中文摘要：

指出了极端温度预测问题受到多种因素的共同影响,具有非线性和高度复杂性。为了提高非线性时间序列在预测模型中的准确性,提出了一种利用最小二乘法优化的BP神经网络预测方法。该方法通过采用最小二乘法对数据样本集进行拟合,用BP算法进行优化,构建了两者结合的预测模型。应用1989-2009年的密云市温度数据资料,分别建立了基于最小二乘优化的BP神经网络和单一BP神经网络模型,并对预测结果进行了分析对比。结果表明：最小二乘法优化的BP神经网络具有更好的泛化能力,对平均最低温度的预测更加稳定,预测精度高于单一的BP神经网络。该模型可以对气候变化中气候变化中的平均最低气温具有较好的预测能力。

英文摘要：

The prediction of extreme temperature is affected by many factors, which has nonlinearity and high complexity. In order to improve the accuracy of nonlinear time series in the prediction model, the article proposes a BP neural network prediction method which is optimized by least squares method. This method builds a combination forecasting model by using the least square method for fitting sample data set and using BP algorithm for optimizing. Taking the temperature data of Miyun from 1989 to 2009 as data source,the article establishes the BP neural network model based on least squares optimization and the single BP neural network model respectively. Then the article analyzes and compares the forecasting results. The results show that the BP neural network model which is optimized by the least squares exhibits better generalization ability than single BP neural network model. The forecast of average minimum temperature is more stable by the BP neural network model which is optimized by the least squares and the prediction accuracy is higher than that by single BP neural network model. The developed model can be expected to predict the average minimum temperature in the climate change with high predictive ability.