Energy Consumption Prediction for Hydraulic Systems Based on the Fusion of Mechanistic Models and Graph Neural Networks

However, owing to complex energy-transfer stages and loss mechanisms, mathematical models yield limited prediction accuracy due to time-varying parameters, whereas data-driven approaches lack physical interpretability generalization. This study proposes an energy-consumption prediction model that integrates an AMESim simulation with a heterogeneous graph neural network (HGNN).

First, the system configuration and drive principle are analyzed, and a mechanistic model describing the evolution of energy consumption in the hydraulic drive system is established from the perspective of energy flow. A physics-based simulation model is then developed in AMESim together with a communication interface for external algorithms, thereby ensuring broad applicability at the mechanistic level. Next, the component and circuit topology of the hydraulic system and the relationships among key state variables are represented by a graph neural network.