Study On Establishment Method Of Environmental Model Based On Neural Network

The onboard electronic equipment cabin is an important part of the aircraft to perform more functions.The normal operation of internal electronic equipment is the basis for its stable operation.Thermal failure is the most likely failure mode for electronic devices,so the acquisition of accurate heat transfer model in pods is of great significance for thermal prediction and thermal management design of airborne electroniccabin.At present,most of the thermal model modeling methods are based on the traditional thermal network model(Thermal Network Model,TNM)established by the node network method.This method is often used to analyze transient thermal response.Because of the idea of lumped parameters,the electronic equipment is equivalent to a thermal node with uniform internal physical properties,which can not well characterize the nonlinear temperature change process in the cabin.Moreover,because the prior physical parameters of the thermal network model cannot be accurately obtained,it is easy to cause multiple solutions of parameter identification.Neural network with its good nonlinear fitting ability has become a major research direction of heat transfer process modeling.In order to realize high precision fast thermal modeling,the random vector functional connection(Random Vector Functional Link,RVFL)neural network method is studied in this paper.The input variables and output variables of the RVFL neural network are determined by analyzing the heat transfer heat network relationship.The filtering algorithm is used to preprocess the temperature measurement point data collected by the experiment to eliminate the negative impact of noise on the accuracy of the model.In order to solve the problem of model failure caused by variable thermal load in the cabin,the sliding time window method and neural network thermal model are introduced to quickly establish the thermal response model of electronic equipment cabin and predict the equipment temperature in the subsequent form.The prediction accuracy of each sliding time window is monitored in real time.Once it does not meet the requirements,the thermal model output weights will be corrected using the data from the next sliding window.The experimental results show that the prediction accuracy of RVFL thermal model is higher than that of traditional thermal network model and BP thermal model.This study can provide an effective way to describe a complex dynamic heat transfer process adaptively and accurately,which may help the thermal control scheme design.