基于CFSFDP-RBF神经网络的加拿大区域气候预测

Regional Climate Prediction for Canada Based on CFSFDP-RBF Neural Network

南极冰川融化、飓风不断增加、海平面逐渐上涨等现象的出现,使人们意识到全球气候变暖给人类生存带来的极大的挑战,针对全球变暖引起的气温变化不定问题,为了准确预测气温变化情况,以加拿大部分地区为例,通过对加拿大10个省份数据预处理,最终选取四个数据保留较完整的省份数据。对此提出了一种改进径向基(RBF)神经网络气候预测模型。该模型采用密度峰值快速聚类(CFSFDP)算法和自适应矩估计(Adam)对RBF神经网络进行优化,首先利用CFSFDP算法聚类出中心簇来确定RBF神经网络径基中心,避免随机选择带来的误差。再利用Adam算法对目标函数进行迭代微分,调整权值,同时自适应地改变学习率,提高预测准确性。针对模型的准确性检验,通过与BP神经网络、RBF神经网络、K-means优化RBF神经网络及论文算法进行对比实验发现本模型具有较高的准确率。针对结果的准确性检验,分别利用改进整合移动平均自回归模型(ARIMA)、向量自回归模型(VAR)与CFSFDP-RBF神经网络算法对气候进行预测,三种模型的结果均得到相似结论,表明该算法预测结果可信。实验结果表明,未来25年平均气温达到15.0470℃,未来25年平均降水量达到2.0984 mm,预测准确率达95%以上。

The melting of Antarctic glaciers,increasing hurricanes and rising sea levels have made people realize that global warming is posing a great challenge to human survival.To address the problem of variable temperature change caused by global warming and to accurately predict the temperature change,this paper takes some regions of Canada as an example and proposes an improved radial basis function(RBF)neural network climate prediction model by preprocessing the climate data of 10 Canadian provinces and finally screening out the data of 4 provinces with more complete data retention.The model uses the Clustering by Fast Search and Find of Density Peaks(CFSFDP)algorithm and Adaptive Moment Estimation(Adam)to optimize the RBF neural net-work.The CFSFDP algorithm is first used to cluster out the central clusters to determine the RBF neural network path base centers to avoid the errors caused by random selection.Then the Adam algorithm is used to iteratively differentiate the objective function and adjust the weights,while adaptively changing the learning rate to improve the prediction accuracy.The accuracy of the model is test-ed by comparing it with BP neural network,RBF neural network,K-means optimized RBF neural network and the algorithm of this paper,and the accuracy of the model is found to be quite high.To test the accuracy of the results,the improved integrated Autore-gressive Integrated Moving Average model(ARIMA),Vector Autoregressive Model(VAR)and CFSFDP-RBF neural network algo-rithm were used to predict the climate,and the results of the three models were similar,indicating that the prediction results of this algorithm are reliable.The experimental results show that the average temperature and precipitation will reach 15.0470℃and 2.0984 mm respectively in the next 25 years,with a prediction accuracy of more than 95%.