In the face of the severe situation that carbon emissions account for nearly half of the total carbon emissions in China’s construction sector, traditional energy-saving design methods are facing bottlenecks because they are difficult to cope with dynamic environments and complex energy scenarios. Based on artificial intelligence technology, a hybrid prediction model combining Long Short-Term Memory Network (LSTM) and Particle Swarm Optimization (PSO) algorithm is constructed in this paper. By capturing the time series characteristics of energy consumption and the ability of global optimization, the high-precision prediction of building energy consumption is realized. Experiments show that the mean square error (MSE), the mean absolute error (MAE) and the coefficient of determination (R2) of the LSTM-PSO model reach 11.29 kWh2, 2.15 kWh and 0.948, respectively, which are significantly improved compared with the traditional model. Based on the prediction results, a dual-path energy-saving strategy of “passive priority and active optimization” is proposed, and an energy-saving economic analysis model covering the whole life cycle is constructed by combining the dynamic optimization of thermal performance of envelope with the hierarchical regulation of active system, which provides a data-driven scientific paradigm for building environment design under the goal of carbon neutrality.