基于AWOA-TEFN的热负荷预测模型
天津商业大学, 信息工程学院, 天津 300134
| 摘 要： | 在节能减排的战略背景下，区域供热系统的智能优化运行成为提升能源利用效率、降低碳排放的关键环节，而精准的热负荷预测是实现这一目标的核心技术支撑。为应对热负荷序列非线性、非平稳性显著及多影响因素耦合的复杂特性，解决现有预测模型对不确定性特征捕捉不足、超参数依赖人工调试、多变量时序信息融合不充分等问题，助力供热系统实现节能降耗与智能运行升级，本文提出一种基于自适应鲸鱼算法优化时间证据融合网络（AWOA-TEFN）的热负荷预测方案。该方案以证据理论与信息融合技术为核心，通过基本概率分配、期望融合及时间归一化-反归一化三大模块，从时间与通道双维度精准捕捉多变量时间序列的不确定性特征；数据预处理阶段采用三次样条插值填补缺失值、四分位数法剔除异常值，结合皮尔逊相关系数筛选核心输入特征，并利用AWOA算法自适应优化输入序列长度、预测步长等关键超参数，显著提升模型鲁棒性与适配性。以安阳换热站小时级热负荷实测数据为基础，将所提方案与GRU、RFR、LSTM、LightGBM等主流预测算法进行性能对比。实验结果表明，AWOA-TEFN方案的平均绝对百分比误差等各项评估指标均优于对比算法，能够精准追踪热负荷动态变化趋势，为供热系统的优化配置、智能调控及能效提升提供了可靠的技术支撑。 |
| 关 键 词： | 热负荷预测; AWOA算法; TEFN网络; 深度学习; 数据预处理 |
| DOI： | 10.57237/j.jest.2025.04.001 |
View:
Downloads:
收稿日期: 2025-12-05
接受日期: 2025-12-24
在线出版日期: 2025-12-29
Heat Load Forecasting Model Based on AWOA-TEFN
School of Information Engineering, Tianjin University of Commerce, Tianjin 300134, China
| Abstract: | Against the strategic backdrop of energy conservation and emission reduction, the intelligent optimization of regional heating systems has become a key link in improving energy utilization efficiency and reducing carbon emissions, with accurate heat load forecasting serving as the core technological support to achieve this goal. To address the complex characteristics of heat load sequences, such as nonlinearity, pronounced non-stationarity, and the coupling of multiple influencing factors, and to overcome the shortcomings of existing forecasting models—namely insufficient capture of uncertainty features, hyperparameter dependence on manual tuning, and inadequate integration of multivariate time series information—this paper proposes a heat load forecasting scheme based on an Adaptive Whale Optimization Algorithm-optimized Time Evidence Fusion Network (AWOA-TEFN). This scheme centers on evidence theory and information fusion technology, and, through three key modules—basic probability assignment, expectation fusion, and time normalization-denormalization—it accurately captures the uncertainty characteristics of multivariate time series from both temporal and channel dimensions. During the data preprocessing stage, cubic spline interpolation is used to fill in missing values, the quartile method is employed to remove outliers, and core input features are selected based on Pearson correlation coefficients. Additionally, the AWOA algorithm adaptively optimizes critical hyperparameters, such as input sequence length and prediction step size, significantly enhancing model robustness and adaptability. Based on hourly measured heat load data from the Anyang heat exchange station, the proposed scheme is compared with mainstream forecasting algorithms, including GRU, RFR, LSTM, and LightGBM. Experimental results indicate that the AWOA-TEFN scheme outperforms the comparative algorithms across all evaluation metrics, including mean absolute percentage error, accurately tracking the dynamic trends of heat load changes. This provides reliable technical support for the optimal configuration, intelligent regulation, and energy efficiency improvement of heating systems. |
| Keywords: | Heat Load Forecasting; Adaptive Whale Optimization Algorithm (AWOA); Temporal Evidence Fusion Network (TEFN); Deep Learning; Data Preprocessing |
基金
| 1. | 大学生创新训练计划项目 (202510069001) |
参考文献
| [1] | 黄润智. 基于机器学习的换热站热负荷预测控制研究[D]. 吉林化工学院, 2024. https://doi.org/10.27911/d.cnki.ghjgx.2024.000058 |
| [2] | Fan H, Li Y, Wang G, et al. Heat load prediction model based on LSTM-GRU neural network method [J]. Journal of Building Engineering, 2025, 114: 114456-114456. https://doi.org/10.1016/j.Jobe.2025.114456 |
| [3] | 孙一文. 热负荷预测中基于皮尔逊相关系数挑选相关因素[J]. 科学技术创新, 2025(22): 1-4. |
| [4] | Zhang X, Luo H, Pei Y, et al. Mechanism-guided short-term heat load prediction of district heating system based on a hybrid data-driven model [J]. Applied Thermal Engineering, 2025, 280(P2): 128113-128113. https://doi.org/10.1016/j.Applthermaleng.2025.128113 |
| [5] | 孙嘉旺. 区域供热系统的节能预测和智能调控方法研究[D]. 天津理工大学, 2023. https://doi.org/10.27360/d.cnki.gtlgy.2023.000244 |
| [6] | 韩英杰, 徐媛媛, 胡蓉. 基于PSO-SVM的多参数条件下供热负荷预测研究 [J]. 科技创新与应用, 2025, 15(08): 73-76. https://doi.org/10.19981/j.CN23-1581/G3.2025.08.016 |
| [7] | 王新雨, 郭振伟, 于丹, 等. 基于PCA-PSO-BP神经网络的住宅供热逐时负荷预测 [J]. 暖通空调, 2023, 53(03): 138-142+160. https://doi.org/10.19991/j.hvac1971.2023.03.25 |
| [8] | 杜占强, 鄢烈祥, 罗勇, 等. 基于LSTM神经网络的热力站短期热负荷预测仿真 [J]. 计算机仿真, 2025, 42(10): 125-129. |
| [9] | 张龙龙, 刘建军, 李冲, 等. 基于HNN和SDA的超短期热负荷预测研究 [J]. 暖通空调, 2025, 55(08): 174-180+93. https://doi.org/10.19991/j.hvac1971.2025.08.24 |
| [10] | 薛贵军, 赵广昊, 史彩娟. 基于改进黏菌算法优化BiLSTM的短期供热负荷控制预测 [J]. 沈阳工业大学学报, 2024, 46(04): 434-441. |
| [11] | 刘立巍, 周建新, 刘培栋, 等. 基于KNN-LSTM的区域热负荷短期预测及在机组热电可行域的应用研究 [J]. 热能动力工程, 2023, 38(03): 91-97. https://doi.org/10.16146/j.cnki.rndlgc.2023.03.012 |
| [12] | 杨博, 李晨晓, 徐嘉骋, 等. 面向非平稳热负荷序列的多模态分解与双路径预测框架 [J]. 热能动力工程, 2025, 40(11): 191-198. https://doi.org/10.16146/j.cnki.rndlgc.2025.11.021 |
| [13] | Zhan T, He Y, Deng Y, et al. Time Evidence Fusion Network: Multi-Source View in Long-Term Time Series Forecasting [J]. IEEE transactions on pattern analysis and machine intelligence, 2025, PP. https://doi.org/10.1109/tpami.2025.3596905 |
| [14] | 孔芝, 杨青峰, 赵杰, 等. 基于自适应调整权重和搜索策略的鲸鱼优化算法 [J]. 东北大学学报 (自然科学版), 2020, 41(01): 35-43. |
| [15] | Mirjalili S, Lewis A. The whale optimization algorithm [J]. Advances in engineering software, 2016, 95: 51-67. |
