| The feed water pump system of the boiler is the heart of the vapor-water hydrodynamic circulation.It is one of the most important auxiliary systems for industrial production processes such as coal-fired power plants.The research on the fault detection and diagnosis of the feed pump system is of great significance to ensure the safe and stable operation of the generator set and the more reasonable and finer load schedule of the power grid.In the method of fault diagnosis of the feed pump system,the method based on signal analysis has a narrow coverage of fault types,and the model-based or knowledge-based methods have difficulties in highprecision modeling or knowledge acquisition and utilization.Therefore,the research and simulation verification of the fault detection and diagnosis of the feed pump system from the data-driven direction,the main research contents include:1.A simulation model of the feed pump system that can simulate multiple faults is established to solve the problem of lack of fault data.On this basis,the faults of some common faults of physical objects and sensors in the system are modeled.Subsequently,the simulation model was built based on SimulationX.The accuracy of the normal model was verified by the actual running data of the pumping system of a power plant.Simulation data for normal and fault conditions is generated,providing data for fault detection and diagnostic model construction and testing.2.A fault detection method for feed water pump system based on state space principal component analysis network(SSPCANet)is proposed.This method introduces the PCANet deep learning network structure into the field of fault diagnosis,and combines with the multistatistical process monitoring method,making full use of the advantages of excellent feature extraction and full parameter monitoring.In addition,the state space model is added as a dynamic layer in the network structure to adapt to the dynamics of the variable pumping system.The output layer is redesigned to fit the target of fault detection.Finally,simulation tests for fault detection were performed on the TE process and the feed pump system and compared with similar methods.3.Combining SSPCANet with Least Squares Support Vector Machine(LSSVM),the SSPCANet is used to extract features from the original data,and then sent to the multi-class LSSVM model for fault type determination.In order to simplify the selection and tuning of LSSVM’s complex hyperparameters,the method of searching for hyperparameters by genetic algorithm and k-fold cross-validation is studied.Finally,a simulation test of fault diagnosis was performed on the feed pump system and compared with the traditional LSSVM. |
