根据柴油机振动信号的特性，使其在相空间里重构，再应用组合神经网络，对柴油机振动信号进行拟合和预测。该组合神经网络是一个两级系统，第一级有两个神经网络的预报——一个多目标前馈网络和一个函数耦合神经网络，用模糊反向传播算法进行训练；第二级是由第一级产生的两个预测结果混合得到的组合模型，采用Karmarkar的线性规划算法进行训练。实际应用证明了该方法的有效性。 According to the characteristic of diesel engine vibration signal, it is reconstructed in the phase space, then the combined neural network is used to fit and predict the diesel engine vibration signal. The combined neural network is a two-level system. The first level has two forecasting of neural networks - a multi-objective feedforward network and a function-coupled neural network, which is trained by the fuzzy back propagation algorithm. The second level consists of the A combined model of the two predicted results produced by the first stage is trained by Karmarkar’s linear programming algorithm. Practical application proves the effectiveness of this method.