BP神经网络模型在电磁环境预测中的应用

Application of BP neural network prediction model to the electromagnetic fields

利用所建立BP神经网络模型的预测功能,基于析因实验设计方法系统分析了所选取因素对电磁环境的影响程度,并结合中心复合实验设计方法,采用响应曲面图直观地呈现了电磁环境的空间分布情况。以实例监测数据为样本建立了BP神经网络模型,电磁环境的模拟系数分别为0.999、0.998和0.841,预测值与实际值的相对误差为0.89%~12.19%。析因实验设计结果表明,工程本身特性(包括工况负荷、垂直高度、水平距离)对电磁环境存在显著的影响,环境因素(温度、湿度、风速)对电磁环境影响不显著,对筛选的显著因素进行进一步的中心复合实验设计,利用响应曲面法直观反映了不同工况负荷条件下,电磁环境在不同垂直高度和不同水平距离的空间分布,可知直流线路对垂直距离小于30 m、水平距离小于10 m区域内的居民住宅影响最大,该研究为直流工程电磁环境污染控制提供理论支持,同时为建立换流站及交流输变电工程附近区域的电磁环境预测模型提供了新思路。

BP neural network model, established to integrate factors with electromagnetic fields, was combined with factorial experiment design and central composite design. The significance of selected factors which affected the electromagnetic fields was analyzed, and then the response - surface method was employed to visually display the spatial distribution of electromagnetic fields based on the central composite design method. Firstly, the BP neural network model was briefly developed using monitoring data, and Nash - Suttclife coefficients were 0. 999 ,0. 998 and 0.841 for the largest ground total electric field strength, the 80% value of ground total electric field strength and magnetic flux density respectively, the relative error of experimental and predictive values cal- culated was between 0.89% and 12.19%, which indicated the accuracy prediction of the model. Secondly, the results of the factorial experiment design demonstrated that the factors, such as temperature, humidity and wind speed, were not of the significance, and the factors including load, height and horizontal distance, were of the significance on the value of electromagnetic fields. Further more, the significant factors were used for central composite design, and the response - surface method was employed to visually display the spatial distribution of electromagnetic fields with the variation of load, height and horizontal distance, which showed that the focus of attention should be placed on the residents living in the area of the vertical dimension less than 30m and horizontal length less than 10m. In conclusion, this research could provide the theoretical support for controlling and protecting the pollution of electromagnetic. Meanwhile, a new idea was provided for BP neural network model of convertor station and substation communication engineering.