Model parameter optimization for chiller fault diagnosis using differential evolution and genetic algorithms

To reduce operating costs and prevent equipment damage through early fault detection in chillers, this study proposes an advanced diagnostic strategy integrating the Radial Basis Function (RBF) model with Exponentially Weighted Moving Average (EWMA) residual charts. While RBF models are proficient in handling nonlinear problems, they are often susceptible to overfitting and performance degradation under data disturbances. These limitations are overcome to optimize model parameters by the Differential Evolution (DE) and Genetic Algorithm (GA), achieving not only higher coefficient of determination coefficients (R2) but also low root mean square errors for two common types of nonlinear faults such as Refrigerant Leakage (RF) and Condenser Fouling (CF). Analysis using the ASHRAE RP-1043 dataset shows that the DE-RBF-EWMA model achieves a diagnostic confidence of 98.48% for the LMTDcd parameter and 98.56% for the Epsilon parameter. This strategy effectively identifies RF and CF across four severity levels, providing a robust tool for real-time chiller monitoring. The true alarm rate of the DE-RBF-EWMA model is the best, especially important at high serious level of faults, reaching 94.30% of RF and 74.90% of CF.