为了提升炉内燃烧优化控制技术,炉内燃烧辐射能信号的在线获取是重要的。本文提出一种基于深度学习的炉内燃烧辐射能信号在线预测方法,以用于弥补辐射能信号检测技术在清洁探头期间无法获得炉内辐射能信号的不足。在一台350 MW燃煤锅炉上进行实验研究,利用火焰探测器连续获得辐射能信号,与负荷等运行参数构建数据集。之后,利用训练集建立并训练多层感知器(MLP)神经网络模型。最后,将运行参数输入到预测模型中以输出辐射能,并进行性能评估及误差分析。结果表明,利用运行参数可以准确预测辐射能,并且使用不同运行参数作为输入时建立的模型性能不同,最优模型的预测精度为97.8%。不同时间段下预测辐射能的标准差最小为0.0028,均小于实测辐射能,表明预测辐射能波动更小,确保控制系统的稳定性。

To improve the optimization control technology of furnace combustion, it is important to obtain the radiative energy signal of furnace combustion online. The study proposes an online prediction method for furnace combustion radiative energy signals based on deep learning to make up for the deficiency of radiative energy signal detection technology that cannot obtain the furnace radiative energy signal during the cleaning of the probe. An experimental study was conducted on a 350 MW coal-fired boiler. The flame detector was used to continuously obtain the radiative energy signal and construct a data set with operating parameters such as load. Afterwards, the training set was used to build and train a multilayer perceptron neural network model. Finally, the operating parameters were input into the prediction model to output the radiant energy, and performance evaluation and error analysis were performed. Results illustrate that the radiative energy can be accurately predicted using the operating parameters, and the performance of the models established when different operating parameters are used as input is different. The prediction accuracy of the optimal model is 97.8%. The minimum standard deviation of the predicted radiative energy in different time periods is 0.0028, which is smaller than the measured radiative energy, indicating that the predicted radiative energy fluctuates less, which is conducive to ensuring the stability of the control system.