Article Preview
Top1. Introduction
In recent years, the electric power industry has become one of the core industries in China, and its position is increasingly stable. The most basic requirement of qualified power system is to provide users with more safe, reliable, high-quality and economical power service. With the increasing scale and complexity of the power system, the accuracy of short-term load prediction of the power system directly affects the normal operation of the power system, which plays a key role in effectively reducing generation costs and implementing optimal control of the power system in various regions.
Realizing short-term forecasting of power system workload is the basis for system planning and decision-making, and it is also the key work for grid to achieve safety and economic indicators. Z Lan (2013) proposed thata complex nonlinear relationship between load demand and various influencing load factors. Select and process the factors affecting the load, and fit the intrinsic relationship between the load and various selected factors. Bisoi R, Dash P K and Das P P (2018) proposed that load forecasting work is not only the nature of work, but its implementation is very difficult. For better accuracy of short-term load forecasting, a GRU-CNN hybrid neural network model which combines the gated recurrent unit (GRU) and convolutional neural networks (CNN) was proposed by Wu L (2020). The proposed model was tested in a real-world experiment, and it can make more fully use data and achieve more accurate short-term load forecasting.
Selecting similar days plays an important role in improving the accuracy of load forecasting. The biggest feature of short-term load forecasting is that it has periodicity. When the external factors of the two different days have similar effects on the short-term load, the changes of the load curve for the two days are similar. Dash S K and Patel D (2016) analyzed the effect of objective factors such as time, temperature and daily type on load size reasonably based on the principle of “near large and far small” and “periodicity” of load. The prediction model of the load correction algorithm based on the similar day theory is constructed and improved. Smolen J, Landewé R B, and Mease P (2013) proposed the linear extrapolation method can obtain the trend of future load changes based on the characteristics of the load cycle change. The similar day method is combined with the linear extrapolation method to form a short-term load forecasting algorithm based on the similar linear extrapolation theory. This algorithm can estimate the future load change trend, but the accuracy of load forecast is easily affected by the drastic changes of external factors.
Selakov A, Cvijetinović D, and Milović L (2014) a hybrid model of short-term power load forecasting based on particle swarm optimization (PSO) and support vector machine (SVM). The effect of predicting the load over a significant period of temperature change is better. Tharwat A, Hassanien A E, Elnaghi B E (2017) and Liu Y, Wen K, Gao Q, et al (2018) proposed that Support vector machine is a machine learning algorithm based on structural risk, which has better generalization performance and prediction accuracy. Compared with support vector machines, fuzzy support vector machines is proposed by Zarbakhsh P, Demirel H (2017) and Lakshmi G, Panicker J R, Meera M (2017) that can use different learning intensities for different training samples. Therefore, the fuzzy support vector machine is used for load forecasting, which is more in line with the reality.