Abstract: The uncertainty of new energy output such as wind and solar makes it difficult to accurately track and quantify carbon emissions on the power supply side, and the participation of dispatchable loads will further increase the complexity of carbon emission optimization. Therefore, a stochastic carbon flow optimization method under source-load interaction is proposed. First, combined with the working characteristics of wind power, photovoltaic and dispatchable load, the fuzzy C-means clustering method is improved from the perspective of membership function and distance function, and the stochastic output scenario set of the power grid in stochastic environment is efficiently and reasonably aggregated. Then, based on the carbon emission flow theory, an incremental carbon flow network model considering network loss is established to track the changes of carbon flow in a timely and rapid manner during the stochastic fluctuation of source load output. Finally, taking into account the economy, environmental protection and user satisfaction, a multi-objective stepped node marginal carbon emission model is established, and marginal carbon parameters are proposed instead of average carbon emission factors, and the carbon emission reduction potential of the load side is finely quantified. The superiority of the source-load interaction optimization method proposed in this paper is verified by example simulation. Compared with the traditional carbon flow theory scheduling method, the comprehensive cost is reduced by 2.50% and the carbon emission is reduced by 7.11%.