Abstract

Background: Under the "Dual Carbon" strategy, the new-type power system integrates various complex equipment, especially the addition of wind power, photovoltaic power generation, distributed energy storage and other systems, which not only brings clean and efficient energy to the power system, but also leads to potential risks in the power grid.

Objective: Traditional methods for assessing the risks of power systems mainly focus on the characteristics of traditional power grids. With the emergence of new features in the new power system, these traditional assessment methods are no longer effective, which could affect the safe and stable operation of the power system. To address this issue, this article proposes a new vulnerability index system for power systems. It also introduces the analytic hierarchy based on Moody's order graph (MAHP) quantification analysis method, and develops the Tabular data risk assessment methodology(TabRAM) risk assessment model based on an improved Transformer method.

Method: MAHP is a comprehensive evaluation method combining Moody's Chart and Analytic Hierarchy (AHP), which has high comprehensiveness and operability. TabRAM uses Bayesian neural networks (BNN) to complete prior data fitting and model complex feature dependencies and potential causal mechanisms on tabular data. The network is trained using an improved Transformer, and after pre-training, it can approximate new prior probabilistic inference in a forward pass to achieve prediction and classification tasks for new datasets.

Results: After multiple rounds of iterations and parameter adjustments, the model achieved the desired performance and was successfully applied to the risk assessment of the new power system.

Conclusion: According to the experimental results, the proposed TabRAM model demonstrates superior performance in the field of risk assessment for the new-type power system, as compared to traditional deep learning and machine learning algorithms.