.png){kind=icon}
.png){kind=icon}
Return
Attention-gated hybrid ANN–TCN–BiLSTM framework with explainable AI for operational efficiency classification in PV–EV microgrids
.png){kind=icon}
OA
AI
Y
DOI：10.1038/s41598-026-67083-y.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The increasing integration of photovoltaic (PV) generation and electric vehicle (EV) charging infrastructure into microgrids creates operational challenges arising from renewable intermittency, stochastic charging demand, and varying grid dependence. This study proposes an explainable attention-gated hybrid framework for binary operational efficiency classification in PV–EV microgrids. The framework integrates an artificial neural network (ANN) branch for static and slowly varying operational variables, a temporal convolutional network (TCN) branch for short-term local variations, and a bidirectional long short-term memory (BiLSTM) branch for sequential dependencies. An attention-gated fusion mechanism adaptively combines the representations learned by these branches before classification. Experiments were conducted using a publicly available synthetic or semi-simulated dataset containing 12,500 samples recorded at 15-minute intervals. A chronological 70:15:15 training, validation, and test split was employed to preserve temporal order and prevent information leakage. The proposed model achieved an accuracy of 0.962, precision of 0.957, recall of 0.966, F1-score of 0.961, area under the receiver operating characteristic curve (ROC-AUC) of 0.989, area under the precision–recall curve (PR-AUC) of 0.984, and Matthews correlation coefficient (MCC) of 0.923. It outperformed ANN, recurrent, TCN, attention-based BiLSTM, and lightweight Transformer baseline models. Shapley additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME) showed that solar generation, solar irradiance, grid energy supply, battery storage capacity, and EV charging demand were the main factors influencing classification decisions. Because the target variable is rule-based and the dataset does not contain measurements from a field-deployed microgrid, the findings should be interpreted as a controlled methodological benchmark. External validation using measured PV–EV microgrid data remains necessary before practical deployment.
Keywords:
PV–EV microgrid
Operational efficiency classification
Attention mechanism
Temporal convolutional network
Explainable artificial intelligence
Journal
IF:
3.9
Papers:
27.1W
Citations:
83.5W
Organization
Cited Papers