A New Low-Cost Internet of Things-Based Monitoring System Design for Stand-Alone Solar Photovoltaic Plant and Power Estimation

Abstract

The increasing demand for solar photovoltaic systems that generate electricity from sunlight stems from their clean and renewable nature. These systems are often deployed in remote areas far from urban centers, making the remote monitoring and early prediction of potential issues in these systems significant areas of research. The objective here is to identify maintenance requirements early and predict potential problems within the system. In this study, a cost-effective Internet of Things-based remote monitoring system for solar photovoltaic energy systems is presented, along with a machine learning-based photovoltaic power estimator. An Internet of Things-compatible data logger developed for this system gathers critical data from the photovoltaic system and transmits them to a server. Real-time visualization of these data is facilitated through web and mobile monitoring interfaces. The measured data encompass current, voltage, and temperature information originating from the photovoltaic generator and battery, alongside environmental parameters such as temperature, radiation, humidity, and pressure. Subsequently, these acquired data are employed for photovoltaic power estimation using machine learning techniques. This enables the estimation of potential issues within the photovoltaic system. In the event of a problem occurring within the photovoltaic system, users are alerted through a mobile application. Early detection and intervention assist in preventing power loss and damage to system components. When evaluating the results according to performance assessment criteria, it was observed that the random forests algorithm yielded the best results with an accuracy rate of 87% among the machine learning methods such as linear regression, support vector machine, decision trees, random forests, and k-nearest neighbor. When prediction models using other algorithms were ranked in terms of success, decision trees exhibited an accuracy rate of 81%, k-nearest neighbor achieved 79%, support vector machine reached 67%, and linear regression achieved 64% accuracy. In conclusion, the developed monitoring and estimation system, when integrated with web and mobile interfaces, has been demonstrated to be suitable for large-scale photovoltaic energy systems.