基于遗传算法优化的BP神经网络的组合预测模型方法研究

摘要

[Objective] The combined forecasting model for studying the classic swine fever morbidity was proposed. [Method] The data was processed by ARIMA and GM(1,1) initially, then the results were used as the inputs of the tnajorizing BP neural network. [Result] The combined model was used to analyze the monthly data from 2000/01 to 2008/05, and the accuracy of the forecasting results from 2008/06 to 2009/06 was 97.379%. The prediction accuracy of the combined model increased by 5.469%, 3.499%, and 1.188%, respectively, compared with BP neural network, ARIMA, GM(1,1), which suggest that the combined model is more steady than traditional methods. [Conclusion] This research has supplied an efficient analytical tool for animals diseases forecasting work, verified the feasibility of the combined model in animal diseases forecasting research, and also has provided references to other animal diseases.%[目的]提出以传统猪瘟发病率为对象的组合预测模型.[方法]利用ARIMA模型以及灰色模型GM(1,1)进行数据初始化处理,将初步处理结果作为优化后的BP神经网络输入构建组合模型.[结果]利用组合模型对2000年到2009年的月度发病数据进行实例分析,结果表明预测数据精度达到97,379％,较ARIMA模型,灰色模型、BP神经网络模型分别提高了5.469％、3.499％、1.188％,模型平稳性增强,预测结果良好.[结论]本研究为动物疫情测报提供了有效的分析手段,验证了组合模型在动物疫情研究中的可行性,并可为其它动物疫病提供借鉴和参考.