广义回归神经网络在煤灰熔点预测中的应用

为了提高估算煤灰熔点的精度,采用广义回归神经网络(GRNN)对求解煤灰熔点问题进行了建模.将煤灰组分作为网络输入,煤灰软化温度作为网络输出,采用实验数据训练网络,训练完成的网络作为模型预测煤灰熔点.仿真结果表明,GRNN的预测值与实验值的最大相对误差为2.81%,而反向传播神经网络(BPNN)预测煤灰熔点的相对误差为3.62%.由于GRNN可应用于小样本问题的学习,GRNN比BPNN对煤灰熔点具有更好的预测和泛化能力.GRNN具有设计简单与收敛快的优点,并提高了实时处理与反映最新运行工况参数的预测能力.

Application of general regression neural network in prediction of coal ash fusion temperature

A general regression neural network (GRNN) was employed to model the coal ash fusion temperature for obtaining better predicting performance. The coal ash compositions were employed as the inputs of GRNN, and the measured ash fusion temperature were used as the outputs of the neural network. The modeling work employing the back-propagation neural network (BPNN) was also conducted to make a comparison with the GRNN. The results show that the maximum predicting error of GRNN was 2.81%, and that of BPNN was 3.62%. Compared to BPNN, the predicting result of GRNN is more accurate for the ash fusion temperature prediction. For GRNN has the learning ability in small training sample size, it can give better predicting and generalization performance under various conditions. The design of GRNN is simpler than that of BPNN, and the calculation time needed by GRNN for convergence is significantly shorter than that needed by BPNN.