1540 framework (e.g., Apache Airflow), ensuring continuous learning with minimal manual intervention. 4. Model Replacement Criterion A newly trained model replaces the deployed model only if it achieves a validation Root Mean Square Error (RMSE) improvement exceeding 5%, thereby preventing performance degradation due to noise or overfitting. This adaptive mechanism enhances long-term usability by ensuring that the deployed model remains both accurate and operationally relevant. 2.4 Feature Set The machine learning models leverage a comprehensive set of features derived from vehicle telemetry, road characteristics, and physical constraints. These features are selected to capture the key factors influencing vehicle speed and cycle time,such as Road and Geometry Features (road grade and road curvature), Vehicle and Load Characteristics (payload weight), Driver Control Inputs (throttle position and brake usage), Powertrain and Fuel Metrics (engine RPM and fuel consumption rate), Temporal and Contextual Features (speed in the previous road segment), Physical and Safety Constraints (physicsbased maximum feasible speed), and Road Condition Indicators (road roughness, quantified as the variance of vehicle acceleration). Collectively, these features enable the model to integrate physical feasibility, operational behavior, and environmental conditions, thereby improving both prediction accuracy and real-world applicability. 2.5 Model Output The trained machine learning model produces operationally interpretable outputs that serve as direct inputs to downstream decision-support components. Specifically, the model generates the following predictions for each road segment and time interval: • Optimal Speed (¡optimal) This output represents the recommended vehicle speed that optimally balances safety constraints, road geometry, payload conditions, fuel efficiency, and driving smoothness. The value is dynamically adjusted based on current operating conditions and learned performance patterns. • Predicted Segment Cycle Time (j[æï[ÅM_WNWÑ[_MTï[) This output estimates the expected time required to traverse the current road segment under the given operational context. It enables short-term performance forecasting and supports higher-level planning functions such as dispatch optimization and productivity monitoring. These outputs are designed to be numerically stable, interpretable, and suitable for realtime consumption by operator-facing systems. 2.6 Operator Guidance Engine To bridge the gap between model predictions and practical operator actions, the system incorporates an operator guidance engine that translates continuous numerical outputs into clear, actionable coaching messages. This component is implemented as a lightweight, rule-based layer to ensure transparency, predictability, and ease of validation in safety-critical environments. The guidance engine evaluates real-time vehicle telemetry in conjunction with the model outputs and triggers contextual feedback based on predefined operational rules.
RkJQdWJsaXNoZXIy MTM0Mzk2