With the continuous expansion of water resource allocation projects, accurate electricity consumption forecasting is crucial for energy conservation, cost control, and construction efficiency. Traditional forecasting methods, such as long short-term memory (LSTM) networks and Transformers, often struggle to capture both short-term and long-term dependencies in complex time-series data. To address this challenge, this paper proposes an xLSTM (extended long Short-term memory) model for multi-regional power consumption forecasting. The xLSTM model combines the short-term dependency modeling capability of sLSTM with the long-term dependency learning capacity of mLSTM, enabling effective analysis of power consumption data across multiple regions while considering temporal correlations among regions. Experimental results show that xLSTM achieve superior predictive performance, with a mean square error (MSE) of 0.0030 and a mean absolute error (MAE) of 0.035, outperforming competing models. The proposed model provides effective technical support for precise electricity demand forecasting and offers practical value for decision-making and intelligent scheduling management in large-scale water resource allocation projects.

The State Grid has continuously improved its material procurement management level and refined its online procurement processes. However, inaccurate estimation of procurement plans, has led suppliers to engage in price games using the general bidding and tendering mechanism during the bidding process. This has resulted in increased procurement costs of the power grid company. Therefore, it is of great significance to establish an accurate and effective electricity material demand forecasting model. In respose to the instability, volatility and intermittency of power material sequences, this paper proposes a forecasting method for power material demand based on parameter-optimized variational mode decomposition (VMD) and long short-term memory neural network (LSTM). Typical power materials from the State Grid e-commerce zone platform were selected. VMD, optimized by using the whale optimization algorithm(WOA) parameters, was adopted to perform modal decomposition on the original sequence. LSTM models were then constructed for each modal component obtained from the decomposition. Finally, the predicted values of each mode were superimposed and reconstructed into the predicted value of power materials. Experimental results show that the proposed method achieves higher prediction accuracy compared to LSTM, EMD-LSTM, VMD-LSTM, PSO-VMD-LSTM and SSA-VMD-LSTM. This approach holds practical significance for the forecast of power grid material purchase.