A novel multi-task learning framework TBM with shared and task-specific features fusion in regional wind-solar-load joint forecasting

作者
Liu Puzhuang,Nor Azwan Mohamed Kamari,Mohd Hairi Mohd Zaman,Mohd Asyraf Zulkifley
出处
期刊:Energy Conversion And Management: X [Elsevier BV]
卷期号:28: 101346-101346 被引量:2
标识
DOI:10.1016/j.ecmx.2025.101346
摘要

As the global energy transition accelerates, the joint forecasting of regional wind power, solar generation, and load demand is critical for enhancing renewable energy utilization, optimizing energy system configurations, and maintaining grid stability. The complex spatiotemporal dynamics and nonlinear interactions among these variables present significant challenges for existing forecasting methods. This study introduces a novel multi-task learning framework TBM (TFT-BiGRU-MultiAttention) that employs shared task experts and Cross-Stitch units to dynamically balance shared and task-specific feature representations. Experimental results demonstrate that the TBM framework achieves R 2 values of 99.227%, 93.958%, and 99.676% for wind power, photovoltaic generation, and load forecasting, respectively, outperforming traditional models such as GRU and MoME. In particular, in load forecasting, the framework records a Mean Absolute Error (MAE) of 57.389 MW and a Root Mean Square Error (RMSE) of 77.700 MW, with substantial improvements over the baseline methods. Additionally, a dropout-based uncertainty quantification approach yields the lowest Continuous Ranked Probability Scores (CRPS) across all tasks, ensuring reliable prediction intervals. Cross-Stitch matrix analysis further reveals task interdependencies effectively captured through adaptive feature sharing, underscoring the framework’s robustness and adaptability. These advances provide critical support for the optimization of renewable energy systems and intelligent scheduling. • Multi-task learning enhances wind, solar, load forecasting with dynamic feature balance. • High-precision wind, solar, load forecasting optimize scheduling and grid stability. • Dynamic feature fusion improves adaptability, robustness for complex energy forecasting. • Cross-Stitch matrix optimizes task interactions, boosting model stability, performance. • Uncertainty quantification ensures reliable predictions for energy risk management.

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
BAEKHYUNLUCKY发布了新的文献求助10
刚刚
刚刚
华无剑的应助被荼蘼采纳,获得10
刚刚
所所的应助被荼蘼采纳,获得10
1秒前
科研通AI65的应助被荼蘼采纳,获得10
1秒前
Maestro_S的应助被荼蘼采纳,获得10
1秒前
奔跑的应助被荼蘼采纳,获得10
1秒前
1秒前
1秒前
大个的应助被流星采纳,获得10
1秒前
科研通AI6.3的应助被荼蘼采纳,获得10
1秒前
科研通AI6.3的应助被荼蘼采纳,获得10
2秒前
科研通AI6.4的应助被荼蘼采纳,获得10
2秒前
余香的应助被荼蘼采纳,获得10
2秒前
科研通AI6.3的应助被荼蘼采纳,获得10
2秒前
3秒前
思源的应助被方法采纳,获得10
4秒前
sssssss完成签到,获得积分10
4秒前
龙眼肉发布了新的文献求助10
4秒前
bkagyin的应助被咪咪里采纳,获得10
5秒前
水bluer完成签到,获得积分10
5秒前
叶叶叶完成签到,获得积分10
5秒前
6秒前
GUYIMI完成签到,获得积分10
6秒前
eye完成签到,获得积分10
6秒前
trap发布了新的文献求助10
7秒前
AA发布了新的文献求助10
7秒前
8秒前
tanhuan完成签到 ,获得积分10
9秒前
xinling发布了新的文献求助10
9秒前
ZC7发布了新的文献求助10
10秒前
10秒前
Ali完成签到,获得积分10
11秒前
11秒前
11秒前
烟花的应助被路飞采纳,获得10
11秒前
12秒前
12秒前
Hello的应助被龙眼肉采纳,获得10
14秒前
悦耳糜完成签到 ,获得积分10
15秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
八旗制度の研究 600
plantTFDB 508
Resource Recovery Handbook 500
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8175267
求助须知:如何正确求助?哪些是违规求助? 9567035
关于积分的说明 21789830
捐赠科研通 7769741
什么是DOI,文献DOI怎么找? 3431840
关于科研通互助平台的介绍 2556149
邀请新用户注册赠送积分活动 2477241