With the deepening of power market reform, small and medium-sized users have gradually been included in the agency power purchase mechanism. The contradiction between price fluctuation risks, load uncertainty and cost control faced by these users has become increasingly prominent. These users are numerous, but their loads are scattered and their risk tolerance is weak. The traditional static electricity price model is difficult to adapt to the dynamic changes of the market and the differentiated demands of users. Therefore, this paper proposes a dynamic adjustment framework for the agency power purchase electricity prices of small and medium-sized users, which includes market perception, risk quantification, optimization decision-making, and execution feedback. This framework is based on the price risk quantification model, multi-objective optimization function, and improved dynamic weight algorithm, integrating historical price trend weighting, risk cost trade-off, and real-time load response feedback, to achieve the collaborative optimization of power purchase costs, risk losses, and user satisfaction. Experimental results show that this method performs well in three typical scenarios of stability, fluctuation, and extreme conditions: the price risk loss is reduced by 32.7% to 41.2%, the average power purchase cost is stable at 0.38 yuan/kWh to 0.42 yuan/kWh, user satisfaction reaches 92.3%, and the comprehensive performance is significantly superior to the traditional fixed electricity price method, single cost optimization method, and static risk control method. In scenarios of price sudden change and large load peak-valley difference, the risk loss is still lower than 8.5%, the response delay is controlled within 50 ms, demonstrating good robustness and real-time performance. This research provides an efficient and adaptive electricity price adjustment technical path for agency power purchase of small and medium-sized users, which can be applied to various types of power market environments, differentiated user groups, and dynamic load demands.