Abstract: In the process of identifying abnormal data in power marketing, any power marketing data is selected to calculate an abnormal threshold value, which is then used to compute the Mahalanobis distance between this threshold value and the abnormal data. Due to the influence of hidden features, it is difficult to clearly distinguish between normal and abnormal data, leading to inconsistencies between identified abnormal data points and actual data points in power marketing. This results in inaccurate labeling of abnormal data intervals, affecting the accuracy of power marketing identification. Therefore, an intelligent recognition technology for abnormal data in power marketing based on hidden feature mining has been designed. According to the normal distribution, the power marketing dataset is sorted to determine the significant level abnormal threshold value of the power marketing data. Electricity usage patterns, equipment health status, and abnormal association rules are used as raw data, which are mapped into a hidden feature space. The critical skew space distance of abnormal data is measured based on hidden feature mining, thereby achieving intelligent recognition of abnormal data in power marketing. The final recognition results show that the identified abnormal data points are consistent with the actual data points, accurately labeling the intervals of abnormal data. The intelligent recognition effect of abnormal data is good, playing a crucial role in improving marketing efficiency.