为进一步提高锅炉系统水冷壁温度的预测精度，提出一种基于变量优化和改进鲸鱼算法优化长短期记忆神经网络的水冷壁温度预测模型。首先，通过互信息算法(MI)进行变量选择，消除初始数据中的冗余变量；其次，使用经验模态分解算法(EMD)对变量选择后的数据进行特征分解，在提取变量有效特征信息的同时降低噪音干扰；最后，使用由非线性递减因子和自适应权值改进后的鲸鱼优化算法(Improved Whale Optimization Algorithm，IWOA)确定长短期记忆神经网络(LSTM)的超参数，得到一种新型锅炉系统水冷壁温度预测模型(MIEMDIWOALSTM)。实验结果表明，相比传统的最小二乘支持向量机(LSSVM)预测模型，MIEMDIWOALSTM模型的均方根误差(RMSE=0.306 8)和平均绝对百分比误差(MAPE=0.054 6)最低，能够实现对锅炉系统水冷壁工质温度的精准预测。

In order to further improve the prediction accuracy of the water wall temperature of the boiler system, a water wall temperature prediction model based on variable optimization and improved whale algorithm optimized long shortterm memory (LSTM) neural network was proposed. Firstly, the mutual information (MI) algorithm was used to select variables to eliminate redundant variables in the initial data; secondly, the empirical mode decomposition (EMD) algorithm was used to decompose the data after variable selection, and the noise interference was reduced while extracting the effective feature information of the variables; finally, the improved whale optimization algorithm (IWOA) improved by nonlinear decreasing factor and adaptive weight was used to determine the hyperparameters of long shortterm memory neural network, and a new boiler system water wall temperature prediction model (MIEMDIWOALSTM) was obtained. The experimental results show that compared with the traditional least squares support vector machine ( LSSVM ) prediction model, the MIEMDIWOALSTM model has the lowest root mean square error (RMSE) of 0.306 8 and mean absolute percentage error (MAPE) of 0.054 6, which can realize the accurate prediction of the working medium temperature of the water wall of the boiler system.

TP301.6

山西省高等学校科技创新项目(2020L0718)；山西省高等学校教学改革创新项目（J2020383）