ScholarWorks@성균관대학교

초록

The HVAC system is an important part of the building system, and sensor failures can lead to a series of problems in the HVAC system, so sensor reliability has a pivotal role in intelligent building systems. Early sensor failures are characterized by hidden propagation and are very likely to evolve into global multivariate failures due to system thermal inertia and time delay effects. The aim of this study is to develop an intelligent co-simulation platform-based sensor fault prediction and tracing methodology for HVAC systems to address the propagation of hidden sensor faults caused by thermal inertia and time-delay effects. The method combines the WOA with a Long Short-Term Memory (LSTM) network to construct an optimal prediction model and introduces TE to quantify the time delay under the optimal spatio-temporal game conditions. And eight fault case studies corresponding to different sensors under the building energy system are conducted to fully measure the accuracy of the method for prediction and traceability. The experimental results show that the prediction accuracy reaches 97.5 %, and it can accurately localize the pre-injected sensor faults of various sizes and effectively differentiate between input and output sensor fault categories. These results demonstrate the method's ability to accurately predict and reliably trace early sensor faults under complex operating conditions, and the proposed method provides a basis for predictive maintenance and operation optimization in intelligent building management. This research innovatively integrates meta-heuristic algorithms, deep learning, and information theoretic analysis into a unified framework for HVAC sensor fault prediction and tracing.

키워드