دانلود مقاله ترجمه شده مبدل های UPS تک فازی شبکه عصبی کنترل شده با پاسخ گذرا و سازگار با بارهای مختلف – مجله IEEE

فهرست مطالب:

چکیده
مقدمه
کنترل شبکه عصبی پیشنهادی مبدل
شبیه سازی
نتیجه گیری

بخشی از ترجمه:

این مقاله یک طرح کنترلی شبکه عصبی را برای مبدل های عرضه توان غیر قابل انقطاع UPS جهت بهبود پاسخ موقت آن ها و سازش با بارهای مختلف خارجی پیشنهاد می کند. دو مدل شبیه سازی برای حاصل کردن الگوهای نمونه برای آموزش شبکه عصبی ایجاد می شود. یکی کنترلگر حلقه-بازخورد-چندگانه برای بارهای خطی و دیکری کنترل گر باخورد-جریان-بار ایده ال برای بارهای غیر خطی می باشد.کنترل گر نوع دوم در مدل بار مرجع ایجاد شده و جریان بار برای ردگیری این مرجع تحت فشار قرار می گیرد.الگوهای نمونه تحت شرایط بارگذاری متعدد در آموزش آفلاین شبکه عصبی انتخاب شده استفاده می شوند که برای کاهش زمان مخاسبه به سادگی ایجاد شده اند. با کامل شدن آموزش، شبکه عصبی برای کنترل آنلاین مبدل UPS استفاده می شود. تشکیل الگوهای نمونه و آموزش شبکه عصبی با استفاده از MATLAB و SIMULINK انجام می شود و صحت یابی نتایج با استفاده از PSpice انجام می شود.پی برده شده است که مبدل کنترل شده شبکه-عصبی پیشنهادی می تواند ولتاژ خروجی سینوسی با اغتشاش هارمونیک کل پایین تحت شرایط بارگذاری مختلف و نیز واکنش های خوب در صورت تغییر بار تولید می کند

۱ مقدمه

با توقف توان بهره روی، عرضه توان غیر قابل انقطاع توان الکتریکی را برای سیستم های بحرانی نظیر سیستم های مخابراتی، سیستم های کامپیوتری و سیستم های پشتیبانی حیات فراهم می کند.مبدل UPS برای تولید ولتاژ خروجی ثابت سینوسی با حداقل اغتشاش هارمونیک کل برای طیف وسیعی از بارهای خارجی از مقاومتی تا القایی و از خطی تا غیر خطی نیاز می باشند. در عین حال، واکنش سریع موقت، پایداری خوب و اطمینان پذیری بالا نیز نیاز می باشد.
در یک مبدل UPS، یک مبدل pwm مودلاسیون پالس-عرض و یک فیلتر خروجی LC درجه دوم وجود دارد. کنترل گر های مختلف برای تنظیم مطلوب استفاده شده اند.کنترل گر ها را می توان به دو گروه تقسیم کرد: آنالوگ بنیان و دیجیتال بنیان. اکثریت کنترل گرهای آنالوگ بنیان بر اساس آنالیز های قطبی- فرکانسی کلاسیک و مدل خطی طراحی می شوند. برخر ار راهبرد های کنترل حلقه بازخورد چندگانه ارائه شده اند. چنین طرحی در جریان رسانا یا ظرفیت سنج فیلتر LC موسوم به سیگنال کنترل یک حلقه بازخورد داخلی می باشد.حلقه بازخورد ولتاژ خارجی برای حصول اطمینان از ولتاژ خروجی سینوسی القا می شود. به دلیل استفاده از حلقه بازخورد جریان درونی، مانع خروجی فیلتر مبدل کاهش یافته و سفتی دینانیکی افزایش پیدا می کند.

بخشی از مقاله انگلیسی:

Abstract – This paper proposes a neural-network control scheme for the inverters of Uninterruptible Power Supplies (UPS) to improve their transient response and adaptability to various loads. Two simulation models are built to obtain example patterns for training the neural network. One is a multiple-feedback-loop controller for linear loads, and the other is an idealized load-current-feedback controller specially designed for nonlinear loads. The latter controller has a built-in reference load model, and the load current is forced to track this reference. Example patterns under various loading conditions are used in the off-line training of a selected neural network, which is made as simple as possible to reduce the calculation time. When the training is completed, the neural network is used to control the UPS inverter on-line. The development of example patterns and training of the neural network are done using MATLAB and Sm and the verification of results is done using PSpice. It is found that the proposed neural-networkcontrolled inverter can provide good sinusoidal output voltage with low Total Harmonic Distortion (THD) under various loading conditions, and good transient responses when the load changes. I. INTRODUCTION When utdity power fails, Unintemptible Power Supplies (UPS) should continue to provide electric power to critical systems such as communication systems, computer systems, and life support systems. The UPS inverter is required to produce constant sinusoidal output voltage with minimum Total Harmonic Distortion (THD) for a wide load range: from resistwe to inductive or capacitive, and from linear to nonlinear. At the same time, fast transient response, good stability and hgh reliability are also required. In a typical UPS inverter there is a Pulse-Width Modulation (PWM) converter and a second-order LC output filter. Various controllers were used to achieve regulation. The controllers can be classified into two groups: analogbased and dgital-based. Most of the analog-based controllers were designed based on linearized model and traktional frequency-domain analysis. Some multiple-feedback-loop control strateges were proposed in [ 1,2]. Such schemes sense * * Department of Electrical Engineering Zhejiang University Hangzhou 310027, China the current in the inductor (or capacitor) of the LC filter as the control signal of an inner feedback loop. An outer voltage feedback loop is incorporated to ensure a sinusoidal output voltage. Owing to the introduction of the inner current feedback loop, the output impedance of the inverter filter is decreased and the dynamic stiffness enhanced. The controlled inverter can produce a satisfactory sinusoidal output voltage within a certain load range. However, a uniform performance cannot be obtained under extreme loading conditions, and THD increases sigmficantly when there is a nonlinear load. In addition, such schemes are analog-based, which suffer from all the drawbacks of analog circuits, such as limited accuracy, uncompensated thermal drift, and need of recalibration. Artificial Neural Network (ANN) techniques have grown rapidly in recent years. An ANN is an interconnection of a number of artificial neurons that simulates a real biologcal brain system, which can compute very fast in a parallel and distributed manner. It also has the ability to approxjmate an arbitrary nonlinear mapping. Compared with other digtal algorithms, although the training process of an ANN is timeconsuming and complex, the actual real-time computation from input to output is fast and simple [3]. ANN has already been employed in various control and signal processing applications in power electronics and dnves [4][5]. In the control of DC-AC inverters, ANN has been introduced in the current control of inverters for AC motor drives, where the ANN receives a phase-current error and generates PWM logic signals to dnve the inverter switches [6]-[9]. The neural network has been used in the harmonic elimination of PWM converters where the ANN replaced a large and memorydemanding look-up table to generate the switching angles of a PWM converter for a given modulation index [lO][ll]. The neural network was also introduced into UPS inverter control [ 121. In this method, the neural network is trained off-line, and the trained network can be employed for on-line control. The inputs of the neural network are time, present output voltage and last sampled output voltage. However, the ANN used may not have enough information to ensure a sinusoidal output voltage under extreme loading conditions. This paper proposes a neural network control scheme for UPS inverters, which leams the control law from This project was supported by The Hong Kong Polytechnic University and Zhejiang University. o-7803-5769-8/99~$10.000 1999IEEE 865 Authorized licensed use limited to: Hong Kong Polytechnic University. Downloaded on July 7, 2009 at 04:00 from IEEE Xplore. Restrictions apply. representative example patterns obtained from two simulation models. One model is a multiple-feedback-loop controller for linear loads, which has been widely used in the control of power converters, the other is an idealized load-currentfeedback controller specially designed for nonlinear loads. The latter controller has a built-in reference load model, and the actual load current is forced to track this reference. This idealized model is not easy to implement experimentally since the load model cannot be obtained beforehand. A selected neural network is trained off-line with the patterns obtained from simulations. The inputs to the neural network are the present capacitor current, previously sampled capacitor current, present load current, present output voltage, and the error between the reference voltage and the output voltage. The output of the neural network is a part of modulation signal to the PWM generator. The whole modulation signal consists of the output of the neural network and a sinusoidal voltage reference. When the training is completed, this neural network is used to control the inverter on-he. Development of the example patterns and training of the neural network are done using h4ATLAE3 and SIMULINK, and performance verifications are done using PSpice. Simulation results have proven that the proposed neural-network-controlled inverter can provide a good sinusoidal output voltage with low THD under various loading conditions, and good transient responses w:hen the load changes.