| Summary: | Anticipating changes in system inertia is crucial for maintaining the stability and reliability of new power systems. While machine learning prediction models can be effective in this regard, many of these models are “black-box” models that lack interpretability, making it difficult to understand the factors that contribute to changes in system inertia. In order to address this issue, we propose a short-term prediction method for power system inertia using eXtreme Gradient Boosting (XGBoost) and examine its interpretability. Our method involves using XGBoost to create a prediction model based on operational data from the power system as input features. The XGBoost explanation mechanism utilizes the SHAP attribution method to decompose prediction results into various dimensions, allowing us to examine the influence of various features on the predicted inertia value. By utilizing this method, we are able to gain a better understanding of how the prediction was made and how each feature contributed to the final result. We conduct simulations on a realistic photovoltaic system and find that the proposed method is effective in explaining the influence of input features on short-term predicted inertia values and providing clear explanations for the model’s prediction results on individual samples. By accurately predicting changes in system inertia and understanding the factors that contribute to these changes, we can take proactive measures to prevent potential weaknesses in the system and ensure its overall stability and reliability.
|
|---|