Optimal Allocation of Fixed and Switched Capacitors for Unbalanced Radial Distribution Feeders Using Artificial Intelligence-Based Approach*


*J.C. Miras, Undergraduate Student Project, Department of Electrical and Electronics Engineering, University of the Philippines, Diliman,(2004)


I. Implementation, Testing and Results

The proposed methodology of optimal allocation of fixed and switched capacitors for unbalanced radial distribution feeders was implemented using the trial edition of Borland C++ Builder 6.0. A matrix header file matrix.h was downloaded from the internet [5] to assist the matrix operation in the project. This matrix header file, also known as Matrix TCL Lite which is a free version of Matrix TCL Pro, provides easy to use functions and can handles large matrices operations. 

The proposed method was tested and applied to two distribution test feeders for five test cases. The program was run using 650 MHz Pentium III microprocessor. The first test system is the 24 kV, 34 bus test distribution feeder with systems peak load at 1100H, and the second is the 13.2 kV Tarlac Electric Cooperative II or TARELCO II with 166 buses and a systems peak load at 2000H-modified by [6] for demonstration purposes. Both are characterized by unbalanced three phase, two-phase and single-phase loads with single phase, two-phase and three-phase lateral feeders. Both of the two test systems were used by [6] for the same problem using Sequential Quadratic Programming method.

Test case 1
The first test case was the implementation of the proposed method to find the optimal fixed capacitor compensation for IEEE 34 bus test distribution feeder. For practical reason, only balanced three-phase capacitor compensation is placed at any location with all of the three-phase wires present. The objective of the test is to verify the effectiveness of GA on finding the near global optimal solution. The same problem was solved by [6] using Sequential Quadratic Programming (SQP).

Representation
Optimal capacitor allocation is a combinatorial optimization problem. Using GA, each possible solution was represented by nxl bit string. Since our goal in this case is to obtain an optimal fixed capacitor allocation, there is only one load level or one switching time for the whole 24 hour load duration. 

Candidate locations Load level 1
Time duration: 0000H-0023H
Location 1 Capacitor compensation 1
Location 2 Capacitor compensation 2
Location 3 Capacitor compensation 3
Location n Capacitor compensation n

Table 4-1. Representation for any possible solution for fixed capacitor compensation

For the purpose of comparison, the standard values used in [6] was used in this test case. Using binary gray code representation, the capacitor sizes was encoded in table 4-2. 

Capacitor sizes(Kvar) [A,B,C] Bit representation
0 000
50 010
75 011
100 001
150 101
200 111
0 110
0 100

Table 4-2. 3 bit gray code representation of compensation capacitors for test case 1

 Using sensitivity analysis, 22 candidate location was selected. Thus, if capacitor sizes are represented in 3 bits, any possible solution or individual has a string length of 66 bits. Note that for 22 candidate locations, 1 load level, and 6 capacitor sizes, there are 622 possible solutions. And that would take a longer time to search for the best combination of compensation if no defined method for the search process was used! 

Parameters
The cost constant for this problem are Ke = 0.05 $/kWh, Kp = 168 $/kW-year, and Kc = 4 $/kW-year. A population size of 100 individuals was randomly initialized. Since the population is relatively small, a linearly normalized fitness ranking was employed. The mutation rate is 1/66 or 0.01515. 

Results
Initially, before compensation, the system suffers from an annual loss of $70,469.19 and $48,698.02 from annual energy loss and annual peak power loss, respectively. The results of the load flow before compensation is summarized in table 4-3.

System losses before compensation

Daily energy loss (kWh) 3,861.325
Annual energy loss (kWh) 1,409,383.862
Peak power loss (kW) 289.869
Minimum voltage (pu) 0.915 (29-c)

Table 4-3. System losses and minimum voltage of IEEE 34 bus test feeder before compensation

For illustration purposes, the program was terminated only after 200 generations. The program was run for five trials. Each run takes around one and a half hours to finish 200 generations. The graph and the summary of the results of the five trials are shown in table 4-4 and table 4-5 and 
figure 4-1.

Trials Peak power loss (KW) Annual energy loss (KWh) Compensation
(Kvar)
Annual Saving ($)
1 231.779 1,106,566.231 1,875 17,400.087
2 231.779 1,106,566.231 1,875 17,400.087
3 231.756 1,106,625.136 1,875 17,400.958
4 231.779 1,106,566.231 1,875 17,400.087
5 231.768 1,106,622.200 1,875 17,399.096
Mean 231.772 1,106,589.206 1,875 17,400.063

Table 4-4. Summary of the results of load flow after compensation in test case 1.

Figure 4-1. Convergence curve of the 5 trials for 100 generations in test case 1. For illustration purposes, 101st - 200th generation was not plotted. Moreover, savings did not significantly increases after 100 generations. 

Candidate locations

Capacitor Compensation in Kvar (A,B,C)

Trial 1 Trial 2 Trial 3 Trial 4 Trial 5
28 50 50 50 50 50
27 100 100 100 100 100
31 75 75 75 75 75
26 75 75 75 75 75
23 50 50 50 50 75
20 75 75 100 75 50
19 75 75 0 75 75
17 0 0 50 0 0
16 50 50 50 50 50
15 75 75 75 75 75

Table 4-5. Output optimal fixed capacitor compensation of the five trials

Based on the above figures, although GAs are base on stochastic processes, the savings after several generations for five trials are almost the same.

The best solution of the five trials ,3rd trial, was selected and compared to the results obtained by [6] using SQP. The comparison of the solution of the two methods is given as follows;

After Compensation

SQP

AI

Peak power loss

235.31

231.756

Energy Loss(kWh/year)

1,116,786

1,106,625.136

Peak Power Loss($/year)

39,532.08

38,934.99

Energy Loss($/year)

55,839.3

55,331.257

Peak Power Loss Reduction(kW

54.569

58.113

Energy Loss Reduction(kWh/year)

291,740.485

302,758.725

Peak Power Loss Reduction Savings($/year)

9,167.59

9,763.02175

Energy Loss Reduction Savings($/year)

14,587.02

15,137.94

Capacitor Cost($/year)

6,600

7,500

Total Savings($/year)

17,154.61

17,400.958

Min. Voltage(pu)

0.935 (bus 29-c)

0.937 (bus 22-c)

Table 4-6. Comparison of the output between SQP and AI based approach of capacitor allocation

Notice from table 4-6 that the capacitor allocation using GA gives a higher savings than using SQP. This shows that for this particular test case, AI performs much better than the method used by [6].

Introduction
Problem Formulation and Solution
Implementation, Testing and Results
Testing and Results (continued1)
Testing and Results (continued2)
Conclusion and References

 

Sponsored Links