摘要
文章旨在解决传统单一任务学习模型无法全面考虑设备状态相互影响和耦合关系的问题。文章通过分析和测试,揭示在不同时间跨度下模型的表现及潜在的优化方向。研究结果表明,随着时间跨度的增加,模型的均方根误差与均值绝对误差呈现整体下降趋势,尤其在短时间跨度下表现更为精确。针对长时间跨度数据,可能需更多的迭代次数以适应数据的复杂性。
The aim of this study is to address the limitations of traditional single-task learning models that fail to fully consider the mutual influence and coupling relationships among equipment states.This paper reveals the model's per‐formance and potential optimization directions across different time spans through analysis and testing.The research findings indicate that as the time span increases,the root mean square error(RMSE)and mean absolute error(MAE)of the model generally decrease,with particularly accurate performance in short time spans.For long time span data,more iterations may be necessary to adapt to the complexity of the data.
作者
潘科
Pan Ke(State Grid Jinhua Power Supply Company,Jinhua 321001,China)
出处
《办公自动化》
2025年第2期126-128,共3页
Office Informatization
关键词
多任务学习
变电站设备
设备状态评估
multi task learning
substation equipment
equipment status assessment
