基于数据空间可靠域的多智能体互补电力系统暂态稳定评估

Multi-agent Complementary Power System Transient Stability Assessment Based on Data Space Reliability Domain

将人工智能应用于电力系统暂态稳定评估受到研究人员的广泛关注,然而如何评价人工智能模型的结果可靠性是该方法在实际应用中面临的关键问题。为此,该文引入数据空间可靠域的相关概念,提出基于数据空间可靠域的多智能体互补电力系统暂态稳定智能评估方法。首先基于SHAP理论,得到经shapley值加权的特征样本空间,再根据训练集中模型正确预测样本与误判样本在该空间上的分布情况确定该模型的数据空间可靠域、数据空间不确定域和异常域,用于对模型输出结果的可靠性评估,并针对后期积累的增量数据,提出持续性可靠域更新方法,从而实现可靠域的持续趋优,最后基于该可靠域,提出多智能体互补的电力系统暂态稳定评估方法。算例测试结果表明数据空间可靠域可有效划分模型能正确预测样本的范围,在所确定样本上的预测准确率可达到1;可靠域更新方法能够实现对可靠域的不断更新优化,提高可靠域的精准度;采用多智能体互补模型可有效弥补单智能体模型在高维空间上确定样本范围较小的缺点,并仍然保持极高的预测准确率。

The application of artificial intelligence to power system transient stability assessment has been extensively focused.However,how to evaluate the reliability of artificial intelligence models is a key issue in the practical application of this method.For this reason,this paper introduced the related concepts of the data space reliability domain,and proposed a multi-agent complementary power system transient stability intelligent assessment method based on the data space reliability domain.Firstly,based on the SHAP theory,a feature sample space weighted by shapley values was obtained,and then according to the distribution of the correct prediction sample and misjudged sample in this space,the data space reliability domain,the data space uncertainty domain and the unusual domain were determined,which were used to evaluate the reliability of the model.And according to the incremental data accumulated in the later stage,a continuous reliability domain update method was proposed to realize the continuous optimization of the reliability domain.Finally,based on the reliability domain,a method of multi-agent complementary power system transient stability assessment was proposed.The test result shows that the data space reliability region can effectively divide the range of the sample that can be correctly predicted,and the prediction accuracy on the determined sample can reach 1.The reliability domain update method can realize the continuous update and optimization of the reliability domain and improve the accuracy of the reliability domain.The multi-agent complementary model can effectively compensate for the shortcomings of the single-agent model in determining sample,and it has a very high prediction accuracy.