Applications Of Partial Least Squares Regression Analysis In Short-term Load Forecasting

Short term load forecasting ,which is a prerequisite for network planning,.has an important significance for system stability , economy and the system's control and optimization.Existing short-term load forecasting methods have traditional forecasting methods and modern forecasting methods, the trend of short-term load forecasting is the upgrading of existing methods, optimizing and weighting average of several synthesis's predicted results. In this paper,a new approach called partial least squares regression algorithm , which bases statistical and mathematical theory , will finish power system short term load forecasting. Partial least squares regression analysis is a kind of multivariate statistical data analysis methods ,which is a novel in recent years,it can be considered that it is the combination of multiple linear regression analysis, principal component analysis, and canonical correlation analysis,it can facilitately realize data preparation and pretreatment, while the number of independent variables is small and the independent variables have a strong multiple correlation, it can aiso finish flexible model, it extracts the input values, and the extracted part have the fine features of linearly independent. In the time of extraction of principal components , not only to consider arguments information but also to consider the dependent variable information, so that the extracted principal component can reflect the independent variables and the dependent variable information,at last we can achieve good results in the regression and prediction. As partial least squares regression achieves a data simplification when model is finished, it can observe the characteristics of the multidimensional in two-dimensional data, which makes partial least squares regression analysis powerful in graphics capabilities and conducive for engineering staff to have an analysis applications.The level of short-term load has many external factors, in addition to historical load data, they also include many weather factors, such as temperature, wind speed and rainfall, which will have an important impact on the load level.In this paper, the values of all these factors is called eigenvalue, firstly establish regression equation between characteristic quantity and the dependent variable, and then make a short-term load forecasting. The example shows partial least squares regression analysis can complete load forecasting of high precision and reliable results.There are 2 electric load peaks: early, late, and a day minimum load value. In this paper, " the weather factors," and " sunrise time " and "sunset time " are given ,through partial least squares regression analysis ,we can regress the"daily maximum load value","daily minimum load values", and the moment they appear,then use these factors, regress "daily average load values.".A total of 5 results are predicted by these factors at last. Through the final computation of"automatic filter out important factors, and synthesis new factors "of partial least squares regression, we find that the impact of various factors on the load is not the same, partial least squares regression analysis gives a kind of method, which analyses the relationship between feature extraction and load ,at last establishes a reasonable model of forecasting.