Solar Radiation Intensity Imputation in Pyranometer of Automatic Weather Station Based on Long Short Term Memory

Abstract

Automatic Weather Station (AWS) experienced problems in the form of component damage and communication system failure, resulting in incomplete parameter data. Component damage also occurs in pyranometers. Decreased pyranometer performance results in deviations, uncertainty in measuring solar radiation intensity, and data gaps. Data imputation is one solution to minimize measurement deviations and the occurrence of missing AWS pyranometer data. This research aims to design and analyze the accuracy performance of the multisite AWS pyranometer solar radiation intensity data imputation model when a data gap occurs. This research attempts to utilize the spatio-temporal relationship of multisite AWS solar radiation intensity in the imputation model. Long-Short Term Memory (LSTM) algorithm is used as an estimator in the multisite AWS pyranometer network. Data imputation modeling stage includes data collection, data pre-processing, creating missing data scenarios, LSTM design and model testing. Overall, LSTM-based imputation model has ability of filling gap data on AWS Cikancung pyranometer with maximum missing sequence of 12 hours. Imputation model has MAPE 1.76% - 5.26% for missing duration 30 minutes-12 hours. It still it meet WMO requirement for solar radiation intensity measurement with MAPE<8%.