top of page

MATLAB Implementation of Neural Network Control of Shunt Active Filter

Writer: lms editorlms editor

Introduction to Nonlinear Loads and Their Impact

Nonlinear loads, such as rectifiers or semiconductor devices, are commonly used in modern electrical systems. These devices cause irregularities in the power system, as their electrical characteristics are nonlinear. When such devices are connected to the system, they affect the quality of the source current. Normally, when a linear load is connected, the current waveform is sinusoidal. However, when a nonlinear load is connected, the current becomes distorted, deviating from the ideal sinusoidal form.

This distortion can result in reactive power consumption, which negatively affects both the grid and the connected loads. To mitigate these issues, compensation techniques are required to maintain a stable and sinusoidal source current.

The Role of Shunt Active Filters

A Shunt Active Filter (SAF) is an essential component in mitigating the problems caused by nonlinear loads. The SAF operates by injecting a compensating current into the system. This injected current helps to eliminate harmonics and restore the source current to a sinusoidal waveform.

The SAF consists of a three-phase inverter and a capacitor connected in parallel to the grid. The system continuously measures the grid voltage and current, as well as the voltage across the capacitor. Using this information, it generates pulses for the inverter, allowing it to inject compensating currents into the system.

Power Calculation for Compensation

For the active filter to effectively compensate for reactive power, it needs to calculate the real and reactive power of the system. This is achieved by converting the grid's three-phase voltage and current into Alpha-Beta quantities. The conversion allows for the accurate calculation of real and reactive power in the system, which are essential for generating the compensating current.

By using a matrix form for these calculations, the system can determine the required power for compensation. The calculated real power is then compared with the reference voltage to adjust the system's operation.

Control Algorithm: Managing the Compensation Process

To achieve effective harmonic compensation, a sophisticated control algorithm is necessary. The algorithm involves several key steps:

  1. Voltage and Current Conversion: The first step is converting the grid voltage (VA, VB, VC) and the load current into Alpha-Beta coordinates.

  2. Real and Reactive Power Calculation: Using these converted quantities, the system calculates the real and reactive power in the power system.

  3. Compensating Current Generation: The calculated power is used to generate the compensating current, which is then injected into the system via the inverter.

The process ensures that even when the load current is nonlinear, the source current remains sinusoidal, improving overall power quality.

Integrating Neural Network Control

To further enhance the performance of the SAF, a Neural Network (NN) controller can be integrated into the system. Unlike traditional Proportional-Integral (PI) controllers, which rely on predefined rules, the NN controller learns from the system’s behavior and adapts to changes in real-time.

The neural network receives two inputs: the reference voltage and the error voltage. By processing these inputs, the NN controller generates the necessary power for compensating reactive power. This dynamic approach helps improve system performance, especially in complex or unpredictable scenarios.

Training the Neural Network for Power Compensation

Training a neural network involves providing input and output data for the system. The input data consists of reference voltage and error voltage, while the output data is the desired loss power to be compensated. The network uses this data to learn how to generate the appropriate control signals for the inverter.

The neural network controller is trained using an iterative process, and the performance is evaluated based on the correlation coefficient (R-value). A high R-value indicates that the neural network has been successfully trained and can accurately model the system’s behavior. After training, the NN controller is used in the SAF to manage real-time power compensation.

Comparing PA Controller and Neural Network Controller

A key advantage of using a neural network controller is its ability to reduce Total Harmonic Distortion (THD) more effectively than traditional PI controllers. In simulations, the neural network controller consistently outperforms the PA controller in reducing THD levels, resulting in a more efficient harmonic compensation system.

  • PA Controller Performance: With the PA controller, the THD of the source current was measured at 4.63% in the system.

  • Neural Network Controller Performance: The neural network reduced the THD to 4.53%, proving that the NN controller provides a more precise and adaptive solution for harmonic compensation.

Conclusion: The Future of Power System Compensation

The integration of Shunt Active Filters with Neural Network Controllers offers a promising solution for managing reactive power and mitigating harmonic distortion in power systems. By continuously adapting to changes in the load and power system, the neural network controller ensures that the system operates at optimal efficiency, maintaining a sinusoidal source current and improving the overall quality of power.

As power systems become more complex with increasing nonlinear loads, advanced control techniques like Neural Network-based SAFs will play a crucial role in ensuring system stability and power quality. This approach marks a significant step forward in the field of power electronics, providing a more intelligent and responsive way to manage power system compensation.

 
 
 

Comments


bottom of page