基于神经网络的建筑能耗混合预测模型

为了提升建筑能耗预测的精度、鲁棒性和泛化能力，提出树种算法（TSA）优化的径向基函数（RBF）神经网络与长短时记忆（LSTM）神经网络结合的混合预测模型. 采用基于自适应噪声的完全集成经验模态分解算法，将建筑能耗数据分解为1组本征模态函数（IMF）分量和1个残余分量，利用样本熵算法将各分量划分为高频分量和低频分量. 采用最小绝对收缩与选择算子（LASSO）方法进行特征选择. 分别利用TSA算法优化后的RBF模型与LSTM模型对低频分量和高频分量进行预测，并叠加重构得到最终预测结果. 模型评估结果表明，混合预测模型的精度为98. 72%. 相比于RBF、TSA-RBF、LSTM模型，所提模型的预测效果更好，且具有较强的鲁棒性和泛化能力，能够更为有效地用于建筑逐时电力能耗预测.

Hybrid prediction model of building energy consumption based on neural network

A hybrid prediction model which combining radial basis function (RBF) neural network optimized by tree-seed algorithm (TSA) and long short-term memory (LSTM) neural network was proposed, in order to improve the accuracy, robustness, and generalization ability of building energy consumption prediction. Firstly, the complete ensemble empirical mode decomposition with adaptive noise algorithm was used to decompose the building energy consumption data into a group of intrinsic mode function (IMF) components and a residual component, and the components were divided into high-frequency components and low-frequency components by sample entropy algorithm. Then, least absolute contraction and selection operator (LASSO) method was used for feature selection. Finally, the RBF model optimized by the TSA algorithm and the LSTM model were used to predict the low-frequency components and high-frequency components respectively, and the final prediction results were obtained by superposition and reconstruction. Model evaluation results showed that the accuracy of the hybrid prediction model was 98.72%. Compared with RBF, TSA-RBF, and LSTM models, the prediction effect of the hybrid model is better. Meanwhile, the model has strong robustness and generalization ability, and can be more effectively used for hourly building electricity consumption prediction.