To address the challenge faced by power grid companies in accurately detecting changes in user industry information, which has been complicated by the increasing variability of industry characteristics in recent years, a data-driven approach for identifying anomalies in load characteristics is proposed. Initially, a two-stage methodology for developing typical load patterns for various industries is presented. The hierarchical density-based spatial clustering of applications with noise (HDBSCAN) technique is utilized to extract typical daily load curves for users under different scenarios. Subsequently, these extracted daily load curves are clustered using an improved K-means algorithm to establish typical load patterns for the respective industries. In the second phase, a multidimensional intelligent diagnostic method for load characteristic anomalies is introduced. User load characteristics are constructed, and the entropy weight method is employed to evaluate the relative significance of typical industry scenarios. The one-class support vector machine (OCSVM) algorithm is then utilized to quantify the degree of anomaly present in user load characteristics across each scenario. Comprehensive suspicion scores are calculated and ranked to accurately identify users exhibiting abnormal load characteristics. The effectiveness of the proposed method is validated through the analysis of actual user data from a specific region. The results demonstrate that the method is both feasible and practical for constructing typical industry load scenarios and for the identification of load characteristic anomalies.