Top

An Improved PointNeXt Mixture of Experts Model for Compressed Air Energy Storage Cavern Pressure Prediction Using Injection and Production Histories and Geological Monitoring Data

Authors

Pietro De Luca
Department of Civil, Environmental and Architectural Engineering, University of Bologna, Bologna, 40136, Italy
Riccardo Amato
Department of Civil, Environmental and Architectural Engineering, University of Bologna, Bologna, 40136, Italy

Files

Abstract

Pressure forecasting for compressed air energy storage caverns requires a consistent treatment of changing air throughput, delayed thermal response, and spatially uneven geological observations. This paper proposes a PointNeXt mixture of expert’s architecture that combines a causal operation encoder with a hierarchical point representation of monitoring locations. Observation quality indicators accompany every measurement, while a sparse gate allocates the fused representation to operating condition specialists. A thermodynamic reference supplies the pressure trend, and a constrained residual head represents departures associated with monitored conditions. The evaluation examines pressure trajectories and forecast errors across operating conditions. Across 1,200 forecast origins and six regimes, the proposed prediction series has a root mean square error of 0.043 MPa, compared with 0.064 MPa for the PointNeXt comparison and 0.056 MPa for the unguided expert comparison. The difference relative to the PointNeXt comparison is 32.8%, providing a quantitative basis for the component sensitivity assessment. Missing observation scenarios increase the error from 0.043 to 0.072 MPa as removal rises from zero to 40%. The analysis specifies leakage resistant partitioning, interval assessment, component removal tests, and operational rejection criteria needed for subsequent empirical evaluation. The contribution is an implementable forecasting design and an auditable evaluation framework with explicit criteria for subsequent independent validation.

Details