Analysis of Influencing Factors of Ultra-Short Term Load Forecasting based on Time Series Characteristics

Abstract

Background: With the institutional reform of the power market and the need for demandside response, the requirements for load forecasting accuracy are getting higher and higher. In order to deeply explore the influencing factors of load forecasting accuracy and improve the forecasting accuracy of ultra-short-term load forecasting, a method for analyzing the influencing factors of ultra-short-term load forecasting considering time series characteristics is proposed in this paper. Firstly, based on the analysis of four different types of load characteristics, eight-time series characteristic parameters that can characterize the characteristics of the load curve and may be related to the prediction accuracy of the prediction model are extracted. These characteristic parameters include dispersion coefficient, slope, daily load rate, daily peak-valley difference, deviation, variance, skewness coefficient and kurtosis coefficient. Secondly, three kinds of load forecasting models are established, including the Autoregressive Integrated Moving Average model (ARIMA), grey system and support vector machine (SVM), and then forecast the load in Anchorage, Alaska, USA. The effects of eight time series features on the prediction accuracy of the three load forecasting models are analyzed. The results show that the discrete coefficient, slope difference, daily load rate and peak-valley difference greatly influence the load forecasting results and have different affects on different forecasting models. When the historical data is small, ARIMA model is suitable for shortterm load forecasting with small slope difference, large daily load rate and small daily peak-valley difference. The grey model is suitable for short-term load forecasting with small discrete coefficients of historical data. The SVM model is suitable for most short-term load forecasting when there is a lot of historical data. With the institutional reform of the power market and the need for demand-side response, the requirements for load forecasting accuracy are getting higher and higher Objective: In order to deeply explore the influencing factors of load forecasting accuracy and improve the forecasting accuracy of ultra-short-term load forecasting, this paper proposes a method for analyzing the influencing factors of ultra-short-term load forecasting considering time series characteristics. Methods: Based on the analysis of four different load characteristics, 8 kinds of time series characteristics such as the dispersion coefficient, slope and daily load rate of the daily load curve are extracted. And three kinds of load forecasting models are established, including autoregressive integrated moving average model (ARIMA), grey system and support vector machine (SVM), and then forecast the load in Anchorage, Alaska, USA. The effects of these 8-time series features on the prediction accuracy of the three load forecasting models are analyzed. Conclusion: The results show that when there are few historical data, the ARIMA model is suitable for short-term load forecasting with a small slope difference, large daily load rate and small daily peak-to-valley difference characteristics. The gray system is suitable for short-term load forecasting with a small discrete coefficient of historical data. The SVM model is suitable for most short-term load forecasting with many historical data.