The system is based on machine learning algorithms (ARIMA, linear regression, LSTM neural networks, XGBoost, etc.). Predictions are generated in two modes: for the next day and for the next hour. The training data includes hourly readings from individual installations, weather conditions, and more. The system generates over 300 AI models daily, achieving an accuracy of approximately 95%. The models have been deployed and integrated with other systems via API.
The system is built on a hybrid architecture: data is collected and integrated on-premise, then anonymized and sent to the cloud. AI model training takes place in the cloud, after which the trained models are copied back to on-premise systems (where de-anonymization occurs). The cloud infrastructure is used exclusively for training the AI models.
