基于改进秃鹰搜索算法优化门控循环单元的短期建筑冷负荷预测模型

Short-term building cooling load prediction model based on improved bald eagle search algorithm for optimizing gated recurrent unit

摘要:

准确地预测建筑冷负荷对空调系统节能优化控制具有重要作用，因此提出一种改进秃鹰搜索算法(BES)优化门控循环单元(GRU)的短期冷负荷预测模型；首先采用完全噪声辅助聚合经验模态分解(CEEMDAN)算法，将建筑冷负荷数据分解为不同频率的分量，采用随机森林结合递归特征消除法为不同频率的分量选取对应的特征；最后采用改进BES算法对GRU模型进行参数寻优，针对BES算法不足进行改进，引入Sobol序列初始化种群、采用非线性控制因子平衡BES算法搜索能力和自适应t分布策略提升算法寻优能力；实验结果表明，与GRU和改进BES算法优化后的GRU相比，提出的预测模型均方根误差下降34.27，22.41、平均百分比误差下降2.72%，2.63%、平均绝对误差下降27.25，25.26；相较于其他预测模型，提出的预测模型具有更高的预测准确度，在实际工程应用中更具优势。

Abstract:

Accurately predicting the cooling load of buildings is crucial for energy-efficient control of air conditioning systems, therefore, an improved bald eagle search (BES) algorithm is proposed to optimize the short-term cooling load prediction model of gated recirculation units (GRU); firstly, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm is used to decompose the building cooling load data into components with different frequencies, and a random forest combined with the recursive feature elimination method is used to select the corresponding features for the components with different frequencies; the improved BES algorithm is used to optimize the GRU model. Finally, the improved BES algorithm is used to optimize the parameters of the GRU model, and the BES algorithm is improved to address the shortcomings of the BES algorithm, by introducing the Sobol sequence to initialize the population, adopting the nonlinear control factor to balance the search capability of the BES algorithm, and the adaptive t-distribution strategy to enhance the algorithm's ability to find the optimal; the experimental results show that compared to the GRU and the optimized GRU, the optimized GRU is more efficient and effective than the optimized GRU. The experimental results show that compared with GRU and GRU with improved BES algorithm, the root mean square error of the proposed prediction model decreases by 34.27, 22.41, the average percentage error decreases by 2.72%, 2.63%, and the average absolute error decreases by 27.25, 25.26; compared with other prediction models, the proposed prediction model has a higher prediction accuracy, which is more advantageous in practical engineering applications.