Modeling and Multi-Objective Optimal Scheduling of Park Source, Grid, Load, Storage Integrated System Based on Digital Twin

With the construction and advancement of new types of power systems, the source-grid-load-storage integrated system has evolved into a complex, large-scale system with intricate structures, numerous devices, and multi-source data interactions. Traditional simulation methods are challenging to meet the operational needs of such systems. To address this issue, this paper utilizes digital twin technology to model and optimize the source-grid-load-storage integrated system of a specific industrial park, based on the CloudPSS platform. First, establish a digital twin model of the system to simulate the working state of the physical entity in a panoramic way. Secondly, a long short-term memory (LSTM) neural network power prediction model based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and kernel principal component analysis (KPCA) is proposed. The neural network power prediction model can accurately predict the power of photovoltaic and wind power based on meteorological forecast data. Furthermore, an improved multiobjective particle swarm optimization (MOPSO) algorithm is designed to solve the system optimization scheduling model. Simulation experiments demonstrate that the proposed prediction model enhances the accuracy of photovoltaic and wind power forecasting by $36.3 \%$ and $22.5 \%$, respectively. Meanwhile, the optimized scheduling strategy reduces system operation costs and environmental costs by $6.3 \%$ and $29.92 \%$, respectively. Through digital twin technology, this research successfully establishes a closed-loop optimization mechanism encompassing ‘sensing-prediction-optimization-feedback’, realizing dynamic coupling between virtual and physical systems. This provides a feasible technical framework for the collaborative optimization of source, grid, load, and storage.