Search

This study proposes a short-term load prediction method of a bidirectional long short-term memory network based on feature mining of the power consumption big data in combination with the attention mechanism (AT) of Bayesian optimization to address the problems that a considerable amount of feature factors exist and the feature relationship is obscured in the historical power consumption big data. The method comprehensively considers the global features of the power consumption data in space and the local features in time. First, the Cen-CK-means clustering method is used to cluster the electricity consumption data of users, and the statistical, combination, and time category characteristics are extracted according to the meteorological factors related to load over multiple time scales. Second, the Bayesian and bidirectional long and short memory networks are combined to extract the temporal and spatial characteristics of the load data itself. Meanwhile, the AT is introduced to automatically assign the corresponding weights to the hidden layer state of the bidirectional long and short memory. This task is carried out to distinguish the importance of the different time load series, which can effectively reduce the loss of historical information and highlight information about key historical time points. Finally, taking the first type of load as an example, compared with the SVP, RBPNN, BiLSTM, and BO-BiLSTM algorithms, the MAPE index is reduced by 1.05%, 1.75%, 0.52%, and 0.26%, respectively. RMSE decreased by 186.61, 154.93, 91.88, and 15.76 MW, respectively, while R2 increased by 0.04, 0.07, 0.03, and 0.03, respectively. In the one-week forecast time, MAPE index decreased by 1.97%, 2.44%, 1.21%, and 0.6%, respectively; RMSE decreased by 271.18, 305.7, 183.13, and 97.91 MW, respectively; and R2 increased by 0.12, 0.08, 0.04, and 0.03, respectively. [ABSTRACT FROM AUTHOR]

To address the challenges associated with the intricate selection of decomposition technical parameters and the adverse impact of high-frequency non-stationary components on prediction accuracy, a differential evolution (DE) and sparrow search algorithm (SSA) parameter optimization model is formulated. Through the optimization of variational mode decomposition (VMD) and bidirectional short and long-term memory (BiLSTM) neural networks, enhancements are made to the overall performance of the prediction method. Subsequently, DESSA–VMD is employed to quadratically decompose the high-frequency unstable component, mitigating the detrimental effects of randomness and volatility on the prediction accuracy of the primary component. Following this, the DESSA–BiLSTM neural network is applied to predict the high-frequency strong non-stationary component, whereas the autoregressive integrated moving average is utilized for predicting the low-frequency periodic component. Finally, a case analysis is conducted using actual data comprising 2000 sets of wind power in a province. When the test set consisted of 600 groups, the root mean square error index value of the proposed prediction model exhibited reductions of 1.95 MW, 2.86 MW, and 1.16 MW, respectively, compared with the BP, SVM, and LSTM methods. Additionally, the mean absolute percentage error index showed decreases of 2.26 %, 2.96 %, and 1.73 %, respectively. [ABSTRACT FROM AUTHOR]