A Multi-Source Time Series Forecasting Framework Based on Informer for Climate Finance and Industrial Economy

As the impact of climate change on financial and industrial systems intensifies, building a high-precision and strong generalization prediction model has become an important research direction for intelligent economic management. This paper proposes a multi-step time series prediction method based on the Informer model, which integrates multi-source heterogeneous data such as meteorology, finance and industry to achieve medium- and short-term predictions of key variables such as carbon prices and industrial output value. The constructed model retains the advantages of the Transformer structure and introduces the ProbSparse Attention mechanism and the time embedding module, which significantly improves the efficiency of long sequence modeling and the ability to identify nonlinear relationships. Experimental results show that Informer is superior to traditional machine learning and deep learning models in terms of prediction accuracy, stability and multi-step fitting ability, and is suitable for multi-variable and multi-scale time series modeling needs in complex systems. This study provides theoretical support and algorithmic basis for the application of intelligent prediction models in scenarios such as green finance, energy scheduling and industrial early warning.