生物质灰软化温度预测模型构建及应用研究

Study on neural network-based prediction model for biomass ash softening temperature

生物质灰的软化温度可以反映锅炉结渣效率与传热效率,因此建立生物质灰软化温度的预测模型能够更好地探究生物质锅炉结渣规律。研究基于反向传播(Back Propagation,BP)神经网络与Elman神经网络建立了生物质灰软化温度预测模型。首先,收集各生物质灰中氧化物的质量分数与灰软化温度并作为原始数据集,确定模型的训练参数;其次,对模型进行测试;最后,对模型的适应性进行评价。研究对比分析了BP神经网络与Elman神经网络的决定系数R^(2)、均方误差根E_(RMS)、平均绝对误差E_(MA)以及平均绝对百分比误差E_(MAP)。之后,使用实验室制取的杨木灰样对Elman网络预测模型进行了验证。研究结果显示,Elman神经网络针对生物质灰软化温度具有更好的预测效果,其预测软化温度与实际软化温度的相对误差不超过20%,预测效果较好。

The softening temperature of biomass ash serves as an indicator of boiler slagging and heat transfer efficiency.Therefore,establishing a predictive model for the softening temperature of biomass ash can contribute to a better understanding of the slagging characteristics in biomass boilers.This paper presents the development of a prediction model for biomass ash softening temperature,leveraging the Back Propagation neural network and Elman neural network.Firstly,the original dataset comprised the mass fraction of oxide in biomass ash and its corresponding softening temperature.This dataset was utilized to establish the training parameters for the model.Subsequently,the model underwent rigorous testing.Finally,an evaluation of the model's adaptability was conducted,wherein the determination coefficient(R~2),Root Mean Square Error(E_(RMS)),Mean Absolute Error(E_(MA)),and Mean Absolute Percentage Error(E_(MAP)) of both the BP neural network and the Elman neural network were compared and analyzed.Subsequently,the Elman network prediction model was validated using laboratory-obtained poplar ash samples.The results indicate that for the test set data,the R~2 values of the softening temperature prediction models for biomass ash using both the BP neural network and Elman neural network are 0.998 87 and 0.999 96,respectively.Additionally,the Elman neural network model exhibits smaller values for E_(RMS),E_(MA) and E_(MAP).The Elman neural network demonstrates superior predictive capabilities for the softening temperature of biomass ash.Specifically,at residence times of 2 h and ash formation temperatures of 600 ℃,800 ℃,and 1 000 ℃,the relative error of the Elman neural network in predicting the softening temperature of poplar ash remains below 20%,indicating a robust predictive performance.This study delves into the correlation between seven components of biomass ash and softening temperature,providing valuable insights for establishing prediction models and achieving accurate predictions of the softening temperature of biomass ash.