几种改进BP神经网络算法预测飞灰含碳质量分数效果的比较

分别介绍了动量梯度下降法、L-M数值优化算法和贝叶斯算法,建立了预测模型,对某电厂飞灰含碳质量分数进行预测。通过预测结果的分析,对比了不同算法对预测精度的影响。结果表明:在该模型下,当训练样本量为30甚至更多时,动量梯度下降法不能收敛,L-M算法和贝叶斯算法的收敛速度比动量梯度下降算法要快很多,而贝叶斯算法的预测精度最高。

Comparison of Prediction Effects of Several Improved BP Neural Network Algorithms on Carbon Content in Fly Ash

An introduction is being presented to the momentum gradient descent algorithm, the L-M numerical optimization algorithm, and the Bayes algorithm, for the purpose of predicting the carbon content by mass fraction in the fly ash of a power plant, based on a newly proposed prediction model. Analysis and comparison results indicate that when the number of training samples is 30 or more, the momentum gradient descent algorithm tends not to converge; the convergence rates of L-M numerical optimization algorithm and Bayes algorithm are much higher than that of the momentum gradient descent algorithm. The prediction precision of Bayes algorithm, however, is the highest.