非确定环境下电厂运行趋势的动态预测方法

Dynamic prediction method of power plant operation trend in uncertain environment

非确定环境下的运行趋势预测是提高电厂设备运行安全性与可靠性的重要研究方向之一。传统的预测方法通常需要建立系统级的精确解析模型,建模难度大,预测准确性也不够理想,但其优势是计算速度快,实时性强。新兴的基于大数据驱动的运行趋势预测方法通过数据挖掘发现电厂运行趋势的新规律,弥补了人工建立解析模型过程中对设备运行规律描述的完备性问题,预测准确性相对于传统方法有明显改善,但多层神经网络的训练需要较长的运算时间,在运行方式发生变化的情况下难以满足实时性要求。动态预测方法研究了实时运行数据、基于运行大数据驱动的预测方法与解析模型预测方法间的内在联系,得出以下结论:变工况运行趋势预测可以采用一般精度的解析模型,仿真计算过程中若累计误差超过阈值,可以利用实时运行数据作为新的初始条件进行矫正;设备故障预测与健康管理可以采用基于数据驱动的运行趋势预测方法,解析模型可以大量模拟故障工况作为训练数据;多层神经网络需要定期离线训练后更新在线系统的模型参数。研究结论可以为高性能、高精度、高可靠性的电厂运行趋势预测系统开发提供新的理论基础。

Running tendency forecasting in uncertain environment is one of the important research directions to improve the safety and reliability of power plant equipment.Traditional forecasting method require the system-level accurate analytical models.It is difficulty in modeling and the forecasting accuracy is not well,but this method has fast computation speed and strong real-time capability.Based on the running tendency forecasting method by big data driven,the new roles of power plant running trend are found through data mining.It makes up for the completeness of equipment running rules by the artificial analytical model,and compared with traditional methods,the accuracy of forecasting is improved obviously.But this method will take longer hours for training multilayer neural networks,it is difficult to meet the requirement in real-time running tendency forecasting when the operation mode changes.The dynamic method studies the internal relationship between real-time running data,the running tendency forecasting method by big data driven and the accurate analytical models forecasting method,and reach the following conclusion:the analytical model with general accuracy can be used during operations under off-design conditions,and in the process of simulation,real-time running data can be used as a new initial condition for check parameters if the cumulative error exceeds the threshold.The running tendency forecasting method by big data driven can be used for equipment prognostics and health management(PHM),and the analytical model can simulate a large number of fault conditions as training data.The multilayer neural network needs to update the model parameters of the online system after that is trained offline.These conclusions can provide a new theoretical basis for the development of high performance,high precision and high reliability power plant running tendency forecasting system.