基于自适应权重与情境感知机制的电力交易风险评估方法

许道强; 王江辉; 许梦晗; 詹天明; 吕从东

doi:10.11959/j.issn.1000-0801.DXKX250283

您当前的位置：

首页 >

文章列表页 >

基于自适应权重与情境感知机制的电力交易风险评估方法

研究与开发 | 更新时间：2026-06-17

- 基于自适应权重与情境感知机制的电力交易风险评估方法
- A risk assessment method for electricity trading based on adaptive weights and context-aware mechanisms
- 电信科学 2026年42卷第5期页码：143-154
- 作者机构：
  
  1.国网江苏省电力有限公司，江苏南京 210024
  2.江苏方天电力技术有限公司，江苏南京 211100
  3.南京审计大学计算机学院，江苏南京 211815
- 作者简介：
  
  许道强（1978- ），男，国网江苏省电力有限公司正高级工程师，主要研究方向为电力营销数字化技术研究级应用。
  王江辉（1992- ），男，江苏方天电力技术有限公司工程师，主要研究方向为电力营销数字化技术管理。
  许梦晗（1990- ），女，国网江苏省电力有限公司高级工程师，主要研究方向为电力营销。
  詹天明（1984- ），男，博士，南京审计大学计算机学院教授、博士生导师，主要研究方向为大数据分析、人工智能。
  吕从东（1987- ），男，博士，南京审计大学计算机学院计算机系主任、副教授，主要研究方向为人工智能算法及其应用、人工智能安全、智能审计、信息安全等。
- 基金信息：
  
  国家自然科学基金资助项目(61976117)
- DOI：10.11959/j.issn.1000-0801.DXKX250283
  中图分类号： TP393
- 收稿：2025-05-06，
  
  修回：2025-06-17，
  
  录用：2025-06-27，
  
  纸质出版：2026-05-20
- 稿件说明：
移动端阅览
许道强,王江辉,许梦晗等.基于自适应权重与情境感知机制的电力交易风险评估方法[J].电信科学,2026,42(05):143-154.

Xu Daoqiang,Wang Jianghui,Xu Menghan,et al.A risk assessment method for electricity trading based on adaptive weights and context-aware mechanisms[J].Telecommunications Science,2026,42(05):143-154.
许道强,王江辉,许梦晗等.基于自适应权重与情境感知机制的电力交易风险评估方法[J].电信科学,2026,42(05):143-154. DOI： 10.11959/j.issn.1000-0801.DXKX250283.

Xu Daoqiang,Wang Jianghui,Xu Menghan,et al.A risk assessment method for electricity trading based on adaptive weights and context-aware mechanisms[J].Telecommunications Science,2026,42(05):143-154. DOI： 10.11959/j.issn.1000-0801.DXKX250283.

摘要

提出一种基于自适应权重与情境感知机制的电力交易风险评估方法，通过多模型融合策略显著提升预测精度与风险管理能力。该方法结合历史表现与时间衰减因子，动态优化线性回归、随机森林及长短期记忆（long short-term memory，LSTM）网络权重的自适应权重调整机制，并根据市场波动、极端天气或政策事件触发规则，动态调整模型权重，以应对突发场景，构建情境感知切换逻辑；采用梯度提升决策树（gradient boosting decision tree，GBDT）整合基础模型预测与市场特征，提升复杂场景鲁棒性的元学习器优化策略。实验表明，混合模型在均方误差（mean square error，MSE）为21.3、平均绝对误差（mean absolute error，MAE）为3.1和决定系数（coefficient of determination，

²）为0.89时优于单一模型，其中自适应权重贡献性能提升10%，情境感知的误差降低5%。在电价单日波动率大于20%、连续高温或政策突变等极端场景下，模型通过动态权重调整显著降低了误差，在提升收益的同时降低了风险。该方法为电力市场提供了科学的风险管理工具，增强了市场稳定性与经济效益，未来可将其应用拓展至金融风险评估领域，并在联邦学习、实时优化与规则自动化等方向进一步探索。

Abstract

An electricity trading risk assessment method based on an adaptive weighting and context-aware mechanism was proposed

which significantly enhanced prediction accuracy and risk management capabilities through a multi-model fusion strategy. The method incorporates an adaptive weight adjustment mechanism that dynamically optimizes

the weights of linear regression

random forest

and long short-term memory(LSTM) models by combining historical performance with time decay factors. Additionally

it employs a context-aware switching logic triggered by market volatility

extreme weather

or policy events to dynamically adjust model weights in response to sudden scenarios. A meta-learner optimization strategy using gradient boosting decision tree(GBDT) integrates the predictions of base models with market features to improve robustness in complex scenarios. Experimental results demonstrate that the hybrid model outperforms individual models across key metrics

including mean square error(MSE) of 21.3

mean absolute error(MAE) of 3.1

and

² of 0.89. The adaptive weighting mechanism contributes to a 10% performance improvement

while context-aware switching further reduces errors by 5%. In extreme scenarios such as daily price volatility exceeding 20%

prolonged high temperatures

or sudden policy changes

the model significantly mitigates errors through dynamic weight adjustments. The hybrid model not only enhances returns but also reduces risks. This method provides a scientific risk management tool for electricity markets

enhancing market stability and economic efficiency. It holds potential for extension to financial risk assessment domains and future exploration in federated learning

real-time optimization

and rule automation.

关键词

Keywords

references

Marček D . Hybrid ARIMA/RBF framework for prediction BUX index [J ] . Journal of Computer and Communications , 2015 , 3 ( 5 ): 63 - 71 .

Ahlawat S . Benchmarking machine learning models [M ] // Statistical Quantitative Methods in Finance . Berkeley, CA : Apress , 2025 : 253 - 285 .

Čatloch D , Chovancová E , Matašová S . Comparison of different machine learning algorithms for stock price prediction [C ] // 2025 IEEE 23rd World Symposium on Applied Machine Intelligence and Informatics (SAMI) . Piscataway : IEEE Press , 2025 : 483 - 488 .

樊荣 . 基于动态类权重的卷积神经网络攻击检测模型 [J ] . 电信科学 , 2025 , 41 ( 8 ): 176 - 185 .

Fan R . Convolutional neural network attack detection model based on dynamic class weight [J ] . Telecommunications Science , 2025 , 41 ( 8 ): 176 - 185 .

Lan Z L , Gu J X , Zheng Z M , et al . A study of dynamic meta-learning for failure prediction in large-scale systems [J ] . Journal of Parallel and Distributed Computing , 2010 , 70 ( 6 ): 630 - 643 .

张冬慧 , 孙波 , 王鹏 , 等 . 权值自适应调整的多分类器融合算法 [J ] . 计算机工程 , 2008 , ( 10 ): 28 - 30 .

Zhang D H , Sun B , Wang P , et al . Multi-classifiers fusion algorithm of adaptive weight adjustment [J ] . Computer Engineering , 2008 , 34 ( 10 ): 28 - 30 .

Kumar A , Sunitha R . MuSeFFF: Multi-stage feature fusion framework for traffic prediction [J ] . Intelligent Systems with Applications , 2023 , 18 : 200227 .

王继禾 , 吴颖 , 迟恒喆 , 等 . 基于Transformer结构增强的神经网络架构搜索性能预测器 [J ] . 计算机学报 , 2024 , 47 ( 7 ): 1469 - 1484 .

Wang J H , Wu Y , Chi H Z , et al . An architecture-enhanced performance predictor for transformer-based NAS [J ] . Chinese Journal of Computers , 2024 , 47 ( 7 ): 1469 - 1484 .

胡润文 , 项世军 , 李晓龙 , 等 . 基于CNN计算局部复杂度的可逆信息隐藏算法 [J ] . 计算机学报 , 2024 , 47 ( 4 ): 776 - 789 .

Hu R W , Xiang S J , Li X L , et al . CNN based local complexity estimation for reversible data hiding [J ] . Chinese Journal of Computers , 2024 , 47 ( 4 ): 776 - 789 .

Serreli L , Fadda M , Girau R , et al . A generative adversarial network (GAN) fingerprint approach over LTE [J ] . IEEE Access , 2024 , 12 : 82083 - 82094 .

Rao F . Parallel adaptive window dynamic time warping attention transformer network model for time series forecasting [C ] // 2025 40th Youth Academic Annual Conference of Chinese Association of Automation (YAC) . Piscataway : IEEE Press , 2025 : 2623 - 2626 .

Han J Y , Wang H B , Niu D , et al . Multi-source data fusion network for sea fog detection [C ] // 2024 36th Chinese Control and Decision Conference (CCDC) . Piscataway : IEEE Press , 2024 : 5755 - 5760 .

Mo T X , Li S S , Li G D . An interpretable machine learning model for predicting cavity water depth and cavity length based on XGBoost-SHAP [J ] . Journal of Hydroinformatics , 2023 , 25 ( 4 ): 1488 - 1500 .

Gilani S A H , Syed M H , Anjum A . Effective diabetes prediction: integrating ensemble learning with LIME for robust results [C ] // 2024 International Conference on Frontiers of Information Technology (FIT) . Islamabad, Pakistan . Piscataway : IEEE Press , 2024 : 1 - 6 .

Wahid A , Faiud I , Mason K . Integrating renewable energy in agriculture: a deep reinforcement learning-based approach [M ] // Machine Learning and Principles and Practice of Knowledge Discovery in Databases . Cham : Springer Nature Switzerland , 2025 : 324 - 336 .

Ahmad S , Lavin A , Purdy S , et al . Unsupervised real-time anomaly detection for streaming data [J ] . Neurocomputing , 2017 , 262 : 134 - 147 .

Cheng Y , Liu Y , Chen T J , et al . Federated learning for privacy-preserving AI [J ] . Communications of the ACM , 2020 , 63 ( 12 ): 33 - 36 .

Xia H S , Tian Y Q , Zhang J Z , et al . Exploring the impact of online news sentiment and relevance on stock market risks: a signalling theory perspective [J ] . Expert Systems , 2025 , 42 ( 1 ): e13364 .

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

随机森林在运营商大数据补全中的应用

融合多源数据与机器学习的Smart CAPEX无线网络规划方法

RFA-XGBoost模型在移动网络潜在投诉用户预测中的应用

面向算力网络的多方安全协同线性回归研究

基于时频变换与动态图注意力的多模态序列推荐方法