摘要
在总结现有神经网络方法缺陷的基础上,提出了模型的思路:预测网络小型化;实时学习;多次预测取均值;加入规则辅助神经网络预测。相对于传统的神经网络模型来讲,该模型突出了动态学习、动态预测的特色,增加了辅助预测的3大规则(异常处理规则、再学习规则和取均值规则)。给出了该模型的工作流程,并以一个实际问题说明了该模型训练、预测的全过程。数据实例表明,该模型是正确的、可行的。同时和其他5种模型预测结果的对比表明,该模型的预测结果是最优的,这充分体现了模型的有效性、先进性。
Based on the limitation of existing neural network model, this paper puts forward the thought of this model： miniaturizes the forecasting network, trains the network in real-time way, adopts the average of abundant forecasting, adds some rules to assistant forecasting. Relative to the traditional neural network model, this model focuses on dynamic training and dynamic forecasting, increases three rules （rule of dealing with abnormity, rule of retraining, and rule of adopting the average） to assistant forecasting. It presents the computing flow of this model and explains the whole process of this model with an example. Numerical example suggests the correctness and feasibility of this model. The contradistinctive result of this model and other five models indicates the validity and superiority of this model.
出处
《计算机工程》
CAS
CSCD
北大核心
2006年第12期199-201,204,共4页
Computer Engineering
基金
国家自然科学基金资助项目(70272002)
关键词
时间序列预测
神经网络
实时学习
多规则
Time series forecasting
Neural network
Real-time training
Multi-rule
