Short-term forecasting of building heating load based on MVMD-SSA-LSTM

Abstract

A short-term heating load forecast for buildings is a critical step in the subsequent control of energy systems, directly impacting system energy consumption. However, given that heating load and its influencing factors constitute volatile time series data, noise interference within the data significantly limits prediction accuracy and stability. To address this issue, this paper proposes a novel MVMD-SSA-LSTM model for building heating load forecasts, which integrates Multivariate Variational Mode Decomposition (MVMD), Sparrow Search Algorithm (SSA), and Long Short-Term Memory (LSTM) neural networks. Initially, a correlation analysis of the factors influencing building heating load is conducted to identify the key determinants. Subsequently, MVMD is employed to decompose the multidimensional dataset into several modes. A correlation analysis is then performed on these decomposed modes to extract supplementary features, which are combined with the original data to form a new dataset, thereby reducing feature redundancy. Finally, an LSTM neural network is utilized as the core predictive model, with the SSA algorithm optimizing three critical parameters: The maximum training iterations, the number of hidden units, and the initial learning rate. The predicted outputs of each heating load mode are aggregated to obtain the final forecast. Results demonstrate that the MVMD-SSA-LSTM model effectively mitigates the uncertainty in heating load sequence forecasts, overcoming noise disturbances and exhibiting superior performance compared to other commonly used models, with significantly higher accuracy and stability.