A novel architecture for a high-accuracy solar irradiance meter with artificial neural network and IoT integration

Abstract

This paper presents a novel architecture for a portable, rechargeable, and cost-effective solar irradiance meter for wireless integration into photovoltaic systems. Designed as an alternative to costly commercial devices, the meter's performance was validated through LTspice simulations using a PV cell and an amplifier circuit to evaluate behavior under varying solar irradiance and temperature conditions. An Artificial Neural Network was trained on the simulation data to predict solar irradiance based on the amplifier output voltage and PV cell temperature. The ANN model was embedded in an ESP8266 microcontroller mounted on a custom-designed printed circuit board and powered by a lithium battery. Acting as a server, the ESP8266 samples voltage and temperature upon client requests to calculate and transmit solar irradiance values. The system can operate continuously 24 h a day and supports simultaneous connection to multiple PV systems. Experimental tests against a commercial meter demonstrated a mean square error of 9.96 W/m², a normalized root mean square error of 1.7%, a correlation coefficient R = 0.9992, and a coefficient of determination R² = 0.9984. Compared with previously reported irradiance meters in the literature, the proposed system achieves higher accuracy and real-time wireless access for multiple PV panels, demonstrating both superior measurement performance and enhanced operational flexibility. The solar irradiance meter can be used in PV systems for multiple-user access, improvement of Maximum Power Point Tracking algorithms, and monitoring of panel performance. It also has potential applications in agriculture, remote monitoring, and smart city systems.