Cross-season short-term load prediction method and system for electric power spot market

By using variational mode decomposition and particle swarm optimization algorithms to optimize the long short-term memory network and random forest model, the problems of multi-scale coupling feature separation and price signal fusion of load data in the electricity spot market are solved, and high-precision and stable short-term load forecasting is achieved.

CN122136812APending Publication Date: 2026-06-02GUODIAN ZHEJIANG POWER SALES CO LTD
1 Cites 0 Cited by

Patent Information

Application Number
CN202610208191.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-02-12
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately separate multi-scale coupling characteristics in the electricity spot market, neglect the price signal guidance mechanism, and fail to adaptively adjust model parameters according to seasonal characteristics, resulting in insufficient accuracy and stability in short-term load forecasting.

Method used

The variational mode decomposition algorithm is used to decompose the load data into three modal components: random fluctuation, intraday cycle and seasonal trend. The hyperparameters of the long short-term memory network and random forest are optimized by combining the particle swarm optimization algorithm to construct seasonally differentiated sub-models. The short-term load forecast results are obtained by weighted ensemble.

Benefits of technology

It achieves a cross-seasonal forecast determination coefficient of over 0.95 and an average absolute percentage error of less than 2%, maintaining stable performance in different seasonal scenarios, improving forecast accuracy and robustness, and demonstrating strong adaptability, enabling it to accurately capture load fluctuation patterns.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to an electric power spot market-oriented cross-seasonal short-term load prediction method and system. The method comprises the following steps: firstly, collecting power load, weather and time feature data to construct a multi-dimensional input feature set, and dividing a training set and a verification set; original load data is decomposed into three types of modal components through variational modal decomposition, and three types of input feature sets are correspondingly split; then, with maximization of a validation set decision coefficient as a target, seasonal differentiation hyper-parameter optimization is carried out through a particle swarm optimization algorithm, and a time sequence dependence prediction sub-model and a multi-feature association prediction sub-model are constructed based on the optimized long and short-term memory network and the random forest respectively; respectively inputting the three types of input feature sets into corresponding sub-models, and outputting two types of prediction results; and finally, performing weighted integration on the two types of results to obtain a short-term load prediction result. Compared with the prior art, the method has the advantage that the prediction precision is remarkably improved.
Need to check novelty before this filing date? Find Prior Art