基于GA-WNN的极化SAR海洋溢油检测方法研究

海洋溢油对海洋生态和人类生活带来严重的影响。由于合成孔径雷达(Synthetic Aperture Radar,SAR)具有全天时全天候的工作能力,在海洋溢油检测中发挥重要作用。目前,极化SAR是SAR探测技术的先进手段。本文利用6个极化特征进行溢油检测,通过对比分析这些特征对不同溢油的检测能力,得出单一极化特征在溢油检测中存在不足。通过J-M特征优选方法,提取出溢油检测识别度较高的特征影像,并利用遗传算法优化的小波神经网络(Genetic Algorithm-Wavelet Neural Network,GA-WNN)进行溢油检测。利用2套Radarsat-2全极化数据进行了方法验证,结果表明,该方法优于其他检测方法,溢油检测精度分别达到90.31%和95.42%。     

边坡稳定性预测的模糊神经网络模型

根据边坡稳定问题具有的模糊性,提出了一种判定边坡稳定性的模糊神经网络模型。该系统仅从期望输入输出数据集即可达到获取知识、确定模糊初始规则基的目的。再利用神经网络学习能力便不难修改规则库中的模糊规则以及隶属函数和网络权值等参数,这样大大减少了规则匹配过程,加快了推理速度,从而极大程度地提高了系统的自适应能力。最后用收集到的边坡数据样本训练和测试模糊神经网络模型,结果表明该模糊神经网络预测边坡稳定性是可行的、有效的。     

Development of a regional neural network for coastal water level predictions

This paper presents the development of a Regional Neural Network for Water Level (RNN_WL) predictions, with an application to the coastal inlets along the South Shore of Long Island, New York. Long-term water level data at coastal inlets are important for studying coastal hydrodynamics sediment transport. However, it is quite common that long-term water level observations may be not available, due to the high cost of field data monitoring. Fortunately, the US National Oceanographic and Atmospheric Administration (NOAA) has a national network of water level monitoring stations distributed in regional scale that has been operating for several decades. Therefore, it is valuable and cost effective for a coastal engineering study to establish the relationship between water levels at a local station and a NOAA station in the region. Due to the changes of phase and amplitude of water levels over the regional coastal line, it is often difficult to obtain good linear regression relationship between water levels from two different stations. Using neural network offers an effective approach to correlate the non-linear input and output of water levels by recognizing the historic patterns between them. In this study, the RNN_WL model was developed to enable coastal engineers to predict long-term water levels in a coastal inlet, based on the input of data in a remote NOAA station in the region. The RNN_WL model was developed using a feed-forwards, back-propagation neural network structure with an optimized training algorithm. The RNN_WL model can be trained and verified using two independent data sets of hourly water levels.The RNN_WL model was tested in an application to Long Island South Shore. Located about 60–100 km away from the inlets there are two permanent long-term water level stations, which have been operated by NOAA since the1940s. The neural network model was trained using hourly data over a one-month period and validated for another one-month period. The model was then tested over year-long periods. Results indicate that, despite significant changes in the amplitudes and phases of the water levels over the regional study area, the RNN_WL model provides very good long-term predictions of both tidal and non-tidal water levels at the regional coastal inlets. In order to examine the effects of distance on the RNN_WL model performance, the model was also tested using water levels from other remote NOAA stations located at longer distances, which range from 234 km to 591 km away from the local station at the inlets. The satisfactory results indicate that the RNN_WL model is able to supplement long-term historical water level data at the coastal inlets based on the available data at remote NOAA stations in the coastal region.     

Development of a Neural Network Algorithm for Retrieving Concentrations of Chlorophyll, Suspended Matter and Yellow Substance from Radiance Data of the Ocean Color and Temperature Scanner

An algorithm is presented to retrieve the concentrations of chlorophyll a, suspended pariclulate matter and yellow substance from normalized water-leaving radiances of the Ocean Color and Temperature Sensor (OCTS) of the Advanced Earth Observing Satellite (ADEOS). It is based on a neural network (NN) algorithm, which is used for the rapid inversion of a radiative transfer procedure with the goal of retrieving not only the concentrations of chlorophyll a but also the two other components that determine the water-leaving radiance spectrum. The NN algorithm was tested using the NASA's SeaBAM (SeaWiFS Bio-Optical Mini-Workshop) test data set and applied to ADEOS/OCTS data of the Northwest Pacific in the region off Sanriku, Japan. The root-mean-square error between chlorophyll a concentrations derived from the SeaBAM reflectance data and the chlorophyll a measurements is 0.62. The retrieved chlorophyll a concentrations of the OCTS data were compared with the corresponding distribution obtained by the standard OCTS algorithm. The concentrations and distribution patterns from both algorithms match for open ocean areas. Since there are no standard OCTS products available for yellow substance and suspended matter and no in situ measurements available for validation, the result of the retrieval by the NN for these two variables could only be assessed by a general knowledge of their concentrations and distribution patterns. This revised version was published online in July 2006 with corrections to the Cover Date.     

Neural network-based GPS GDOP approximation and classification

In this paper, the neural network (NN)-based navigation satellite subset selection is presented. The approach is based on approximation or classification of the satellite geometry dilution of precision (GDOP) factors utilizing the NN approach. Without matrix inversion required, the NN-based approach is capable of evaluating all subsets of satellites and hence reduces the computational burden. This would enable the use of a high-integrity navigation solution without the delay required for many matrix inversions. For overcoming the problem of slow learning in the BPNN, three other NNs that feature very fast learning speed, including the optimal interpolative (OI) Net, probabilistic neural network (PNN) and general regression neural network (GRNN), are employed. The network performance and computational expense on NN-based GDOP approximation and classification are explored. All the networks are able to provide sufficiently good accuracy, given enough time (for BPNN) or enough training data (for the other three networks).     

Feed forward Artificial Neural Network model to predict the average summer-monsoon rainfall in India

In the present research, possibility of predicting average summer-monsoon rainfall over India has been analyzed through Artificial Neural Network model. In formulating the ANN — based predictive model, three-layer network has been constructed with sigmoid non-linearity. The monthly summer monsoon rainfall totals, tropical rainfall indices and sea surface temperature anomalies have been considered as predictors while generating the input matrix for the ANN. The data pertaining to the years 1950–1995 have been explored to develop the predictive model. Finally, the prediction performance of neural net has been compared with persistence forecast and Multiple Linear Regression forecast and the supremacy of the ANN has been established over the other processes.     

Prediction of Chemical Oxygen Demand (COD) Based on Wavelet Decomposition and Neural Networks

The chemical oxygen demand (COD) parameter of a wastewater treatment plant is predicted based on wavelet decomposition, entropy, and neural networks (NN) for rapid COD analysis. This paper also describes the usage of wavelet and NNs for parameter prediction. Data from a wastewater treatment plant in Malatya, Turkey, were used. This dataset consists of daily values of influents and effluents for a year. To reduce the dimension of input parameters and to decrease the NN training time, wavelet decomposition and entropy were used. Test results were presented graphically. The test results of the trained model were found to be closer to the measured COD values.     

Application of LPW and SAI SAFNWC/MSG satellite products in pre-convective environments

The Spinning Enhanced Visible and Infrared Imagery (SEVIRI) instrument, on board the Meteosat Second Generation (MSG), is a radiometer with eight infrared (IR) spectral bands. Seven of these channels are used to retrieve Layer Precipitable Water (LPW) and Stability Analysis Imagery (SAI). Both products are the PGE07 and the PGE08 of SAFNWC (Satellite Application Facility on support to Nowcasting and Very Short-Range Forecasting). The authors at Instituto Nacional de Meteorología (INM) have developed the LPW and SAI algorithms, in the SAFNWC framework. Both products are retrieved using statistical retrieval based on neural networks. The main advantage of these algorithms versus physical retrieval algorithms is the independence from the Numerical Weather Prediction (NWP) models. The LPW provides information on the water vapor contained in a vertical column of unit cross-section area in three layers in the troposphere (low, middle and high) and in the total layer in cloud free areas. The SAI provides estimations of the atmospheric instability in cloud free areas, in particular the Lifted Index (LI).The stability and precipitable water obtained with both products are routinely generated every 15 min at a satellite horizontal resolution of 3 km in NADIR. A significant advantage of these MSG products, compared to traditional measurements such as radiosondes, is their ability to measure high resolution temporal and spatial variations of atmospheric stability and moisture in pre-convective environments. The main disadvantage is that they do not have the vertical resolution of radiosonde. The MSG moisture and stability time trend fields are especially useful during the period preceding the outbreak of convection due to the high resolution. Once the outbreak of convection occurs, the products calculated in the clear air pixels surrounding the convective system will allow to foresee the evolution of the convection.     

Sounding-derived indices for neural network based short-term thunderstorm and rainfall forecasts

A neural network-based scheme to do a multivariate analysis for forecasting the occurrence and intensity of a meteo event is presented. Many sounding-derived indices are combined together to build a short-term forecast of thunderstorm and rainfall events, in the plain of the Friuli Venezia Giulia region (hereafter FVG, NE Italy).For thunderstorm forecasting, sounding, lightning strikes and mesonet station data (rain and wind) from April to November of the years 1995–2002 have been used to train and validate the artificial neural network (hereafter ANN), while the 2003 and 2004 data have been used as an independent test sample. Two kind of ANNs have been developed: the first is a “classification model” ANN and is built for forecasting the thunderstorm occurrence. If this first ANN predicts convective activity, then a second ANN, built as a “regression model”, is used for forecasting the thunderstorm intensity, as defined in a previous article.The classification performances are evaluated with the ROC diagram and some indices derived from the Table of Contingency (like KSS, FAR, Odds Ratio). The regression performances are evaluated using the Mean Square Error and the linear cross correlation coefficient R.A similar approach is applied to the problem of 6 h rainfall forecast in the Friuli Venezia Giulia plain, but in this second case the data cover the period from 1992 to 2004. Also the forecasts of binary events (defined as the occurrence of 5, 20 or 40 mm of maximum rain), made by classification and regression ANN, were compared. Particular emphasis is given to the sounding-derived indices which are chosen in the first places by the predictor forward selection algorithm.     

FORECAST OF PREFERRED FAULT BASED ON NEURAL NETWORK

基于优势面区域稳定性评价理论和人工神经网络 (ANN)的原理和方法 ,探讨了基于 ANN的优势断裂预报神经网络算法及模型 ,并结合实例检验表明应用反传 (BP)神经网络模型判定优势断裂的新方法是有效的 ,且取得了理想的结果。