Multivariate Outliers Detection for Assessing Data Quality and Enhancing Interpretability in Photovoltaic Power Predictions

Abstract

Efficient energy prediction is essential for maintaining a balance between demand and supply. The accuracy of the prediction process relies significantly on accurate data processing. Within this process, outliers detection stands out as a critical step, ensuring the reliability and precision of energy predictions. This study focuses on the integration of a Multivariate Outliers Detection (MOD) approach within the Long-Short Term Memory (LSTM) model for predicting Photovoltaic (PV) power data. Encompassing 31 PV arrays with diverse technologies and structures. Through this investigation, we aim to shed light on the impact of multivariate outliers in predicting the PV power using LSTM model, offering valuable insights for advancing the reliability and explainability of PV power predictions.