针对传统BP神经网络算法在单螺杆制冷机组中存在的收敛速度慢、易陷入局部极小的缺点,虽然能够通过动量-自适应神经网络对其进行优化,但其效果并不理想,因此本文建立可拓神经网络的诊断模型。首先可拓神经网络诊断模型改变了网络的拓扑结构,采用双权连接输入层与输出层的方式;然后利用可拓理论的物元模型确定初始权值和利用距离作为测度工具去调整类中心和权值中心的方法;最后再通过实例数据验证,证明该方法相比于动量-自适应神经网络能够更精确、快速地诊断出制冷机组的故障,以达到预期效果。 Aiming at the shortcomings of the traditional BP neural network in the single screw refrigeration unit, the convergent speed is slow and easy to fall into the local minima. Although it can be optimized by the momentum-adaptive neural network, its effect is not satisfactory. Therefore, Diagnostic Model of Extension Neural Network. Firstly, the diagnostic model of BP neural network changes the topology of the network, and adopts the way that the dual weights connect the input layer and the output layer. Then, using the matter-element model of extension theory, the initial weight and the distance are used as the measurement tools to adjust the center and Finally, the example data is used to verify that the proposed method can diagnose the chiller failure more accurately and quickly than the momentum-adaptive neural network to achieve the expected results.