Electric quantity prediction system and prediction method thereof

Abstract

The invention relates to an electric quantity prediction system and a prediction method thereof, and belongs to the field of electric quantity prediction. Because the electric quantity measurement model is improved and the quantile regression model is introduced, the valuable electric quantity reference interval can be predicted more accurately, the interference of data noise is effectively reduced, the characteristics of the electric quantity are further extracted from time and space by combining the advantages of the long and short-term memory network and the convolutional neural network, and the attention mechanism is used to update the weight and optimize the feature extraction effect, and the quantile regression model is used to realize probability prediction, so that a credible electric quantity prediction reference range can be provided for researchers on the basis of ensuring the accuracy of point prediction.

Description

technical field [0001] The invention relates to the field of electricity forecasting, in particular to an electricity forecasting system and a forecasting method thereof. Background technique [0002] Power companies usually count the line loss rate to reflect their own management level and economic benefits. However, due to the connection of distributed energy sources to the grid and the phenomenon of different periods of power supply and sales, the statistics of the line loss rate are more difficult. Therefore, the realization of short-term power forecasting is crucial to reduce the statistical error of line loss. [0003] With the development of mathematical theory and modern technology, further improving the accuracy of short-term power forecasting means considerable economic benefits to society, and these potential benefits promote the continuous improvement of power forecasting models. Among them, regional power forecasting can reduce the statistical error of line los...

AI Technical Summary

Problems solved by technology

[0004] Some of the existing related technologies use statistical methods, such as ARIMA (Autoregressive Integrated Moving Average model, differential integrated moving average autoregressive model), this method realizes the prediction of future electricity according to the historical change law of electricity, but it is to find from itself law and cannot be combined with other features for analysis, and at the same time, the fitting effect for nonlinear data is not good

In addition, more neural network methods are used, for example, there is a method of using LSTM (Long Short-Term Memory, long-term short-term memory network) to predict power, but it does not consider the correlation in the feature space, resulting in the capture of features that affect power insufficient

There is also a method of using CNN (Convolutional Neural Networks, Convolutional Neural Networks) to predict power, but it does not take into account the temporal context, resulting in the inability to extract the periodic influence of time

In power forecasting, in addition to using a suitable model, building accurate data features can also effectively improve the prediction accuracy of the model, but the existing feature processing methods cannot find out the influential features better and more comprehensively.

Moreover, most of the existing related technologies are based on point forecasting. However, when there is a huge fluctuation in power, the accuracy of point forecasting often cannot meet the requirements. In order to provide a more reasonable reference value, it is necessary to present the probability interval of possible power generation. However, the existing probabilistic prediction technology is still in its infancy, and the accuracy of the prediction interval is insufficient. Too large or wrong intervals will provide negative reference value

Images