The trend of decarbonization of energy structure makes the low-carbon operation of the main grid network a key part of power system optimization. This paper takes the distribution grid under distributed power access as the research object, based on the theory of distribution grid structure and carbon emission flow. The carbon emission characteristics of distributed thermal power, wind power, gas, photovoltaic, and energy storage units are elaborated in turn, and the corresponding carbon emission mathematical model is established. Due to the uncertainty of renewable energy itself, the operation of the decarbonized power grid system has certain security risks. In order to predict the unexpected situation of the decarbonized grid system in advance to formulate the scheduling strategy, an interval prediction model based on gated recurrent neural network (GRU) is constructed. The model provides data support for the system’s low-carbon operation strategy by providing uncertainty information of the grid system’s contingency targets. The contingency situation is divided into two phases: the day-ahead and intraday phases, and the day-ahead phase utilizes the data information obtained from the prediction for intraday strategy generation. In the intraday phase, the corresponding adjustment method is used to optimize the scheduling according to the actual situation. The designed two-phase scheduling strategy reduces carbon emissions by 11.03% in the control variable scenario by adjusting the day-ahead phase, which has a significant reduction effect.
