Novel LMS-Neural Network based DSTATCOM for Improving Power Quality M.Balasubramanian 1, Dr.P.Selvam 2, Dr.S.Gopinath3, Anna baby4, sreehari.S5, Jenopaul.P6
1Assistant Professor, Department of EEE, Government College of Engineering, Tirunelveli, Tamilnadu, India. Email id: [email protected]
2Assistant Professor, Department of EEE, Government College of Engineering, Tirunelveli, Tamilnadu, India. Email id: [email protected]
3 AssociateProfessor, Department of Electrical Engineering, Annasaheb Dange College of Engineering and Technology, Ashta, Maharashtra, India Email id: [email protected]
4 Assistant Professor, Department of EEE, Adi Shankara institute of Engineering and technology, kerala, India. Email id: [email protected]
5Assistant Professor, Department of EEE, Adi Shankara institute of Engineering and technology, kerala, India. Email id: [email protected]
6 Professor, Department of EEE, Adi Shankara institute of Engineering and technology, kerala, India.
Email id: [email protected]
ABSTRACT
The Distribution Static Compensator (DSTATCOM) consistency in a 3-phase linear & non- linear load fed by a 3-phase supply is investigated in this paper using a Novel LMS-based neural network control approach. The DSTATCOM is generally implemented in distribution networks to minimize power quality problems. The Novel LMS-based neural networks are preferable for reducing computation complexity. Power Quality(PQ) issues arise when non-linear load connections are made. The load current will produce harmonics, and the voltage at the point of common coupling (PCC) will also be distorted. The controlling mechanism mitigates these PQ issues. The switching pulses for DSTATCOM's voltage source converter will be generated by using a Novel LMS-based control technique to generate reference source current. For generating reference current, the control algorithm has several steps. Estimation of weighting values, voltage control around the DC capacitor, and generation of reference current are the stages involved. The source current is compared to the reference current, and these values are then used to generate pulses using a hysteresis current controller. The main objective of this paper is to preserve the THD value of source current in accordance with IEEE-519 requirements for harmonic limit.
Keywords: Power Quality, DSTATCOM, Point of Common Coupling(PCC), Voltage Source Converter (VSC).
1. INTRODUCTION
Nowadays the major source of power consumption in distribution network is reacting loads which consume lagging current lead to low power factor. Also unbalanced loads are commonly present in distribution network. The above factor increases the requirement of reactive power, feeder losses and reduces active power flow in distribution feeder system. In addition to that the usage of nonlinear load in distribution system increases the distortion in supply voltage due to consumption of nonlinear load current. Hence the Power Quality(PQ) problems arises in distribution system are poor power factor, unbalance and harmonics. Then passive filter, active filters and hybrid filters were developed to compensate this PQ issues. In which passive filters are not much effective in reducing the PQ problems when used alone. The active filter are power electronics based converters which provide effective solution for PQ problems nowadays. These active filters are connected to the distribution system in series or shunt or in both. In these Shunt Active Filters(SAF) provide better solution for currently related PQ issues. The power electronic converter based filters used in distribution system are also called as custom power devices
(DSTATCOM, DVR & UPQC) in which DSTATCOM shunt connected device. The effectiveness of DSTATCOM in compensating PQ problems depends on its topology and control technique used for generating reference current. The conventional control techniques for DSTATCOM are theories based on instantaneous PQ, synchronous reference frame and so on. Other control methods deadbeat control method, SOGI best control method, direct power control strategy and so on. Other recent control methods using neural network are based on conductance estimation based algorithm and back propagation control. The advantages of this adaptive filtering technique are automatic adjustment of system parameters based on dynamics and easy implementation. Two major classification of adaptive filtering techniques are Recursive Least Square(RLS) algorithm and Least Mean Square(LMS) algorithm. Out of these LMS based control technique [13] provides quick convergence and improved dynamic response. In this paper a novel LMS- neural network based control algorithm is developed and implemented in DSTATCOM to mitigate PQ issues in distribution networks.
2. DESIGN OF DSTATCOM
The DSTATCOM is a shunt-based control system with three legs. It's connected to the PCC between load & source. There's also a dc capacitor, a three-stage inverter (IGBT), a coupling transformer, and a control circuit. The voltage source converter(VSC) converts the dc voltage into a 3-phase AC voltage at PCC, is an important component of the DSTATCOM. The DSTATCOM operates the inverter such that the inverter voltage and line voltage is balanced, enabling VSC to produce compensating current at PCC. Reactive power compensation, power factor correction, and voltage regulation are all possible with the DSTATCOM. A DSTATCOM is a shunt-related control electronic control based unit that provides compensating current. DSTATCOM uses a VSC and a capacitor on the DC side to generate filter current. A coupling inductor connects the device to the PCC as a shunt compensator. After all is said and done, the DSTATCOM is capable of power factor correction, reactive power compensation and balancing the unbalance load.
2.1. DC Capacitor Voltage
The PCC voltage is used to estimate DC voltage ( ) for effective PWM control of VSC of DSTATCOM, and it must be more relevant than other factors of the AC mains voltage. The DC reference voltage is specified as,
Where m-modulation index, m=0.8 by default, and is the AC line output voltage of the DSTATCOM.
2.2. DC Bus Capacitor
When loads are added to the DC bus, its voltage drops and when loads are removed, the DC bus voltage rises. The equation for is as follows, based on the principle of energy conservation.
where is the DC bus's minimum voltage level, is the overloading factor, is the phase voltage, is the VSC's phase current, is the time it takes to restore the DC bus voltage.
2.3. AC Inductor
The ripple current, , and switching frequency, , are used to determine the AC inductance. The AC inductance is calculated as follows:
where is the AC inductance. The new ripple is estimated to be 1%.
3. LEAST MEAN SQUARE ALGORITHM (LMS)
The acronym LMS stands for Least Mean Squares. It's a calculation for learning. The LMS calculation, introduced by Widrow and Hoff in 1960, is a flexible calculation that uses an inclination-based steepest average technique. In comparison to other calculations, this one is fairly simple; it does not require relationship work computation or lattice reversals.
The procedure of the LMS algorithm is depicted in Figure 3.1. It use the steepest descent calculation to precisely determine the vector called J(n) inclination at any regular cycle. If the stage size parameter is set correctly, the tap weight vector can also be resolved. The ideal wiener arrangement will be linked to the stage size decision and the optimally specified tap weight vector.
For example, tap input relationship network R and cross link vector p between the information and desired reaction as creation information on the two registered lattices. To get a gauge of J(n) another gauge of the relationship grid R and the cross connection vector p in the condition is fundamental. This is derived as,
The use of instant estimates for R and p in conjunction with the different discrete magnitude values of the tap input vector and the appropriate response, as defined by, is a natural choice of predictors.
Gradient vector value can be defined as,
In the aforementioned steepest descent algorithm, replace the estimation of the gradient vector J(n) with the following:
The tap weight vector was used to differentiate it from the values derived using the above algorithm. Alternatively, the result can be expressed using the following three basic relations:
1. Filter output:
2. Error signal:
3. Weight adaptation:
The LMS calculation's unpredictability is mainly explained by reference conditions. The information sources needed by the calculation are the error vector, input vector, and so on, should be current and crisp. The input is given in terms of a stochastic run, beginning with a single cycle operation and progressing further, is non-deterministic in nature and can be thought of as a collection of genuine angle vector headings. The feedback model for defining the steepest descent algorithm is very similar to this model.
Figure 3.1 Simple structure of LMS
4. PROPOSED SYSTEM
The present power quality problems in distribution networks are poor power factor, poor voltage regulation, unbalancing etc. The DSTATCOM mitigate these power quality issues. The framework is divided into three sections: a 3-phase source, loads, and DSTATCOM. This paper makes use of a 3-leg VSC that serves as a DSTATCOM. Figure 4.1 depicts the proposed system's block diagram. DSTATCOM is linked to the PCC through the interfacing inductor . The ripple filters are designed to filter out high frequency switching harmonics. The DSTATCOM make the source current balanced and reduce the harmonics in it using compensating current.
Figure 4.1 Block diagram for distribution system with DSTATCOM
4.1 Control Algorithm
A DSTATCOM control objective is to supply the compensate currents to balance supply current and reduce PQ issues. The synchronous reference frame and instantaneous PQ theory are two of the several control algorithms for the control of DSTATCOMs. In this paper, the reference currents are generated using Novel LMS-based Neural Network(NN) Control Algorithm shown in Figure 4.2 and actuate the PWM pulses to the VSI using hysteresis controller.
Figure 4.2 Novel LMS-based NN Control Algorithm
4.2 Unit Templates Generation
The phase voltages (Va, Vb, and Vc) are determined from PCC voltage and the positive sequence of phase voltages is determined. The peak of point of common coupling voltages is determined using this positive voltage series. These amplitudes are calculated as follows:
The unit templates are evaluated for in phase and quadrature quantity as,
4.3 Loss Component
The error in of the VSC ‘ ’ is defined as the difference between the reference DC bus voltage and the measured DC bus voltage at the tth sampling moment.
It's referred to as the "true loss" portion. This is fed into a PI controller. The PI controller's output maintains a constant DC bus voltage at the tth sampling moment.
Where is the d-axis dimension of supply currents,
The voltage error of the measured PCC voltage and its reference value are fed to an AC PI controller. At the tth sampling moment, the AC voltage voltage error is,
It's referred to as the reactive loss part. is the formula for keeping the PCC voltage constant at the tth instatnt is,
where and are the gain of PI controller respectively, and ) are the amplitudes of the quadrature component of the fundamental reference current at the tth and (t-1)th instants, respectively.
4.4 Novel LMS-based Neural Network(NN) Control Algorithm
The weight value of active and reactive components is extracted using the Novel LMS based Neural Network(NN) Control Algorithm. It is used to derive weight for d-axis component of load current, which is then trained using the Adaline NN control algorithm. The d-axis part weights of 3- phase load current is calculated:
The q-axis part weights can have determined as,
Where it is the rate of convergence is the fundamental d-axis part of reference supply currents has an average weight of
The q-axis part of reference supply currents has an average weight of,
The supply currents' fundamental 3-phase reference d-axis components are measured as
The supply currents' fundamental 3-phase reference q-axis components are given as
Total reference currents is the sum of and ,
4.5 Hysteresis Current Controller
Describing upper hysteresis band and lower hysteresis band limit is needed for this control strategy. If the load is variable, the variation of DC voltage is a common issue. The two-level hysteresis current control scheme is the most common hysteresis control technique. It is a nonlinear technique that is dependent on the current error. The reference current and the band limit given to it are correlated in this technique. When it reaches the upper band limit, the switches turn off., when current crosses the lower band limit, turns on. The reference supply currents ( , , ) and measured supply currents ( , , ) are fed into a hysteresis current controller in this system, which generates the VSC gating pulses.
5. SIMULATION AND RESULTS
The simulation model, which includes a source, load, DSTATCOM, and control block, is shown in Figure 5.1. The nonlinear load is a diode bridge rectifier, and the linear load is a series combination of resistance and inductance for each phase.
Figure 5.1 Simulation diagram of proposed system 5.1 Simulation Output
After connecting a non-linear load, the harmonics are increased in load current and the source current becomes distorted. The controller is used to minimize harmonics while maintaining good efficiency. The results of the simulation are shown below.
Figure 5.2 Source voltage before compensation
The phase source voltages are depicted in the Figure 5.2 above. The voltage level reaches nearly 320V peak before compensation. The grid voltages are initially not affected by any loads.
When non-linear loads are connected, the source current is distorted, and the voltages at PCC are also get distorted.
Figure 5.3 Source current before compensation
The above Figure 5.3 represents distorted source current after the connection of non-linear load. Here, it gets current value of 17A nearly. Non-linear load means, it doesn’t consume a sinusoidal current because of impedance changes. So that source current will also be distorted.
Figure 5.4 Load voltage before compensation
Figure 5.5 Load current before compensation
Above figures Figure.5.4 & Figure 5.5 denote load voltage and load current before compensation respectively. The load line voltage is nearly 580V and the current value is reached nearly 18A before the compensation. These load voltage and current are not sinusoidal because of non-linear load. These distorted parameters also affect source parameters and point of common coupling parameters.
Figure 5.6 Positive sequence voltage (Va, Vb, Vc)
The above Figure 5.6 show the positive sequence voltages are generated for reference current. It is generated from point of common coupling voltages. It is used for unit templates generation. It is reaches the 320V voltage without distortion.
Figure 5.7 Unit templates for real component
Figure 5.8 Unit templates for reactive component
The above Figure 5.7 & Figure 5.8 represents unit templates of real and reactive component.
This is generated by sensing PCC voltage. These unit templates is used to generate control signal of reference current value.
Figure 5.9 DC link voltage
The control signals are produced, and the DSTATCOM is connected to the switching signal.
The DSTATCOM controls and maintains the capacitor's dc voltage. The DC link voltage is maintained nearly 680V as shown in Figure 5.9.
5.2 THD Analysis
THD is defined as the ratio of the rms value of the harmonic content to the rms value of the fundamental quantity.
Figure 5.10 THD for source current before compensation
When the load is paired, the source current is distorted. It has harmonics in it. The Figure5.10 shows the THD value of source current before compensation. Since the THD value is about 14.48%, it clearly indicates that the presence of harmonics in the source current. To reduce the harmonics, compensation is needed.
Figure 5.11 THD for Source current with compensation
The Figure 5.11 represents THD analysis of the source current after compensation provided by Novel LMS-NN based control technique. Here, the THD values are reduced to below 5% in the system because of compensation which meets IEEE standard also.
. Figure 5.12 THD value for load current
The Figure 5.12 shows THD value of load current. Actually, non0linear load has some harmonics in the current. These current THD is nearly 13.72%. So, it is express the harmonic present in load current.
Figure 5.13 THD value for source voltage
Figure 5.13 express the THD analysis of source voltages after compensation. THD value of source voltage is about 5%. In non-linear load, the current doesn’t sinusoidal when connected to the sinusoidal source voltage. But source voltage doesn’t disturb. It shows harmonic reduction by control mechanism.
6. CONCLUSION
Simulation of Novel LMS-based neural network based control technique aimed at DSTATCOM to improve the power quality of 3-phase distribution systems subject to unbalance &
nonlinear load is done. The Novel LMS-based NN system has a simple design that simplifies computation and makes it simple to use. To extract fundamental active and reactive components from load current and to calculate reference supply current and various error signals, the proposed control algorithm was used. The DSTATCOM has been used to generate VSC switching pulses derived by comparing reference supply current with actual. As a result, power quality issues are decreased, and THD for source current remains below IEEE-519 requirements.
REFERENCES
[1] M. Badoni, A. Singh, and B. Singh, ‘Variable forgetting factor recursive least square control algorithm for DSTATCOM’, IEEE Trans. Power Del., vol. 30, no. 5, pp. 2353–2361, Oct. 2015.
[2] Bhim singh, ‘Power quality and mitigation techniques’, 1st edited in John wiley publisher, 2015.
[3] Arindam Ghosh, Gerard Ledwich, ’Power quality enhancement using custom power devices’, Kluwer Academic Publishers, 2002.
[4] C Dugan, Mark F. McGranaghan, surya santoso, H. wayne Beaty , ’Electrical power systems quality’, 2nd McGraw-Hill, 2004.
[5] Z.-Q. Wang, M. T. Manry, and J. L. Schiano, ‘LMS learning algorithms: Misconceptions and new results on converence’, IEEE Trans. Neural Netw., vol. 11, no. 1, pp. 47–56, Jan. 2000.
[6] N. Mohan, T. M. Undeland, and W. P. Robbins, ‘Power Electronics: Converters, Applications, and Design’. New Delhi, India: Wiley, 2003.
[7] S. Mishra, ‘Neural-network-based adaptive UPFC for improving transient stability performance of power system’, IEEE Trans. Neural Netw.,vol. 17, no. 2, pp. 461–470, Mar. 2006.
[8] A. M. Munoz, ‘Power Quality Mitigation Technologies in a Distributed Environment’, Londn, U.K.: Springer, 2007.
[9] H. Akagi, E. H. Watanabe, and M. Aredes, ‘Instantaneous Power Theory and Applications to Power Conditioning’. New York, NY, USA: Wiley-IEEE Press, Jul. 2009.
[10] B. Chen, Y. Zhu, and J. Hu, ‘Mean-square convergence analysis of ADALINE training with minimum error entropy criterion’, IEEE Trans. Neural Netw., vol. 21, no. 7, pp. 1168–1179, Jul.
2010.
[11] B. Singh, S. S. Murthy, R. R. Chilipi, S. Madishetti, and G. Bhuvaneswari, ‘Static synchronous compensator-variable frequency drive for voltage and frequency control of small- hydro driven self excited induction generators system’, IET Generat., Transmiss. Distrib., vol. 8, no. 9, pp. 1528–1538, Sep. 2014.
[12] C. Kumar and M. K. Mishra, ‘Operation and control of an improved performance interactive DSTATCOM’, IEEE Trans. Ind. Electron., vol. 62,no. 10, pp. 6024–6034, Oct. 2015.
[13] Bhim singh, ‘Application of LMS-Based NN Structure for Power Quality Enhancement in a Distribution Network Under Abnormal Conditions’,vol 29, IEEE Trans. Neural Netw., May 2018 [14] M. Srinivas, I. Hussain, and B. Singh, ‘Combined LMS-LMF based control algorithm of DSTATCOM for power quality enhancement in distribution system’, IEEE Trans. Ind. Electron., vol. 63, no. 7, pp. 4160–4168, Jul. 2016.
