Short-term electricity price prediction method based on long-term and short-term memory network

Abstract

The invention discloses a short-term electricity price prediction method based on a long-term and short-term memory network. The electricity price prediction method comprises the following steps of carrying out periodic preprocessing on abnormal data and missing data in an original electricity price; constructing a long-term and short-term memory network model; performing long-term and short-termmemory network model training, selecting model parameters, and building a short-term electricity price prediction model; and carrying out short-term electricity price prediction, and carrying out reverse normalization processing on prediction result data to obtain an actual electricity price prediction value. According to the electricity price prediction method, the historical electricity price data is preprocessed through mean interpolation and wavelet transform, so that the noise of a training set is reduced; based on a long-term and short-term memory network prediction model, short-term electricity price prediction is carried out by taking historical influence electricity price factor data as a test set, and reasonable transaction decision making is carried out on market participants, so that the significance of reducing the cost, avoiding the risk and improving the profit rate is great.

Description

technical field [0001] The invention relates to the technical field of electric power market, in particular to a short-term electricity price prediction method based on long-short-term memory network. Background technique [0002] Since the reform of the power system, the domestic power market has undergone major changes, and various market players have faced a new and fierce market competition environment in the process of power trading. As an effective tool for market participants to participate in electricity market transactions, electricity price forecasting is conducive to market participants to make reasonable trading decisions to reduce costs, avoid risks and increase profit margins. Therefore, how to make accurate day-ahead electricity price forecasts to make optimal trading decisions has become an important research topic. Existing research mainly focuses on the point prediction of electricity prices based on historical electricity price data, using support vector ...

AI Technical Summary

Problems solved by technology

Existing research mainly focuses on the point prediction of electricity prices based on historical electricity price data, using support vector machines and artificial neural network models, which have problems such as difficult to deal with large-scale data, easy to fall into local optimum, and less consideration of internal laws of time series.

The cyclic neural network can better consider the time series law, but after multi-stage propagation, there may be a problem that the gradient tends to disappear or explode

There is a lack of a short-term electricity price prediction method that can process time series, overcome long-term dependence on data, and provide market participants with optimal trading decisions

Images