This paper presents a novel approach to short-time load forecasting by the application of nonparametric regression. The method is derived from a load model in the form of a probability density function of load and load affecting factors. A load forecast is a conditional expectation of load given the time, weather conditions and other explanatory variables. This forecast can be calculated directly from historical data as a local average of observed past loads with the size of the local neighborhood and the specific weights on the loads defined by a multivariate product kernel. The method accuracy relies on the adequiite representation of possible future conditions by historical data, but a measure to detect any unreliable forecast can be easily constructed. The proposed procedure requires few parameters that can be: easily calculated from historical data by applying the cross-validation technique.
Due to the deregulation of the power system, the electric power industry is undergoing a transformation in terms of its planning and operation strategies. Because of the importance in reducing financial and operational risk, improving load forecasting accuracy is paramount. In some load forecasting applications, K-means clustering is used to group customers prior to forecasting. This method has been shown to improve the accuracy of load predictions. However, there are situations where K-means clustering reduces load forecasting accuracy. This paper studies the factors that affect the performance of K-means clustering. Additionally, several strategies have been proposed to tackle the lower estimation accuracy problems for this clustering algorithm. The data used for validating the proposed strategies associated with the factors is from Consolidated Edison Company of New York, Inc. (ConEdison). The mean absolute percent error (MAPE) and relative mean square error (RMSE) are utilized to evaluate the forecasting results of K-means based least squares support vector machines (LS-SVM) and preprocessed K-means based LS-SVM. Additionally, the outperformance of preprocessed K-means based LS-SVM is demonstrated via the data results.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.