提出一种融合多阶段夏普利值分解（SHAP）特征分析和气象参数预测的遗传-反向传播（GA-BP）负荷预测模型。通过多阶段SHAP分析,探究气象参数在不同时间尺度上对负荷的影响,识别关键时间尺度以优化输入变量选择,减少特征冗余并降低过拟合风险。此外,结合K-均值聚类和复合模型架构,设计高精度气象参数预测方法并采用GA-BP模型实现负荷预测。结果显示,该模型能有效提升预测精度,最高可达到R2为0.86、均方误差（MSE）为0.0961,相较于单一的GA-BP负荷预测模型MSE降低40.12%,可在建筑供能系统启动前实现较高精度预测。
Abstract
This study proposes an advanced load forecasting model based on GA-BP, which integrates multi-stage SHAP feature analysis and meteorological parameter prediction. The proposed approach employs multi-stage SHAP analysis to investigate the influence of meteorological parameters on building loads across different time scales. This enables the identification of critical time scales, optimization of input variable selection, reduction of feature redundancy, and mitigation of overfitting risks. Additionally, a high-precision meteorological parameter prediction framework is developed by combining K-means clustering with a composite model architecture. The GA-BP model is then utilized to achieve accurate load forecasting. Experimental results demonstrate the effectiveness of the proposed model, achieving a maximum coefficient of determination of 0.86 and a mean squared error（MSE） of 0.0961. Compared to a standalone GA-BP model, the proposed approach reduces MSE by 40.12%, significantly improving prediction accuracy. This high-precision forecasting capability, available prior to the activation of the building energy supply system.
关键词
太阳能建筑 /
热负荷 /
神经网络 /
负荷预测 /
BIPV /
PV/T /
夏普利值分解
Key words
solar buildings /
thermal load /
neural networks /
load forecasting /
BIPV /
PV/T /
Shapley additive explanations （SHAP）
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] PEREIRA R, AELENEI L.Optimization assessment of the energy performance of a BIPV/T-PCM system using Genetic Algorithms[J]. Renewable energy, 2019, 137: 157-166.
[2] YANG T T, ATHIENITIS A K.A review of research and developments of building-integrated photovoltaic/thermal (BIPV/T) systems[J]. Renewable and sustainable energy reviews, 2016, 66: 886-912.
[3] 周泽楷, 侯宏娟, 孙莉, 等. 基于CNN和BiLSTM神经网络模型的太阳能供暖负荷预测研究[J]. 太阳能学报, 2024, 45(10): 415-422.
ZHOU Z K, HOU H J, SUN L, et al.Research on solar heating load forecasting based on CNN and BiLSTM neural network model[J]. Acta energiae solaris sinica, 2024, 45(10): 415-422.
[4] 李博文, 张丽玮, 冯洪庆. 基于多元非线性分布式能源系统的运行预测研究[J]. 太阳能学报, 2023, 44(6): 137-144.
LI B W, ZHANG L W, FENG H Q.Research on operation prediction of distributed energy system based on multivariate nonlinear analysis[J]. Acta energiae solaris sinica, 2023, 44(6): 137-144.
[5] KUZLU M, CALI U, SHARMA V, et al.Gaining insight into solar photovoltaic power generation forecasting utilizing explainable artificial intelligence tools[J]. IEEE access, 2020, 8: 187814-187823.
[6] WANG R, WANG J Y, XU Y Z.A novel combined model based on hybrid optimization algorithm for electrical load forecasting[J]. Applied soft computing, 2019, 82: 105548.
[7] AGÜERA-PÉREZ A, PALOMARES-SALAS J C, GONZÁLEZ DE LA ROSA J J, et al. Weather forecasts for microgrid energy management: review, discussion and recommendations[J]. Applied energy, 2018, 228: 265-278.
[8] XU L, WANG S W, TANG R.Probabilistic load forecasting for buildings considering weather forecasting uncertainty and uncertain peak load[J]. Applied energy, 2019, 237: 180-195.
[9] 任海波. 新型BIPVT建筑及其供能系统节能特性研究[D]. 北京: 北京工业大学, 2021.
REN H B.Study on energy-saving characteristics of new BIPVT building and its energy supply system[D]. Beijing: Beijing University of Technology, 2021.
[10] 薛东, 段立强, 高统彤, 等. 考虑多重特征与不确定性度量的综合能源系统负荷预测研究[J]. 太阳能学报, 2024, 45(7): 379-388.
XUE D, DUAN L Q, GAO T T, et al.Study of integrated energy system load forecasting considering multiple characteristics and uncertainty measures[J]. Acta energiae solaris sinica, 2024, 45(7): 379-388.
[11] NAWAR M, SHOMER M, FADDEL S, et al.Transfer learning in deep learning models for building load forecasting: case of limited data[C]//SoutheastCon 2023. Orlando, FL, USA, 2023: 532-538.
[12] BACANIN N, JOVANOVIC L, ZIVKOVIC M, et al.Multivariate energy forecasting via metaheuristic tuned long-short term memory and gated recurrent unit neural networks[J]. Information sciences, 2023, 642: 119122.
[13] WU K L, GU J, MENG L, et al.An explainable framework for load forecasting of a regional integrated energy system based on coupled features and multi-task learning[J]. Protection and control of modern power systems, 2022, 7(1): 24.
[14] MOON J, RHO S, BAIK S W.Toward explainable electrical load forecasting of buildings: a comparative study of tree-based ensemble methods with Shapley values[J]. Sustainable energy technologies and assessments, 2022, 54: 102888.
[15] MITRENTSIS G, LENS H.An interpretable probabilistic model for short-term solar power forecasting using natural gradient boosting[J]. Applied energy, 2022, 309: 118473.
基金
国家重点研发计划、政府间国际科技创新合作重点专项项目（2022YFE0118500）; 北京工业大学“城市碳中和”“城市更新”科技创新基金（052000514124561）
PDF(2270 KB)
