Prediction Of ERMS Radiation Data Based On Feature Fusion Network

Nuclear power is an essential means for China to achieve the carbon-peaking and carbon-neutrality targets.The most essential peripheral surveillance system for ensuring nuclear power operation is the environment radiation monitoring system.The public’s perception of risk is a major element influencing the growth of nuclear power.This topic comes from the project of Fujian Provincial Nature Fund.The purpose is to apply the deep learning network model to extract and analyze the features of ERMS time series data,and to mine the feature impact factors related to radiation dose rate.Then,the cyclic neural network(LSTM,GRU)and feature fusion neural network(PLSTM-FCN,PGRU-FCN)are utilized to construct a real-time prediction model of HPIC dose rate value,with the goal of improving anomaly detection capabilities.Firstly,the time series data of ERMS platform,Meteorological Bureau website and space environment prediction center are constructed into a data set,the characteristic factor set related to radiation data is established,and the training set and experimental set data for network training are divided;Then,the real-time prediction of dose rate of high-voltage ionization chamber detector is realized by using the cyclic neural network prediction model based on long-term and short-term memory network and gated cyclic unit network.Numerical experiments show that compared with GB prediction algorithm,this neural network model can capture the characteristics between time series data,ensure the accuracy of HPIC dose rate prediction,and reduce the impact of external noise to a certain extent;Finally,in view of the influence of numerical fluctuation caused by external factors in the process of feature extraction,the high-dimensional features of HPIC dose rate in space and time sequence are mined in parallel based on short-term memory complete convolution network and parallel cyclic neural network complete convolution network feature fusion neural network,so as to realize the real-time prediction of dose rate.In this study,the data training set is input at the same time,and the real-time prediction model is composed of LSTM or GRU network layer and full convolution network layer in parallel.The extracted features of the above two types of networks are fused at the same time,and the built model will predict the dose rate according to the fused features.Experiments show that this method has high feature grabbing ability and strong anti-interference ability for radiation environment data.Numerical experiments verify the feasibility of HPIC dose rate prediction,and provide a new idea for abnormal data analysis of HPIC dose rate.