Safe and economic operation of power system is based on load forecasting, and how to increase
forecasting accuracy is the premise of power dispatching and economic analysis. Present paper
establishes SVM (support vector machine) forecasting model based on fast K-medoids clustering
algorithm and data accumulative pre-processing. FKM (fast K-medoids clustering algorithm) is
applied to extract similar days by dividing all samples into k clusters, and respective forecasting of k
clusters can realize the forecasting of a whole object. Before inputting the data into SVM system, the
original data is preprocessed by accumulation to weaken the irregularity disturbance and strengthen
sequence regularity. Due to existing unexplained component in forecasting error, GARCH
(generalized autoregressive conditional heteroskedasticity) model is employed to forecast the error
with non-white noise. According to its results, error correction is applied to the forecasted daily peak
load. The forecasting effect of the proposed model is compared with other models in the given
example, which verifies that SVM model based on fast K-medoids clustering algorithm and GARCH
model has the characteristic of effectiveness, superiority and universality. The accuracy of daily peak
load forecasting is enhanced significantly
Relation:
Journal of Applied Science and Engineering 19(3), pp.249-258
