基于经验模型可能最优效果受数据质量限制的事实,提出了“受限最优模型”的概念,定量分析了噪音强度、样本规模对受限最优模型效果的影响.提出了利用受限最优模型期望效果进行优化建模思想,基于该思想,提出了一种基于噪音信息指导的神经网络优化建模方法,仿真试验表明,该建模方法切实可行,效果明显优于传统方法。为了客观地评价模型效果,还提出了一种新的模型评估指标——误差平均功率,分析了它和常用的模型评估指标——误差均方之间的关系,指出了其应用意义。
According to the fact that the possible optimal effect of empirical model is restricted by data quality, the concept of "restricted optimal model" was proposed, the influence of data noise and sample size on the effect of restricted optimization model was quantitatively analyzed. An idea modeling supervised by noise information was proposed, based on this idea, an artificial neural network modeling method supervised by noise information was proposed. Simulation experimental results verify that the approach can work with good performance. To evaluate objectively the performance of model, a new evaluation index named error average power was proposed, its relation to common index named mean square deviation was analyzed, and its application significance was pointed out.