Research On Health Diagnosis And Prediction Of Plain Reservoir Based On Data Driven

In recent years,with the increasing development of the economy and society,the contradictions between human beings and water resources are increasingly prominent.A healthy water environment is an essential requirement for ensuring people's livelihood and improving people's living standard,and also a basic condition for the harmonious development of economy,society and ecological environment.As a communication channel between human beings and nature,the research of plain reservoir health diagnosis and prediction has become one of the hot fields.With the rapid promotion of data-driven technology in the water conservancy industry as well as gradual rise of the plain reservoir health management concept,a massive health monitoring data of reservoir has accumulated on the plain reservoir health monitoring platform.To better analyze and process these date,it is important to combine the data driven technology combined with plain reservoir health management.It is of great significance to the plain reservoir health diagnosis and prediction using data driven technology.This study analyzes the characteristics of reservoir health monitoring data and USES neural network method and deep learning method are applied to diagnose and predict the health of plain reservoir.The BP neural network,RBF neural network and particle swarm optimization grey neural network are programmed by MATLAB software to predict the reservoir health.LSTM and GRU were programmed by TensorFlow platform to predict reservoir health.By means of analysis and comparison,the appropriate methods are selected for the health diagnosis and prediction of the reservoir,in order to ensure the accuracy of the diagnosis and prediction of the health condition of the reservoir,providing and the reservoir managers with some indications on the health condition of the reservoir.The main conclusions are as follows:(1)In terms of the health diagnosis of the plain reservoir,the BP neural network,RBF neural network and the grey BP neural network optimized by particle swarm optimization algorithm are used for the diagnosis and analysis of the reservoir health.Through analysis and comparison,RBF neural network is faster than BP neural network in training accuracy,and the training accuracy is better than BP neural network.It is more simple and effective for the diagnosis of reservoir health.Particle swarm optimization of grey BP neural network makes up for the deficiency of the BP neural network and grey model,the parameters are further optimized by the particle swarm optimization(pso),comparing to the single BP neural network and RBF neural network can embody more advantages and advanced nature,the particle swarm algorithm to optimize the gray BP neural network in the reservoir of health diagnosis performs better,to achieve the purpose of the health diagnosis of reservoir.(2)In the health of the plain reservoir prediction,BP neural network and RBF neural network do not have the function of health indicators to predict reservoir to an unknown number of days,the particle swarm algorithm to optimize the gray BP neural network using the grey model and BP neural network model has predicted the future 15 days reservoir health status,and the predicted results are generally in accordance with the change law of reservoir health.It can be used as reference for the reservoir management in the study of reservoir health analysis.(3)In the deep learning method,both short-term and long-term memory network(LSTM)and neural network(GRU)door cycle five indicators of the health monitoring data are carried on the depth study,detailed health of reservoir comprehensive evaluation index,structural safety,the ecological environment function,social function health index and stability index were analyzed,respectively,for the five indicators,to predict the change of the five indicators over the next 15 days.Through analysis and comparison,it is found that the root-mean-square error of GRU algorithm in the test set is generally smaller than that of LSTM algorithm,and the fitting degree is better than that of LSTM algorithm.In the prediction of the next 15 days,the GRU algorithm has a better effect,which is in line with the law of data changes and achieves the purpose of reservoir health prediction.The above research methods are able to effectively predict the health diagnosis of the plain reservoir.The established neural network model and deep learning model can better serve the data analysis of reservoir health and ensure the healthy,efficient and sustainable operation of the reservoir.