Research On Modeling Of Steam Generator And Water Level Control Strategy Based On Multi-Agent

The steam generator(SG)is an important equipment connecting the primary coolant system and the secondary loop system of nuclear reactor.Its main function is to transfer the heat from the primary loop’s coolant to the secondary loop fluid through heat transfer tubes,and produce saturated steam for the steam turbine to generate electricity.In order to maintain the normal operation of the equipment,the water level in SG must be controlled within the specified range.When the water level is too low,the bend part of the heat transfer tube will be overexposed to the steam,shortening its service life.When the water level is too high,the separation effect of the steam separator will be affected,and the wet steam generated will corrode the turbine blades and reduce the efficiency of power generation.The SG is composed of a large number of heat transfer tubes and related components,and its structure is complex.How to control the water level of the steam generator accurately is always a concern problem.The multi-agent technology is a kind of distributed modeling technology,which is capable of decision and control tasks of complex systems.In this paper,the modeling and control strategies of SG are studied based on multi-agent technology,including SG water level control framework,power level monitoring method,SG modeling method and water level control method.The main research contents of the paper are as follows:(1)Aiming at the problems of low data processing efficiency and the design concept not meeting the requirements of advanced control methods,a SG water level control framework based on multi-agent technology is proposed to achieve the optimal control of SG water level in the full power range.In the high-performance CPU and GPU hardware environment,as well as Windows system,the simulation system is deployed to realize the data acquisition and processing,water level control and intelligent algorithm application,so as to verify the feasibility of the scheme at the simulation level.(2)The lumped parameter modeling method and distributed parameter modeling method of SG were studied.In the lumped parameter model modeling,Irving transfer function model was established to analyze the " false water level " phenomenon under the transient state of the supply water flow step and the steam flow step.In the distributed parameter modeling,a SG of Hualong 1 was taken as the research object.On the Simscape platform,the control module,the steam separator module and the water supply regulating valve module were designed by using the fluid mechanism model,and the SG model was established based on the modular modeling idea.The accuracy of SG model is verified under steady and transient conditions,and the formation process of " false water level " is analyzed in detail.(3)To provide a faster response speed of strategic scheduling and reduce the influence of time delay,a power monitoring model DDRNN based on deep denoising residual neural network was proposed to achieve a breakthrough in the power level monitoring mode of SG.To improve the robustness of the model,the training method of traditional deep neural network was improved to make the model have the performance of noise reduction,and the adaptive learning rate method was applied to improve the training process.The proposed method can realize power level monitoring and has good generalization performance and anti-noise performance.(4)Aiming at the difficult problem of parameter tuning of traditional PI controller,a control method based on deep deterministic strategy gradient algorithm,DDPGCPI,was proposed.Taking the lumped parameter model as the research object,the deep reinforcement learning algorithm DDPG was used to realize the controller parameters self-learning,which got rid of the dependence on the expert knowledge.At the same time,the segmented reward function was constructed to improve the training effect of the model.The proposed method not only greatly improves the performance of water level regulation,but also reduces the difficulty of adjusting the controller parameters.(5)To improve the efficiency of controller design for distributed parameter model,an imitative learning control method(ILCPI)was proposed to reduce the difficulty of controller design while ensuring the control effect.A fuzzy logic algorithm was proposed to solve the problem of uncertainty for the feature imitation of operating parameters.To improve the performance of fuzzy logic model,an improved genetic algorithm was adopted to solve the problem of parameter optimization.(6)The anti-interference control method was studied.To solve the problem of poor antiinterference performance of the ILCPI method at low power,an adaptive gain scheduling antiinterference control method(GSAI)was proposed.The proposed method can realize the adaptive scheduling of the controller parameters and improve the anti-interference effect of the controller in the full power range.(7)To make full use of the operation data of SG,a multi-task fusion control method(BMTFC)is proposed.By analyzing the correlation between power level monitoring and control process information,this paper expounds the relationship between different tasks,and uses data mining technology(data fusion and multi-task learning)to extract power level and control strategy related characteristic information from massive data.BMTFC method integrates multiple feature information,and can simultaneously complete complex tasks that can only be completed by multiple agents working together,including power level monitoring task,gain scheduling task and strategy optimization task,so as to achieve better water level regulation performance and anti-interference performance,and show the control effect comparable to the original framework.The research results of this paper provide an important method for the design and operation characteristics analysis of SG water level controller,which has a great application reference value.With the support of multi-agent technology,it is expected to promote the standardization,networking,intelligence and openness of nuclear power system.