Research Of Fault Detection Methods In HVAC System Based On Data Statistics Characteristics

Energy consumption of the HVAC system is huge and in large part at present,because system fault can’t timely be detected,thus effective fault detection is the key to guarantee the system performance and energy consumption.HVAC system fault detection has four difficult problems(parameter drift,multiple modes,incipient failures and pre-existing failures).Traditional fault detection methods are difficult to meet requirements.Considering the characteristics of HVAC system contains a lot of similar units,from the viewpoint of mathematical statistics,this paper presents a fault detection method based on statistical properties of similar unit parameters,and the characteristics is analyzed,failures in the units is checked.In this paper,a special class of hybrid system mathematical models is introduced to describe such multi-unit systems.The state of each unit was estimated by Multimode filter method based on particle filter.Then according to the state estimation we can establish the Gaussian mixture distribution,and adopt hybrid clustering algorithm to estimate the unknown parameters of the Gaussian distribution.The hybrid clustering algorithm improves the iteration speed by using granular computing method.And the rival penalization mechanism is proposed when the number of clusters is unknown.Then each unit parameter mapping into the common parameter clustering space,and we can obtained each unit’s state.Finally,according to the statistical features of unit with the same operation mode,proposed the residual chi-square method based on the sequential probability ratio decision to detect whether there are failures in the units.Finally,the paper finished the simulation experiment by using the simulated air conditioning system experimental platform and MATLAB simulation.Firstly,the results compared the hybrid clustering with traditional clustering from three aspects of the accuracy,rapidity and stability of the clustering.Secondly,fault detection method proposed in this paper can detect various faults effectively.