基于回归BP神经网络误差分级迭代法的逐时气温预测方法研究

Research on Hourly Air Temperature Forecasting Method Based on Regression BP Neural Network Error Classification Iterative Method

气温预测是天气预测中的一项主要内容,由于气温的影响因素多而复杂,要想达到精细化预测目的,仍是十分复杂的科学难题.当前学术界的一般方法是假设数学模型对温度物理过程进行研究,建立了BP神经网络模型、温度与相对湿度之间的回归模型,最终在回归模型的基础上通过改进的BP神经网络建模,即利用BP神经网络误差分级迭代法建模,通过历史温度进行逐时气温预测,全样本误差达到0.617℃.

Forecasting air temperature is one of the main contents for the weather forecasting.Due to the complexity characteristics of the influence factors of temperature, it is still a complicated scientific difficult problem to meet the requirements of refined prediction. The general method of current academic circles is to study physical process of temperature based on mathematical model. In this paper, the BP neural network model and regression model between the temperature and relative humidity are established. Finally, the paper establishes improved BP neural network model on the basis of the regression model, which is named the error classification iterative method. The model is used to forecast time-time air temperature by using the historical temperature, which result of the sampling error is 0.617℃.