A Load Aggregate Grouping Prediction Method Based on Gated Recurrent Unit Networks

Abstract

The present invention relates to a load aggregate grouping prediction method based on a gated cyclic unit network, which includes the following steps: using an adaptive distributed spectral clustering algorithm to cluster user load data, thereby obtaining multiple users with similar load characteristics Electric groups, and obtain the load characteristic matrix of each group; build three GRU networks, and train the three GRU networks by extracting the timing characteristics of the groups, and obtain the prediction models of the three GRU networks, and use the random forest algorithm to analyze the three GRU networks. The network performs model fusion to obtain the load forecasting model of each group; the characteristics of the time to be predicted are input into the load forecasting model, and the load prediction value of each group is obtained separately, and the prediction values ​​of different groups are summed to obtain the prediction of the final load aggregate value; the present invention introduces group forecasting, deep neural network, and model fusion methods, so that user load characteristics and changing rules can be grasped, and the forecasting accuracy is high and the applicability is strong.

Description

technical field [0001] The invention belongs to the technical field of electric power system load forecasting, and in particular relates to a grouping forecasting method of load aggregates based on a gated cycle unit network. Background technique [0002] Accurate and fast load forecasting plays an important role in the safe and economical operation of the power system. Traditional load forecasting is divided into levels based on the physical structure of power system measurement, such as system level, bus level, substation level, microgrid level, etc. Usually, the load forecasting method developed for a specific level cannot be applied to other levels. In recent years, with the popularity of smart meters, power companies have been able to obtain massive amounts of fine-grained user load data. Based on smart meter data, it can get rid of the limitation of power system measurement structure, and can flexibly divide load aggregates of different scales according to demand and ...

AI Technical Summary

Problems solved by technology

However, LSTM has the disadvantage of long training time, and because the load aggregate contains multiple load characteristics, different neural network structures are applicable to different load characteristics.

Images