Enhancing BIPV Building Energy Autonomy in PV systems through predictive energy control and real-time error reduction

The integration of photovoltaic (PV) systems into power grids poses challenges due to the intermittent nature of solar energy, which increases grid dependency and limits energy autonomy. Accurate ultra-short-term (UST) forecasting and intelligent energy management are essential to enhance PV self-consumption. This study introduces the Energy Control Optimization Predictive Intelligent Management System (ECO-PIMS), an advanced framework that integrates hybrid forecasting models, internal adaptive real-time error correction, and dynamic energy allocation via a logical control function (LCF). ECO-PIMS employs advanced signal-decomposition techniques, including Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), Sample Entropy (SE), and Variational Mode Decomposition (VMD). Two novel hybrid models are developed: ICEEMDAN-VMD-CNN-LSTM-BiGRU (IV-CNNLSTM-BiGRU) for PV-production forecasting and ICEEMDAN-VMD-LSTM-BiGRU (IV-LSTM-BiGRU) for energy-demand forecasting. In the forecasting process, Convolutional Neural Networks (CNN) extract spatial features and Long Short-Term Memory (LSTM) networks learn temporal patterns for PV production, while a Bidirectional Gated Recurrent Unit (BiGRU) refines residual errors. ECO-PIMS optimizes real-time energy management by dynamically allocating power between PV and the grid based on UST predictions. Simulations over seven days demonstrate that the PV–Battery–Grid configuration with error correction reduces grid contribution to 17.7%, compared to 29.7% without correction and 32.4% in a PV–Grid setup with correction. Moreover, the proposed models achieve a 20% reduction in Mean Absolute Error (MAE) compared to conventional CNN and LSTM approaches. This study uses real measured data from a positive energy building (PEB) in Poschiavo, Switzerland. The proposed methodologies were implemented in MATLAB and Simulink, underscoring their practical applicability.