A support vector machine (SVM) forecasting model based on rough set (RS) data preprocess was proposed by combining the rough set attribute reduction and the support vector machine regression algorithm, because there a...A support vector machine (SVM) forecasting model based on rough set (RS) data preprocess was proposed by combining the rough set attribute reduction and the support vector machine regression algorithm, because there are strong complementarities between two models. Firstly, the rough set was used to reduce the condition attributes, then to eliminate the attributes that were redundant for the forecast, Secondly, it adopted the minimum condition attributes obtained by reduction and the corresponding original data to re-form a new training sample, which only kept the important attributes affecting the forecast accuracy. Finally, it studied and trained the SVM with the training samples after reduction, inputted the test samples re-formed by the minimum condition attributes and the corresponding original data, and then got the mapping relationship model between condition attributes and forecast variables after testing it. This model was used to forecast the power supply and demand. The results show that the average absolute error rate of power consumption of the whole society and yearly maximum load are 14.21% and 13.23%, respectively, which indicates that the RS-SVM forecast model has a higher degree of accuracy.展开更多
基金Project(70901025) supported by the National Natural Science Foundation of China
文摘A support vector machine (SVM) forecasting model based on rough set (RS) data preprocess was proposed by combining the rough set attribute reduction and the support vector machine regression algorithm, because there are strong complementarities between two models. Firstly, the rough set was used to reduce the condition attributes, then to eliminate the attributes that were redundant for the forecast, Secondly, it adopted the minimum condition attributes obtained by reduction and the corresponding original data to re-form a new training sample, which only kept the important attributes affecting the forecast accuracy. Finally, it studied and trained the SVM with the training samples after reduction, inputted the test samples re-formed by the minimum condition attributes and the corresponding original data, and then got the mapping relationship model between condition attributes and forecast variables after testing it. This model was used to forecast the power supply and demand. The results show that the average absolute error rate of power consumption of the whole society and yearly maximum load are 14.21% and 13.23%, respectively, which indicates that the RS-SVM forecast model has a higher degree of accuracy.
