In order to achieve material classification and cooperating distribution under the unified supply management system of emergency materials in the event of a public emergency, the importance of emergency materials was divided into several levels according to the specific emergency situation. In accordance with the principle of priority distribution of important materials, a model for the hierarchical and cooperating distribution of materials between multiple warehouses was been established, which could integrate and optimize transportation vehicles and various emergency materials throughout the region. The model took the shortest total distribution time as the optimization objective, combined cooperating distribution with time series decision-making, considered the delivery process of all vehicles between warehouses and demand points as the time series decision-making process of multi-agent collaboration, reduced the computational complexity of multi-agent and multi-task assignment problem, and made the algorithm of time series decision-making model still applicable in the case of large-scale problems. In addition, on the basis of improving the variable input and output dimensions of LSTM (Long Short Term Mermory) network, combined with the theoretical framework of genetic algorithm (GA), the LSTM-GA algorithm for this problem was designed, and a study simulation was carried out. It was found that the convergence speed and stability of LSTM-GA algorithm were improved compared with single algorithm. The results show that the LSTM-GA algorithm can realize the variable dimensions of LSTM network receiving and output information, and is an effective method to study the hierarchical and cooperating distribution of emergency materials.