| Small Modular Reactors(SMRs)require fault tolerance measures to ensure safe and stable operation in the event of a sensor or other s ystem component failure.SMR conditions change frequently,and if the sensor fails under transient conditions,which are by nature a disturbance,it is difficult to achieve fault tolerance of the sensor.At the same time,SMR is a complex,time-varying,strongly coupled nonlinear system,and the system process parameters are sensitive to changes in load,which places higher demands on the control system of SMR.Therefore,it is necessary to conduct research related to the application of intelligent fault-tolerant methods for SMR.A real-time simulation and anal ysis s ystem was established for the Integrated Pressurized Water Reactor IP200.For the case of sensor failure,An improved h ybrid neural network model is proposed to reconstruct the anomalous sensing values in this paper,while the traditional threshold setting method for fault detection is improved-Bayesian estimation method.Replacing the actual measured value with the reconstructed value after diagnosing the fault y sensor ensures the accuracy of the value of each process parameter transmitted to the control and protection s ystem,thus achieving fault tolerance of the sensor.In response to the failure of SMR s ystem equipment,a SMR sub-control s ystem including reactor power control s ystem,steam generator feedwater control s ystem and regulator pressure control s ystem based on the classical PID controller is established in this paper.Based on this,the linear active disturba nce rejection controller(LADRC)is combined into the above sub-control s ystem as an improvement of the PID controller,and the parameters are adjusted with the deep deterministic policy gradient algorithm(DDPG)to achieve the design optimization of the control system.Finall y,a joint simulation of sensor fault tolerance and s ystem faul t tolerance control is performed to verif y the effectiveness and feasibilit y of the SMR intelligent fault tolerance control method proposed in this paper.Simulation results show that for sensor fault tolerance,the improved neural network model can accuratel y reconstruct the normal sensing values with higher reconstruction accuracy than the Long Short-Term Memor y(LSTM)neural network.The Bayesian threshold estimation method is also able to detect fault y sensors within a short time after the occurre nce of a fault,and has a better robustness b y reducing the false alarm rate compared to the common threshold setting method.For system fault-tolerant control,the control performance of the DDPG combined with the LADRC controller is better than that of t he classical PID controller,reducing the impact of s ystem-level faults on the SMR.Therefore,the intelligent fault-tolerant control method designed in this paper can ensure the safe and stable operation of SMR even in the event of sensor failure and s yst em equipment failure.The research results can provide some theoretical basis and methodological reference for the stud y of fault-tolerant control of SMR. |
