首页 | 本学科首页   官方微博 | 高级检索  
相似文献
 共查询到20条相似文献，搜索用时 536 毫秒
1.
This paper presents the development of a multiple‐station neural network for predicting tidal currents across a coastal inlet. Unlike traditional hydrodynamic models, the neural network model does not need inputs of coastal topography and bathymetry, grids, surface and bottom frictions, and turbulent eddy viscosity. Without solving hydrodynamic equations, the neural network model applies an interconnected neural network to correlate the inputs of boundary forcing of water levels at a remote station to the outputs of tidal currents at multiple stations across a local coastal inlet. Coefficients in the neural network model are trained using a continuous dataset consisting of inputs of water levels at a remote station and outputs of tidal currents at the inlet, and verified using another independent input and output dataset. Once the neural network model has been satisfactorily trained and verified, it can be used to predict tidal currents at a coastal inlet from the inputs of water levels at a remote station. For the case study at Shinnecock Inlet in the southern shore of New York, tidal currents at nine stations across the inlet were predicted by the neural network model using water level data located from a station about 70 km away from the inlet. A continuous dataset in May 2000 was used for the training, and another dataset in July 2000 was used for the verification of the neural network model. Comparing model predictions and observations indicates correlation coefficients range from 0·95 to 0·98, and the root‐mean‐square error ranges from 0·04 to 0·08 m s?1 at the nine current locations across the inlet. Copyright © 2007 John Wiley & Sons, Ltd.  相似文献   

2.
Temporally weighted average curve number method for daily runoff simulation   总被引：1，自引：0，他引：1  
Nam Won Kim  Jeongwoo Lee 《水文研究》2008,22(25):4936-4948
The modified Soil Conservation Service curve number (CN) method is widely used in long‐term continuous models to predict daily surface runoff. However, it has been shown that this method gives poor results in reproducing peak flows in high rainfall periods. This is because there is an inaccuracy stemming from the model algorithm as it adjusts the daily runoff curve number as a function of soil moisture content at the end of the previous day. This paper proposes an alternative daily based curve number technique that can provide better prediction of daily runoff during the high flow season. The proposed method uses the temporally weighted average curve number (TWA‐CN) to estimate daily surface runoff, while considering the effect of rainfall during a given day as well as the antecedent soil moisture condition. To test the applicability of the TWA‐CN method, it was incorporated with the long‐term, continuous simulation watershed models SWAT and SWAT‐G. Simulations were conducted for the Miho River watershed located in the middle of South Korea. The graphical displays and statistics of the determination coefficient (R2) and the Nash–Sutcliffe model efficiency (NSE) of the observed and simulated daily runoff indicated that the modified SWAT with the TWA‐CN method may provide better runoff prediction (R2 = 0·837, NSE = 0·833) than the original SWAT (R2 = 0·815, NSE = 0·824). Likewise, the determination coefficient (R2 = 0·816) and the Nash–Sutcliffe efficiency (NSE = 0·834) for the modified SWAT‐G are also higher than the original version (R2 = 0·782, NSE = 0·825). It is expected that the improved capability in predicting surface runoff using the suggested CN estimate method will provide a sound contribution to the accurate simulations of water yield. Copyright © 2008 John Wiley & Sons, Ltd.  相似文献   

3.
Özgür Kişi 《水文研究》2009,23(2):213-223
This paper reports on investigations of the abilities of three different artificial neural network (ANN) techniques, multi‐layer perceptrons (MLP), radial basis neural networks (RBNN) and generalized regression neural networks (GRNN) to estimate daily pan evaporation. Different MLP models comprising various combinations of daily climatic variables, that is, air temperature, solar radiation, wind speed, pressure and humidity were developed to evaluate the effect of each of these variables on pan evaporation. The MLP estimates are compared with those of the RBNN and GRNN techniques. The Stephens‐Stewart (SS) method is also considered for the comparison. The performances of the models are evaluated using root mean square errors (RMSE), mean absolute error (MAE) and determination coefficient (R2) statistics. Based on the comparisons, it was found that the MLP and RBNN computing techniques could be employed successfully to model the evaporation process using the available climatic data. The GRNN was found to perform better than the SS method. Copyright © 2008 John Wiley & Sons, Ltd.  相似文献   

4.
Fish habitat and aquatic life in rivers are highly dependent on water temperature. Therefore, it is important to understand andto be able to predict river water temperatures using models. Such models can increase our knowledge of river thermal regimes as well as provide tools for environmental impact assessments. In this study, artificial neural networks (ANNs) will be used to develop models for predicting both the mean and maximum daily water temperature. The study was conducted within Catamaran Brook, a small drainage basin tributary to the Miramichi River (New Brunswick, Canada). In total, eight ANN models were investigated using a variety of input parameters. Of these models, four predicted mean daily water temperature and four predicted maximum daily water temperature. The best model for mean daily temperature had eight input parameters: minimum, maximum and mean air temperatures of the current day and those of the preceding day, the day of year and the water level. This model had an overall root‐mean‐square error (RMSE) of 0·96 °C, a bias of 0·26 °C and a coefficient of determination R2 = 0·971. The model that best predicted maximum daily water temperature was similar to the first model but excluded mean daily air temperature. Good results were obtained for maximum water temperatures with an overall RMSE of 1·18 °C, a bias of 0·15 °C and R2 = 0·961. The results of ANN models were similar to and/or better than those observed from the literature. The advantages of artificial neural networks models in modelling river water temperature lie in their simplicity of use, their low data requirement and their good performance, as well as their flexibility in allowing many input and output parameters. Copyright © 2008 Crown in the right of Canada and John Wiley & Sons, Ltd.  相似文献   

5.
Evapotranspiration (ET) is one of the basic components of the hydrologic cycle and is essential for estimating irrigation water requirements. In this study, an artificial neural network (ANN) model for reference evapotranspiration (ET0) calculation was investigated. ANNs were trained and tested for arid (west), semi‐arid (middle) and sub‐humid (east) areas of the Inner Mongolia district of China. Three or four climate factors, i.e. air temperature (T), relative humidity (RH), wind speed (U) and duration of sunshine (N) from 135 meteorological stations distributed throughout the study area, were used as the inputs of the ANNs. A comparison was conducted between the estimates provided by the ANNs and by multilinear regression (MLR). The results showed that ANNs using the climatic data successfully estimated ET0 and the ANNs simulated ET0 better than the MLRs. The ANNs with four inputs were more accurate than those with three inputs. The errors of the ANNs with four inputs were lower (with RMSE of 0·130 mm d?1, RE of 2·7% and R2 of 0·986) in the semi‐arid area than in the other two areas, but the errors of the ANNs with three inputs were lower in the sub‐humid area (with RMSE of 0·21 mm d?1, RE of 5·2% and R2 of 0·961. For the different seasons, the results indicated that the highest errors occurred in September and the lowest in April for the ANNs with four inputs. Similarly, the errors were higher in September for the ANNs with three inputs. Copyright © 2008 John Wiley & Sons, Ltd.  相似文献   

6.
This study was conducted under the USDA‐Conservation Effects Assessment Project (CEAP) in the Cheney Lake watershed in south‐central Kansas. The Cheney Lake watershed has been identified as ‘impaired waters’ under Section 303(d) of the Federal Clean Water Act for sediments and total phosphorus. The USDA‐CEAP seeks to quantify environmental benefits of conservation programmes on water quality by monitoring and modelling. Two of the most widely used USDA watershed‐scale models are Annualized AGricultural Non‐Point Source (AnnAGNPS) and Soil and Water Assessment Tool (SWAT). The objectives of this study were to compare hydrology, sediment, and total phosphorus simulation results from AnnAGNPS and SWAT in separate calibration and validation watersheds. Models were calibrated in Red Rock Creek watershed and validated in Goose Creek watershed, both sub‐watersheds of the Cheney Lake watershed. Forty‐five months (January 1997 to September 2000) of monthly measured flow and water quality data were used to evaluate the two models. Both models generally provided from fair to very good correlation and model efficiency for simulating surface runoff and sediment yield during calibration and validation (correlation coefficient; R2, from 0·50 to 0·89, Nash Sutcliffe efficiency index, E, from 0·47 to 0·73, root mean square error, RMSE, from 0·25 to 0·45 m3 s?1 for flow, from 158 to 312 Mg for sediment yield). Total phosphorus predictions from calibration and validation of SWAT indicated good correlation and model efficiency (R2 from 0·60 to 0·70, E from 0·63 to 0·68) while total phosphorus predictions from validation of AnnAGNPS were from unsatisfactory to very good (R2 from 0·60 to 0·77, E from ? 2·38 to 0·32). The root mean square error–observations standard deviation ratio (RSR) was estimated as excellent (from 0·08 to 0·25) for the all model simulated parameters during the calibration and validation study. The percentage bias (PBIAS) of the model simulated parameters varied from unsatisfactory to excellent (from 128 to 3). This study determined SWAT to be the most appropriate model for this watershed based on calibration and validation results. Copyright © 2008 John Wiley & Sons, Ltd.  相似文献   

7.
High‐resolution, spatially extensive climate grids can be useful in regional hydrologic applications. However, in regions where precipitation is dominated by snow, snowmelt models are often used to account for timing and magnitude of water delivery. We developed an empirical, nonlinear model to estimate 30‐year means of monthly snowpack and snowmelt throughout Oregon. Precipitation and temperature for the period 1971–2000, derived from 400‐m resolution PRISM data, and potential evapotranspiration (estimated from temperature and day length) drive the model. The model was calibrated using mean monthly data from 45 SNOTEL sites and accurately estimated snowpack at 25 validation sites: R2 = 0·76, Nash‐Sutcliffe Efficiency (NSE) = 0·80. Calibrating it with data from all 70 SNOTEL sites gave somewhat better results (R2 = 0·84, NSE = 0·85). We separately applied the model to SNOTEL stations located < 200 and ≥ 200 km from the Oregon coast, since they have different climatic conditions. The model performed equally well for both areas. We used the model to modify moisture surplus (precipitation minus potential evapotranspiration) to account for snowpack accumulation and snowmelt. The resulting values accurately reflect the shape and magnitude of runoff at a snow‐dominated basin, with low winter values and a June peak. Our findings suggest that the model is robust with respect to different climatic conditions, and that it can be used to estimate potential runoff in snow‐dominated basins. The model may allow high‐resolution, regional hydrologic comparisons to be made across basins that are differentially affected by snowpack, and may prove useful for investigating regional hydrologic response to climate change. Published in 2011 by John Wiley & Sons, Ltd.  相似文献   

8.
Forecasting river flow is important to water resources management and planning. In this study, an artificial neural network (ANN) model was successfully developed to forecast river flow in Apalachicola River. The model used a feed‐forward, back‐propagation network structure with an optimized conjugated training algorithm. Using long‐term observations of rainfall and river flow during 1939–2000, the ANN model was satisfactorily trained and verified. Model predictions of river flow match well with the observations. The correlation coefficients between forecasting and observation for daily, monthly, quarterly and yearly flow forecasting are 0·98, 0·95, 0·91 and 0·83, respectively. Results of the forecasted flow rates from the ANN model were compared with those from a traditional autoregressive integrated moving average (ARIMA) forecasting model. Results indicate that the ANN model provides better accuracy in forecasting river flow than does the ARIMA model. Copyright © 2004 John Wiley & Sons, Ltd.  相似文献   

9.
Tropospheric (ground‐level) ozone has adverse effects on human health and environment. In this study, next day's maximum 1‐h average ozone concentrations in Istanbul were predicted using multi‐layer perceptron (MLP) type artificial neural networks (ANNs). Nine meteorological parameters and nine air pollutant concentrations were utilized as inputs. The total 578 datasets were divided into three groups: training, cross‐validation, and testing. When all the 18 inputs were used, the best performance was obtained with a network containing one hidden layer with 24 neurons. The transfer function was hyperbolic tangent. The correlation coefficient (R), mean absolute error (MAE), root mean squared error (RMSE), and index of agreement or Willmott's Index (d2) for the testing data were 0.90, 8.78 µg/m3, 11.15 µg/m3, and 0.95, respectively. Sensitivity analysis has indicated that the persistence information (current day's maximum and average ozone concentrations), NO concentration, average temperature, PM10, maximum temperature, sunshine time, wind direction, and solar radiation were the most important input parameters. The values of R, MAE, RMSE, and d2 did not change considerably for the MLP model using only these nine inputs. The performances of the MLP models were compared with those of regression models (i.e., multiple linear regression and multiple non‐linear regression). It has been found that there was no significant difference between the ANN and regression modeling techniques for the forecasting of ozone concentrations in Istanbul.  相似文献   

10.
Hydrological processes of lowland watersheds of the southern USA are not well understood compared to a hilly landscape due to their unique topography, soil compositions, and climate. This study describes the seasonal relationships between rainfall patterns and runoff (sum of storm flow and base flow) using 13 years (1964–1976) of rainfall and stream flow data for a low‐gradient, third‐order forested watershed. It was hypothesized that runoff–rainfall ratios (R/P) are smaller during the dry periods (summer and fall) and greater during the wet periods (winter and spring). We found a large seasonal variability in event R/P potentially due to differences in forest evapotranspiration that affected seasonal soil moisture conditions. Linear regression analysis results revealed a significant relationship between rainfall and runoff for wet (r2 = 0·68; p < 0·01) and dry (r2 = 0·19; p = 0·02) periods. Rainfall‐runoff relationships based on a 5‐day antecedent precipitation index (API) showed significant (r2 = 0·39; p < 0·01) correspondence for wet but not (r2 = 0·02; p = 0·56) for dry conditions. The same was true for rainfall‐runoff relationships based on 30‐day API (r2 = 0·39; p < 0·01 for wet and r2 = 0·00; p = 0·79 for dry). Stepwise regression analyses suggested that runoff was controlled mainly by rainfall amount and initial soil moisture conditions as represented by the initial flow rate of a storm event. Mean event R/P were higher for the wet period (R/P = 0·33), and the wet antecedent soil moisture condition based on 5‐day (R/P = 0·25) and 30‐day (R/P = 0·26) prior API than those for the dry period conditions. This study suggests that soil water status, i.e. antecedent soil moisture and groundwater table level, is important besides the rainfall to seasonal runoff generation in the coastal plain region with shallow soil argillic horizons. Copyright © 2011 John Wiley & Sons, Ltd.  相似文献   

11.
S. C. Bird 《水文研究》1987,1(4):321-338
Suspended solids contamination caused by runoff below a working colliery in the Upper Clydach catchment in South Wales, U.K., was investigated in relation to hydrological controls. Field studies over a 16 month period found that concentrations below the colliery ranged from 4 to 8028 mg 1?1. Simple correlation and linear regression analysis of spot and storm event samples taken below the colliery gave a correlation coefficient of 0·39 between flow and suspended solids concentration. Because of the lack of explained variance, a multiple linear regression model of within-storm concentrations was derived using four selected independent variables. X1 the time relation of the sample to the storm peak; log X2 the stormflow at the time of sampling; log X3 the baseflow at the time of sampling; and log X4 an index of the storm intensity. Analysis of the entire dataset gave an R2 of 0·34. When the results from three atypical events were excluded however, the R2 value improved to 0·65. Beta coefficients indicated that rising limb conditions (X1) and intense storms (log X4) along with dry antecedent conditions (log X3) represent the worst combination of hydrological factors for producing suspended solids contamination.  相似文献   

12.
Özgür Kişi 《水文研究》2008,22(20):4142-4152
This paper proposes the application of a neuro‐wavelet technique for modelling monthly stream flows. The neuro‐wavelet model is improved by combining two methods, discrete wavelet transform and multi‐layer perceptron, for one‐month‐ahead stream flow forecasting and results are compared with those of the single multi‐layer perceptron (MLP), multi‐linear regression (MLR) and auto‐regressive (AR) models. Monthly flow data from two stations, Gerdelli Station on Canakdere River and Isakoy Station on Goksudere River, in the Eastern Black Sea region of Turkey are used in the study. The comparison results revealed that the suggested model could increase the forecast accuracy and perform better than the MLP, MLR and AR models. Copyright © 2008 John Wiley & Sons, Ltd.  相似文献   

13.
Prem B. Parajuli 《水文研究》2010,24(26):3785-3797
The climatic processes such as changes in precipitation, temperature and atmospheric CO2 concentration can intensify the effects on water resources. An assessment of the effects of long‐term climate change on water resources is essential to the development of water quality improvement programs. This study was conducted in the Upper Pearl River Watershed (UPRW) in east‐central Mississippi to assess the effects of long‐term potential future climate change on average mean monthly stream flow from the five spatially distributed U. S. Geological Survey (USGS) gage stations in the UPRW using the Soil and Water Assessment Tool. The model was calibrated (January 1981 to December 1994) and validated (January 1995 to September 2008) using monthly measured stream flow data. The calibrated and validated model determined good to very good performance for stream flow prediction (R2 and E from 0·60 to 0·86) between measured and predicted stream flow values. The root mean square error values (from 14 to 37 m3 s?1) were estimated at similar levels of errors during model calibration and validation. The results showed that long‐term (50 years) average monthly stream flow sensitivity due to climate change effects was found the greatest as a result of percentage change in the precipitation followed by carbon dioxide (CO2) concentration and temperature. The long‐term model simulation scenarios as compared with the base scenario for all five spatially distributed USGS gage stations in the UPRW estimated an average monthly stream flow decrease (from 54 to 67%) and average monthly stream flow increase (from 67 to 79%) depending on the spatial characteristics of the USGS gage stations. Overall, the results indicate that the UPRW hydrology is very sensitive to potential future climate changes and that these changes could stimulate increased streamflow generation from the watershed. Copyright © 2010 John Wiley & Sons, Ltd.  相似文献   

14.
Autoregressive neural network (AR-NN) models of various orders have been generated in this work for the daily total ozone (TO) time series over Kolkata (22.56°N, 88.5°E). Artificial neural network in the form of multilayer perceptron (MLP) is implemented in order to generate the AR-NN models of orders varying from 1 to 13. An extensive variable selection method through multiple linear regression (MLR) is implemented while developing the AR-NNs. The MLPs are characterized by sigmoid non-linearity. The optimum size of the hidden layer is identified in each model and prediction are produced by validating it over the test cases using the coefficient of determination (R 2) and Willmott’s index (WI). It is observed that AR-NN model of order 7 having 6 nodes in the hidden layer has maximum prediction capacity. It is further observed that any increase in the orders of AR-NN leads to less accurate prediction.  相似文献   

15.
This study simulates how spatial variations in particle‐size emissions from a playa affect bulk and size‐resolved dust concentration profiles during two contrasting wind erosion events (a small local and a large regional event) in the Channel Country, Lake Eyre Basin, Australia. The regional event had higher dust concentration as a result of stronger frontal winds and higher erodibility across the playa. For each event, two emission scenarios are simulated to determine if measured size‐resolved dust concentration profiles can be explained by spatial variability in source area emissions. The first scenario assumes that particle‐size emissions from source areas occur at a uniform rate, while the second scenario assumes that particle‐size emissions vary between and within source areas. The uniform emission scenario, reproduced measured bulk dust concentration profiles (R2 = 0·93 regional and R2 = 0·81 local), however simulated size‐resolved dust concentration profiles had poor statistical fits to measured size‐resolved profiles for each size class (the highest were R2 = 0·5 regional and R2 = 0·3 local). For the differential particle‐size emission scenario, the fit to the measured bulk dust concentration profiles is improved (R2 = 0·97 regional and R2 = 0·83 local). However, the fit to the size‐resolved profiles improved dramatically, with the lowest being R2 = 0·89 (regional) and R2 = 0·80 (local). Particle‐size emission models should therefore be tested against both bulk and size‐resolved dust concentration profiles, since if only bulk dust concentration profiles are used model performance may be over‐stated. As the source areas in the first 90 m upwind of the tower were similar for both events, the percentage contributions of each particle‐size class to total emissions can be compared. The contribution of each particle‐size class was similar even though the wind speed, turbulence and dust concentrations were significantly different; suggesting that the contribution of each particle‐size to the total emitted dusts is not related to wind speed and turbulence. Copyright © 2012 John Wiley & Sons, Ltd.  相似文献   

16.
A process‐based, spatially distributed hydrological model was developed to quantitatively simulate the energy and mass transfer processes and their interactions within arctic regions (arctic hydrological and thermal model, ARHYTHM). The model first determines the flow direction in each element, the channel drainage network and the drainage area based upon the digital elevation data. Then it simulates various physical processes: including snow ablation, subsurface flow, overland flow and channel flow routing, soil thawing and evapotranspiration. The kinematic wave method is used for conducting overland flow and channel flow routing. The subsurface flow is simulated using the Darcian approach. The energy balance scheme was the primary approach used in energy‐related process simulations (snowmelt and evapotranspiration), although there are options to model snowmelt by the degree‐day method and evapotranspiration by the Priestley–Taylor equation. This hydrological model simulates the dynamic interactions of each of these processes and can predict spatially distributed snowmelt, soil moisture and evapotranspiration over a watershed at each time step as well as discharge in any specified channel(s). The model was applied to Imnavait watershed (about 2·2 km2) and the Upper Kuparuk River basin (about 146 km2) in northern Alaska. Simulated results of spatially distributed soil moisture content, discharge at gauging stations, snowpack ablations curves and other results yield reasonable agreement, both spatially and temporally, with available data sets such as SAR imagery‐generated soil moisture data and field measurements of snowpack ablation, and discharge data at selected points. The initial timing of simulated discharge does not compare well with the measured data during snowmelt periods mainly because the effect of snow damming on runoff was not considered in the model. Results from the application of this model demonstrate that spatially distributed models have the potential for improving our understanding of hydrology for certain settings. Finally, a critical component that led to the performance of this modelling is the coupling of the mass and energy processes. Copyright © 2000 John Wiley & Sons, Ltd.  相似文献   

17.
The spatial scale effect on sediment concentration in runoff has received little attention despite numerous studies on sediment yield or sediment delivery ratio in the context of multiple spatial scales. We have addressed this issue for hilly areas of the Loess Plateau, north China where fluvial processes are mainly dominated by hyperconcentrated flows. The data on 717 flow events observed at 17 gauging stations and two runoff experimental plots, all located in the 3906 km2 Dalihe watershed, are presented. The combination of the downstream scour of hyperconcentrated flows and the downstream dilution, which is mainly caused by the base flow and is strengthened as a result of the strong patchy storms, determines the spatial change of sediment concentration in runoff during flood events. At the watershed scale, the scouring effect takes predominance first but is subordinate to the downstream dilution with a further increase in spatial scale. As a result, the event mean sediment concentration first increases following a power function with drainage basin area and then declines at the drainage basin area of about 700 km2. The power function in combination with the proportional model of the runoff‐sediment yield relationship we proposed before was used to establish the sediment‐yield model, which is neither the physical‐based model nor the regression model. This model, with only two variables (runoff depth and drainage basin area) and two parameters, can provide fairly accurate prediction of event sediment yield with model efficiency over 0·95 if small events with runoff depth lower than 1 mm are excluded. Copyright © 2011 John Wiley & Sons, Ltd.  相似文献   

18.
The quantification of debris‐flow hazard requires estimates of debris‐flow frequency and magnitude. Several methods have been proposed to determine the probable volume of future debris flows from a given basin, but most have neglected to account for debris recharge rates over time, which may lead to underestimation of debris‐flow volumes in basins with rare debris flows. This paper deals with the determination of debris recharge rates in debris‐flow channels based on knowledge of debris storage and the elapsed time since the last debris flow. Data are obtained from coastal British Columbia and a relation is obtained across a sample of basins with similar terrain and climatic conditions. For Rennell Sound on the west coast of the Queen Charlotte Islands, the power‐law relation for area‐normalized recharge rate, Rt, versus elapsed time, te was Rt = 0·23te?0·58 with an explained variance of 75 per cent. A difference in recharge rates may exist between creeks in logged and unlogged forested terrain. The power function for undisturbed terrain was Rt = 0·20te?0·49, while the function for logged areas was Rt = 0·30te?0·77. This result suggests that for the same elapsed time since the last debris flow, clearcut gullies tend to recharge at a slower rate than creeks in old growth forest. This finding requires verification, particularly for longer elapsed times since debris flow, but would have important implications for forest resource management in steep coastal terrain. This study demonstrates that commonly used encounter probability equations are inappropriate for recharge‐limited debris flow channels. Copyright © 2005 John Wiley & Sons, Ltd.  相似文献   

19.
The present study aims to develop a hybrid multi‐model using the soft computing approach. The model is a combination of a fuzzy logic, artificial neural network (ANN) and genetic algorithm (GA). While neural networks are low‐level computational structures that perform well dealing with raw data, fuzzy logic deal with reasoning on a higher level by using linguistic information acquired from domain experts. However, fuzzy systems lack the ability to learn and cannot adjust themselves to a new environment. Moreover, experts occasionally make mistakes and thus some rules used in a system may be false. A network type structure of the present hybrid model is a multi‐layer feed‐forward network, the main part is a fuzzy system based on the first‐order Sugeno fuzzy model with a fuzzification and a defuzzification processes. The consequent parameters are determined by least square method. The back‐propagation is applied to adjust weights of network. Then, the antecedent parameters of the membership function are updated accordingly by the gradient descent method. The GA was applied to select the fuzzy rule. The hybrid multi‐model was used to forecast the flood level at Chiang Mai (under the big flood 2005) and the Koriyama flood (2003) in Japan. The forecasting results are evaluated using standard global goodness of fit statistic, efficient index (EI), the root mean square error (RMSE) and the peak flood error. Moreover, the results are compared to the results of a neuro‐genetic model (NGO) and ANFIS model using the same input and output variables. It was found that the hybrid multi‐model can be used successfully with an efficiency index (EI) more than 0·95 (for Chiang Mai flood up to 12 h ahead forecasting) and more than 0·90 (for Koriyama flood up to 8 h ahead forecasting). In general, all of three models can predict the water level with satisfactory results. However, the hybrid model gave the best flood peak estimation among the three models. Therefore, the use of fuzzy rule base, which is selected by GA in the hybrid multi‐model helps to improve the accuracy of flood peak. Copyright © 2009 John Wiley & Sons, Ltd.  相似文献   

20.
Growing interest in the use of artificial neural networks (ANNs) in rainfall‐runoff modelling has suggested certain issues that are still not addressed properly. One such concern is the use of network type, as theoretical studies on a multi‐layer perceptron (MLP) with a sigmoid transfer function enlightens certain limitations for its use. Alternatively, there is a strong belief in the general ANN user community that a radial basis function (RBF) network performs better than an MLP, as the former bases its nonlinearities on the training data set. This argument is not yet substantiated by applications in hydrology. This paper presents a comprehensive evaluation of the performance of MLP‐ and RBF‐type neural network models developed for rainfall‐runoff modelling of two Indian river basins. The performance of both the MLP and RBF network models were comprehensively evaluated in terms of their generalization properties, predicted hydrograph characteristics, and predictive uncertainty. The results of the study indicate that the choice of the network type certainly has an impact on the model prediction accuracy. The study suggests that both the networks have merits and limitations. For instance, the MLP requires a long trial‐and‐error procedure to fix the optimal number of hidden nodes, whereas for an RBF the structure of the network can be fixed using an appropriate training algorithm. However, a judgment on which is superior is not clearly possible from this study. Copyright © 2004 John Wiley & Sons, Ltd.  相似文献   

设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号