Distillation integrated mode sensing algorithm and system for unbalanced power data

Abstract

The invention relates to an unbalanced electric power data-oriented distillation integrated mode sensing algorithm, which comprises the following steps of S1, carrying out situation space reconstruction processing on electric power data with unbalanced situation distribution by adopting a DGS-based oversampling technology; s2, constructing a power data deep learning mode perception model based on a CLSTM neural network by adopting a model compression technology and combining an LN mechanism and an Attention mechanism; and S3, proposing a distillation type integrated learning strategy and performing optimization to solve the redundancy problem of the base learning machine cluster. The method has the advantages of being stable in algorithm efficiency, high in efficiency and small in scale, and meanwhile the redundancy problem of the base learning machine cluster can be solved.

Description

Technical field [0001] The invention is the field of power data mining technology, which involves a distilled integration mode perception algorithm and system for distillation integration models for uneven power data. Background technique [0002] With the rapid development of the development of smart grids and the energy Internet, the integration process of modern information technology such as artificial intelligence and traditional power grid technology has also increased, becoming an auxiliary power grid manager to quickly perceive the inherent laws of the poly source data of the power system, implement accurate control and decision -making Efficient means. At present, cutting -edge artificial intelligence technology represented by integrated learning and deep learning shows good application value in terms of power system security and control, equipment operation and maintenance and failure diagnosis, and load management. Services and promotion of energy transformation in my ...

AI Technical Summary

Problems solved by technology

The situation space reconstruction method represented by the SMOTE oversampling algorithm can avoid coupling the situation space reconstruction process with the pattern recognition process, and is more widely used. However, the traditional SMOTE method ignores the distribution characteristics of the sample set, and the reconstructed The situation space is easily distorted, and it is difficult for the algorithm to maintain a stable sampling effect in a high-ratio unbalanced scene, and even interfere with the classification process of the majority class samples

By learning the power data situation distribution law, the intelligent generative model represented by generative confrontation network can obtain artificial samples with high fidelity, but the complexity of the algorithm is high, and the computational cost is high, which is difficult to meet the actual application requirements.

Images