Class picture conversion method in power load prediction

Abstract

The invention discloses a class picture conversion method in power load prediction. A new method for converting time series data into data of a class color picture structure is provided. The converteddata is inputted into an improved convolutional neural network using a special-shaped convolution kernel according to a certain logic, and the capacity of extracting local and overall implicit feature rules in a time * feature matrix is enhanced. In an actual power short-term load prediction experiment, the training time is shortened, and the prediction precision is improved in comparison with amainstream method.

Description

technical field [0001] The invention belongs to the technical field of neural network and picture recognition, relates to a method for converting similar pictures, in particular to a method for converting similar pictures in power load forecasting. Background technique [0002] Short-term power load forecasting mainly refers to forecasting the power load in the next few hours, one day to several days. Power load forecasting is an important part of the energy management system. Short-term load forecasting not only provides guarantee for the safe and economical operation of the power system, but also provides a Under the basis of scheduling scheduling plan, power supply plan, and transaction plan (document [1]). At present, the application of artificial neural network in the field of load forecasting has been relatively mature, such as the common BP neural network (backpropagation, BP), recurrent neural network (Recurrent Neural Network, RNN), long short-term memory network (L...

AI Technical Summary

Problems solved by technology

[0005] (2) In the actual application of the model, although the common fully connected neural network has a certain predictive effect, it has no memory ability at all and cannot learn dynamic time-series behavior. Although RNN and LSTM have memory ability, this kind of memory will change with The time axis is elongated and weakened, and the ability to learn the combination of different features at different times is weak. The representation learning ability of the convolutional neural network has its research value, but it is necessary to consider how to convolve the feature time series

Images