长江口区基于BP算法的表层悬沙浓度计算模型

Surface Layer Suspended Particle Matter Concentration Calculation Model Based on BP Algorithm in Yangtze Estuary

传统的悬浮泥沙浓度测量方法存在着各种不足,而ADCP(声学多普勒流速剖面仪)具有观测悬浮泥沙浓度的潜能。文章目的在于利用BP神经网络强大的非线形处理能力和黑箱模型的优势探讨长江口区快速测量悬浮泥沙浓度的可能性。根据2006年7月在长江口区的20个站位使用RD300K型宽幅ADCP测量得到的实测资料,提取其回声强度信息,把回声强度、温度、盐度、流速作为输入层变量,利用BP算法反演悬浮泥沙浓度,结果表明此方法精度较高(平均相对误差为18.64%),说明利用4-4-1模式的BP人工神经网络算法拟合长江口区表层悬浮泥沙浓度是可行的,但拟合效果与流向无关,而是与采样现场的海况有很大关系。海况越好,拟合效果也越好。在拟合除表层以外其他水层时需要考虑声波传输过程中各水层物理性质的变化。

There are kinds of limitations among the traditional measurement methods of suspended particle matter concentration, aiming at seeking to provide fast and accurate method for measuring it. The investigation was carried out during June 2006 using RD300K ADCP, echo intensity extracted from ADCP, temperature, salinity and current velocity were selected as input layer variables, suspended particle matter concentration of surface layer was calculated based on BP algorithm. The results showed that the calculated results were validated and had a high accuracy; however, the accuracy was negative correlation with sea conditions and independent of current direction. The physical character must be considered when retrieval other layers except surface layer.