| Small nuclear reactor has multiple functions such as power generation,heating,hydrogen production,seawater desalination and power propulsion.It is of strategic significance to ensure our energy diversification and national defense security.At present,the control and protection system of small reactor mostly follows the design concept of commercial nuclear power plant.In the critical moment of anomaly or accident,the operator’s experience judgment and timely decision are still indispensable.In recent years,intelligent operation of small reactor based on autonomous control technology has become one of the research hotspots.Based on traditional automatic control device,it enables the target object to have functions such as performance optimization,health diagnosis and fault-tolerant control through deep data processing and fusion analysis without or with little help of personnel.The core of autonomous operation is to use artificial intelligence technology to simulate the operator’s judgment and decision-making behavior,so that the reactor has "intelligent capability".In order to solve the problem of fault warning and controllable operation after accident that may be faced in the autonomous operation of small reactors,this paper proposes an autonomous control scheme with intelligent characteristics based on machine learning algorithms such as long and short term memory neural network(LSTM),which has intelligent diagnosis and intelligent control functions.Considering the different fracture area,location and occurrence time corresponding to each same accident,the phenomenon characteristics and event sequence after the accident are different,the knowledge and experience of operators in the performance procedures can be deduced by intelligent algorithm,which has good applicability.Firstly,the model of each module is established by historical data,and combined with the typical data samples generated by the simulation machine during the operation of the unit under various working conditions,the parameter warning function based on Autoencoder neural network(AENN)is realized.By learning from historical data,AENN can generate a model of a typical device that can be used to monitor real-time data and detect anomalies.Then,taking the primary cooling accident under SGTR condition as an example,LSTM algorithm and transfer learning were used to establish the model,and the automatic control module in the procedure was established with the adaptive PID of the model parameters,and the general process of establishing the model for the controlled quantity was given.LSTM algorithm is used to process and predict time series data of nuclear power plants.Transfer learning solves the problem of limited data in nuclear power plants.The simulation results show that the steady-state error of the intelligent control module gradually decreases after 1600s,the average relative error of the intelligent control system is 2.6667℃/h,and the actual cooling rate is 4.7%of the set value of the cooling rate.Under the same conditions,the average relative error of manual operation is 11.1807℃/h,and the actual cooling rate is 20%of the set value of cooling rate.It can be seen that compared with manual operation,the intelligent control method proposed in this paper has higher control precision and faster adjustment.speed.It can be considered to replace the operator’s partial manual operation after the accident handling,which is conducive to the optimization of operation procedures. |
