基于电力时序大模型的时序数据少样本生成方法、装置、设备、介质和产品
By constructing a small-sample cue vector and a conditional cue vector based on a large-scale power time series model, and combining a joint loss function and a multi-scale constraint function for multi-round training, the problem of insufficient historical data in new power systems is solved, and the stability and efficiency of time series data generation are improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA SOUTHERN POWER GRID ARTIFICIAL INTELLIGENCE TECHNOLOGY CO LTD
- Filing Date
- 2026-04-15
- Publication Date
- 2026-07-17
AI Technical Summary
In new power systems, especially in newly built industrial parks, newly commissioned power plants, the initial stage of distributed power generation integration, and extreme weather conditions, insufficient historical data or incomplete sample distribution make it difficult for traditional methods to maintain stable performance in scenarios with few samples, and they are prone to overfitting.
A method based on a large power time series model is adopted. By acquiring enhanced feature vectors, constructing few-sample cue vectors and conditional cue vectors, and combining joint loss function and multi-scale constraint function for multi-round training, power time series data is generated, and consistency correction and operation boundary verification are performed.
It improves the stability of time series data generation in scenarios with scarce data or cold start, reduces the need for historical samples, and reduces redundant modeling and computing power consumption.
Smart Images
Figure CN122045827B_ABST