An Advanced Load Simulation Methodology Incorporating Dynamic Correlation Analysis for Diverse Electricity Users

Abstract

Real-time load simulation plays a pivotal role in power system dispatching and safety assessments. However, the heterogeneous composition of load groups complicates the trends of load changes. Traditional simulation methods often fall short in capturing the intricate patterns of load variations and fluctuations, as well as the interconnections among different load groups. To address these shortcomings, this paper enhances the granularity of load simulation by mapping the relationships between typical load processes and individual user loads. We introduce a clustering neural network specifically designed for the typical load processes of multiple users. This network utilizes a fluctuation attention mechanism and a deep embedding clustering algorithm to identify diverse typical fluctuation processes across different load types. Furthermore, we propose a causality analysis method for various load groups and processes, using a convergent cross mapping algorithm to detect potential causal links among different load users and their processes. Additionally, we establish a multi-task learning-based neural network model for simulating multiple loads, enabling parallel, high-precision simulations of typical multi-user load processes. The effectiveness of our proposed methods is validated using electricity data from a city in North China, demonstrating their capability to accurately capture typical output characteristics and the interrelations among different users, thereby enhancing the accuracy of load simulations.