Design Of Generator Failure Prediction Platform Based On Improved LSTM

The energy problem has become one of the three basic problems in the modern society.The development of industry and the maintenance of equipment are inseparable from mini generators.The development of a complete fault prediction system for mini generators is the most effective way to neutral carbon,emission reduction and reduce operation and maintenance costs.At present,the field of mini generator failure prediction has problems such as prediction untimely,time-consuming failure prediction,and high development cost of failure prediction system.In response to the above problems,this thesis proposes an improved Long Short-Term Memory neural network model,tests the algorithm using a mini generator operating state data set,and finally designs a complete mini generator fault prediction system.The main contents of this thesis are as follows:1.This thesis introduced a fault weight layer into the traditional LSTM network to establish a complete fault prediction model for mini generators.According to the characteristics of the mini generator operating state data set,a fault weighted layer adds to the traditional LSTM network by using attention mechanism,which improves the consistency of prediction.At the same time,to solve the problem of too long training time,this thesis has carried out a design of parallel network to speed up the network fitting speed.By comparing the improved algorithm with the vanilla LSTM network and Bi-LSTM network,the results prove that compared with other algorithms,the improved algorithm has better performance in fitting time,accuracy,and kappa coefficient.The overall performance of the coefficients is better.The accuracy,loss and kappa coefficient of the algorithm are 94.335%,0.1917,and 0.67,respectively.2.This thesis designed a fault prediction system for mini generators,and it uses the C/S architecture to build the system.The server uses Docker to deploy the algorithm.The client software uses the cross-platform Flutter framework.The mini generator fault prediction system includes functions such as user registration and login,operation and maintenance management,fault prediction,electronic fences,etc.,and finally this thesis realizes a fully multi-platform application.3.This thesis tested the fault prediction system for mini generators.First it sets up the test environment required for the test,designs different test methods for the server and the client,combines the functional requirements of the mini generator fault prediction software,designs the corresponding test cases in accordance with the System and Software Engineering Testing Standard.it used the JMeter test tool to perform functional and performance test work.According to the test results,the operation and maintenance personnel can use the functions of the system normally,and the stability and performance of the system also meet the national software standards,which finally verifies the strong availability of the mini generator fault prediction system.