Source-load joint prediction method based on iTransformer-LSTM for port microgrid

Abstract

The green transformation of ports has led to the construction of a high proportion of renewable energy port microgrids becoming a key path for port decarbonization. High precision prediction of ship shore power load and high endowment renewable energy is an important prerequisite for achieving low-carbon economic operation of port microgrids. Therefore, this paper considers the “source tracing load” coupling characteristics of port microgrids and proposes a source-load joint prediction method based on inverted Transformer (iTransformer) and long short-term memory (LSTM) for port microgrids. Firstly, grey relationship analysis is used to analyse the coupling characteristics between sources and loads in port microgrids, as well as their correlation with influencing factors, in order to achieve feature dimensionality reduction. Secondly, using iTransformer for important features differentiation extraction, and utilizing self-attention mechanism to effectively capture multidimensional features dependency relationships. Finally, LSTM is used for nonlinear dynamic modeling of time series, and the weights of different prediction tasks are adaptively balanced through a multidimensional output gating mechanism to output the source-load joint prediction results. Verified by the actual operational data of Hukou Port Area in Jiujiang, Jiangxi Province, the results show that the proposed source-load joint prediction model can reduce the prediction errors of various prediction tasks and has higher prediction accuracy compared to existing mainstream models.