16 from IEEE Xplore. Restrictions apply. - C’s, O’s and S’s are the capital costs, operation and maintenance costs and salvage values of system’s units respectively. - wlife, slife, dlife, blife and rlife are the life cycles of system’s units. α w , α s , α d and α b - are the resource availability of system’s units. - LFn is the load factor in year n. H is the number of hours per year. - dod and η are the deep discharge limit and the battery efficiency respectively. - Wt is the present worth factor of year t. B. Emission functions: The emissions of CO2, SO2 and NOx are of interest in this work. For example, the emission function for CO2 can be expressed as follows: min[T dlife+ j −1] X H LFi FCO2 Dc ∑ ∑ dj j =1 i= j T where Et is the energy demand. For example, the energy produced by the wind turbine in the first year is X w1 α w LF1 H . The energy extracted from the battery is modeled as follows. The battery will be charged during light load and discharged during peak load. Assuming that the battery is cycled each 12 hours, the energy supplied to the demand during the peak period, from 0 to H / 2 , is written as follows: E w1 + E s1 + E d 1 + E bd = D '1 H LF1 2 H η dod 2 Ebd = X b (5) (6) where - E w1 , E s1 and E d 1 are the energy generated by WT, PV and DU respectively during the peak period. - E bd is the discharge energy from the battery. ' (2) - D 1 is the peak demand in the peak period and LF1 is the load factor during the same period. Where FCO2 is the weight of CO2 emitted due to burning 1kg fuel; Dc is the fuel consumption of the Diesel unit in kg/ kWh. Similar factors are used to calculate the quantities of SO2 and NOx emitted. During the light load period, from H / 2 to H , the energy supplied to the demand can be written as below: [kg] C. Planning and Operational Constraints Constraints are needed to ensure the stability of the system, avoid over-sizing of system components and guarantee demand supply at any time. i. Reliability Constraint: E w 2 + E s 2 + E d 2 − Ebc = D ' 2 H LF2 2 H dod 2 Ebc = X b where - Ew2 , Es 2 (7) (8) and Ed 2 are the energy generated by WT, PV and DU during light load period. ' These constraints ensure that the available capacity is higher than the load thus allowing for a given reserve margin R: t ∑X n =1 wn t t t n =1 n =1 n =1 α w + ∑ X snα s + ∑ X dnα d + ∑ X bn α b t + ∑ X rn α r ≥ D0 × (1 + inc ) (1 + R ) t - D 2 is the peak demand in this period and LF2 is the load factor during the same period. - E bc is the energy charged to the battery during the same period. (3) t = 1, T n =1 where D0 is the present peak demand, inc is the rate of load increase per year and R is the reserve. From the above equations, we can get the energy flow from the battery as below η − 1 Eb = Ebd - Ebc = X b H dod 2 (9) The energy produced by PV panels and diesel units is calculated in the same way. ii. Energy Constraint: The energy generated by all system units has to demand at all times. t t t × + + α α X LF X LF X dn α d LFt ∑ ∑ t sn s t ∑ wn w n =1 n =1 n =1 t η −1 dod LFt ∑ X bn α b 2 n =1 t = 1, T equal the iii. Renewable energy contribution limit: For reliability purpose, the whole system at any time will not operate with more than 40% of renewable contribution. + The concern here is to prevent system collapse due to stability H = E t considerations when suddenly loosing the renewable resource. t t t X + X sn ≤ Rl D0 (1 + inc ) , t = 1, T (10) wt (4) ∑ n =1 ∑ n =1 Authorized licensed use limited to: IEEE Xplore. Downloaded on March 19, 2009 at 11:16 from IEEE Xplore. Restrictions apply. where Rl is the maximum limit for renewable contribution in the whole system. In [7] Rl is taken from 30% to 40% of the the following load curve and increasing by 7% each year as shown in Fig.2 and Fig. 3. demand. In this thesis, it is taken equal to 40%. If the load does not require quality supply, renewable resources should be utilized as possible and therefore this constraint should be removed. However, if the load requires quality supply as in this thesis, Rl must be set to the mentioned values. After setting Rl to 40%, and referring to the energy constraint, it is found that renewable units supply around 10 % of the load which is acceptable and the role of renewables is fuel saver. Fig.1. Ksara Wind Speed Distribution. iv. The Battery constraint: It is assumed that the battery capacities must be less than the difference between the maximum load and the minimum load during the worst day in the year. The ratio of minimum load over maximum load Min / Max is constant during all the project period because the load is increasing at a constant rate. Taking into account the deep discharge limit dod for lead-acid battery and its efficiency, the size of batteries is limited as follows: t ∑X n =1 ≤ (1 + dod ) × D0 × (1 + inc ) (1 − Min / Max ) / η , t bn (11) t = 1, T III. SOLVING APPROACH The practical approach followed to reach a solution is based on three steps: First, LP is used to find a system with minimum cost but with no consideration for the emissions. Then LP is used again to find the system with minimum emissions without any limit on cost. Since these objectives are conflicting in nature, any improvement in one objective will lead to a degradation of the other objective. Many techniques were developed to solve multi-objective problems including the ε-constraint and the goal attainment methods. The ε-constraint method is used where the objective is to minimize the total cost but the emission objectives are now added to the above described constraints. There is a difficulty in choosing the best εi, i=1,2,3, for the emission objectives in order to get the best solution. Applying goal attainment method to this problem will give a solution more tolerant than the ε-constraint method. Fig. 2. Load Duration Curve. Demand 1200 1000 80 K 60 40 20 0 1 The above described methodology was applied to a hypothetical site in Lebanon whose wind speed distribution is shown below. The availability of solar resource is assumed to be equal to 25 percent with maximum solar irradiance equal to 611.5 Wh/m2. The load to be met is 500 kW varying according to 5 7 yea 9 Fig. 3. A 10 year demand forecast. The capital and operation and maintenance costs for all system components are listed in the appendix A. The operation and maintenance cost for diesel generators is calculated according to the following equation: O&M = IV. SIMULATION RESULTS 3 FixedO & M * Rated + VariableO & M * Rated * CF * Hours Rated * CF * hours (12) Where:- Rated is the rated power for the generator. - CF is the capacity factor of the generator and it is assumed to be equal to 0.6. - Hours is the number of hours per year. The fuel cost is added to the above O & M to get the total operation and maintenance cost. Authorized licensed use limited to: IEEE Xplore. Downloaded on March 19, 2009 at 11:16 from IEEE Xplore. Restrictions apply. The wind resource availability is calculated in the usual way using the Weibull distribution of wind speed in conjunction with the turbine power characteristics. A. Minimizing Cost Only (MCO) Using LP The results of minimizing cost only using LP, without any constraints on the emissions are shown in Fig4. The emissions in this case are, as expected lower than in the MCO problem, at 28.3 thousand Tons. C. ε-Constraint Method In order to minimize the total cost and the emissions in the same time, the ε-constraint method is applied. First of all ε, our emission target, has to be determined by plotting the curve of cost versus ε as shown in Fig. 6. Installed Units Each Year 900 800 700 Reserve KW 600 Battery 500 Diesel 400 Solar 300 Wind 200 100 0 1 2 3 4 5 6 Year 7 8 9 10 Fig.4. MCO Base Case. As seen, no PV units are added because wind power is less expensive. Also, the addition of wind turbines is limited by the wind resource contribution constraint, and no batteries are adopted because of their high cost. The minimum present worth cost is $4.7 millions, which corresponds to an energy price of $0.13/ kWh. The CO2 emissions were at 28.9 thousand Tons. Decreasing the cost of wind turbine by 20% does not cause an increase in the added capacity of WT because the stability constraint (8) is biding in this case preventing any new renewable resource from being added. B. Minimizing Emission Only (MEO) Using LP The results of minimizing emissions only using LP, without any constraint on the cost, are shown in Fig 5. Fig. 6. Cost Versus Emission Rates. As seen, we have higher cost with a lower discount rate and vice-versa. Also the program starts to give a feasible solution when ε is equal to 0.80894 kg of CO2 per Kwh, 0.0097 for SO2 and 0.001166 for NOx. At this limit, the base case total cost is $49.591m. If the renewable contribution limit is relaxed to 75%, the model produces feasible solutions after 0.7305 kg of CO2 per kWh (Fig. 7). But as a result, the total cost increases as shown in Fig. 7. Installed Units Each Year 1200 1000 Reserve KW 800 Battery Diesel 600 Solar 400 Wind 200 0 1 2 3 4 5 6 Year 7 8 9 10 Fig. 5. MEO Base Case. Here, the model takes less diesel units in order to reduce emissions. To compensate for the energy lost, solar panels rather than wind turbines are adopted . This is because the availability of the solar resource is 25% whereas the availability of wind is around 21%. As expected the cost here is higher at $5.8 millions, which corresponds to $0.17/ kWh. Fig. 7. Relaxing Renewable contribution limit. Authorized licensed use limited to: IEEE Xplore. Downloaded on March 19, 2009 at 11:16 from IEEE Xplore. Restrictions apply. From Fig. 6, it is obvious that minimum cost and minimum emissions are obtained when ε is equal to 0.80894 kg of CO2 per kWh. The base case capacity additions are shown in Fig. 8. lower. This is indicated in the cost of the GA method being lower than the cost in the ε-constraint. Also the model installs more PVs than WTs because the availability of solar power is higher than that of wind power. In contrast to the LP and ε-constraint methods, the goal attainment adds batteries. V. CONCLUSION Fig. 8. ε-Constraint Base Case. It is to be noted that the total cost is higher than that of minimizing cost only objective due to higher installation of PVs, and is less than that of minimizing emissions only. The emissions are the same as those resulting from the base case of minimizing emissions only because ε is set at its minimum value that gives feasible solution. D. Goal Attainment Method The goal attainment method (GA) tries to balance between the emissions and the cost objectives. The results of goal attainment show that more emissions are obtained as compared with MEO and ε-constraint method by a certain percentage and the cost is increased as compared with MCO by the same percentage which is the attainment factor. The results of the GA method in the base case are shown in Fig. 9. This paper has presented a methodology to design a hybrid wind photovoltaic system with diesel generators and battery banks over a certain planning horizon. First, linear programming is used to minimize the total system cost while satisfying the energy demand, reliability, stability and battery constraints. The results of this technique are an optimum set of wind turbines, photovoltaic panels, diesel generators and battery banks. Second, linear programming is used to minimize the total emissions of carbon dioxide (CO2), sulfur dioxide (SO2) and nitrogen dioxide (NOx) subject to the constraints used in the first exercise. The results make another optimal hybrid renewable energy system. Minimizing all the precedent objectives at the same time forms a multi-objective problem which is solved by the εconstraint and the goal attainment methods. The ε-constraint method minimizes the total cost while meeting the same constraints but with total emissions being less than a certain value ε determined by the linear programming when minimizing emissions only or by the designer. The goal attainment method tries to balance all the objectives and make them as close as possible to the initial goals determined by the minimum cost and minimum emissions sub-problems. VI. Installed Units Each Year 900 800 700 Reserve KW 600 Battery 500 Diesel 400 Solar 300 Wind 200 100 0 1 2 3 4 5 6 Year 7 8 9 10 Fig. 9. Goal Attainment Base Case. First, the model installs more diesel units than those of the ε-constraint method which increases the emissions and reduces the total cost as expected. The DUs added in all years according in the ε-constraint method are the maximum allowable for the specified emissions. In the GA, the specified emission constraints are slightly relaxed and more DUs are added in this case. The rest of demand is supplied from WTs and PVs rather than PVs alone as in the εconstraint method because the cost of such a combination is APPENDIX Parameter Wind turbine price Solar panel price Diesel engine price Battery price Wind O&M Solar O&M Diesel total O&M Battery O&M Project life span Wind turbine life span Solar panel life span Diesel life span Battery life span Battery efficiency Battery depth of discharge Fuel cost Discount rate Escalation rate Carbon % in diesel Sulfur % in diesel Nitrogen % in diesel Value 1015 $/KW 4401 $/KW 813 $/KW 1600 $/KW 0.01 $/KWh 0.0046 $/KWh 0.1402 $/KWh 0.02 $/KWh 10 years 20 years 20 years 7 years 5 years 0.85 0.45 1.82 $/ gallon 0.1 0.03 86.4 2 0.1 VII. REFERENCES [1] [2] R. Chedid and S. Rahman:’Unit Sizing And Control Of Hybrid WindSolar Power Systems’, IEEE Transactions On Energy Conversion, Vol.12, No.1, March 1997. W. Kellogg, M. Nehrir, G. Venkataramanan and V.Gerez:’Generation Unit Sizing And Cost Analysis For Stand- Alone Wind, Photovoltaic, And Hybrid Wind/PV Systems’, IEEE Transactions On Energy Conversion, Vol.13, No.1, March 1998. Authorized licensed use limited to: IEEE Xplore. Downloaded on March 19, 2009 at 11:16 from IEEE Xplore. Restrictions apply. [3] [4] [5] [6] [7] S. Karaki, R. Chedid and R. Ramadan:’ Probabilistic Performance Assessment Of Autonomous Solar-Wind Energy Conversion Systems’, IEEE Transactions On Energy Conversion, Vol.14, No.3, March 1999. E. Gavanidou and A. Bakirtzis; Design of a Stand Alone System with Renewable Energy Sources Using Trade-Off Methods. IEEE Transactions On Energy Conversion, Vol.7, No.1, March 1992. C. Protogeropoulos, B. Brinkworth and R. Marshall:’ Sizing and Techno-Economical Optimization For Hybrid Solar Photovoltaic/Wind Power Systems With Battery Storage’, International Journal Of Energy Research, Vol.21, 1997. T. Morgan, R. Marshall and B. Brinkworth:’ “ARES”- A Refined Simulation Program For The Sizing And optimization Of Autonomous Hybrid Energy Systems’, Solar energy, Vol.59, Nos.4-6, 1997. G. Seeling-Hochmuth,:’ A Combined Optimization Concept For The Design And Operation Strategy Of Hybrid-PV Energy Systems’, Solar Energy, Vol.61, No.2, 1997. VIII. BIOGRAPHIES Riad Chedid was born in Lebanon in 1960. He got his M.Sc degree (with distinction) in electrical engineering from Moscow Power Engineering Institute in 1986, and his Ph.D from Imperial College of Science, Technology and Medicine, U.K. in 1992 He joined the American University of Beirut in 1992 where he is currently a Professor of Electrical and Computer Engineering. His research interests include energy planning and policy, modeling of renewable energy systems, and numerical simulation of electromagnetic fields. Sami H. Karaki is professor of electrical engineering at the American University of Beirut (AUB), Lebanon. He joined AUB in 1991 and contributed to the development of its Electric Power Engineering program. From 1981 to 1990 he was with Kuwait Institute for Scientific Research, Kuwait where he contributed to two regionally leading projects on the power system interconnection of Arabic countries. He obtained his BE from AUB in 1975 and his Ph.D. from the University of Manchester Institute of Science and Technology, UK, in 1980. He is presently teaching courses in digital systems design, computer programming, power electronics, and power systems. His main research interests are in modeling and analysis of renewable energy systems, power system planning under competition, short-term load forecasting, and artificial intelligence applications in electric power systems. Ahmad Al-Rifai was born in Lebanon (Arsal) in June 20, 1977. He obtained his Bachelor of Electrical Engineering from Damascus University in 2001. He graduated with Honors and was awarded the Bassel Assad Prize in 1999-2000. He obtained his Master of Engineering degree: major Electrical Engineering in June 2004 from AUB. He is currently working with Saudi Oger in the high-current design office in Riyadh, Saudi Arabia. Authorized licensed use limited to: IEEE Xplore. Downloaded on March 19, 2009 at 11:16 from IEEE Xplore. Restrictions apply.