基于冲量权值的ART神经网络及其在地震预报中的应用

Impulse Force Based ART Network with GA Optimization and its Application to Earthquake Prediction

在传统的ART(AdaptiveResonanceTheory)网络结构中忽略了样本属性重要性的不同对分类结果产生的影响。然而实际应用中,需要在网络预测阶段对此予以重视。该文提出了基于冲量权值的ART网络IFWART(ImpulseForceWeightbasedAdaptiveResonanceTheory)。它引入冲量权值表示属性的重要性,通过进化算法的优胜劣汰机制优化冲量权值,并将量化的权值结果分配到网络的比较层中,从而提高网络预测精度。在UCI标准数据集上将IFWART与其他有监督ART网络进行了比较实验,验证了IFWART的有效性。最后,将其应用于地震强震时间序列预报中,取得了很好的效果。

The different affects of input attributes on category results in supervised ART(Adaptive Resonance Theory)network,quite important during the predictive stage in the application,have been ignored by traditional researches.In fact,some of the attributes will have larger effect than the others on category results.However,even for the experts in that field,it is difficult to evaluate the effect.In this paper a novel supervised ART network named Impulse Force Weight based Adaptive Resonance Theory(IFWART) network is proposed.It enhances the prediction accuracy of the supervised ART network by using Genetic Algorithm optimized Impulse Forces on attributes.Then some experiments on benchmark data sets are given to show its good performance.Finally,it is applied to predict the earthquake time serial in mainland of China.The result is satisfactory.