It is crucial to establish accurate models for the identification of genes interactions. In our study, the ensemble model (EM) of flexible neural tree (FNT) and ordinary differential equations (ODEs) is used to infer gene regulatory networks (GRNs) more precisely. During the train of FNT, probabilistic incremental program evolution (PIPE) is employed to optimize the architecture, and particle swarm optimization (PSO) is employed to evolve the parameters. Similarly, the ODEs also use this method to optimize. Finally, the experimental results show that the model can be more accurate compared with some earlier methods. The mean squared error (MSE) of gene expression time series forecasting in this paper is smaller than prior work.