SCADA master station load prediction method based on deep learning

Abstract

The invention discloses an SCADA master station load prediction method based on deep learning. The master station load prediction method mainly comprises the following steps: generating original data;preprocessing the original data, wherein the preprocessing comprises three times of Hermite interpolation calculation and feature extraction and feature normalization one by one, wherein the Hermiteinterpolation calculation comprises box-type graph abnormal value elimination and conformal segmentation; training the sample data obtained through normalization in the preprocessing by adopting a deep learning LSTM network, and obtaining an optimal model parameter through cross validation; and carrying out real-time load prediction by using the optimal model parameters. A self-adaptive nonlinearprocessing technology is adopted, a large amount of historical data is used for training, the short-term load prediction precision is improved, and dispatchers can accurately know the change fluctuation of the future load in real time.

Description

technical field [0001] The invention relates to the technical field of electric power systems, in particular to a load forecasting method for a SCADA master station based on deep learning. Background technique [0002] Traction power supply load forecasting can be divided into long-term, medium-term and short-term load forecasting, and short-term load forecasting has become the basic work of system scheduling analysis and optimization. Accurate load forecasting can not only ensure the stability of traction power supply, reduce power consumption costs, but also improve power supply quality , It also contributes to the safe operation of the power supply system. Load data is not only related to historical data, but also related to other factors, such as weather, temperature, rainfall and so on. The relationship between load data and these factors is highly complex and nonlinear, and the quantitative expression of this relationship has not been established at this stage. The l...

AI Technical Summary

Problems solved by technology

The relationship between load data and these factors is highly complex and nonlinear, and the quantitative expression of this relationship has not been established at this stage

The load distribution of the traction power supply system is becoming more and more complicated and irregular, which makes the load forecasting of the power supply system very difficult

[0003]At present, the load forecasting of traction power supply system mostly adopts linear forecasting methods, such as linear regression, autoregressive moving average model (ARMA), autoregressive integral moving average model (ARIMA) etc. These linear forecasting methods are mostly suitable for medium and long-term load forecasting. For short-term non-stationary and nonlinear forecasting situations, it is difficult to simulate the actual relationship between data, and these methods consider few factors and have low forecasting accuracy, so they are not suitable for short-term load forecasting. Load Forecasting

Traditional neural networks and support vector machines can better solve short-term complex situations. This type of method has strong nonlinear mapping capabilities, but the neural network model is easily trapped in local minima and affected by overfitting during the training process. The support vector The machine is a shallow feature learning model, which is difficult to implement for large-scale training samples, and there is no memory unit in the network of these two methods, and the analysis of the temporal correlation of time series data is lacking.

Images