Research On Hypertension Prediction Model Based On LSTM-CNN

Hypertension is one of the most pressing public disease challenges worldwide and is considered the largest contributor to the global burden of disease prevention,as well as a major contributor to cardiovascular disease.Continuous and regular blood pressure monitoring is essential for the early diagnosis and prevention of cardiovascular disease.In general,our human blood pressure fluctuates over time and is influenced by several factors,such as stress,mood,food,exercise,and medication use.Continuous blood pressure monitoring,not just on specific Blood pressure monitoring at different time points is necessary for early detection and treatment of hypertension.The blood pressure time series has the characteristics of periodicity and nonstationarity,and it is difficult for a single prediction model to accurately predict it.Based on this,this thesis focuses on the related research of feature fusion and blood pressure signal decomposition combined with deep learning method for short-term blood pressure prediction.This thesis firstly proposes a bidirectional long-short term memory network(Bi LSTM)-convolutional neural network(CNN)short-term blood pressure prediction model fused with the Attention mechanism.The model converts the original ABPMean blood pressure sequence into a smooth time series through wavelet transform noise reduction,analyzes its correlation with other influencing factors(systolic blood pressure,diastolic blood pressure,respiration,etc.),and selects features with greater correlation as Bi LSTM The input features of the model,and then generate a new data sequence,send the new data sequence to the attention mechanism to strengthen the temporal features,and then extract the spatial features through the convolutional neural network,and finally,the data will be sent to the last hidden layer,resulting in final predicted value.Secondly,a combined prediction model of At-Bi LSTM-CNN based on Improved Ensemble Empirical Mode Decomposition(EEMD)is proposed.The model decomposes the blood pressure data into several intrinsic eigenmode function(IMF)subsequences through the improved EEMD,and then uses Bi LSTM and CNN to extract temporal and spatial features,respectively.These features are trained and fused through a specific fusion module.The fusion features are input into the model for prediction,and the prediction results of each time series are obtained.Finally,the prediction results of all IMF component subsequences are superimposed and combined to obtain the prediction results.In order to verify the prediction effect of the model proposed in this thesis,we select the MIMIC II physiological signal data set for experiments,and use the root mean square error(RMSE),mean absolute error(MAE),mean absolute percentage error(MAPE)and fitting score R~2 to conduct experiments.The results are evaluated.The experimental results show that the model proposed in this thesis is better than other methods under various evaluation indicators.The model proposed in this thesis has the smallest error in the prediction value and the highest prediction accuracy,it can more accurately predict the trend of blood pressure changes,and provide effective prevention and control measures for the prevention of sudden blood pressure rise.