我国气温变化的神经网络预测模型

A NEURAL-NETWORK-BASED FORECASTING MODEL FOR TEMPERATURE CHANGE IN CHINA

利用神经网络系统工具,把CO2浓度、太阳辐射、全球冰量和气溶胶作为输入,把气温作为输出,结果发现模拟值和目标值吻合较好,比多元回归拟合值的精度高。在此基础上,进行了两个控制性试验,即CO2浓度每年分别增加1%和0.5%,试验结果表明:随着CO2浓度的增加,21世纪前30年每10年中国气温的增加速度分别为0.4℃和0.2℃;到21世纪末,中国气温将分别增加3.8℃和2.4℃,神经网络模拟结果和气候系统模式模拟结果具有较好的一致性,从而进一步说明,神经网络预测模型在分析气候变化方面具有可信性和可行性。

As a system tool, neural network was used to simulate the climate change with CO2 concentration, solar radiation, global ice concentration and aerosol as input and temperature as output. It is found that the simulated value is consistent with target value and it has more accurate prediction than the multiple linear regression does. Based on this, two controlled experiments were done with annual CO2 concentration increasing by 1% and 0.5%, respectively. The results show that with the CO2 concentration increasing, the temperature in China will have increases of 0.4 ℃ and 0.2 ℃ every 10 years respectively in the early 30 years of the 21st century. Furthermore, it will increase 3.8℃ and 2.4℃ by the end of the century. Noticeably, the neural network has the same simulation results as climate system models do. As can be seen from the above results, the neural network has the reliability and feasibility in analyzing the climate change.