Electric load forecasting method based on topology and temporal convolutional network

Abstract

The invention provides a power load prediction method based on topology structure and time convolution network, which is used for predicting the electric power consumption value of various electric appliances in the power system to realize the prediction of electric power load. The electrical appliance power data within a certain period of time is preprocessed; the preprocessed electrical appliance power data is subjected to topology processing based on a predetermined electrical appliance topology processing method to obtain a plurality of power topology data sorted in time series as a series of time series data; The pre-trained power load prediction network processes the time series data to obtain the spectral space prediction vector output by the power load prediction network; the spectral space prediction vector is restored into a graph structure using the inverse Fourier transform to obtain the predicted power consumption values ​​of various electrical appliances. Therefore, the correlation between the power consumption of electrical appliances can be used to effectively improve the prediction accuracy of electrical appliance power during prediction.

Description

technical field [0001] The invention belongs to the technical field of power operation and maintenance, relates to a power load forecasting method, and in particular relates to a power load forecasting method based on a topology structure and a time convolutional network. Background technique [0002] Power load forecasting is to predict its future value based on the past and present of the power load. The power load forecasting can infer the development trend and possible status of the load, and improve economic and social benefits. [0003] The current general direction of power load forecasting methods is divided into traditional forecasting methods and modern forecasting methods. Modern forecasting methods mainly include the following: forecasting methods based on CNN models, methods using LSTM models combined with time series to forecast power system loads, and using Neural network methods such as direct convolution using multi-dimensional data such as electricity data,...

AI Technical Summary

Problems solved by technology

[0004] However, there are still some problems in the current modern prediction method. First, the current method does not analyze the potential correlation between various electrical appliances, such as using a dishwasher after using a microwave oven, and using an air conditioner while using a computer; Second, when using the CNN convolutional neural network, the time series cannot be reflected, that is, the time information of the data cannot be used. When using the LSTM model, the network structure will choose to forget some information that the network structure considers unimportant, resulting in the possibility of missing potentially important information; Third, the deep learning models used in the current method are difficult to perform convolution on large time series, and the receptive field is small, so it is difficult to obtain high-accuracy results for model training that requires long-term dependence

Images