Neural Predictive Model Control for Intelligent Universal Transformers in Advanced Distribution Automation of Tomorrow

Transcription

1 Neural Predictive Model Control for Intelligent Universal Transformers in Advanced Distribution Automation of Tomorrow Maryam Sadeghi, Majid Gholami Dept. of Electrical Engineering Islamic Azad University Islamshahr branch PO Box: 3335/369 Sayad Shirazi Ave, Namaz Sqr, Islamshahr IRAN Abstract: - Intelligent Universal Transformer (IUT) is a key point introducing as an Intelligent Electrical Devices (IED) for Advanced Distribution Automation (ADA) in future. ADA is the state of art, comprising flexible electrical architecture contributed with open communication construction for the tomorrow s distribution automation. IUT is based on a power electronic transformer employing the new technology of high voltage-low current solid-state devices to cope with the current transformer deficiencies. Solid-State devices in IUT topology Including rectifiers, converters and PWM inverters in input output stages which will be controlled trough the intelligent control fashion leads to robust control strategy. In this regards predictive control technique using artificial neural networks investigated for a three phase power PWM converters with current and voltages regulators. Neural Predictive Controller (NPC) is realized for a non linear optimizer, equipping Focused Time Delay Neural Network (FTDNN) for system modeling and optimization procedure. For prediction and control two strategic parts are considered. The first is FTDNN carrying out for power inverter dynamics system model and the other is optimizer unit subjected for minimizing the optimization index for performing the duty cycle of inverters as a control signals. In this approach NPC current source and voltage source controllers in input output stage of proposed four layers IUT topology results the smooth regulation in IUT output voltages and input current and improve the system characteristics under load and source disturbances. Key-Words: - ADA, IUT, IED, power electronic, DER, NPC, ANN Introduction ADA describes multi-function (energy exchange) system comprising from flexible electrical architecture and open communication construction for exchanging both Data and information between participants and system component instead of a traditional static fashion with single function concerning only on delivering energy to end-users [3-4]. ADA equips with the full automation and control functionality integrating throughout the distributed devices leads to the new approach in control and management for distributed automation systems. ADA open and flexible architecture together with the dynamic system monitoring with an interoperable network of electronic devices improve the functionality, reliability, performance and system power quality. Re developed electrical architecture with open communication construction, facilitate exchanging Data and information in a dynamic manner instead of a traditional static fashion. These two elements synergistically empowered each other and will comprise the future distribution system Fig.. Fig. ADA topology ADA debates new modern technologies evolving Distributed Energy Resources (DER), Intelligent Electronic Devices (IED), new sensors and new powerelectronic appliances [-2] like Intelligent Universal Transformer, a multi-functional equipment in lieu of traditional transformers [6-9]. ISBN:

2 Different topologies, design methods and applications are described for Intelligent Universal Transformers [4-6]. Fig. 2, illustrates the difference IUT utilization in near future [7]. Fig. 3 IUT HV-LC power electronic base transformer Main component diagram Fig. 2 IUT utilization in smart grid of future IUT four layers topology are discussed in the next section. Section III concerns with the prediction and control strategy. Simulation results are shown in section V and last section elucidates the conclusion with the prospective features. 2 Intelligent Universal Transformer Concept and Topology IUT basic concept comprises semiconductor devices rather than traditional transformer composed of copper and iron [-3]. Power electronic devices keep the ability for intelligently control of a high-frequency switching inverter to desired voltage which is adapted to the variety of changes occurring in the utility distribution line and the secondary voltages supplied to end users and inspire of all above mentioned the sinusoidal primary current is maintained [0]. In IUT topology (Fig 3), AC input voltage is modulated first by the static converter to the High Frequency (HF) square wave passing through the HF transformer and then demodulated by synchronous converter. As the transformer size is dependent inversely on the frequency, the HF IUT will have noticeably reduction in physical dimension, weight and stress factor, in contrast with the traditional transformer. In the other word, IUT is the High Voltage High Frequency (HV-HF) oil free transformer results in oil elimination, physical dimension and weight reduction and leads to maintenance free equipment. IUT four layers topology is based on seven main individual blocks comprising the power electronic equipment and high frequency transformer Fig. 4 [5]. Figure. IUT Multi-Converter-Based transformer four layers topology At the first stage a multilevel rectifier (r) rectifies the AC HZ frequency input (fig 4). Multilevel Inverter (i) produces the High Frequency - High Voltage (HF-HV) square wave. DC capacitor banks are in the stage (2). Next stage is high frequency transformer (3) with the ratio of :, isolating two sides from each other. Rectifiers and filters (4), producing the DC buses outputs. Full main bridge inverter with split DC buses 20/240-V 60-Hz output (5) is the fifth stage. Full auxiliary bridge inverter with single DC bus for 400-Hz output is in the next stage (6). Stage (7) is the buck DC/DC converter, for 48-VDC output. IUT expands the capabilities of distribution transformer to DC voltage option, automatic sag correction, real time voltage regulation, reliable diverse power as 400Hz service option for communication usage, three-phase power from a single phase line, Harmonic Filtering, Flicker mitigation, options for energy storage and dynamic system monitoring. 3 Control Strategy Neural controllers are widely used in nonlinear control system. It introduces control methodology that could be applied for identification and control; especially in cases that parameters of system are unknown and dynamic of system is nonlinear. In this regards the Neural Predictive Controller (NPC) is more advanced in compare with the last neural control ISBN:

3 strategies which have been used for IUT controllers [23, 24]. Control strategy is based on NPC in input-output stages of IUT. It will be trained to elaborate the switching angles for controlling the PWM inverters in IUT. As IUT is connected directly to the grid in primary stage, to prevent the harmonic distortion, input current should be sinusoidal and in phase with the input voltage. In this regard the Input current is sensed, predicted and controlled by a current source controller with NPC. The other controllers are in the output stage in the DC/AC converters Fig. 8-a. This leads to constant output in spite of different loads. Each inverter in the output stage will be controlled by a voltage source controller with NPC. 4 Predictive Control 4. Basic concept Predictive control comprises two strategic parts prediction and optimization units [9], [2-22]. introduced to characterize the application of Recursive ANN (RANN). Like the ARX models (Autoregressive by exogenous inputs), the network input signals are associated to its own input and past outputs. This structure represents the characteristic of TDNN. Focused Time Delay Neural Network (FTDNN) is the forward daynamic network including a tapped delay line at input so the dynamic appear only at input layer (fig 6). Figure 3. Focused Time Delay Neural Network structure In the above figure, weights w and w2 represent the weights of the connections between layers -2 and between layers 2-3. bj are the bias hidden neurons and S represents a sigmoid function in the output of each hidden neurons. b2 is the bias of output linear neuron. Figure 2. Neural Predictive Controller general structure first part is charged for predicting the future values of outputs via a plant model and the second has duty to generate a future control based on the predicted values. In this regards the neural network is novel method applying for identification and modeling the nonlinear dynamic system [20]. Control algorithm is based on the predicted output rather than the real output and is computed by errors between the reference signal and predicted values resulting from the model. The proposed neural network plant model is a Time Delayed Neural Network (TDNN) and trained offline. Nonlinear function f could be considered as a function of output [y(k), y(k-),..., y(k-n)] and input [u(k),u(k-),..., u(k-m)]. So the proposed nonlinear model will be expressed by: ( k + ) = f [ y( k), y( k ),..., y( k n), u( k), u( k ),..., u( k m) ] () The variable n and m are auto regressors and the number of exogenous regressors. For modeling the dynamic systems feedback effect of the ANN is y Figure 4. FTDNN construction in NPC in discrit time domain The generic expression of the proposed TDNN model is: N (2) y( k + ) = bs + w (, i). S( Xi) (3) And at last the following equation generalizes the differential of TDNN: (4) e ( k + ) = r ( k + ) y ( k + ) y( k + ) = u ( k ) n N i= i= 2 ( i,) + w ( i, j). y( k j + ) + w ( i, n + j). u( k j + ) Xi = b j= w 2 (, i). S ( X i ). w ( i, n + ) The proposed predictive control models could predict N steps forward so that the optimization function optimizes the whole trajectory of future control in a horizon of N steps ahead [8]. This means that the predictive controller is not limited to one point and m j= (5) ISBN:

4 expands its duty to the entire vector of N predicted errors so the optimization index is expressed by: r r T (6) J = [ E. E ] r 2 r U ( k + ) = U ( k) λ r J (7) U ( k) And r E = [ e( k + ), e( k + 2),..., e( k + T )] (8) An optimization index represents the function that is minimized via control procedure for optimizing the future performance of plant. 5 Matlab Simulation In this approach, neural predictive controllers for three-phase power converters with current and voltage regulators are used to predict the future output voltage of converters response and control the line current and output voltages. TDNN is considered with six hidden layers N, four auto regressors n (four future control input) and four exogenous regressors m (four past plant states and four past control inputs) and use the hyperbolic tangent as sigmoid function. Four layers IUT topology in Matalb is demonstrated in fig 8. It is connected directly to the 500V, 3 phase sinusoidal input voltage. The proposed scheme is utilized for mitigating the power-line disturbances resembling voltage sags and swells in lower voltage devices. IUT outputs are 3 phase sinusoidal and 240V DC. In the first stage IGBT rectifiers, rectify the input voltage and produce the DC bus. Input current should be sinusoidal and in phase with input voltage. This is done trough the NPC in input stage. The IGBT Inverters in the next stage convert DC voltage to HF square wave. This HF wave will pass via HF transformer with ratio : for isolating two stages from each other. Four AC/DC converters rectify and produce four DC buses for four outputs. Then four inverters in outputs convert DC buses to four IUT outputs (48V DC, two 240V AC 60 HZ outputs and 20V AC 400 HZ for communication usage). Figure 5. (a) input current regulation by NPC and Four NPC s for IUT outputs reulation, (b) IUT four layer simulation model To cope with the nonlinearity and became in adaptability to the changes in parameters the adaption law is considered as below: 20 2 λ = λ ( ) 5 / 0. y( k) π (9) With considering λ 0 = 6 x 0-4, and using adaptive algorithm, λ is computed in each step, and the system adapts to minimize the error. NPC current source controller in input stage predicts and regulates the input current and prevents from ISBN:

5 harmonic distortion. Four NPC in output, control the output voltages. In response to set of trial control inputs, neural network predicts four future states of inverters indicating the four outputs corresponding to the next four seriate future states of the inverters which will keep in four memories. Four past states of inverters are kept in four memories in NPC construction. The NPC predict the future outputs of inverter in conjunction with the four past states of inverter and four future states the inverter and the future control inputs. Simulation results are shown in Fig Conclusion Moderated infrastructure technology has been raised in response to the necessity for the oncoming distribution automation motivating the new advanced technologies as IED s and Distributed Energy Resources (DER s). IUT is a Pioneer IED s for the Advanced Distribution Automation with a major advantages which are summarized in part II. Five NPC s in input - output stages regulate the input current and output voltages, guaranteeing the system stability in case of input - output disturbances. 48V DC service options together with three other AC outputs are available with the proposed four layers IUT topology. NPC indicates the adaptive strategy for controlling the plant especially in case of nonlinear applications such as unexpected distortion in system dynamic. In this case, NPC produces the predictive control in conjunction with the neural network model of IUT inverters for predicting the future parts of control inputs for adjusting them to receiving the desired amounts. The future control is selected by optimization function. The proposed adaption rules for demonstrated the best tracking for achieving to the favorite stages. Stability, reliability and efficiency are all enhance trough the IUT topology and the mentioned predictive control scheme. The applicability of IUT is described with the full network functionality especially in tomorrow s smart grid. In this regards optimization in cost for Industrialization of IUT is a part of innovation which could be raised in power electronic technologies and control algorithm for enhancing the efficiency and improving the system performance which could be stated in IUT forthcoming projects. Acknowledgment Authors want to thank the research deputy of Azad University, Islamshahr branch for their efforts and their financial support of this research. Figure 6. (a) Thre phase input signal, (b) Modulation Index, (c) Control Sygnal, (d) V refrence, 48 V DC Output and predictive voltage by NPC Fig (9-a), illustrates Three phase input voltage applies on IUT; Fig (9-b) is modulation index in DC/DC converter. Control action is depicted on Fig (9-c). DC output voltage, reference voltage and predicted voltage are shown in last figure (9-d). Voltage reaches the steady state of 48V DC at 5.5 second. This is the α degree for IGBT inverters in DC/DC converters. References: [] EPRI Report 0095, Technical and System Requirements for Advanced Distribution Automation, [2] M. McGranaghan, F. Goodman, Technical and System Requirements for Advanced Distribution Automation, 8th International Conference on Electricity Distribution, CIRED, Turin, 6-9 June [3] IEEE Power Engineering Society, Research Plan for Advanced Distribution Automation, General Meeting, [4] Energy and Environmental Economic Inc, Value of Distribution Automation applications prepared, [5] EPRI Product ID # 00956, Feasibility Study for the Development of High-Voltage, Low-Current ISBN:

6 PowerSemiconductor Devices, 2003 Strategic Science and Technology Project, [6] S. Ratanapanachote, Application of an electronic Transformer in a power distribution system, Texas A&M University, 2004 [7] H.Akagi, The next generation medium voltage power conversion systems, Journal of the Chinese Institute of Engineers, 2007 [8] H. Krishnaswami, V. Ramanarayanan, Control of high-frequency AC link electronic transformer, Indian Institue of Sience, 2005 [9] D. Wang, C. Mao, J. Lu, S. Fan, F. Peng, Theory and application of distribution electronic power transformer, Electric Power Systems Research Journal, vol 77, 2007, pp [0] H. Iman-Eini,, JL. Schanen, Sh. Farhangi, J. Barbaroux, JP. Keradec, A Power Electronic Based Transformer for Feeding Sensitive Loads, IEEE, 2008 [] A. Hefner, Silicon-Carbide Power Devices for High-voltage, high-frequency Power Conversion, National Institute of Standards and Technology, Gaithersburg, 2007 [2] D.Aggler,J.Biela, J.W.Kolar, Solid State Transformer based on Sic JFETs for Future Energy Distribution Systems, [3] Marc C. Robbinson, Solid state Universal Intelligent Transformer, [4] J. S. Lai,A. Maitra,A. Mansoor,F. Goodman, Multilevel Intelligent Universal Transformer for medium voltage applications, IEEE Industry Application Conf., 2005 [5] F.Goodman, Intelligent Universal Transformer Technology Development, EPRI 2006 [6] A Maitra & A Sundaram, Universal Intelligent Transformer Design and applications, Power Distribution Conference, 2009 [7] A. Maitra, A. Sundaram, M. Gandhi, S. Bird, Sh. Doss, Intelligent Universal Transformer Design and Applications, CIRED 20th International Conference on Electricity Distribution, Prague, 8- June [8] L. Schnitman, A. B. Fontes, The Basic Ideas of Neural Predictive Control, Proceedings of the 7th Mediterranean Conference on Control and Automation (MED99) Haifa, Israel - June 28-30, 999 [9] Garcia, C.E.; Prett, D.M. and Morari, M, Model predictive control: theory and practice - a survey, IFAC Workshop on Model Based Control, Atlanta- USA, pg June, 989. [20] Ahmed, M.S. and Anjum, M.F, Neural-net-based direct self-tuning control of nonlinear plants, International Journal of Control. Vol. 66, nº, pp [2] Clarke, D.W.; Mohtadi, C. and Tuffs, P.S, Generalized predictive control. Part I: the basic algorithm, Automatica, Vol. 23 nº 2, pp [22] Clarke, D.W.; Mohtadi, C. and Tuffs, P.S. Generalized predictive control. Part II: extensions and interpretations, Automatica, Vol. 23 nº 2, pp [23] M. Sadeghi, M. Gholami, Intelligent Universal Transformers Online Adaptive Control via Artificial Neural Network, Adavanced material research journal., in press. [24] M. Sadeghi, M. Gholami, Developing Adaptive Neuro-Fuzzy Inference System for controlling the Intelligent Universal Transformers in ADA, Adavanced material research journal., in press. ISBN: