基于概率神经网络的机组状态多步预报方法

为了解决由单步预报递推运算获得的多步预报存在的预报误差的迭代累积问题，提出了相空间动力学轨道的相似多步预报概念，利用概率神经网络合理分配相似算子，构造了多步预报的概率神经网络结构．然后，以模拟振动数据比较了单步预报神经网络、多步预报神经网络和多步预报概率神经网络的预报能力，并预报了燕山石化大机组停车概率的变化趋势，实践表明该方法具有良好的多步预报能力．

Probabilistic Neural Networks Applied to Multi-Step Forecast of Machinery Behavior

Multi step forecast of machinery behavior is obtained from the recurrence operation of the single step method. Because of the iteration cumulation error, requirement of the multi step prediction cannot be satisfied. The concept of multi step forecast phase space dynamical trace is proposed, the similar factors are allocated rationally using probabilistic neural networks (PNN). Predictions based on the single step and multi step PNN are compared. The variation in trends for the large rotating machinery shut down probabilities in Yanshan Petrochemical Company is predicted.