This work presents a comprehensive design and implementation of an AI-enabled Smart Power Factor Correction (PFC) System that integrates advanced technologies such as Machine Learning (ML), Deep Learning, IoT, Edge Computing, and Blockchain with conventional PFC hardware. The proposed system intelligently compensates reactive power and improves power factor in real time by dynamically switching capacitor banks based on load predictions and power quality analysis. At the hardware level, the system utilizes components such as Arduino Uno, ACS712 current sensor, LM358 op-amp, single-channel relays, and ceiling fan capacitors, while more advanced processing is supported through ESP8266/ESP32 modules for connectivity and Jetson Nano or Raspberry Pi for edge AI inference. The ML algorithms, trained using historical load data and power quality parameters, run either on embedded microcontrollers (TinyML) or edge devices for low-latency decision-making. Additionally, a smart capacitor bank is used to provide fine-grained control over reactive power compensation, and system logs are securely recorded through a lightweight blockchain node to ensure transparency in smart grid environments. The integrated ThingsBoard and Node-RED dashboard enables remote monitoring and real-time analytics for system adaptation and performance tracking. Simulation and hardware results demonstrate a significant improvement in power factor correction accuracy and response time compared to conventional fixed or manually switched capacitor systems. The proposed AI-driven model not only adapts to dynamic and nonlinear load conditions but also reduces over- or under-compensation through predictive switching. Comparative analysis confirms enhanced Total Harmonic Distortion (THD) reduction, power factor stabilization, and improved system resilience under varying load profiles. The integration of AI and smart technologies thus marks a promising advancement toward intelligent, autonomous, and transparent power quality enhancement in next-generation smart grids.
Prakash et al. (Wed,) studied this question.