基于改进的LSTM-CNN高血压组合预测模型

Combined Forecast Model of Hypertension Based on Improved LSTM-CNN

据国内外研究表明,高血压是危害人生命健康的重要疾病,如何及时的对血压升高进行预警成为一个研究的热点问题。为解决这一问题提出了一种融合注意力(Attention)机制的长–短期记忆网络(Long Short-Term Memory, LSTM)–卷积神经网络(Convolutional Neural Network, CNN)短期血压预测方法,该方法可以减少历史信息的丢失,实现短期血压预测。将此预测方法与单一的LSTM模型、CNN模型和LSTM-CNN组合模型进行对比,实验采用均方根误差(RMSE)、平均绝对误差(Mean Absolute Error, MAE)和平均绝对百分比误差(MAPE)对其进行评价,结果我们的方法误差最小,预测准确度最高,验证了我们模型的有效性和可扩展性。

According to studies at home and abroad, hypertension is an important disease endangering peo-ple’s life and health. How to give early warning of high blood pressure has become a hot issue. To solve this problem, a short-term blood pressure prediction method based on long short-term memory (LSTM)-convolutional neural network (CNN) and attention mechanism is proposed. This method can reduce the loss of historical information and achieve short-term blood pressure pre-diction. This prediction method is compared with single LSTM model, CNN model and LSTM-CNN combination model. Root mean square error (RMSE), mean absolute error (MAE) and mean abso-lute percentage error (MAPE) are used to evaluate this prediction method. As a result, our method has the smallest error and the highest prediction accuracy, which verifies the effectiveness and scalability of our model.