Solution of Combined Economic Emission Dispatch Problem with Valve-Point Effect Using Hybrid NSGA II-MOPSO Solution of Combined Economic Emission Dispatch Problem with Valve-Point Effect Using Hybrid NSGA II-MOPSO

This chapter formulates a multi-objective optimization problem to simultaneously mini- mize the objectives of fuel cost and emissions from the power plants to meet the power demand subject to linear and nonlinear system constraints. These conflicting objectives are formulated as a combined economic emission dispatch (CEED) problem. Various meta-heuristic optimization algorithms have been developed and successfully implemented to solve this complex, highly nonlinear, non-convex problem. To overcome the shortcomings of the evolutionary multi-objective algorithms like slow convergence to Pareto-optimal front, premature convergence, local trapping, it is very natural to think of integrating various algorithms to overcome the shortcomings. This chapter proposes a hybrid evolu- tionary multi-objective optimization framework using Non-Dominated Sorting Genetic Algorithm II and Multi-Objective Particle Swarm Optimization to solve the CEED prob- lem. The hybrid method along with the proposed constraint handling mechanism is able to balance the exploration and exploitation tasks. This hybrid method is tested on IEEE 30 bus system with quadratic cost function considering transmission loss and valve point effect. The Pareto front obtained using hybrid approach demonstrates that the approach converges to the true Pareto front, finds the diverse set of solutions along the Pareto front and confirms its potential to solve the CEED problem.


Introduction
In order to operate the power system economically and also to protect the environment from pollution the power system operator has to carry out optimal scheduling of active power to simultaneously minimize the fuel cost and the emissions from the fossil fuel-fired power plants.
These objectives are desirable to obtain great economic benefit [1] and to reduce the nitrogen oxide (NO x ), sulfur oxide (SO x ) and carbon dioxide (CO 2 ) pollutants which cause harmful effect on human beings [2]. These conflicting objectives can be formulated as a multi-objective combined economic emission dispatch (CEED) problem. This CEED problem can be solved using traditional mathematical programming techniques such as lambda iteration, gradient search [1] and can also be solved using modern heuristics optimization techniques. The numerous advantages of solving the CEED problem using heuristic optimization methods compared to the traditional mathematical programming techniques are they are population-based, do not require any derivative information, do not use gradient information in search process, use stochastic operators in search process, they are simple to implement and flexible, have inbuilt parallel architecture and they are scalable and are also computationally quick.
A single optimal solution cannot be obtained for a multi-objective CEED problem which simultaneously minimizes the conflicting objectives of fuel cost and emission. Thus the simultaneous minimization of conflicting objectives in a multi-objective optimization problem (MOP) gives rise to a set of tradeoff solution called as Pareto-optimal (PO) solutions [3] which needs further processing to arrive at a single preferred solution. In literature domination based framework using multi-objective evolutionary algorithms (MOEA) which simultaneously minimizes the fuel cost and emission have been employed to solve the CEED problem. These population-based approaches can obtain the multiple non dominated solutions in a single simulation run. These non-dominated solutions portray the tradeoff between fuel cost and emission objectives of CEED problem. Modern meta-heuristic optimization algorithms like Genetic Algorithm [4,5], Biogeography Based Optimization [6], Particle Swarm Optimization [7], Bacterial Foraging Algorithm [8], Scatter Search [9], Teaching Learning Based Optimization [10], Differential Evolution [11] and Harmony Search Algorithm [12] have been developed and successfully implemented to solve this complex, highly nonlinear, non-convex CEED problem.
The multiple objective CEED problem can also be transformed into a single objective problem using a weighted sum approach and h parameter values. The h parameters are used to overcome the dimensionality problem when combining multi-objectives and the converted single objective problem is then solved using evolutionary algorithms [13][14][15]. Another technique to solve CEED problem without the h parameter is to normalize the fuel cost and emission components [6] and solve the single objective function using evolutionary algorithms (EA). In these approaches for the chosen value of weights will give one particular PO solution at a time. However, the disadvantage of these methods is that it requires multiple runs to find the set of PO solutions.
Each evolutionary algorithm has its own characteristics and merits; therefore it is natural to think of integrating these different algorithms to handle a complex problem like CEED. In the research field of Evolutionary Algorithms merging of two or more optimization algorithms into a single framework is called hybridization. In [16][17][18][19][20][21] hybrid multi-objective optimization algorithms have been successfully applied to solve CEED, various complex engineering problems, and standard test functions. The results indicate that the hybrid algorithms are effective, can exchange elite knowledge within the hybrid framework, can do parallel processing, can improve the exploration and exploitation capabilities and can yield more favorable performance than any single algorithm.
In order to obtain a globally optimal solution without being trapped in local optima requires a tradeoff between exploration and exploitation task in the search process. Exploration phase in any algorithm is important to search every part of the solution domain to provide an estimate of the global optimal solution. On the other hand exploitation phase in any algorithm is important to improve the best solutions found so far by searching in their neighborhood. In this chapter, a hybrid framework using Non-Dominated Sorting Genetic Algorithm II (NSGA II) [22] and Multi-objective Particle Swarm Optimization (MOPSO) [23] is used to solve the CEED problem. This hybrid framework integrates the desirable features of the NSGA II and MOPSO while curbing their individual flaws. These population-based approaches use different techniques for exploring the search space and when they are combined will improve the tradeoff between the exploration and exploitation tasks to converge around the best possible solutions. The main purpose of this hybridization technique is to obtain a well-spread and well-diverse PO solution. When the proposed hybrid algorithm is used to solve the highly complex CEED problem the PO solution is obtained in less number of iteration and is also computationally fast when compared to MOPSO.
The rest of the chapter is organized as follows. The next section formulates the CEED problem. In Section 3, the transmission loss handling procedure and the constraint handling procedure is explained. In Section 4 the short review of NSGA II and MOPSO is provided. Section 5 is devoted to explaining the hybrid algorithm. In Section 6 the hybrid algorithm is applied on standard IEEE 30 bus systems and it also discusses the simulation results. Finally, the conclusion is drawn in Section 7.

Formulation of combined economic emission dispatch (CEED) problem
The combined economic emission dispatch problem has two conflicting objectives. The first objective can be stated as determining the optimal power generation schedule from a set of online generating units to satisfy the load demand subject to several physical and operational constraints to minimize the fuel cost. The second objective can be stated as determining the optimal power generation schedule from a set of online generating units to satisfy the load demand to minimize the pollutant emissions produced by the generating units. Both the conflicting objectives have to be minimized at the same time because operating the system with minimum cost will result in higher emission and considering only the minimum environmental impact is not practical which results in high production cost of the system. This section formulates the objective functions of the CEED problem along with equality and inequality constraints to maintain rigorous standards to meet the practical requirements of the power system. The goal of this chapter is to find the Pareto-optimal solutions of the CEED problem which minimize both these objectives subject to constraints. The mathematical formulation is as follows.

Objective functions of CEED problem
The general formulation for a multi-objective optimization problem (MOOP) is to minimize the number of objective functions simultaneously. A general mathematical model is represented as follows [21]: where f x ð Þ represents the vector of objectives and f i x ð Þ, i ¼ 1, 2, ⋯, m is a scalar decision variable which maps decision variable x into objective space f i ¼ R n ! R: The n-dimensional variable x is restricted to lie in a feasible region D which is constrained by j in-equality constraint and k equality constraint, i.e.
The decision variable x can be written more suitably as where T is the transposition of the column vector to the row vector. The decision variables are restricted to take a value within a lower x bounds. These bounds are called the decision space [3].
In MO CEED problem the number of objectives m ¼ 2.The mathematical model of CEED is represented as follows: subject to power balance equality constraints h x ð Þ and bounds. The function f 1 x ð Þ represents the minimization of total fuel cost function and the function f 2 x ð Þ represents the minimization of the emissions from the fossil fuel fired plants. The decision variable x consists of the real power generation of the n generating units and can be written as where Pg i is the real power output of the i th generator.
Power plants commonly have multiple valves that are used to control the power output of the units. In a practical generating unit, when steam admission valves in thermal units are first opened, a sudden increase in losses is registered which results in ripples in the cost function. In order to model these ripples accurately, sinusoidal functions are added to the quadratic cost function [24]. The resulting cost function contains higher order nonlinearity and makes the problem non-differentiable and non-convex. Hence there are two versions of the fuel cost function, the quadratic function represented by f 1 x ð Þ and the combination of quadratic and a sinusoidal (valve-point) function represented by f 1, V x ð Þ . The two versions of the fuel cost functions are given below where a i , b i , c i represent the cost coefficients of the generator i. e i and f i are coefficients to model the effect of valve point of the generator i.
The second objective function f 2 x ð Þ is an emission function which takes into account the major pollutants caused by the fossil fuel fired power plants. The main pollutants from the power plants are the sulfur oxides and nitrogen oxides. The sulfur oxide emissions are proportional to the fuel consumed by the power plants and have the same form as that of the fuel cost function given by (6). The sulfur oxide emission function can be stated as follows [7].
The nitrogen oxides emissions are difficult to evaluate as the nitrogen is available in air and also in the fuel. The production of nitrogen gas is related to boiler temperature and air content. The modeling of the nitrogen oxides consists of straight lines and exponential terms. The nitrogen oxides emission function can be stated as follows The total emission function is obtained by adding the coefficients of (8) and (9) which gives the combination of the mixture of sulfur oxides and nitrogen oxides pollutants [7]. The total emission function can be stated as follows The total emission function given by (10) has a quadratic term and an exponential term which makes the function highly nonlinear. In (10) α i , β i , γ i , η i , δ i are the emission coefficients of the generator i. The modeling of the emission function is very important because according to the Amendments of the Clean Air Act regulatory agencies might decide to limit power plant emission in the areas where there are high concentrations of harmful contaminants.

Active power balance equality constraint and bounds
In order ensure that the total real power generation exactly match with the total load demand Pd and transmission loss Pl in the system a power balance equality constraint given in (11) should be satisfied.
The transmission losses in the power network are function of Pg and can be represented using B-matrix coefficients (Kron's loss formula [1]) as follows where B ij , B 0i , B 00 are transmission loss coefficients. There are instances in literature where the power losses in the system is neglected and the power balance equation given by (11) is curtailed as follows The above equations given by (11) and (13) are most common form of power balance equation found in the literature.
The power output of each generator i should lie within its minimum limit (Pg min i ) and maximum limit (Pg max i ) given by

Combined economic emission dispatch
The purpose of the CEED problem is to determine the Pareto-optimal real power generation Ã T that minimize the two conflicting objective given by (7) and (10) while satisfying the real power equality constraint given by (11) and the bounds given by (14). The bi-objective CEED problem can be formulated as In MO CEED problem, the economic and emission objectives will conflict with each other and is not possible to satisfy them simultaneously. There is no way of improving these objectives without degrading at least one of these objectives and the resulting set of non-dominated solutions thus obtained are called Pareto-optimal set. The objective function values of all elements in the PO set in the objective space constitute the Pareto front. When the sufficient number of PO solutions is available for the CEED problem then it is possible to find a convex curve containing these solutions to produce the Pareto front. The two main goals of MO CEED problem: 1. Find a set of non-dominated solutions which lie on the Pareto-optimal front 2. Find a wide spread of non-dominated solutions to represent the entire range of the Paretooptimal front.

Constraint handling mechanism
At any stage of the algorithm whenever a new population is being generated it is very important to make sure that the population lies within the decision space. While solving the CEED problem this implies that the population should satisfy the equality constraints and bounds. If the transmission losses are neglected than the k th variable of the candidate solution Pg k can be calculated by subtracting the sum of the power generations (excluding the Pg i from the power demand Pd. If the power transmission losses are considered, to determine Pg k and to maintain the equality constraint becomes hard. It is done using the following steps.
Step 1. Update the variables belonging to the set α n by normal optimization process of an evolutionary algorithm.
Here rand is a uniformly distributed random number in the range of 0; 1 ½ . The set α n contains all the integers in the range 1; n ½ except k, where k is a randomly generated integer which lies in the range of 1; n ½ Step 2. If updating of the variables is carried out using any other technique then regulate the updated variables which violate the lower bounds as Pg i ¼ Pg min i ; i ∈ α n . Regulate the updated variables which violate the upper bounds as Pg i ¼ Pg max i ; i ∈ α n .
Step 3. Obtain the value of the k th variable of the candidate solution Pg k by solving the following quadratic equation (17) whose coefficients are associated with the variables belonging to the set α n and the transmission loss coefficients [7]. To improve the potential candidate solution and also to improve the flexibility and diversity of the optimization algorithm the value of k is randomly generated integer between 1 and n.
Out of the two roots of the quadratic equation (17), one root will be selected as the value of the variable Pg k using the following procedure. If both the roots of the quadratic equation lie within the bounds then the root which has the minimum value is selected. If only one root lies within the bounds, this root is selected as the value of Pg k and the other root which lies outside the bounds is neglected. If both the roots lay outside the bounds the value of Pg k is set equal to Pg min k .
Step 4. Calculate the residue P RD by subtracting the total system demand Pd and the total system transmission loss Pl from the sum of the total power generation P n i¼1 Pg i .
If P RD j j< tol, then go to step 7; otherwise go to step 5. Here, tol is the demand tolerance usually set as 0:001 p:u: Step 5. Recalculate Pg i using Eq. (16).
Step 6. Repeat step 3, step 4 and step 5 until P RD j j< tol. This step will ensure that the candidate solution will always lie within the decision space.
Step 7. Stop the constraint handling procedure. The main purpose of this constraint handling mechanism is to increase the flexibility and diversity of the algorithm and to make sure that the candidate solution generated at any point of the algorithm always lies within the decision space.

NSGA II and MOPSO algorithms for solving CEED problem
Several Evolutionary Multi-objective (EMO) algorithms like NSGAII, MOPSO, SPEA 2 (Strength Pareto Evolutionary Algorithm), GDE 3 (Generalized Differential Equation) have been designed and used in solving numerous complex real word problems involving two or more objectives. All these algorithms can find the multiple Pareto-optimal solutions in a single run. Out of all these available algorithms, two of the widely used reliable methods for solving bi-objective optimization problems are the NSGA II and MOPSO. This section provides the review of these two EMO algorithms.
NSGA II was proposed in [22] as an improvement of the NSGA proposed in [25]. This NSGA II algorithm was the revised version of NSGA to overcome the following criticisms: • Computational complexity associated with non-dominated sorting.
• Lack of elite-preserving strategy.
• Lack of maintaining diversity among obtained solutions.
The NSGA II algorithm is very efficient for solving multi-objective optimization problems since it incorporates an efficient elitism preserving technique using non-domination sorting. The population is ranked based on non-domination sorting before the selection is performed. All non-dominated individuals are classified into one category. Another layer of non-dominated individuals are considered after the group of classified individuals are ignored. This process is continued until all individuals in the population are classified. NSGA II also uses a mechanism for preserving the diversity and spread of the solutions without specifying any additional parameters (NSGA uses fitness sharing). This crowding distance operator guides the selection process towards a uniformly spread out Pareto-optimal front. The NSGA II algorithm for solving the CEED problem is stated below: • Specify the parameters for the CEED problem In order to handle multiple objectives Pareto dominance is incorporated into PSO algorithm and the MOPSO algorithm is proposed in [23]. The algorithm proposed in [23] uses an external repository of particles to keep a record of the non-dominated vectors found along the search process. At each generation, for each particle in the swarm, by using Roulette wheel selection, a leader is selected from the external repository. This leader then guides other particles towards better regions of the search space by modifying the flight of the particles. A special mutation operator is applied to the particles of the swarm and also to the range of each design variable of the problem to be solved to improve the explorative behavior of the algorithm. The value of the mutation operator is decreased during the iteration. To produce well spread Pareto fronts the MOPSO algorithm in [23] uses an adaptive grid. The MOPSO algorithm for solving the CEED problem is stated below: • Specify the parameters for the CEED problem

Hybrid NSGA II and MOPSO algorithm for solving CEED problem
The mechanism of the proposed hybrid approach for solving the CEED problem is to integrate the desirable features of NSGA II (retaining the elitism feature) and MOPSO (exploitation capability) while curbing the individual flaws (NSGAII--does not have an efficient feedback mechanism, PSO overutilization of resources). The mechanism to explore the search space differs in both the algorithms. GA uses mutation and crossover operators which will enhance the exploration task of the hybrid algorithm. The particles in PSO are influenced by their own knowledge and information shared among swarm members. PSO enhances the exploitation task of the hybrid algorithm by finding better solutions from the good ones by searching the neighborhood of good solutions. In this hybrid algorithm at every generation, the Pareto dominance of the population is computed and based on these values non dominated sorting is performed [19]. In order to avoid premature convergence, the elite upper half of the population are enhanced by NSGA II algorithm while the lower half of the population are considered as swarm particles and are optimized by MOPSO to make them converge around the best possible solutions. The hybrid NSGA II-MOPSO algorithm for solving the CEED problem is stated below: • Specify the parameters for the CEED problem

Numerical tests
In order to validate the proposed hybrid algorithm, the CEED problem was solved for IEEE 30bus system and the results are presented in this section. The fuel cost coefficients with valvepoint loading, emission coefficients, and generator limits are adapted from [26] and is given in Table 1. The transmission loss B-matrix coefficients are obtained by running a load flow program and is in [26] is adapted here and given in Table 2. The total power demand in the system is 2:834 p:u: to the base of 100 MVA. Program in MATLAB was developed for the Hybrid Algorithm to perform CEED and executed on 1:60 GHz, Intel T2050 processor, 1:5 GB RAM HP Pavilion Laptop with WINDOWS 7 operating system. Various test cases are considered to compute the Pareto front of the multi-objective CEED problem. The Pareto-optimal front is obtained using the NSGA II algorithm and also using the MOPSO algorithm given in Section 4. The Pareto front obtained from the hybrid approach given in Section 5 is then compared with the Pareto front obtained using NSGAII and MOPSO algorithm.
In case 1 the fuel cost function is modeled as a quadratic function with sine term to incorporate the valve-point effect. The transmission losses are also considered in this case. The Pareto front obtained using NSGA II, MOPSO, and Hybrid NSGAII-MOPSO is shown in Figures 1, 2 and 3 respectively. In all these figures there is a discontinuity in the Pareto front due to modeling of the valve point loading effect of generators.     Table 3. We can observe from Figure 2 and Table 3 that there are difficulties in MOPSO algorithm in obtaining well spread Pareto front and also very slow convergence to the Pareto front when compared to NSGA II. This can be improved if the proposed hybrid approach is used to solve the CEED problem.
The Parameter setting for the hybrid algorithm is same as those given above expect for the settings provided here Population Size nPop ¼ 200; Maximum number of iteration MaxIt ¼ 50; Repository size nRep ¼ 20. The extreme points of the Pareto front and time for execution of the proposed NSGAII-MOPSO hybrid algorithm are provided in Table 3. From Table 3 it is clear that the extreme points found by the hybrid algorithm are better than NSGA II and MOPSO Figure 3. Pareto-optimal curve for IEEE 30 bus system obtained using Hybrid NSGAII and MOPSO Algorithm. algorithm. Even though the time of execution of the Hybrid algorithm is slower than NSGA II it is able to find well spread Pareto front compared to NSGA II. The hybrid algorithm is far superior to MOPSO in terms of converge speed and also in finding well spread Pareto-optimal front.

Method
In case II the valve point effect is neglected from the fuel cost curve and is solved using the proposed hybrid approach using the same parameters. The Pareto front obtained is shown in Figure 4 and is a continuous curve when compared to the Pareto front shown in Figure 3. In Figure 3 the Pareto front is discontinuous due to the effect of the Valve point loading in the cost curve. Both these case studies indicate that the hybrid approach is effective to solve the CEED problem.

Conclusion
In this chapter, a hybrid multi-objective optimization algorithm based on NSGA II and MOPSO have been proposed to solve the highly nonlinear, highly constrained combined economic emission dispatch problem. At any stage of the algorithm, only feasible solution is created because of the incorporation of the proposed constraint handling mechanism. During every iteration of the hybrid algorithm new population is created and NSGA II is applied on best performing individuals whereas MOPSO is applied on the lower ranked individuals to strengthen the exploration and exploitation capability of the algorithm. This hybrid approach is tested on an IEEE 30 bus system. The results obtained shows that the hybrid approach is efficient for solving CEED problem and is also able to quickly converge to a better Paretooptimal front when compared to MOPSO algorithm. The result obtained by the hybrid approach also demonstrates it is able to yield a wide spread of solutions and convergence to true Pareto-optimal fronts.