Construction of a smart grid load forecasting platform based on clustering algorithm

Abstract

Abstract With the rapid development of economic levels, the demand for electricity usage in various industrial sectors has continued to rise. In order to adequately meet the growth in demand for electricity usage at each user end within the power system, accurate forecasting of electricity load is required. This paper uses the K-Means algorithm as the basis, combined with the Canopy algorithm, whose simple and efficient advantages can be used as the initial clustering step for demanding clustering algorithms. The PCA algorithm is used to reduce the dimensionality of the residential multidimensional electricity consumption feature set, and after optimization, the feature vector weights are calculated to obtain the appropriate selection to determine the training and testing samples. The improved K-Means algorithm carries out five clustering tests, and its average accuracy is above 97%. The average relative error is then used as an indicator to compare the algorithm with the time series method, curve extrapolation method, and WKNR method for analysis. The average relative error of the smart grid load forecasting method proposed in this paper is 0.87%, while the average relative errors of the other three algorithms are 1.64%, 1.57%, and 0.93%, respectively. The distribution of relative errors falls less than 1% more than the other methods. It can be seen that the improved K-Means algorithm has higher prediction accuracy and also makes the smart grid load testing platform more practical due to its simple implementation principle.