Conditional diffusion new energy output scene generation method driven by large language model

By using a conditional diffusion model driven by a large language model, combined with a deep residual network and one-dimensional convolution, and optimizing hyperparameters, the efficiency and accuracy issues of generating new energy power output scenarios were solved. This enabled efficient and accurate generation of new energy power output scenarios, improving the stability and efficiency of power grid dispatch.

CN121580862AActive Publication Date: 2026-02-27DALIAN UNIV OF TECH
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202610073568.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-01-20
Publication Date
2026-02-27
Estimated Expiration
2046-01-20

AI Technical Summary

Technical Problem

Existing methods for generating new energy output scenarios have low computational efficiency on the grid side, limited scenario diversity, and difficulty in accurately depicting the spatiotemporal correlation under multi-energy coupling, resulting in bias and high computational complexity in power system dispatch optimization.

Method used

A conditional diffusion model driven by a large language model is adopted, which combines a deep residual network and a one-dimensional convolution to construct a conditional diffusion model. The model generates scene sets through Markov chains, and the hyperparameters are optimized through a large language model to improve the accuracy and efficiency of scene generation.

Benefits of technology

It significantly improves the computational efficiency and accuracy of generating new energy output scenarios, reduces the Euclidean distance and mean absolute error of the scenario set by 14.9%, reduces the computation time by 87%, optimizes the scheduling decision effect, and reduces the fluctuation of the system's remaining load by 20.5%.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121580862A_ABST
    Figure CN121580862A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of electrics, and particularly relates to a conditional diffusion new energy output scene generation method driven by a large language model. The method comprises the following steps: firstly, constructing a conditional diffusion model suitable for power grid side new energy output scene generation, embedding conditional information to implicitly learn conditional probability distribution of actual new energy output, and generating a scene set based on a Markov chain; secondly, proposing a thinking chain optimization framework driven by a large language model, analyzing a training state by utilizing the semantic reasoning ability of the large language model, and quickly locking an optimal parameter interval under an extremely low calculation budget; an example verification result shows that the average Euclidean distance of the generated scene set is increased by more than 30.7% compared with that of a traditional method; the optimization efficiency and timeliness are greatly improved, the scene generation precision is further improved by about 1% compared with a Bayesian optimization method under extremely few calculation budget which only allows 10 iterations, and the scene set generation time is reduced by 87% compared with a Copula model; and the reliability and timeliness of the scheduling decision are obviously improved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Three-dimensional target reconstruction method based on non-calibration single view

    CN118470221A

  • Renewable energy system operation scene generation method and system

    CN119358219A

  • Extreme scene generation method, system and device and storage medium

    CN120217848A