Juanjuan Yang ^*

• Suzhou Industrial Park Institute of Services Outsourcing, Suzhou, China

In the context of energy resource scarcity and environmental pressures, accurately forecasting energy consumption and optimizing financial strategies in smart grids are crucial. However, practical applications face challenges due to the high dimensionality and dynamic nature of the data, hindering accurate prediction and strategy optimization. This paper aims to proposed a fusion algorithm for smart grid enterprise decision-making and economic benefit analysis, enhancing the accuracy of decision-making and predictive capability of economic benefits. The proposed method combines techniques such as deep reinforcement learning (DRL), long shortterm memory (LSTM) networks, and the Transformer algorithm. LSTM is used to process and analyze time series data, capturing historical patterns of energy prices and usage. DRL and the Transformer algorithm are then employed to further analyze the data, enabling the formulation and optimization of energy purchasing and usage strategies. Experimental results demonstrate that this approach outperforms traditional methods in terms of improving energy cost prediction accuracy and optimizing financial strategies. Notably, on the EIA Dataset, the proposed algorithm achieves a reduction of over 48.5% in FLOP, and a decrease in inference time by over 49.8%, and an improvement of 38.6% in MAPE. This research provides a new perspective and tool for energy management in smart grids and offers valuable insights for handling other high-dimensional and dynamically changing data processing and decision optimization problems.