基于构造混合约束损失函数的Informer- PINN氢动力船舶超短期多步负荷预测框架

doi:10.19595/j.cnki.1000-6753.tces.251906

摘要
图/表
参考文献
相关文章 (13)

全文: PDF (1637 KB)
输出: BibTeX | EndNote (RIS)

摘要

高精度船舶电力系统负荷预测可为氢动力船舶能量管理优化提供参考依据,进而提升船舶运行综合效能。然而,陆上电力系统负荷预测模型直接应用于船舶场景时,难以有效融合多源异构数据的非线性耦合关联,且未充分考虑船舶电力系统的时变物理特性。为此,本文提出一种面向氢动力船舶的超短期多步负荷预测框架,基于物理神经网络（PINN）的 Informer 模型（Informer-PINN）。首先,采用最大信息系数分别筛选与维持负荷、推进负荷高度相关的变量作为模型输入,并通过非均匀相空间重构方法处理船舶运行及水文气象数据序列;其次,引入混沌系统构建损失函数的物理驱动部分,结合混沌系统特征匹配约束与动力学行为约束,与数据驱动的预测损失共同构成混合约束损失函数;最后,基于 “氢舟 1 号” 实际运行数据集开展算例验证。结果表明,所提 Informer-PINN 预测架构的性能优于多种当前先进预测模型,可有效应对内河航道氢动力船舶的负荷波动问题。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	刘星斗
	邹亮
	韩智云
	马良旺
	王睿

关键词 ：超短期多步负荷预测, 氢动力船舶, 物理神经网络, Informer模型, 混合约束损失函数

Abstract：

With the promotion and application of hydrogen energy in the shipbuilding industry, the energy management of hydrogen-powered ships demands high precision and timeliness in ultra-short-term multi-step load forecasting. However, load forecasting models for onshore power systems are difficult to directly adapt to ship scenarios. Ship loads are influenced by a couplingof factors such as navigation conditions and hydro-meteorology, exhibiting strong nonlinearity and dynamic fluctuations. Existing methods often neglect the integration of physical constraints anddata dynamics, resulting in insufficient forecasting accuracy and poor adaptability to varying conditions. This makes it difficult to support efficient energy scheduling for hydrogen-powered ships.
To address this issue, this paper proposes an Informer-PINN ultra-short-term multi-step load forecasting framework for hydrogen-powered ships, based on a constructed hybrid constraint loss function. The framework first selects multi-source variables strongly correlated with load through the Maximum Information Coefficient (MIC) to extract highly relevant features. Subsequently, it employs a non-uniform phase space reconstruction method to process ship operation and hydro-meteorological data, transforming them into feature forms suitable for model learning. Finally, itconstructs a hybrid loss function that integrates physical constraints of chaotic systems with data-driven methods. By combining the long-sequence prediction capability of Informer and the physical law embedding capability of Physical Information Neural Network (PINN), a closed-loop process of "multi-source variable selection - data reconstruction - hybrid constraint prediction" is formed.
Using the actual operating data of "Hydrogen Boat No.1" in the Three Gorges section of the Yangtze River as the validation sample (with a sampling interval of 1 minute, covering both stationary and non-stationary conditions), the proposed framework was compared with models such as ARIMA, XGBoost, GRU, and Transformer. The results showed that in the ultra-short-term predictions of 1/3/5/10 steps, the error metrics such as MAE and RMSE of the proposed framework were superior to those of the comparative models, and the accuracy decay rate under non-stationary conditions was lower. At the same time, the model training convergence speed was significantly accelerated, balancing prediction accuracy and computational efficiency, and verifying the advantage of integrating data-driven and physical constraints.
The Informer-PINN framework effectively addresses the core challenge of adapting onshore load forecasting models to ship scenarios, providing a high-precision basis for energy managementin hydrogen-powered ships. In the future, it can be further extended to complex sea conditions and other operational scenarios, deepening the coupling between the propulsion system and the ship's physical mechanisms, optimizing the weight configuration of the hybrid constraint loss function, enhancing the model generalization and robustness under extreme operating conditions, and facilitating the intelligent and efficient operation of hydrogen-powered ships.

Key words： Ultra-short-term multi-step load forecasting hydrogen powered vessels physical information neural network Informer model mixed constraint loss function

收稿日期: 2025-11-05

PACS:

TK91

基金资助:

国家重点研发计划资助项目（2022YFB4300703）

通讯作者: 邹亮男,1983年生,教授,博士生导师,研究方向为新能源发电与智能微电网、先进电工理论与装备等。E-mail：zouliang@sdu.edu.cn

作者简介: 刘星斗男,1996年生,博士研究生,研究方向为低/零碳船舶电力系统负荷预测与能量管理。E-mail：202320713@mail.sdu.edu.cn

引用本文:

刘星斗, 邹亮, 韩智云, 马良旺, 王睿. 基于构造混合约束损失函数的Informer- PINN氢动力船舶超短期多步负荷预测框架[J]. 电工技术学报, 0, (): 15-. Liu Xingdou, Zou Liang, Han Zhiyun, Ma Liangwang, Wang Rui. Informer-PINN ultra-short-term multi-step load forecasting framework for hydrogen powered ships based on constructing a mixed constraint loss function. Transactions of China Electrotechnical Society, 0, (): 15-.

链接本文:

https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.251906 https://dgjsxb.ces-transaction.com/CN/Y0/V/I/15

[1] 伏云辉.全球150 余国提出碳中和目标[J]. 生态经济,2024, 40(3): 1-4.
Yunhui. More than 150 countries around the world put forward the goal of carbon neutrality[J]. Ecological Economy, 2024, 40(3): 1-4.
[2] 国家能源局.中国氢能发展报告(2025)[R]. 北京:国家能源局,2025.
National Energy Administration. China hydrogen energy development report (2025)[R]. Beijing: National Energy Administration, 2025.
[3] Jiang J, Zou L, Liu X, Han Z, et al.A two-level real-time energy management strategy incorporating load forecasting for hydrogen fuel cell-powered ships[J]. Energy Conversion and Management, 2025, 341: 120032.
[4] 高云,解超,张高航,等. 计及电动汽车接入的园区综合能源系统低碳协调运行调度策略[J]. 电气技术,2025, 26(2): 14-25.
Gao Y, Xie C, Zhang G H, et al.Low-carbon coordinated operation scheduling strategy for park integrated energy system considering electric vehicle access[J]. Electrical Engineering, 2025, 26(2): 14-25.
[5] 刘汉有,范爱龙,夏民杰,等. 氢电混合动力船舶功率分配控制与能量管理策略试验验证[J]. 交通运输工程学报,2025, 25(4): 221-237.
Liu H Y, Fan A L, Xia M J, et al.Experimental verification of power distribution control and energy management strategy for hydrogen-electric hybrid power ships[J]. Journal of Traffic and Transportation Engineering, 2025, 25(4): 221-237.
[6] 王琛,王颖,郑涛,等. 基于ResNet-LSTM 网络和注意力机制的综合能源系统多元负荷预测[J]. 电工技术学报,2022, 37(7): 1789-1799.
Wang C, Wang Y, Zheng T, et al.Multi-load prediction of integrated energy system bBased on ResNet-LSTM network and attention mechanism[J]. Transactions of China Electrotechnical Society, 2022, 37(7): 1789-1799.
[7] Liu X D, Zou L, Zhang L, et al.Load prediction of hydrogen powered ship: A deep prediction framework considering operating environment and error correction[J]. IEEE Transactions on Smart Grid, 2025, DOI: 10.1109/TSG.2025.3615925.
[8] 赵洪山,吴雨晨,温开云,等. 基于时空注意力机制的台区多用户短期负荷预测[J]. 电工技术学报,2024, 39(7): 2104-2115.
Zhao H S, Wu Y C, Wen K Y, et al.Short-term load forecasting for multi-Users in transformer area based on spatiotemporal attention mechanism[J]. Transactions of China Electrotechnical Society, 2024, 39(7): 2104-2115.
[9] 赵洋,王瀚墨,康丽,等. 基于时间卷积网络的短期电力负荷预测[J]. 电工技术学报,2022, 37(5): 1242-1251.
Zhao Y, Wang H M, Kang L, et al.Short-term power load forecasting based on temporal convolutional network[J]. Transactions of China Electrotechnical Society, 2022, 37(5): 1242-1251.
[10] 李云松,张智晟. 考虑综合需求响应的 Transformer-图神经网络综合能源系统多元负荷短期预测[J]. 电工技术学报,2024, 39(19): 6119-6128.
Li Y S, Zhang Z S.Short-term multi-Load forecasting for integrated energy system based on transformer-graph neural network considering integrated demand response[J]. Transactions of China Electrotechnical Society, 2024, 39(19): 6119-6128.
[11] 朱继忠,董瀚江,李盛林,等. 数据驱动的综合能源系统负荷预测综述[J]. 中国电机工程学报,2021, 41(23): 7905-7923.
Zhu J Z, Dong H J, Li S L, et al.Review of data-driven load forecasting for integrated energy system[J]. Proceedings of the CSEE, 2021, 41(23): 7905-7923.
[12] 史佳琪,马丽雅,李晨晨,等. 基于串行-并行集成学习的高峰负荷预测方法[J]. 中国电机工程学报,2020, 40(14): 4463-4472.
Shi J Q, Ma L Y, Li C C, et al.Daily peak load forecasting based on sequential-parallel ensemble learning[J]. Proceedings of the CSEE, 2020, 40(14): 4463-4472.
[13] Chen Y, Chen M, Xu L, Liang Z.Chance-constrained optimization of storage and PFC capacity for railway electrical smart grids considering uncertain traction load[J]. IEEE Transactions on Smart Grid, 2024, 15(1): 286-298.
[14] 肖白,黄钰茹,姜卓,等. 数据匮乏场景下采用生成对抗网络的空间负荷预测方法[J]. 中国电机工程学报,2020, 40(24): 7990-8001.
Xiao B, Huang Y R, Jiang Z, et al.Spatial load forecasting method using generative adversarial network under data-scarce scenario[J]. Proceedings of the CSEE, 2020, 40(24): 7990-8001.
[15] 杨亚宇,邰能灵,黄文焘,等. 船舶中压直流综合电力系统(一):系统结构和电力电子变换器[J]. 电工技术学报,2024, 39(21): 6647-6665.
Yang Y Y, Tai N L, Huang W T, et al.Medium-voltage DC integrated power system for ships (I): system structure and power electronic converters[J]. Transactions of China Electrotechnical Society, 2024, 39(21): 6647-6665.
[16] 谭洪,王宇炜,王秋杰,等. 基于氢能固态运输的电-氢综合能源系统双层调度模型[J]. 电工技术学报,2025, 40(3): 744-758.
Tan H, Wang Y W, Wang Q J, et al.A bi-level scheduling model for electric-hydrogen integrated energy system based on hydrogen solid-state transportation[J]. Transactions of China Electrotechnical Society, 2025, 40(3): 744-758.
[17] Banaei M., Boudjadar J, Khooban M H.Stochastic model predictive energy management in hybrid emission-free modern maritime vessels. IEEE Trans. Ind. Inform.,2021, 17: 5430-5440.
[18] Pang S, Zou L, Zhang L, Wang H, Wang Y, Liu X, Jiang J.A hybrid TCN-BiLSTM short-term load forecasting model for ship electric propulsion systems combined with multi-step feature processing[J]. Ocean Engineering, 2025, 316: 119808.
[19] Tariq A, Uzun B, Deliktas B, Yayli M O.Bending analysis of cantilever microbeams with three porosity distributions using physics-informed neural network and modified couple stress theory[J]. Engineering Applications of Artificial Intelligence, 2025, 159: 111589.
[20] Zamzam A S, Sidiropoulos N D.Physics-aware neural networks for distribution system state estimation[J]. IEEE Transactions on Power Systems, 2020, 35(6): 4347-4356.
[21] 杨珂,王鑫,凌佳杰,等. 基于物理信息神经网络的同步发电机建模[J]. 中国电机工程学报,2024, 44(12): 4924-4932.
Yang K, Wang X, Ling J J, et al.Modeling of synchronous generator based on physics-informed neural network[J]. Proceedings of the CSEE, 2024, 44(12): 4924-4932.
[22] Khodayar M, Liu G, Wang J, Khodayar M E.Deep learning in power systems research: A review[J]. CSEE Journal of Power and Energy Systems, 2021, 7(2): 209-220.
[23] Zhang Z, Li Z, Du Y, Jiang X.A rapid motion forecast strategy for ships in waves using seakeeping and maneuvering modules[J]. Ocean Engineering, 2024, 309: 118539.
[24] Srivastava A K, Singh A K, Singh S P.Groundwater level forecasting using artificial neural network, adaptive neuro fuzzy inference system and support vector machine techniques: A case study of Ganga Alluvial Plain, India[J]. Journal of Hydrology, 2015, 526: 186-201.
[25] Wang J Y, Qiu L, Liang Y J, et al.A spatio-temporal radial basis function collocation method based on Hausdorff fractal distance for Hausdorff derivative heat conduction equations in three-dimensional anisotropic materials[J]. Applied Mathematics and Computation, 2025, 502: 129501.
[26] Cao Q, Wei D Q.Dynamic surface sliding mode control of chaos in the fourth-order power system[J]. Chaos, Solitons and Fractals, 2023, 170: 113420.
[27] McGrane-Corrigan B, de Andrade Moral R, Mason O. Observations on robust diffusive stability and common Lyapunov functions[J]. Systems & Control Letters, 2025, 205: 106263.
[28] Al-Hussein A B A, Tahir F R, Ouannas A, et al. Chaos suppressing in a three-buses power system using an adaptive synergetic control method[J]. Electronics, 2021, 10: 1532.
[29] Bi C, Ren P, Yin T, et al.An informer architecture-based ionospheric foF2 model in the middle latitude region[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19: 1005305.
[30] Liu X D, Zou L, Zhang L, et al.A multi-channel spatiotemporal SegNet model for short term wind power prediction with sequence decomposition and reconstruction[J]. Results in Engineering, 2025, 27: 105797.
[31] Salami M, Sobhani F.A hybrid short-term load forecasting model developed by factor and feature selection algorithms using improved grasshopper optimization algorithm and principal component analysis[J]. Electrical Engineering, 2020, 102(1): 437-460.
[32] Jiang J, Zou L, Zhang L, Wang H, Wang Y, Liu X, Pang S.A three-layer energy management system for hydrogen-powered ships combined instantaneous load forecasting[J]. Electric Power Systems Research, 2025, 243: 111494.
[33] 彭昱,符琛,郭昕,等. 基于FCM-BOA-TCN-GRU 的分布式光伏出力异常检测方法[J]. 电工技术学报,2025, 40(17): 5389-5400.
Peng Y, Fu C, Guo X, et al.Distributed photovoltaic output anomaly detection method based on FCM-BOA-TCN-GRU[J]. Transactions of China Electrotechnical Society, 2025, 40(17): 5389-5400.
[34] 全睿,程功,周宇龙,等. 基于增强型鲸鱼优化算法 CNN-BiGRU-AT 模型的燃料电池衰退预测[J]. 电工技术学报,2025, 40(19): 6342-6358.
Quan R, Cheng G, Zhou Y L, et al.Fuel cell degradation prediction based on enhanced whale optimization algorithm CNN-BiGRU-AT model[J]. Transactions of China Electrotechnical Society, 2025, 40(19): 6342-6358.