Abstract: In the distributed data fusion system, factors such as communication delay can lead to different order of evidence in fusion nodes. Traditional evidence combination rules are difficult to achieve consensus in this case. According to the basic properties of the combination rules, when the combination rules do not satisfy the associativity, the difference in the order of evidence leads to nonconsensus of the combination results. Most of the conflicting evidence combination rules lose associativity in solving the conflict paradox of the Dempster rule. Some of the conflicting evidence combination rules maintain the associativity at the cost of combination result uncertainty, but the result uncertainty seriously affects user decision. This paper starts from the point of conflicting basic belief assignments (BBA) reserving and redistributing to combine disorder conflicting evidence. Previous conflicting BBA are reserved by the free Dezert Smarandache (DSm) model. The previous conflicting BBA and current evidence are used to generate the weights. Then the previous conflicting BBA and current conflicting BBA are redistributed through the weights. The combination rule can reduce the influence of the evidence order on the combination results, and can maintain a good consensus in the case of disorder conflicting evidence. Numerical examples show the efficiency and rationality of the proposed approach.