Non-Integer Order Polynomial Network: A Novel Fault Diagnosis Tool

This paper introduces a novel Non-Integer Order Polynomial Network (NIOPN) designed to achieve higher approximation accuracy by employing fractional moments to approximate nonlinear functions, compared to the integer moments used in traditional Multidimensional Taylor Networks (MTN). The fractional moment polynomials offer richer and more precise mathematical representations, enhancing the accuracy of function approximation. To validate the effectiveness of NIOPN, we applied it to fault diagnosis in Air Handling Units (AHU) within HVAC systems. Fault diagnosis in AHU systems is critical for maintaining operational efficiency and preventing system failures. By utilizing NIOPN, our model achieved a high diagnostic accuracy of 100%, outperforming traditional methods such as random forest, support vector machine, K-nearest neighbors, and the latest residual MTN. These results demonstrate NIOPN's strong potential as a powerful tool for fault diagnosis in HVAC systems. Given its versatility as a general model, NIOPN can also be employed in in other classification, regression, and prediction tasks. This model presents a new universal approach distinct from neural networks, offering a new tool for pattern recognition tasks.