| Thermal power units are and still will be dominant in China’s power industry for a long time.As the vital important part of the unit,the thermal system operation status affects the safe,economic and stable operation of the entire unit.Due to the complex structure of the thermal system,the coupling between equipments,the numerous parameters,and the long-term operation under variable load conditions,the thermal equipment is more prone to failure and difficult to diagnose.If a fault cannot be found and eliminated in time,a chain reaction may occur,causing a larger-scale accident and impacting the stable operation of the power grid.Therefore,it is essential to monitor the operating parameters of the system in real time,to predict and diagnose the possible fault promptly,and remind the operation and maintenance personnel to deal with it timely.The high-pressure feedwater heater system of a 600 MW supercritical unit is taken as the object investigated,and the simulation experiments are carried out with the help of the full-scope simulator of the given unit.The fault-free operation data of the unit under variable load conditions are extracted.The normal performance predictive model of the high-pressure heater system is established with a long short-term memory(LSTM)network,and the model is trained and verified.it is shown that the model is with splendid fitting and high accuracy,and can accurately predict the characteristic parameters of the high-pressure heater system.In order to carry out simulation research and analysis of the typical fault rules of the high-pressure heater system,the operation data under different faults are extracted from the simulator.In the light of the symptom fuzzy calculation method,the standard sample knowledge base of the typical faults of the high-pressure heater system is constructed.The fault diagnosis model of the high-pressure heater system is established with LSTM network,which is trained and validated offline.By Combining the above-established LSTM performance prediction model and with the fault diagnosis model,a real-time fault diagnosis program for the high-pressure heater system is developed on the MATLAB platform.Using the LSTM prediction model to predict the expected value of the characteristic parameters in real time,the program calculates the real-time fault symptoms according to the predicted and actual current values of the characteristic parameters in the light of the symptom fuzzy calculation method.By combining the fault symptom zoom optimization technology and the LSTM fault diagnosis model,the most possible current fault type is searched.Aiming at maximum fault resolution,and the real-time fault diagnosis result is given.By communicating with the simulator,online fault diagnosis simulation tests for the typical faults of high-pressure heaters under steady state and variable load conditions are conducted.It is shown that timely and accurate fault diagnosis under different load conditions can be achieved with the online fault diagnosis model,verifying the effectiveness of the method. |
