Solar Energy Optimization Using Similarity Search

This paper presents a total solar power analysis system providing important information on how effectively solar plants are operating by means of machine learning, natural language processing, and modern online technologies. The system takes in raw plant data. Included in this data are system identifiers, daily and total energy yield, time-stamped power measurements such DC and AC output. A multi-algorithm clustering pipeline is utilised to identify operational patterns and anomalies through the application of KMeans, DBScan, and HDBScan techniques, each providing unique advantages in pattern recognition and density estimation. The system enhances analytical depth by integrating natural language processing, generating 768 -dimensional embeddings through a nomic-text model implemented via Ollama. The embeddings encapsulate latent relationships in the solar data and are stored in a vector-enabled Supabase database, indexed for efficient similarity searches utilising the IVFFlat algorithm. A custom query function facilitates threshold-based similarity matching, enabling accurate identification of related performance profiles.