2results about How to "Prove superiority" patented technology

This invention belongs to the field of smart grid and load monitoring technology, and provides a multi-task non-intrusive load decomposition method system. The method includes: acquiring the total load data sequence at the input port of the power consumption unit, and inputting the total load data sequence in parallel to a power decomposition branch and a state identification branch; the power decomposition branch processes the total load data sequence to generate a power prediction value for the target appliance; the state identification branch processes the total load data sequence to generate an operating state probability for the target appliance; and the power prediction value and the operating state probability are fused to generate the final decomposed power of the target appliance. This invention, through a dual-branch multi-task architecture and output fusion mechanism, collaboratively optimizes the power prediction and state identification tasks, effectively solving problems such as insufficient long-term dependency modeling, inaccurate capture of diverse features, and power false alarms caused by inconsistent prediction results in existing technologies, significantly improving the accuracy and reliability of load decomposition.

This invention discloses a data augmentation method for low-resource news event extraction based on a large language model and generative adversarial networks. The method acquires original news events and extracts a training dataset from them. This dataset undergoes parsing, separation, generator, and discriminator processing, and is then combined with a Sentence-BERT model to obtain a high-quality augmented training dataset. This invention, through a generative paradigm, significantly improves the accuracy scores of both trigger word classification and argument classification in the event extraction subtask, achieving a dramatic performance leap. After generating augmented training data using this invention, the model is trained, ultimately achieving higher accuracy than other methods in the event extraction task, demonstrating the superiority of this data augmentation method.