![]() |
Fusion of Neural Networks, Fuzzy Systems and Genetic Algorithms: Industrial Applications
by Lakhmi C. Jain; N.M. Martin CRC Press, CRC Press LLC ISBN: 0849398045 Pub Date: 11/01/98 |
Previous | Table of Contents | Next |
A modulation technique with low commutation frequency (lower than 1 kHz) enables the converter to work with higher power ratings. It results in increased converter efficiency and reduced switches stress. Therefore, converters may operate with power levels previously suitable only for six-pulse converters.
A modulation technique that falls within these characteristics is the predictive current loop SVM. This modulation technique has been used for VSI converters and DFCs [13][21][22]. The technique is capable of controlling the output line currents keeping their distortion beneath a desired value, usually 4% to 6%. This is done while operating with a maximum commutation frequency of 600 Hz throughout the whole output frequency range (usually under 120 Hz).
Although this technique offers a remarkable converter performance, it poses some problems [21][22]. The technique forces the loads line current space vector Il to stay within the error zone in the α-β plane of the current space vector reference Iref. Whenever Il falls out of the accepted error zone surrounding Iref, the controller predicts the current trend for every converter state. It then selects the state that brings the current back to the accepted error zone for the longest time. This operation is depicted in Figure 6. Naturally, the current prediction is limited by the output frequency at which the converter is operating. Usually, for high output frequencies, the algorithm must be modified to avoid the unacceptable time delay produced by the required processing. Specifically, the current trend is predicted at a different time, not only when the controller detects the current error. The algorithm presents another drawback, which is due to the simplified load model used to actually predict the current trend for the different converter states. In order to assure a proper prediction for the converters state selection, a parameter identification algorithm must be employed, where, usually, the loads inductance, resistance, and back e.m.f. are required. Naturally, this increases the overall control algorithm complexity and reliability.
Figure 6 Predictive-current algorithm operation. Whenever the line current space vector Il falls out of the line current reference space vector Irefs accepted error zone, the converter selects the next converter state by predicting the current trend for every converter state.
In this section a new eXpert knowledge-based SVM (XSVM) technique is presented, which is based upon an expert knowledge of the converters operation. The technique uses a set of rules to determine the next converter state, depending on the input (measured) variables. For output and input current control, only the loads line currents and input phase voltages must be measured. The output line voltages and input line currents are also required. They are obtained by a software waveform reconstruction using the converters transfer functions H and Hi and Equations (10) and (14). This software reconstruction reduces and simplifies the measuring and control circuitry for the XDFC. The presented XSVM technique requires a reduced processing time, being 70 times faster than the predictive current control algorithm employed in voltage source inverters. It is also independent of the loads parameters, eliminating the need for online parameter identification. This independence is achieved by the way in which the next converter state is selected, based only on how the converters input and output variables vary in time.
The SVM presented herein is used to control both input and output currents, thus it uses two different sets of rules. These are based on a predictive current controlled converter operation, and, therefore, present performances similar to that technique. Basically, the XDFC with XSVM simultaneously controls the input and output currents. The load current distortion is kept under 6%, and the input current distortion is diminished. The input-output control slightly increases the converters commutation frequency. However, it keeps the maximum below 850 Hz. The modulation presented allows the XDFC to operate with a unity ac-ac voltage gain. This fact is an important achievement that enables XDFC drives to operate with the systems nominal voltage level, hence eliminating the need for coupling transformers used to counteract the converters voltage loss of other DFCs.
The converter employs a fuzzy logic controller and the expert knowledge based SVM introduced in Section 4. Software waveform reconstruction is performed by using transfer functions H and Hi, which model the converter operation. According to Table I matrix H can take 25 different forms from the 27 different electric states. These are transformed into space vectors using Parks matrix (Table III), which are required by the XSVM used for this converter.
The modeling approach chosen, based on the converter transfer function H, sets two different control objectives. The first one is to control the output line currents of the XDFC, which is the prime objective as they are the loads currents. This control is realized with H using the voltage space vectors. The second objective is to reduce the input line current distortion and, thus, increase the input power factor. This control is also realized with H (or Hi), but this time using the current space vectors in the XSVM. Clearly, there are two completely different control goals for the XDFC, and both must be fulfilled using the same means, i.e., the converters transfer function H. To solve this dilemma, a fuzzy logic controller is used. The fuzzy controller determines which XDFC side has higher priority, either the output-load side or input-utility side. Then, it hands over command of the converter to the XSVM algorithm of the chosen port while the next converter state is selected.
Previous | Table of Contents | Next |