人工神经网络在汽轮机热力参数在线仿真及故障识别中的应用

APPLICATION OF ARTIFICIAL NEURAL NETWORK IN ON-LINE THERMAL PARAMETER SIMULATION AND FAILURE IDENTIFICATION OF STEAM TURBINE

目前汽轮机组中一部分测点传感装置损坏率高 ,使得一些热经济性分析的结果产生较大偏差。基于人工神经网络 ,作者对当前普遍使用的 BP网络模型进行了某些改进、对网络中的一些参数作出了调整 ,以使模型具有较强的自适应能力并使得网络的收敛速度沿着最佳方向进行 ,还编制了相应的程序。作为实例 ,文中对某一实际机组的参数进行了仿真计算 ,绝大多数数据的相对误差在 1.5 %以内 ,可以满足工程实际的需要。文章对输入输出参数之间的关联程度 ,对影响输出结果的精度、收敛速度等因素进行的分析比较可供今后的热力参数在线仿真和负荷预测借鉴。用该模型对动力系统的热力参数进行在线仿真减少了传感的维护量 ,特别是对提高汽轮机组故障诊断技术水平有一定的意义 ,此外还改善了基于参数采集的应用软件的可靠性。

At present because of the high failure rate of some sensors installed at the measuring points,it leads to obvious deviation in the analysis of thermal economy.On the basis of the theory of neural network some improvements of BP network model which is being widely used are carried out and some parameters in the neural network are adjusted,in addition,corresponding calculation program is developed.The improvements and adjustments make the BP model more adaptive and the network converged along optimal direction.As an example some key thermal parameters of a real steam turbine system are simulated and calculated in this paper,the relative errors of most calculated data are less than 1.5% and it is shown that this model is practicable.In this paper the results of the analysis and comparison on the correlation extent between input and output parameters and that on the factors influencing outputs,such as the precision and convergence speed,are available for reference in on-line thermal parameter simulation and load forecasting in the future.With the on-line thermal parameters simulation by this model the workload of sensor maintenance can be reduced and the technique of failure diagnosis of steam turbine is improved,moreover,the reliability of application software based on parameter acquisition is enhanced.