The Evaluation Method for Invoking Electric Power Artificial Intelligence Models

A method for evaluating the invocation of electric power artificial intelligence (AI) models has been proposed. Initially, this method involves collecting information on model performance, invocation details, and effectiveness feedback, while considering the correlation between model speed and accuracy with evaluation metric parameters. Following the order of correlation strength, a set of key evaluation metric parameters are selected to form an invocation evaluation metric set. Subsequently, all parameters within this set are normalized, and different levels along with corresponding scores for each parameter are established. The optimal weight combination for each evaluation metric parameter is then determined using the non-dominated sorting genetic algorithm, and the evaluation score for each model is calculated through weighted summation. Finally, based on the discovery rate of power equipment defects, false detection ratio, and recognition efficiency, an overall effectiveness evaluation is conducted on the model with the highest evaluation score. The superiority or inferiority of the selected model is judged according to the evaluation results. This method of evaluating model invocations effectively addresses technical issues impacting the user experience of AI model systems, achieves quantitative evaluation of models, enhances the accuracy of horizontal comparisons among models, and increases user satisfaction.