<p>The abnormal electricity consumption (EC) prediction is essential in the electric power sector. Precise prediction is crucial for electricity supply regulations. A defining property of electrical energy is the necessity to maintain a consistent balance between consumption and generation, as electricity cannot be stored in substantial amounts and is not easily transported. The key objective is to classify EC patterns as abnormal or normal using deep learning (DL) techniques. The developed abnormal EC detection model follows a systematic workflow to handle data effectively and achieve high classification accuracy. The process begins with data collection using the State Grid Corporation of China (SGCC) dataset for analysing diverse consumption patterns. The Min-Max normalization technique is employed in data preprocessing. The variational autoencoder (VAE) model reduces the high-dimensional data into a compact latent space, preserving key features while minimizing computational complexity. In clustering, density-based spatial clustering of applications with noise (DBSCAN) is used to identify clusters of similar consumption patterns and detect anomalies as outliers, utilizing its ability to handle noise and non-linear cluster shapes effectively. Finally, classification is performed using a capsule network (CapsNet), which captures complex spatial hierarchies and relationships in the data. CapsNet’s performance is further enhanced through hyperparameter tuning using the salp swarm algorithm (SSA). The hyperparameter tuning ensures optimal settings for improved accuracy and efficiency. This integrated approach enabled precise identification of abnormal EC patterns. The research model performance is evaluated with metrics such as accuracy, precision, recall, F1-score and achieved 98.84%, 98.70%, 98.51% and 98.68% respectively. The model outperformed all the compared models in this research with better performances.</p>