Neural Network Control of Shunt Active Filter in MATLAB

Neural Network Control of Shunt Active Filter in MATLAB

Introduction

In power systems, the integration of non-linear loads, such as those using three-phase rectifiers with RL (resistor-inductor) components, can introduce harmonic distortions. These distortions, caused by the non-linear characteristics of semiconductor diodes, affect the source current, leading to non-sinusoidal waveforms. To address this issue, syntactic filters are employed to eliminate harmonics and ensure the source current remains sinusoidal.

Syntactic Power Filter Functionality A syntactic power filter typically comprises a three-phase inverter and capacitor connected to the grid. By measuring the input current and coordinating it with the input voltage, the syntactic filter generates compensating currents. These compensating currents are injected into the power system to counteract the reactive power consumption caused by non-linear loads, thereby maintaining sinusoidal source currents. Need for Compensation When non-linear loads are connected to the power system, reactive power is consumed, leading to undesirable effects such as distorted source currents. To mitigate this, compensating currents are injected into the grid to ensure the source current remains sinusoidal. This is crucial for preserving the overall quality of load currents and preventing adverse impacts on the power system. Working Principle of Syntactic Filter

1. Measurement: Input current and voltage are measured.

2. Conversion: Voltage and current measurements are converted into Alpha-Beta quantities.

3. Power Calculation: Apparent power is calculated using Alpha-Beta quantities.

4. Filtering: A syntactic filter injects compensating currents to maintain sinusoidal source currents.

5. Control Algorithm: Traditional Proportional-Integral (PI) controllers or Neural Network controllers are used for control.

Syntactic Power Filter with Neural Network Control In the presented model, a neural network controller is employed instead of a traditional PI controller to enhance the filtering process. The neural network is trained based on collected data, allowing it to adapt and improve the accuracy of compensating current generation. Neural Network Training The neural network is trained using input data (reference voltage and error) and corresponding output data. The training process involves iterations, adjusting the network's weights to minimize the error between the predicted and target outputs. The success of training is indicated by the correlation coefficient (R), with a value of 1 indicating a perfect match. Simulation Results Simulation results demonstrate the effectiveness of the neural network-controlled syntactic power filter in compensating for harmonic distortions. The source current is shown to closely follow a sinusoidal waveform, and the Total Harmonic Distortion (THD) is reduced compared to traditional controllers. Conclusion The integration of neural network control in syntactic power filters presents a promising approach to mitigate harmonic distortions in power systems. By enhancing the adaptability and learning capabilities of the controller, the system can efficiently compensate for non-linear loads, maintaining the integrity of source currents and overall power quality. This advancement offers potential benefits in reducing THD and improving the performance of power systems under varying load conditions.