Short Term Power Load Combination Forecasting Method Based on Feature Extraction

Abstract

Accurate and fast forecasting of short-term load is conducive to the safe and stable operation of the power system, and a short-term power load combination forecasting method based on feature extraction is proposed. Firefly Sparrow Algorithm (FSA) is applied to find the optimal combination of influencing parameters in Variational Mode Decomposition (VMD) to obtain the signal components with the best effect. Since the signal components contain different influence characteristics and timing information, the Maximum Information Coefficient (MIC) is used to screen the features of each signal component, establish the feature matrix, and use the over-zero rate as an index to determine the high and low frequency signal demarcation points. Based on the different characteristics of high and low frequency signals, the Informer model is used to forecast the high frequency signal components, and the LSTM is used to forecast the low frequency signal components. All the forecasting results are reconstructed to obtain the final forecasting value. Taking the Spanish power load data as an example, considering the actual seasonal factors, and experimentally comparing with other forecasting models, the results show that after the feature screening, the errors are significantly reduced, and the decidability coefficient is significantly improved, which verifies the accuracy and universality of the model proposed in this paper.

Keywords

• Power load forecasting
• feature selection
• hybrid model
• variational mode decomposition
• informer