This section presents a genetic algorithm in conjunction with fuzzy logic (GA-FL) to formulate the capacitor placement and sizing problem in the presence of voltage and current harmonics, taking into account fixed capacitors with a limited number of capacitor banks at each bus [46]. Operational and power quality constraints include bounds for rms voltage, THDv, and the number as well as the size of installed capacitors. In this hybrid global optimization method, fuzzy approximate reasoning is introduced to determine the suitability (fitness function) of each GA chromosome for capacitor placement. This will improve the evolution process of the GA, lowering the probability of getting stuck at local optima. Fuzzy logic has shown good results for capacitor bank allocation when combined with genetic algorithms under sinusoidal operating conditions [1].
Suitability of THDv and voltage are defined for each chromosome based on fuzzy approximate reasoning. These variables are used with a cost index to determine the suitability (fitness function) of each GA chromosome for capacitor placement.
Fuzzy set theory (FST) contains a set of rules that are developed from qualitative descriptions. When suitability of THDv, voltage, and cost of the distribution system are studied, an engineer can choose the most appropriate locations and sizes of shunt capacitor banks. These rules are defined to determine the suitability of each chromosome (Schrom) and may be executed employing fuzzy inferencing. Schrom serves as the fitness function in the GA to further improve the evolution process, reducing the probability of converging to a local optimum.
The chromosome structure of Fig. 10.9 is used.
The suitability level of each chromosome (Schrom) is computed by defining fuzzy membership functions for suitability of total harmonic distortion (μSTHD), voltage (μSV), and cost (μcost). A fuzzy expert system (FES) uses the number of buses with high THDv and voltage values as well as the average of the THDv and voltage deviation (at these buses) to determine the suitability of THDv and voltage.
A concern for the development of fuzzy expert systems is the assignment of appropriate membership functions, which could be performed based on intuition, rank ordering, or probabilistic methods. However, the choice of membership degree in the interval [0,1] does not matter, as it is the order of magnitude that is important [1].
For a given chromosome, define the average unacceptable value of THDv as
where NTHD is the number of buses with high total voltage harmonic distortions (e.g., THDj > THDvmax). The following steps are performed to compute STHD:
• Fuzzification of THDavghigh and NTHD (based on membership functions of Fig. 10.12a and 10.12b, respectively) to compute the corresponding membership values (μTHD and μNTHD).
• Fuzzy inferencing and defuzzification based on the decision matrix of Table 10.5, the membership function of Fig. 10.13, Mamdani max-prod implication, and center average methods [50]:
For a given chromosome, define the average unacceptable value of voltage deviation as
where NΔ|V| is the number of buses with unacceptable (high) values of voltage. Vmax = 1.1 pu and Vmin = 0.9 pu are upper and lower bounds of rms voltage, respectively. The following steps are performed to compute SV:
• Fuzzification of ΔVavg and NΔ|V| (based on membership functions of Fig. 10.14a and 10.14b, respectively) to compute the corresponding membership values (μΔ|V| and ).
• Fuzzy inferencing and defuzzification (using the decision matrix of Table 10.5, Fig. 10.13, and Eq. 10-18) to determine the suitability of voltage (SV).
Compute the cost for all chromosomes (Eq. 10-8) and linearly normalize them into [0, 1] range with the largest cost having a value of one and the smallest one having a value of zero.
Suitability level for a given chromosome is computed from STHD, SV, and cost using the following steps:
• Fuzzification of STHD, SV, and cost (based on Fig. 10.13) to compute the corresponding membership values (μSTHD, μSV, and μcost).
• Fuzzy inferencing and defuzzification (using the decision matrix of Table 10.6, Fig. 10.14b, and Eq. 10-18) to determine the suitability of the chromosome (Schrom).
Table 10.6
Decision Matrix for Determining Suitability of a Chromosome (Schrom)
STHD | ||||||||||
Low | Medium | High | ||||||||
SV | Low | “cost” | “cost” | “cost” | ||||||
Low | Medium | High | Low | Medium | High | Low | Medium | High | ||
Low-medium | Low | Low | Medium | Low-medium | Low | Medium-high | Medium | Low-medium | ||
Medium | “cost” | “cost” | “cost” | |||||||
Low | Medium | High | Low | Medium | High | Low | Medium | High | ||
Medium | Low-medium | Low | Medium-high | Medium | Low-medium | Medium-high | Medium | Medium | ||
High | “cost” | “cost” | “cost” | |||||||
Low | Medium | High | Low | Medium | High | Low | Medium | High | ||
Medium | Low-medium | Low | Medium-high | Medium-high | Medium | High | High | Medium-high |
Three genetic operators including reproduction (based on the “roulette-wheel” mechanism), crossover (with crossover probability of 0.6 to 1.0), and mutation (with mutation probability of 0.01 to 0.1) are used.
Figure 10.15 illustrates the block diagram of the fuzzy expert system for the determination of the suitability of a chromosome for capacitor placement.
The shunt capacitor placement and sizing problem in the presence of linear and nonlinear loads is solved using the GA-FL algorithm of Fig. 10.16 as follows:
Step 1: Input system parameters and the initial population.
Step 2 (Fuzzy Fitness or Suitability): Perform the following steps for all chromosomes (Fig. 10.15):
Step 2a: Run the Newton-based harmonic power flow program.
Step 2b: Compute suitability of THDv and voltage (Fig. 10.13, Table 10.5, and Eq. 10-18).
Step 2c: Compute cost index.
Step 2d: Compute suitability, Schrom (Fig. 10.14b, Table 10.6, and Eq. 10-18).
Step 3 (Reproduction Process): Select a new combination of chromosomes by rolling the roulette wheel.
Step 4 (Crossover Process): Select a random number for mating two parent chromosomes. If it is between 0.6 and 1.0 then combine the two parents, and generate two offspring. Else, transfer the chromosome with no crossover.
Step 5 (Mutation Process): Select a random number for mutation of one chromosome. If it is between 0.01 and 0.1 then apply the mutation process at a random position. Else, transfer the chromosome with no mutation.
Step 6 (Updating Populations): Replace the old population with the improved population generated by Steps 2 to 5. Check all chromosomes and save the one with minimum cost, satisfying all constraints.
Step 7 (Convergence): If all chromosomes are the same or the maximum number of iterations is achieved (Nmax = 40), then print the solution and stop, else go to Step 2.
Apply the hybrid GA-FL method (Fig. 10.16) and the capacitor placement techniques of prior sections (MSS, MSS-LV, fuzzy, and genetic) to the 23 kV, 18-bus, distorted IEEE distribution system of Fig. E10.3.1. The six-pulse rectifier (with P = 0.3 pu = 3 MW and Q = 0.226 pu = 2.26 MVAr) causes a maximum voltage THDv of 8.486% (column 2 of Table E10.7.1).
Table E10.7.1
Simulation Results of MSS, MSS-LV, Fuzzy, Genetic and GA-FL Algorithms for the 18-Bus, Distorted IEEE Distribution System of Fig. E10.3.1 with per Unit VA = 10 MVA, per Unit V = 23 kV, and Swing Bus Voltage = 1.05 pua
Local optimization | (Near) global optimization | |||||
Before optimization | After MSS optimization | After MSS-LV optimization | After fuzzy optimization | After GA optimization | After GA-FL optimization | |
Capacitor bank locations and sizes (pu) | Q2 = 0.105 | Q2 = 0.030 | Q2 = 0.120 | Q2 = 0.015 | Q2 = 0.030 | Q2 = 0.030 |
Q3 = 0.060 | Q3 = 0.090 | Q3 = 0.060 | Q3 = 0.030 | Q3 = 0.000 | Q3 = 0.000 | |
Q4 = 0.060 | Q4 = 0.180 | Q4 = 0.180 | Q4 = 0.060 | Q4 = 0.165 | Q4 = 0.195 | |
Q5 = 0.180 | Q5 = 0.240 | Q5 = 0.240 | Q5 = 0.240 | Q5 = 0.330 | Q5 = 0.300 | |
Q7 = 0.060 | Q7 = 0.120 | Q7 = 0.120 | Q7 = 0.210 | Q7 = 0.105 | Q7 = 0.105 | |
Q20 = 0.060 | Q20 = 0.090 | Q20 = 0.090 | Q20 = 0.090 | Q20 = 0.060 | Q20 = 0.090 | |
Q21 = 0.120 | Q21 = 0.120 | Q21 = 0.120 | Q21 = 0.120 | Q21 = 0.090 | Q21 = 0.075 | |
Q24 = 0.150 | Q24 = 0.000 | Q24 = 0.000 | Q24 = 0.000 | Q24 = 0.015 | Q24 = 0.015 | |
Q25 = 0.090 | Q25 = 0.030 | Q25 = 0.000 | Q25 = 0.000 | Q25 = 0.015 | Q25 = 0.015 | |
Q50 = 0.120 | Q50 = 0.000 | Q50 = 0.000 | Q50 = 0.000 | Q50 = 0.030 | Q50 = 0.000 | |
Total capacitor (pu) | Qt = 1.005 | Qt = 0.900 | Qt = 0.930 | Qt = 0.765 | Qt = 0.840 | Qt = 0.825 |
Minimum voltage (pu) | 1.029 | 1.016 | 1.013 | 0.998 | 1.003 | 1.005 |
Maximum voltage (pu) | 1.055 | 1.056 | 1.059 | 1.050 | 1.050 | 1.050 |
Maximum THDv (%) | 8.486 | 6.370 | 4.720 | 4.899 | 4.883 | 4.982 |
Losses (kW) | 282.93 | 246.43 | 250.37 | 257.46 | 249.31 | 248.18 |
Capacitor cost ($/year) | 1978.20 | 1692.00 | 2206.80 | 1458.30 | 1788.75 | 1817.55 |
Total cost ($/year) | 159853.14 | 139199.94 | 141913.26 | 145120.98 | 140903.73 | 140302.8 |
Benefits ($/year) | 20653.6 | 17939.88 | 14732.16 | 18949.41 | 19550.34 |
a All capacitors were removed before the optimization process.
Simulation results for the MSS, MSS-LV, fuzzy, and genetic algorithms are listed in columns 3 to 6 of Table E10.7.1, respectively. Among these algorithms, the GA approach results in a near global solution with the best yearly benefit of $18949 per year (column 6 of Table E10.7.1).
The GA-FL approach of Fig. 10.16 is also applied to this system for optimal placement and sizing of capacitor banks. Results show a yearly benefit of $19550 per year (last row of Table E10.7.1) and maximum voltage THDv is limited to 4.982% (column 7 of Table E10.7.1). Compared to the GA approach, total yearly benefit is increased by $600 (e.g., 3.2%), and the total allocated capacitance is decreased by 0.015 pu.
The iterations of the GA-FL algorithm are continued until all generated chromosomes become equal or the maximum number of iterations is reached.
At each iteration, a fuzzy expert system (Fig. 10.15) is used to compute the fitness function of each chromosome and to select the parent population. As a result, new generations have more variety of chromosomes and higher probability of capturing the global solution. As demonstrated in Fig. E10.7.1, the relatively narrow search band of the conventional GA does not allow a large variation of chromosomes between iterations and the solution (e.g., benefits) is trapped at a near global point after the 14th iteration. However, the larger search space of the GA-FL approach (allowing large variations of benefits in consecutive iterations) has resulted in a better near global solution at the 35th iteration.
The inclusion of objective function and power quality constraints will automatically eliminate all solutions generating extreme values for voltages and/or currents and prevents fundamental and harmonic parallel resonances.
Apply the GA (Fig. 10.11) and GA-FL (Fig. 10.16) methods to the 4.16 kV, 123-bus IEEE radial distribution network of Fig. E10.8.1. Specifications of this system are given in reference [51]. To introduce harmonic distortion, include 20 nonlinear loads (Table E10.8.1) with different harmonic spectra (Table E10.8.2). Model these loads as harmonic current sources. The coupling between harmonics is not considered and the decoupled harmonic power flow algorithm of Chapter 7 (Section 7.5.1) can be used.
Table E10.8.1
Nonlinear Load Data for the IEEE 123-Bus System (Fig. E10.8.1)
Nonlinear bus | Nonlinear load type (Table E10.8.2) | kW | kVAr |
6 | Six-pulse variable-frequency drive (VFD) | 15 | 10 |
7 | Six-pulse 3 | 15 | 10 |
10 | Six-pulse 2 | 24 | 3 |
19 | Six-pulse 2 | 24 | 3 |
26 | Six-pulse 2 | 15 | 10 |
30 | Six-pulse 1 | 24 | 3 |
33 | Six-pulse 3 | 24 | 3 |
42 | Six-pulse 3 | 15 | 10 |
49 | Six-pulse variable-frequency drive (VFD) | 60 | 45 |
52 | Six-pulse 1 | 24 | 3 |
64 | Six-pulse 1 | 24 | 3 |
71 | Six-pulse variable-frequency drive (VFD) | 15 | 10 |
73 | Six-pulse 3 | 24 | 3 |
87 | Six-pulse 2 | 60 | 45 |
89 | Six-pulse 1 | 24 | 3 |
92 | Six-pulse 1 | 24 | 3 |
94 | Six-pulse variable-frequency drive (VFD) | 15 | 10 |
95 | Six-pulse 1 | 24 | 3 |
106 | Six-pulse variable-frequency drive (VFD) | 15 | 10 |
113 | Six-pulse 2 | 24 | 3 |
Table E10.8.2
Harmonic Spectra of Nonlinear Loads for the IEEE 123-Bus System (Fig. E10.8.1)
Nonlinear load | ||||||||
Six-pulse 1 | Six-pulse 2 | Six-pulse 3 | Six-pulse VFD | |||||
Order | Magnitude (%) | Phase (deg) | Magnitude (%) | Phase (deg) | Magnitude (%) | Phase (deg) | Magnitude (%) | Phase (deg) |
1 | 100 | 0 | 100 | 0 | 100 | 0 | 100 | 0 |
5 | 20 | 0 | 19.1 | 0 | 20 | 0 | 23.52 | 111 |
7 | 14.3 | 0 | 13.1 | 0 | 14.3 | 0 | 6.08 | 109 |
11 | 9.1 | 0 | 7.2 | 0 | 9.1 | 0 | 4.57 | -158 |
13 | 7.7 | 0 | 5.6 | 0 | 0 | 0 | 4.2 | -178 |
17 | 5.9 | 0 | 3.3 | 0 | 0 | 0 | 1.8 | -94 |
19 | 5.3 | 0 | 2.4 | 0 | 0 | 0 | 1.37 | -92 |
23 | 4.3 | 0 | 1.2 | 0 | 0 | 0 | 0.75 | -70 |
25 | 4 | 0 | 0.8 | 0 | 0 | 0 | 0.56 | -70 |
29 | 3.4 | 0 | 0.2 | 0 | 0 | 0 | 0.49 | -20 |
31 | 3.2 | 0 | 0.2 | 0 | 0 | 0 | 0.54 | 7 |
35 | 2.8 | 0 | 0.4 | 0 | 0 | 0 | 0 | 0 |
37 | 2.7 | 0 | 0.5 | 0 | 0 | 0 | 0 | 0 |
The high penetration of these nonlinear loads causes a maximum THDv of 8.03% (column 2, Table E10.8.3). The results of the GA-FL algorithm show a 5.5% increase in total annual benefit compared with those generated by the GA method (last row of Table E10.8.3), whereas the maximum THDv is significantly limited to 2.76%.
Table E10.8.3
Simulation Results of GA (Fig. 10.11) and GA-FL (Fig. 10.16) Algorithms for the 123-Bus, Distorted IEEE Distribution System (Fig. E10.8.1) Using Decoupled Harmonic Power Flow (Chapter 7, Section 7.5.1)
Before optimization | GA optimization | GA-FL optimization | |
Capacitor bank locations and sizes (kVAr) | Q83 = 200 | Q26 = 150 | Q42 = 300 |
Q88 = 50 | Q42 = 300 | Q52 = 300 | |
Q72 = 150 | Q72 = 300 | ||
Q76 = 300 | Q105 = 150 | ||
Total capacitor (kVAr) | Qt = 350 | Qt = 900 | Qt = 1050 |
Minimum voltage (pu) | 0.874 | 0.904 | 0.915 |
Maximum voltage (pu) | 0.980 | 0.980 | 0.980 |
Maximum THDv (%) | 8.0319 | 3.1018 | 2.7603 |
Losses (kW) | 94.96 | 82.021 | 81.28 |
Capacitor cost ($/year) | 125 | 360 | 390 |
Total cost ($/year) | 53114.5 | 46127.72 | 45744.24 |
Benefits ($/year) | 6986.8 | 7370.3 |
The generations (iterations) of the GA-FL algorithm are continued until all chromosomes of the population become equal or the maximum number of generations is achieved. The initial conditions for Eq. 10-8 (e.g., the initial capacitor values) do not usually reside inside the permissible solution region. In this section, a GA-FL approach is used to minimize the objective function while directing the constraints toward the permissible region.
To select the parent population at each generation, a fuzzy expert system (Fig. 10.15) is used to compute the fitness function of each chromosome considering the uncertainty of decision making based on the objective function and constraints. As a result, new generations have a higher probability of capturing the global solution.
As demonstrated in Fig. E10.8.2, the relatively narrow search band of the conventional GA does not allow a large variation of chromosomes between generations, and thus it is more difficult to escape from local optima. For the 18-bus and 123-bus systems, the solution is trapped at a near global point after the 14th and 22nd iterations, respectively. In the GA-FL approach, there is more variety in the generated parent population as the chromosome evaluation is based on fuzzy logic. The larger search space (allowing large variations of benefits in consecutive iterations) has resulted in a better near global solution at the 35th and 29th iterations for the 18-bus and 123-bus systems, respectively.
The inclusion of objective function and power quality constraints will automatically eliminate all solutions that generate extreme values for voltages and/or currents and prevents fundamental and harmonic parallel resonances.
Related to reactive power compensation is the placement and sizing of capacitor banks so that the transmission efficiency can be increased at minimum cost. First reactive power compensation is addressed for sinusoidal conditions. Various optimization methods are described such as analytical methods, numerical programming, heuristic approaches, and artificial intelligence (AI-based) methods. The latter ones comprise genetic algorithms, expert systems, simulated annealing, artificial neural networks, and fuzzy set theory. Furthermore, graph search, particle swarm, tabu search, and sequential quadratic programming algorithms are explained. Thereafter, the optimal placement and sizing of capacitor banks is studied under the influence of harmonics relying on some of the same optimization methods as used for sinusoidal conditions. Thus, the nonsinusoidal case is an extension of the sinusoidal approach. It is recommended that the reader first applies some of the optimization procedures to the sinusoidal case and then progresses to nonsinusoidal conditions.
Eight application examples highlight the advantages and disadvantages of the various optimization methods under given constraints (e.g., THDv, Vrms) and objective functions (e.g., cost). Unfortunately, there is not one single method that incorporates all the advantages and no disadvantages. It appears that the more advanced methods such as fuzzy set theory, genetic algorithms, artificial neural networks, and particle swarm methods are more able to identify the global optimum than the straightforward search methods where the gradient (e.g., steepest decent) is employed guiding the optimization process. This is so because the more advanced methods proceed not along a single trajectory but search for the global optimum within a certain region in a simultaneous manner.