Line loss management efficiency is one of the main indicators used to judge the effectiveness of electric power enterprises’ work, as well as a comprehensive indicator used to reflect their operation level and production technology. In this paper, we analyze the types of line losses in low-voltage station areas, present a method for calculating these losses, examine the influence indices of these losses, and outline the process for collecting line loss data in these areas. The maximum load loss amount is used for outlier construction, and the improved K-Means clustering algorithm is utilized for clustering processing of line loss data, which is combined with the isolated forest algorithm to solve the anomaly scores of the line loss data and then to obtain the positioning of abnormal data of line loss in low-voltage station areas. Then, based on the Markov chain, the probabilistic modeling of the time series random variable data of line loss in the LV station area is carried out, and the OLS-optimised RBF neural network is used to process its time series variation data. Applying the combined model to the detection of line loss anomalies in LV station areas, the global maximum value of its time-voltage parametric difference is only 34.28 V, and the value of the directional electron transmission distance is only 3.41 μm. When the expansion constant is 5.42, the OLS-RBF model’s optimal prediction accuracy and minimum mean square error reach 98.43% and 0.0012, respectively. The average residual mean value for the model prediction results and the actual line loss data is 4.4 kW, and the average relative error is only 0.12%. Therefore, this paper’s method effectively locates anomalies, detects stochastic line loss data in low-voltage stations, and accurately processes time-series random variable data.