To make up for the lack of optimal control on carbon emission in the conventional optimization calculation of power flow of the power grid，this paper presents a Q（λ） learning algorithm based on the semi-Markov decision process by the carbon emission flow model based on the results of power flow.This algorithm converts the constraints，controllable variables to status and action of the algorithm，while using a linear weighted approach to convert carbon emission loss，network loss and the stable component of voltage to rewards to dynamically search for the optimal action strategy via continuous fault testing，retrospecting and iteration.Using an IEEE 57-bus system for case study，this method yields good results，providing an effective and feasible way to solve the problem of multi-objective optimal carbon emission flow.