Central Control System Based on Genetic Algorithms to Improve Power Quality Ryszard Klempka
Maciej Tondos
AGH-University of Science and Technology Cracow, Poland
[email protected]
AGH-University of Science and Technology Cracow, Poland
[email protected]
Abstract—In the distributed systems there are various distributed consumers who negatively influence the power quality indices (parameters), many disturbance-sensitive loads and also various distributed controlled devices are intended for improving power quality. The main objective of the paper is to present a central control system which, on the basis of the set of input signals (mainly voltages and currents measured at selected points of a power system), will generate reference signals for individual controlled, distributed devices to be used for the supply improvement. The authors take on the task of analysis of an example power system and design a power quality improvement system, optimized by means of genetic algorithms. Keywords - power quality; genetic algorithms; central control system
and distributed power sources. There are also other systems dedicated for specific use which, if designed for these purposes, may perform the functions of fundamental component reactive power compensators, voltage stabilizers and high harmonics parallel active filters. A specific role can be played by AC adjustable speed drives (ASDs) which, as common loads in industrial installations, are of particular significance for the quality improvement [1]. To such load can be applied a VSI inverter, which guarantees a sinusoidal input current (necessary to supply the motor with the active power required by mechanical part of load) in all modes of motor operation and it has an option of generating an additional input current - the current component according to the reference generated by a central control system intended for compensation of other parallel disturbing loads (Fig.1).
I. INTRODUCTION On the basis of the current trends for solving complex technical problems, a new concept of power quality improvement is proposed. It consists in creating a distributed system for supply conditions improvement in a given islanding power system, in e.g. geographical terms (with determined points of delivery), or an internal installation system of an industrial consumer. There are various distributed consumers in the considered system who negatively influence the power quality indices (parameters). In the same system there are many disturbancesensitive loads. For the whole system are set the limit values of power quality indices. There are also various distributed controlled devices in the system, intended for improving power quality, i.e. controlled sources of reactive power (static compensators, idle synchronous motors (SM) and generators (SG) with excitation current control, active power filters (APF)
II.
MAIN OBJECTIVE
The main objective is to develop a central control system which, on the basis of the set of input signals (mainly voltages and currents measured at selected points of a power system), will generate reference signals for individual controlled, distributed devices to be used for the supply improvement. The following devices shall be taken into consideration as controlled from the central control system: • Static VAR compensator - e.g. TSC or TCR/FC • synchronous motors with excitation current control • distributed power sources (photovoltaic, wind turbine generators with indirect frequency converters, etc.). • parallel active filter and/or ASDs (VSI type).
III. ...
Is it possible to determine the reference signal for the power quality improvement system on the basis of solely the voltage signals acquired at selected points of the supply system?
APF n
S. motor (SM)
S. generator (SG) CENTRAL CONTROL SYSTEM
Wind turbin
PV system
SVC
ASD 1 ... ASD k
Measurement Control
Figure 1. The concept of distributed system for power quality improvement
RS3
LS3
PCC
D1
LS2
RS1
LS1
LL3 RL1
e D2
Genetic algorithms are becoming a popular optimisation method for technical applications. Basing on a natural selection and the "survival of the fittest" principle, they are a powerful tool used in cases of large optimisation spaces with complex mathematical description, as well as in the event of impossibility of formulation of such description, and consequently, of employing the known optimisation methods. The waveform of the active filter (i) reference current which guarantees minimization of the "critical" load RL2 voltage distortion was determined, using the genetic algorithm, from simulations carried out in the Matlab/Simulink environment. TABLE I. THE HARMONICS' AMPLITUDES BEFORE AND AFTER THE RD OPTIMISATION OF THE FILTER'S 3 HARMONIC
D3 RL3
To answer the question formulated this way, an example power system with schematic diagram shown in Figure 2, has been considered. Existing non-linear loads cause voltage distortion in the system's nodes. At one of the nodes a current source is connected – a parallel active filter (i) – whose purpose is the reduction of voltage distortion level at PCC where another load, "sensitive" to high harmonics (RL2), is connected. In practical industrial situations, the role of the active filter can be played by an ASD with indirect frequency converter, which, if not fully mechanically loaded or, if oversized in design phase, can also play the role of active parallel compensator. The spectrum of the load RL2 before to the active filter (i) connection is shown in Figure 3 (THDu (RL2) = 31.8%). In most publications the attempts at solution of the presented problem were based on the analytical solution of the circuit (this implies its full identification, which normally is impracticable) or, with a larger number of compensating devices, on the theory of multiagent systems [5, 6, 7, 8]. In this paper genetic algorithms were employed in order to solve the formulated multi-criterial optimisation task [2].
Other loads
RS2
DETERMINING THE REFERENCE SIGNALS
One of the essential problems in the presented conception is the answer to the question [4]:
APF 1
i
RL2
D4
Figure 2. The example power system with nonlinear load ( e = 230 2 sin(ωt ) , RL1 = 30Ω, RL2 = 30Ω, RL3 = 3Ω, RS1 = 3Ω, RS2 = 3Ω, RS3 = 3Ω, LS1 = 10mH, LS2 = 10mH, LS3 = 100mH, LL3 = 100mH)
Voltage harmonic order (n) 1 3 5 7 9 11 13 15 THDu (RL2) [%]
Voltage harmonics values [%] Before the filter After the filter connection connection 144 140 42 0.6 15 27 5 10 5 3 3 4 1.3 3 1.6 1.5 31.8 20.8
Note: It can be noted that the third harmonic optimisation results in the increase of high-order harmonics.
URL2
150
URL2
150
100
100
50
50
0 0
5
10
15
20
25
30
35
40
45
50
distortion minimization, presented in this paper. The following procedure was adopted:
00
5
10
15
20
n
25
30
35
40
45
50
n
Figure 3. The spectrum of the load RL2 before and after the 3rd harmonic optimisation 150
URL2
URL2
150
100
100
a)
b) 50
50
0
0
5
10
15
20
25
30
35
40
45
50
n
5
0
10
15
20
25
30
35
40
45
50
30
35
40
45
50
n
URL2
URL2
100
100
d)
c)
50
50
00
0
150
150
5
10
15
20
25
30
35
40
45
50
0
0
5
10
15
20
25
n
n
1) In the first step it was assumed that the active filter is the source of solely the third harmonic. The task consists in such selection of the amplitude and phase values of the filter-generated 3rd harmonic, that its content in the load (RL2) voltage would be minimized to the possibly largest extent. The results are presented in Figure 3 and Table 1. 2) In the following steps subsequent harmonics were added to the active filter (i) current and, using the genetic algorithm, their amplitude and phase values were optimised, step by step reducing the load (RL2) voltage harmonic distortion factor. Figure 4 shows the voltage and current spectra obtained after several optimisations. 3) In the next step of the genetic algorithm the minimization of harmonics (3rd, 5th, 7th, 9th and 11th) was made again in loop. Figure 7 shows the graph of harmonic distortion factors THDu obtained in this step. Table 2 summarizes amplitudes of the RL2 load voltage harmonic obtained after subsequent optimisation of harmonics: 3rd, 5th, 7th, 9th and 11th. Figure 5 shows the comparison of amplitudes of the RL2 voltage harmonics obtained in subsequent steps of the presented procedure. For each optimisation a slight increase can be observed in the amplitudes of adjacent harmonic components. The tendency toward decreasing the THDu value (Fig. 6) is maintained during the subsequent optimisations.
Figure 4. Spectrum of the RL2 load voltage after optimisation of subsequent harmonics in the active filter current. (a) 3rd and 5th harmonic, (b) 3rd, 5th and 7th harmonic, (c) 3rd, 5th, 7th and 9th harmonic, (d) 3rd, 5th, 7th, 9th, and 11th harmonic.
TABLE II.
AMPLITUDES OF HARMONICS AFTER THE OPTIMISATION OF VALUES AND PHASES OF SUBSEQUENT HARMONIC COMPONENTS
Voltage harmonic order (n) 1 3 5 7 9 11 13 15 THDu (RL2) [%]
40
URL2
35 30 25 20 15
RL2 voltage harmonics values [%] after the optimisation of harmonics 3
3+5
3+5+7
3+5+7+9
3+5+7+9+11
140 0.6 27 10 3 4 3 1.5
139 4.7 1.3 18.7 7.3 2.1 2.1 2.4
138 6.7 3 1.5 12 5.4 1.6 1.3
138 7.5 5 1.3 0.5 8.1 3.5 1
138 8 6.2 2.9 1.5 0.05 5.1 2
20.8
14.6
11.2
9.5
9
10
3
5
7
9
11
13
15
n Figure 5. Amplitudes of the load (RL2) voltage harmonics after optimisation of subsequent harmonics in the active filter current (dark blue – no filter, blue – 3rd, light blue –3rd +5th, yellow - 3rd +5th +7th, red - 3rd +5th +7th +9th, brown 3rd +5th +7th+9th +11th)
In general case the objective can be the minimization of reactive power flows, energy losses in the system, and stabilization of voltages at the system nodes. These tasks can be approached independently or, on their basis, can be formulated the task of global optimisation where they will be used as constraints in solving the optimisation task, e.g. the
35
22
30
20
THDu [%]
0
THD [%]
5
25 20
18 16 14
15 12 10
10
5 0
8 0
3
5
7
9
11
Optimisation of harmonic
Figure 6. THDu factors after each subsequent harmonic optimisation. The optimisation denoted with "0" refers to the system before the filter connection
6
0
5
10
15
20
25
30
35
Subsequent optimisations
Figure 7. Values of the load (RL2) voltage distortion factor THDu obtained in result of optimisation of subsequent harmonics (3rd, 5th, 7th, 9th and 11th) for the algorithm executed in the loop
140
URL2
150
120 100
URL2
100
80 60
50
40 20
0 0
5
10
15
20
25
30
35
40
45
0
50
0
5
15
10
20
25
30
35
40
45
50
n
n
Figure 8. The spectrum of the load RL2 voltage and current after minimizing the selected harmonics (a) two times, and (b) seven times TABLE III. SUMMARY OF PARAMETERS OF THE CURRENT HARMONICS GENERATED BY THE ACTIVE FILTER (FOR MIN. THDU) Order of the current harmonic Amplitude I [A] Phase ϕ [º]
3
5
7
9
11
12.78 117.7
7.41 72.86
4.98 26.67
3.48 -21.05
2.4 -70.7
4) In the fourth step it is assumed that the number of active filters / active current sources is greater than one (Fig. 9). They are: • connected at different electrical distances from the considered "critical" load; • the transmission costs of generated electric power are different (depending on their distance from the protected load); • their capability to participate in the compensation process may change with time (due to e.g. varying mechanical load of an ASD whose input rectifier is used as an active filter); • different looking from their unit costs of generating electric power needed for compensation [3].
TABLE IV.
AMPLITUDES OF VOLTAGE HARMONICS PRIOR TO THE OPTIMISATION AND AFTER TWO AND SEVEN OPTIMISATION LOOPS
Voltage harmonic order n 1 3 5 7 9 11 13 15
RS3
Voltage harmonics values [%] Before the filter After 2 After 7 connection optimisation optimisation loops loops 144 137.5 138 42 2.3 0.9 15 3.2 0.75 5 2.5 0.68 5 1.5 0.55 3 0.5 0.43 1.3 6.7 8 1.6 3.1 4.3
LS3
PCC
D1
RS2
LS2
RS1
LS1
RS1
LS1
RS1
LS1
All these factors should be taken into consideration as the constraints of the solved optimisation task (e.g. by means of introducing weighting coefficient), whose final effect should be determining the reference currents for all compensating devices participating in the compensation process. It can be expected that, first of all, will be activated the sources closest to the critical load. It may happen that some of them will not be activated because of too high cost of generation or too high transmission costs. It depends on the criterion adopted for global optimisation.
D3 RL3
LL3
RL1
Example
RL1 RL2
RL1
e
i1 D2
i2
i3
D4
Figure 9. The example power system with nonlinear load and three active filters 150
URL2
URL2
150
100
100
50
50
0 0
5
10 15 20
25 30 35
n
40 45 50
As clearly seen from figure 7, the subsequent optimisations, although they reduce the values of the optimised harmonics (not all of them, see Fig. 8), may worsen the THDu factor. Minimum THDu value is 6.8%, and was obtained in result of the algorithm loop minimizing the harmonics 3rd, 5th, 7th, 9th and 11th being executed two times. In this case the current source generates the signal whose spectrum is presented in Table III. Subsequent executions of the algorithm's loop do not improve THDu, despite further minimization of the harmonics amplitudes. After seven optimisation loops the THDu factor has increased to 7.26% (Table IV).
00
5 10 15 20 25 30 35 40 45 50
n
Figure 10. The spectrum of the load RL2 before and after harmonic compensation
Existing non-linear load – Figure 9 - cause voltage distortion in the system's nodes. At three nodes a current sources are connected – a parallel active filters (i1,2,3) – their purpose is the reduction of voltage distortion level at PCC where another load, "sensitive" to high harmonics (RL2), is connected. The waveforms of the active filters (i) reference currents which guarantee minimization of the "critical" load RL2 voltage distortion were determined, using the genetic algorithm. The spectrum of the load RL2 before the active filters (i) connection is shown in Figure 10 (THDu(RL2) = 31.8%). Figure 10 shows also the voltage spectra obtained after several optimizations using genetic algorithms (THDu (RL2) = 4.9%). In this case the current sources generate the signals whose spectrums are presented in Table V.
TABLE V.
SUMMARY OF PARAMETERS OF THE CURRENT HARMONICS GENERATED BY THE ACTIVE FILTERS (FOR MIN. THDU) IN PCC
Current source no 1 2 3
Order of the current harmonic generated by the active filter
3.
5.
7.
9.
11.
Amplitude of the filter current [A] Phase of the filter current ϕ [0] Amplitude of the filter current [A] Phase of the filter current ϕ [0] Amplitude of the filter current [A] Phase of the filter current ϕ [0]
1.9 140 9.3 108 0 15.9
0 85 5.2 72 1.5 37
0 -25 1.4 10.5 2.3 21
0 116 3.3 -35 0 -77
0 118 0 -133 1.7 -92
IV.
EXAMPLE MODEL OF POWER SYSTEM
The ability of the proposed control system to improve power quality in an example supply network (Fig. 11) has been proved. There are two controlled current sources (APF) connected at different electrical distances from the considered "critical" point (terminals of voltage source) and their unit costs of generating electric power needed for compensation are different (it was introduce by means of different weighting coefficient). V.
CONCLUSIONS
A power system always contains sources of the first harmonic reactive power and distortion power sources. They are distributed over the whole network and their percentage shares are different. Hence the necessity for preventing the propagation of these disturbances or, in other words, for their compensation. This is a complex task, mainly because of the difficulty of mathematical description of the phenomena in the system, and also because of various constraints and different criteria of optimisation which should be taken into account.
The better the quality of compensation the more expensive it is. This is the compromise to be found, and not the only one. The authors take on the task of analysis of an example power system and design of a power quality improvement system, optimised by means of genetic algorithms. REFERENCES [1] R. Epperly, F. Hoadley, R. Piefer, “Considerations when applying ASDs in continuous process“. IEEE Trans. on Industry Applications, vol. 33, mo. 2, March 1997, pp. 389-396. [2] R. Klempka, Z. Hanzelka, “Distributed system of power quality improvement“, Proceedings of 19th. International Conference and Exhibition on Electricity Distribution, CIRED 2007, Vienna 21-24 May 2007. [3] M. McGranaghan, C. Melhorn, “Economics of different plant ride-through improvement solutions for power system problems“. The Machinery Reliability Conference, Charlotte, USA, 1998. [4] K. Mikołajuk, S. Kwiczak, “Iterative and optimization algorithms for current harmonic estimation“. Archiwum Elektrotechniki (in Polish). [5] T. Jones, E. Petrie, “Spreading the net – Distributed power generation and creating a virtual utility to manage it". ABB Review, no. 3, 2000, p.13-21. [6] L.M. Tolberg, H. Qi, F.Z. Peng, “Scalable multi-agent system for real-time electric power management. Proceedings of IEEE Power Eng. Society Summer Meeting, Vancouver, BC, 15-19 July 2001. Piscataway: IEEE, 2001. vol. 3, p. 1676-1679. [7] K. Vanthounout, G. Deconinck, “The hierarchical-distributed topology for the communication infrastructure of complex embedded automated systems“. Proceedings of International Workshop on Design of Reliable Communication Networks, Budapest, Hungary, 7-10 October 2001. [8] G. Weiss (editor), “Multiagent systems: a modern approach to distributed artificial intelligence“. Cambridge: MIT Press, 1999. This work was performed under the finance support of the Polish State Committee for Scientific Research, grant No. 3 T10A 005 29
Figure 11. The example power system with nonlinear load and two active filters
