TC3.2 Smart Energy-aware Zero-touch Traffic Engineering – ML Inference Engine (TC32_ML)

This repository provides the ML inference engine for the ACROSS TC3.2 use case (Smart Energy-aware Zero-touch Traffic Engineering). The software predicts router power consumption from instantaneous network telemetry-derived metrics to support Network Digital Twin (NDT) environments and energy-aware traffic engineering. The inference service is containerized (Docker/Docker Compose) and uses Kafka for message exchange: it consumes input telemetry features, runs a selected ML model, and publishes predicted power consumption along with feature-based variation rates. The repository also includes a telemetry mock service to replay experiment datasets, validate predictions against ground truth, and generate verification plots. Supported router types: RA and RB. Supported model types: linear regression, polynomial regression, random forest, and deep neural networks (with optional scaling). Models are organized by router type/model type and selected dynamically via environment configuration.