Interpretive Neural Network Power Grid Fault Diagnosis Model For Evaluating The Action Behavior Of Relay Protection Syste

The rise of the concept of "energy internet" and China’s goal of "carbon peaking and carbon neutrality" not only make the scale of power grid more and more massive,but also promote the development of new energy generation scale,bringing more wind,solar and other clean energy into the power grid,putting higher demands on the reliability and stability of power grid operation.The reliability and stability of the grid operation has put forward higher requirements.Especially when a large amount of alarm information comes into the dispatching and monitoring center,it is difficult for dispatchers to process the alarm information and make accurate judgment on the fault in a short time.Therefore,a more efficient and accurate grid fault diagnosis system is required to assist dispatchers to diagnose and handle faults in a timely manner and ensure a continuous and stable power supply.Artificial neural network can handle complex nonlinear problems well with its strong generalization and high fault tolerance,which can make up for the shortage of traditional mathematical methods to describe the fault diagnosis process of power grid,therefore,this paper conducts an in-depth study on power grid fault diagnosis based on artificial neural network.1)The background and significance of the research on grid fault diagnosis are explained,and then the research progress in grid fault diagnosis at home and abroad is discussed.The basic features and functions of artificial neural networks are introduced,and on this basis,back propagation neural networks are focused on,and later differential evolutionary algorithms are introduced to optimize the networks.2)In view of the fact that neural networks cannot achieve the judgment of protection behavior well when applied to grid fault diagnosis,a dual BP neural network grid fault diagnosis model that can evaluate the action behavior of protection systems is proposed.The fault diagnosis model is divided into two stages to diagnose the fault,firstly,the location of the fault is judged based on the alarm information when the fault occurs in the grid,then,the correct action,false action and rejection of the protection and circuit breaker are coded by using 0,1 and-1,and the false action and rejection behavior of the protection and circuit breaker are quantitatively evaluated by combining the fault location and alarm information,which makes the diagnosis process more perfect.Finally,the fault example is simulated in a simple distribution system,and the simulation results surface the correctness of this improved model.3)When the scale of the grid is large,the problem of dimensional disaster arises when using the above method,and the problem of "black-boxing" of the neural network process still exists.Therefore,a general neural network fault diagnosis model is proposed to evaluate the action behavior of relay protection system.By analyzing the action logic relationship among the faulty element,protection and circuit breaker,the mapping relationship among neurons is reconstructed by breaking the fully connected structure of the neural network;especially for the case that the faulty element has multiple remote backup protections and circuit breakers,the neuron parameters are determined by adopting the idea of first overall and then refinement to shorten the training time.training time;then,the expected action states of protection and circuit breaker at fault are analyzed,and the action behaviors of protection and circuit breaker are further explained and analyzed by comparing with the actual alarm information,and the fault diagnosis system is improved.The model is also validated on IEEE30 node and actual power system in Jilin province,and the results show that the model’s can not only accurately diagnose fault elements,but also effectively evaluate protection and circuit breaker rejection or misoperation behaviors,and also has certain portability.