Short-term Load Forecasting Of Power System Considering Real-Time Weather Factors

Short-term load forecasting is a key of power scheduling and market transactions. Its predictions accuracy is relating to the power system security, reliability, economy. So how to improve the accuracy of short-term load forecasting is a hot issue concerned by many scholars. In recent years, with air conditioning, heating and other weather sensitive loads increasing proportion in total electricity load, the key to improve the accuracy of short-term load forecasting is to build high precision load forecasting model considering the influence of meteorological factors on power load.In this paper, the characteristics of the power load were analyzed. According to the nature of the power load, the loads are classified into different type. The daily load characteristics, load characteristics week, month load characteristics and other indicators are selected to analyze the load characteristics. The load is decomposed into four parts of typical components, weather-sensitive components, special events component, random components. The analysis showed that the weather-sensitive component is influenced by temperature and humidity.BP neural network has good robustness for nonlinear, multivariate, non-continuous complex issues, so it is suitable for short-term load forecasting. Detailed analyses of the basic principles of BP neural network are-discussed. The momentum-adaptive learning rate adjustment algorithm is proposed to quickly find the global optimal solutions considering the shortage of tradition BP neural network.24 points data of daily load were used as input vector to design BP neural network. The forecasting results of weekdays and rest days show that BP neural network achieved satisfactory results in the short term load forecasting.Propose a weather forecast temperature correction algorithm based the gray system theory when considering forecasting weather temperature often has some errors. The cumulative effect of temperature will have an impact load, so a correction for the current period temperature is made considering the impact of cumulative effect of temperature in prior periods and the number of days before. With correction for temperature, humidity, the multiple linear regression prediction model considering real-time weather factors is established. The practical examples show that the model can effectively improve the prediction accuracy of short-term load forecasting.