Research On Ventilator Pressure Prediction Based On Machine Learning

In recent years,in the context of COVID-19 sweeping the world,ventilators have received extensive attention as an effective means of artificially replacing the function of spontaneous ventilation.This paper investigates the prediction of airway pressure based on mechanical learning and deep learning methods for mechanical ventilation using ventilators.Achieving high precision airway pressure prediction helps provide decision support for clinicians to configure ventilator parameters,as well as advance the automation of the aforementioned processes.In this paper,experiments use multiple models for airway pressure prediction,including linear regression and its modified model,random forest,GBDT,XGBoost,and deep learning structures containing 1d-CNN and LSTM.And the model effect was also evaluated by MSE,MAE and R~2.The experimental data were selected for some sequence samples of length 80.In the non-deep learning algorithm,the difference,mean and other methods are used to introduce sequence features into a single sample,and the modeling effect is significantly improved compared to that without sequence features.Experiments show that the R~2 index of the best single-model GBDT is 0.9919.In terms of deep learning algorithm,sequence feature extraction is performed on the original features through 1d-CNN,and the R~2 index of the deep learning model with 200 iterations is 0.9949.Based on the concept of model fusion,this paper proposes two methods to improve the Boosting family of models.One is to replace the initial learner of GBDT,so that the R~2 index of the GBDT model is increased to 0.9924.The second is to use convolutional features to model the GBDT model,so that the R~2 index of the GBDT model is increased to 0.9933.The above methods are easy to implement in other sequence prediction problems,and have certain promotion value.