Implementation of Integrating PV System Production Forecasting Using Recurrent Neural Networks in Local Weather Station Prototype

Abstract

This study explores the crucial role of weather stations in measuring, collecting, and reporting weather data, as well as the implementation of modern technologies such as Long Range (LoRa) radio wave modulation technology for real-time data monitoring. Equipped with components like temperature, humidity, solar radiation, and wind sensors, the weather station ensures accurate and efficient data collection. Testing of LoRa technology at the PEM Akamigas Campus demonstrated an effective range of approximately ±85 meters, ensuring optimal connectivity between the Subroto Building and the Energy Laboratory Building. Data consistency from the Message Queue Telemetry (MQTT) protocol server and Haiwell Human-Machine Interface (HMI) confirms the reliability of weather monitoring. Additionally, this research focuses on weather and energy production predictions for the PV system at the Subroto Building, using an Recurrent Neural Network (RNN) deep learning model to enhance the accuracy of solar panel energy production forecasts. Data evaluation from April 1, 2024, to April 22, 2024, highlights the potential. Based on the real-time sensor data installed in the field on a combination of 3 series solar panels, resulted in production forecasting with Root Mean Square Error (RMSE) values of approximately 4.9965 for voltage, and 0.0081 for current. This indicates fairly satisfactory results. For power testing, the RMSE results are still unsatisfactory, highlighting an opportunity for future model improvements. The combination of LoRa technology and the RNN model is expected to provide valuable insights into reliable weather monitoring and energy production at the PEM Akamigas Campus, with improvements to the model parameters for power data, which is inherently derived from the multiplication of voltage and current parameters.