With the widespread usage of nonlinear loads and development of renewable energy in the smart grid, the power-quality disturbances related to the harmonic current and reactive power would be introduced in the power system. These phenomena may result in the misoperation, additional power loss, and reduction in service life for electric devices [
1]. Hence, the compensation for the load current to modify the power quality and meet the system requirements is the important issue for the modern power grid. The commonly seen harmonic current mitigation strategies can be divided into the shunt passive power filter and the SAPF. The shunt passive power filter is a traditional and simple way to mitigate the harmonic distortion, which is composed of inductors and capacitors. In recent years, many researches related to the shunt passive filter have focused on the optimal design. By considering the design limitations, the optimal sizing strategy is proposed in [
2] to deal with the harmonic distortion for the specific components. In [
3], the optimization of size for a novel-type passive filter is developed under non-sinusoidal conditions. To reduce the harmonic disturbances for the demand side, the strategy based on the multi-objective Pareto algorithm is applied in [
4] to perform the optimal design of the passive power filter. However, the shunt passive power filter is only useful for the elimination of certain harmonic components. Moreover, the shunt passive power filter may lead to the series or parallel resonance with the system impedance. Due to the characteristics of the dynamic compensation for the time-varying harmonic, the system regulation for the reactive power, the flexible tuning for the specific harmonic components, and the improvement of power factor, the SAPF has become the mainstream strategy for the regulation of power quality [
5].
From the research literature, it is found that many effective methods have been proposed to implement the control of the SAPF, including the calculation of the reference compensation current, the phase synchronization for the SAPF compensation, the regulation of the DC-link voltage, the detection of fundamental and harmonic components, etc. In general, many researches apply the instantaneous reactive power compensation control technique (p-q method) for the control of the SAPF [
6]. However, the performance of the p-q method may be deteriorated by the deviation of the power system frequency, interharmonics, and distorted source voltage. To solve the drawbacks of the traditional p-q method, the method based on sliding discrete Fourier transform (DFT) is used in [
7] to separate the fundamental positive-sequence and harmonic components. Due to the requirement of at least one complete cycle for the calculation of harmonics, the sliding DFT method would suffer from the problem of delayed compensation response. The second-order generalized integrator combined with a comb filter and the instantaneous reactive power compensation theory are applied in [
8] to perform the extraction of the phase angle, system frequency, and positive/negative sequences of fundamental component to deal with the limitations of the adaptive notch filters. To improve the compensation response of the SAPF, the algorithm based on an adaptive linear neural network is applied in [
5], [
9] to perform the real-time harmonic mitigation. Due to the Fourier series-based structure of the adaptive linear neural network, the compensation performance would be interfered with the interharmonics, which are difficult to include in the solution model in advance. The optimized finite impulse response predictor is implemented in [
10] to compensate the computation delay, where the optimization of the cost function is performed offline. With the development of intelligent control, many advanced neural network-based strategies have been proposed in the literature. In [
11], the recurrent neural network (RNN)-based controller is applied for the approximation of the unknown nonlinear function of the SAPF and modification of the compensation performance. The controller based on fuzzy neural network (FNN) is utilized In [
12] to attenuate the effect of arbitrary external disturbances and modeling uncertainties in the process of the SAPF compensation. A compensation method based on recurrent probabilistic fuzzy neural network and global sliding mode control is proposed in [
13] to enhance the steady-state and dynamic performance of the SAPF. In [
14], the feature selection method with a recurrent neural network is proposed to fit the uncertain function and improve the performance of the conventional global sliding mode control for the SAPF. A self-evolving emotional neural network is applied in [
15] to implement a model-free control system to provide fast convergence and global robustness of the SAPF controller. From the above-mentioned researches, it is found that the proposals are mainly focused on the improvement of the SAPF compensation accuracy to meet the requirement of the harmonic injection to the power system in IEEE Standard 519-2022 [
16]. Therefore, the power system frequency deviation due to the power mismatch between the energy source and demand side, interharmonics, and distorted source voltage is usually not considered in the solution procedure of the SAPF compensation. In this way, the inaccurate analysis for the reference compensation current and the error of the phase synchronization would be introduced and deteriorate the performance of the SAPF.
In this paper, the control strategy of the reference compensation current with the RWFNN is proposed to perform the accurate fundamental positive-sequence extraction and regulation of the DC-link voltage even when the deviation of power system frequency, distorted source voltage, and interharmonics are present. The main characteristics of the proposed control strategy can be summarized as follows.
The organization of this paper can be listed as follows. In
Section 2, the proposed control strategy of the reference compensation current based on the RWFNN controller is introduced, including the fundamental positive-sequence extraction, the calculation of the synchronization phase, the regulation of the DC-link voltage, and the calculation of the reference compensation current. Several comprehensive case studies with a real-time simulator are performed in
Section 3 to examine the effectiveness of the proposed control strategy for the SAPF.