Research On Predictive Control Strategy Of Active Power Filter Based On RBF Neural Network

With the development of power electronics industry, electric power industry has been paid more and more attention. At present, China has gradually formed a large-scale and wide range of high-voltage transmission network. Our requirements have become increasingly high, such as power system security, reliability, stability, power quality, ability to respond to incidents and disasters. But we are enjoying the convenience of the electric power industry, consideration must be given to the negative effects brought by these techniques produce.These effects are especially harmonic. Further exploration and research in the light of the main problems in harmonic suppression, is not only the need of electric power industry theory development in China, but also the high requirement of the practice of the power industry. In the future, the intelligent development will be the trend of electric power industry in China. That how to apply intelligent control theory to power system application, suppress harmonics, improve the power factor, is the effective way to promote the development of electric power industry.The main of this project is research of active power filter harmonic detection and control technology. Because the traditional control strategy does not take into account the system delay problem. The core is:Leading to the concept of predictive control by analyzing the serious influence of delay on the APF compensation performance and using Radical Basis Function neural network to predict the harmonic current. By researching of the learning algorithm of the RBF neural network, especially the K-means clustering algorithm. This project introduces how to use K-means clustering algorithm to compute clustering center layer in the hidden layer, and using the least square method and Gradient descent method to compute the base widths and weights. Based on this learning algorithm, the conception is proposed.That is training the RBF neural network by the input and output sample-reconstruction, so as to achieve optimal predictive effect. And the forecast implementation of the model is described in detail. On this basis, the prediction model based on adaptive filter LMS algorithm is described in detail, focusing on the research of the entire structure of the algorithm. And by using of the two methods for predictive control of the harmonic current, we achieve to predict the next shot current reference, send out the PWM signals and make the main circuit send the compensation current. This method can eliminate the delay, and make matlab simulation and error analysis. The simulation results show that the prediction based on RBF neural network is more accurate, the error is smaller, the real-time performance is better. The simulation results lay the theoretical foundation for the APF harmonic current forecast.Finally, in the overall analysis of the active power filter, the project makes the system Matlab/simulink model. By using ip-iq rules of instantaneous wattless power theory, harmonic is detected. By RBF neural network prediction model, we can get harmonic current reference value. It makes the main circuit send the compensation current, so as to suppress harmonic. The construction of the system model and the effects of good simulation have proved that:RBF neural network predictive control has been the ideal use of active power filter in harmonic control. It has great theoretical significance and practical value. Prediction method of this project proposed opens up a new space for the field of harmonic control.