基于SSAn-SSAl-LSTM的短期空调负荷预测模型
Short-term air conditioning load forecasting model based on SSAn-SSAl-LSTM
摘要：
本文提出了一种奇异谱分析(SSAn)和麻雀搜索算法(SSAl)优化的长短期记忆网络(LSTM)的组合空调负荷预测模型。使用皮尔逊相关系数和主成分分析法对输入特征进行挑选和处理,以消除特征之间的冗余性和相关性。针对空调负荷的波动性和随机性,采用SSAn将空调负荷分解为多个分量。同时针对LSTM超参数设置的问题,采用SSAl对模型进行优化,使用优化后的LSTM对各个分量进行预测,对预测结果进行重构。利用办公建筑和医疗建筑的空调负荷数据对模型进行了验证和分析。研究发现,与其他模型相比,SSAn-SSAl-LSTM模型表现最好,在预测办公建筑空调负荷时决定系数(R2)高达0.996 7,平均绝对百分比误差(MAPE)、平均绝对误差(MAE)和均方根误差(RMSE)分别为0.62%、14.42 kW和18.82 kW,在预测医疗建筑空调负荷时R2高达0.992 7,MAPE、MAE和RMSE分别为0.50%、19.40 kW和25.71 kW。
Abstract：
In this paper,a combined air conditioning load forecasting model based on long short-term memory network (LSTM) optimized by singular spectrum analysis (SSAn) and sparrow search algorithm (SSAl) is proposed.The Pearson correlation coefficient and the principal component analysis are used to select and process the input features to eliminate the redundancy and correlation between the features.In response to the volatility and randomness of the air conditioning load,SSAnis used to decompose the air conditioning load into multiple components.At the same time,aiming at the problem of LSTM hyperparameter setting,SSAlis used to optimize the model,and the optimized LSTM is used to predict each component.The prediction results are reconstructed.The model is validated and analysed using the air conditioning load data from office and medical buildings.It is found that the SSAn-SSAl-LSTM model performs the best compared with other models.When predicting the air conditioning load of the office building,the coefficient of determination (R2) is as high as 0.996 7,the average absolute percentage error (MAPE),the average absolute error (MAE) and the root mean square error (RMSE) are 0.62%,14.42 kW and 18.82 kW,respectively.When predicting the air conditioning load of the medical building,R2is 0.992 7,MAPE,MAE and RMSE are 0.5%,19.40 kW and 25.71 kW,respectively.
Keywords:air conditioning load; forecasting model; singular spectrum analysis (SSAn); sparrow search algorithm(SSAl); long short-term memory network (LSTM)