Rapid short-term forecasting of grid loads can realize more stable and economical load deployment control in grid planning and operation, and improve the efficiency of the dispatching system. In this paper, we first collect load data for distribution bandwidth operation, propose a suitable method for preprocessing, and summarize the load fluctuation law in distribution bandwidth operation. Segmented linear regression analysis is carried out using unsupervised methods of isometric division, isofrequency division, and CART regression tree. It is found that all segmented linear regression models can achieve high accuracy prediction with R2>0.96 in the low-temperature segment, but the R2 in the high-temperature segment of WT and AH is only 0.86, which indicates the limitation of the single-factor segmented model in its applicable interval. The general daily load profile has “morning peak” and “evening peak”, with the “morning peak” peaking around 08:20 to 08:30, the “evening peak” peaking around 18:00 to 19:00, and then the load gradually decreasing and reaching a trough around 04:30 AM. It is appropriate to quantify load fluctuations from a combination of simplicity of form and precision of quantification. The maximum temperature, minimum temperature, and average temperature have a significant effect on the load fluctuation, so the maximum temperature, minimum temperature, and average temperature can be used as inputs to the control strategy. Finally, based on the analysis results, load fluctuation stabilization control strategies such as optimizing the fault judgment logic, improving the closed-loop regulation procedure, and reasonably arranging the operation mode are proposed.