Ultra-short-term power load prediction method, equipment, medium and product

Abstract

The invention provides an ultra-short-term power load prediction method, and the method comprises the following steps: acquiring required original business data, and forming features for the original business data through feature engineering of a system; when the predicted time sequence is non-stationary, carrying out differential processing on the predicted data, and replacing the load of the predicted time point with a differential result; according to the formed features, giving a sampling weight according to the length of a span away from a predicted time point, carrying out the random sampling, and merging the data obtained after random sampling to form modeling data; performing deep learning training and a debugging process by taking features obtained by the feature engineering and target variables obtained by dependent variable conversion as input and output data of a long-short term memory model respectively to obtain the model, and performing differential reduction through an output result of the obtained model to complete ultra-short term load prediction. The problem that the model effect cannot be further improved due to excessive dependence on a deep learning algorithm and no data deep processing method is solved.

Description

technical field [0001] The present invention relates to the technical field of ultra-short-term load forecasting, in particular to an ultra-short-term power load forecasting method, equipment, medium, and product. Background technique [0002] In the current ultra-short-term load forecasting method, when the deep learning algorithm is used to realize the forecasting process and the model effect cannot be improved, in addition to missing value processing and outlier processing, less in-depth data sampling and systematic artificial feature engineering are carried out, and there is no Transform the target variable (dependent variable). Due to the powerful feature extraction capabilities of deep learning algorithms and the ineffectiveness of artificial feature engineering in other fields, developers also focus on the training of deep learning algorithms in the field of load forecasting, resulting in some characteristics of the data itself (such as long-period features) are lost...

AI Technical Summary

Problems solved by technology

[0003] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a method for ultra-short-term power load forecasting, which solves the problem of excessive reliance on deep learning algorithms, without combining data sampling methods, features constructed by artificial experience, and data depth such as conversion and extraction of target variables. Handling method, resulting in the problem that the model effect cannot be further improved

Images