水轮机模型综合特性曲线数值处理方法研究

Numerical treatment of synthetic characteristic curves in turbine modeling

水轮机模型综合特性曲线的数值处理对水轮机选型、水电站优化运行及计算机仿真有着重要的作用。不同的数值处理方法得到的结果可能不同,各结果间的差别及反映的各参数间的相关关系的差异性可能造成水轮机实际运行和预估值间的不同,严重影响水轮机的安全和经济运行。为了比较不同数值处理方法的差异性及适用范围,以A858a-36.6型水轮机为例,对其模型综合特性曲线分别进行高阶曲面拟合法、三维空间曲面插值法及BP神经网络拟合法的数值处理,并以数据处理值和真实样本值间相对误差累加值的大小作为评判基准。结果表明:这3种方法均适用于水轮机模型综合特性曲线的数值处理,但高阶曲面拟合法精度差,适用于短时间内粗略求解各参数间相关关系;三维空间曲面插值法精度最高,适用于纯图像类型或可视化问题的数据处理;BP神经网络拟合法精度在二者之间,可作为三维空间曲面插值法的补充处理,样本容量越多其精度越高。

Numerical treatment of synthetic characteristic curves plays an important role in the selection of turbine types,optimal operation of hydropower plants,and computational simulations,different numerical treatments may result in different outcomes.Because the differences of the outcomes and their impacts on the correlations among the parameters are unknown,this may cause the gap between the actual operation and the predictions,thus influencing the safe and economical operation of the turbine.In order to compare the differences and applicability of different numerical methods,we used A858 a-36.6 turbine as the research object and adopted high order surface fitting,three dimensional spatial surface interpolation,and BP neural network fitting to treat the model characteristic curve respectively.Then the simulation results were judged by the sum of the relative error between the simulation results and the actual operation data.According to the comparison results,these three methods can all be used to deal with the curve,but high order surface fitting method performs poorly in accuracy,so it is only suitable for rough solutions of the correlations among parameters in a short time.Three dimensional spatial surface interpolation gets the highest precision,which is suitable for data processing of the pure image type or visualization problems.The precision of BP neural network fitting method is between these two methods,it can be used as a supplement for three dimensional spatial surface fitting.What’s more,the more samples it gets,the higher the precision becomes.