Search

This paper addresses the difficult question of how to perform meaningful comparisons between neural network-based hydrological models and alternative modelling approaches. Standard, goodness-of-fit metric approaches are limited since they only assess numerical performance and not physical legitimacy of the means by which output is achieved. Consequently, the potential for general application or catchment transfer of such models is seldom understood. This paper presents a partial derivative, relative sensitivity analysis method as a consistent means by which the physical legitimacy of models can be evaluated. It is used to compare the behaviour and physical rationality of a generalised linear model and two neural network models for predicting median flood magnitude in rural catchments. The different models perform similarly in terms of goodness-of-fit statistics, but behave quite distinctly when the relative sensitivities of their inputs are evaluated. The neural solutions are seen to offer an encouraging degree of physical legitimacy in their behaviour, over that of a generalised linear modelling counterpart, particularly when overfitting is constrained. This indicates that neural models offer preferable solutions for transfer into ungauged catchments. Thus, the importance of understanding both model performance and physical legitimacy when comparing neural models with alternative modelling approaches is demonstrated. [ABSTRACT FROM AUTHOR]

Artificial neural networks (ANNs) are very effective statistical models for (1) extracting significant features or characteristics from complex data structures and/or for (2) learning nonlinear relationships involved in any input-output mapping. Another interesting aspect of ANN modeling is the fact that overall performance of these models is not greatly hampered by the presence of error-corrupted values in some input nodes. ANNs have gained interest in remote sensing applications as valuable inverse models that can retrieve physical characteristics of interest, such as precipitation, from remote sensing measurements collected from radars or satellites. The spatial coverage and high resolution of remote sensing measurements relative to ground-based measurements can improve the hydrological modeling of the water cycle at both local and global scales. This review paper intends to present recent advances in artificial neural network modeling of remote sensing applications in hydrology. This paper focuses on precipitation and snow water equivalent (SWE) retrievals from remote sensing data. [ABSTRACT FROM AUTHOR]

In this paper two models are presented based on Data-Driven Modeling (DDM) techniques (Artificial Neural Network and neuro-fuzzy systems) for more comprehensive and more accurate prediction of the pipe failure rate and an improved assessment of the reliability of pipes. Furthermore, a multivariate regression approach has been developed to enable comparison with the DDM-based methods. Unlike the existing simple regression models for prediction of pipe failure rates in which only few factors of diameter, age and length of pipes are considered, in this paper other parameters such as pressure and pipe depth, are also included. Furthermore, an investigation is carried out on most commonly used mechanical reliability relationships and the results of incorporation of the proposed pipe failure models in the reliability index are compared. The proposed models are applied to a real case study involving a large water distribution network in Iran and the results of model predictions are compared with measured pipe failure data. Compared with the results of neuro-fuzzy and multivariate regression models, the outcomes of the artificial neural network model are more realistic and accurate in the prediction of pipe failure rates and evaluation of mechanical reliability in water distribution networks. [ABSTRACT FROM AUTHOR]

This paper describes an application of SVM over the Bird Creek catchment and addresses some important issues in developing and applying SVM in flood forecasting. It has been found that, like artificial neural network models, SVM also suffers from over-fitting and under-fitting problems and the over-fitting is more damaging than under-fitting. This paper illustrates that an optimum selection among a large number of various input combinations and parameters is a real challenge for any modellers in using SVMs. A comparison with some benchmarking models has been made, i.e. Transfer Function, Trend and Naive models, It demonstrates that SVM is able to surpass all of them in the test data series, at the expense of a huge amount of time and effort. Unlike previous published results, this paper shows that linear and nonlinear kernel functions (i.e. RBF) can yield superior performances against each other under different circumstances in the same catchment. The study also shows an interesting result in the SVM response to different rainfall inputs, where lighter rainfalls would generate very different responses to heavier ones, which is a very useful way to reveal the behaviour of a SVM model. [ABSTRACT FROM AUTHOR]

This paper presents a new approach for the real-time, near-optimal control of water-distribution networks, which forms an integral part of the POWADIMA research project. The process is based on the combined use of an artificial neural network for predicting the consequences of different control settings and a genetic algorithm for selecting the best combination. By this means, it is possible to find the optimal, or at least near-optimal, pump and valve settings for the present time-step as well as those up to a selected operating horizon, taking account of the short-term demand fluctuations, the electricity tariff structure and operational constraints such as minimum delivery pressures, etc. Thereafter, the near-optimal control settings for the present time-step are implemented. Having grounded any discrepancies between the previously predicted and measured storage levels at the next update of the monitoring facilities, the whole process is repeated on a rolling basis and a new operating strategy is computed. Contingency measures for dealing with pump failures, pipe bursts, etc., have also been included. The novelty of this approach is illustrated by the application to a small, hypothetical network. Its relevance to real networks is discussed in the subsequent papers on case studies. [ABSTRACT FROM AUTHOR]

As part of the POWADIMA research project, this paper describes the technique used to predict the consequences of different control settings on the performance of the water-distribution network, in the context of real-time, near-optimal control. Since the use of a complex hydraulic simulation model is somewhat impractical for real-time operations as a result of the computational burden it imposes, the approach adopted has been to capture its domain knowledge in a far more efficient form by means of an artificial neural network (ANN). The way this is achieved is to run the hydraulic simulation model off-line, with a large number of different combinations of initial tank-storage levels, demands, pump and valve settings, to predict future tank-storage water levels, hydrostatic pressures and flow rates at critical points throughout the network. These input/output data sets are used to train an ANN, which is then verified using testing sets. Thereafter, the ANN is employed in preference to the hydraulic simulation model within the optimization process. For experimental purposes, this technique was initially applied to a small, hypothetical water-distribution network, using EPANET as the hydraulic simulation package. The application to two real networks is described in subsequent papers of this series. [ABSTRACT FROM AUTHOR]

Considering the complexity of reservoir systems, a model fusion approach is proposed in this paper. According to different inflow information represented, the historical monthly data can be constructed as two time series, namely, yearly-scale series and monthly-scale series. Even grey model (EGM) and adaptive neuro-fuzzy inference system (ANFIS) are adopted for the forecasts at the two scales, respectively. Grey relational analysis (GRA) is subsequently used as a scale-normalized model fusion tool to integrate the two scales' results. The proposed method is evaluated using the data of the Three Gorges reservoir ranging from January 2000 to December 2012. The forecast performances of the individual-scale models are improved substantially by the suggested method. For comparison, two peer models, back-propagation neural network and autoregressive integrated moving average model, are also involved. The results show that, having combined together the small-sample forecast ability of the EGM in the yearly-scale, the nonlinearity of the ANFIS in the monthly-scale, and the grey fusion capability of the GRA, the present approach is more accurate for holistic evaluation than those models in terms of mean absolute percentage error, normalized root-mean-square error, and correlation coefficient criteria, and also for peak inflow forecasting in accordance with peak percent threshold statistics. [ABSTRACT FROM AUTHOR]

Applying feed-forward neural networks has been limited due to the use of conventional gradientbased slow learning algorithms in training and iterative determination of network parameters. This paper demonstrates a method that partly overcomes these problems by using an extreme learning machine (ELM) which predicts the hydrological time-series very quickly. ELMs, also called singlehidden layer feed-forward neural networks (SLFNs), are able to well generalize the performance for extremely complex problems. ELM randomly chooses a single hidden layer and analytically determines the weights to predict the output. The ELM method was applied to predict hydrological flow series for the Tryggevælde Catchment, Denmark and for the Mississippi River at Vicksburg, USA. The results confirmed that ELM's performance was similar or better in terms of root mean square error (RMSE) and normalized root mean square error (NRMSE) compared to ANN and other previously published techniques, namely evolutionary computation based support vector machine (EC-SVM), standard chaotic approach and inverse approach. [ABSTRACT FROM AUTHOR]

Predicting shear stress distribution has proved to be a critical problem to solve. Hence, the basic objective of this paper is to develop a prediction of shear stress distribution by machine learning algorithms including artificial neural networks, classification and regression tree, generalized linear models. The data set, which is large and feature-rich, is utilized to improve machine learning-based predictive models and extract the most important predictive factors. The 10-fold cross-validation approach was used to determine the performances of prediction methods. The predictive performances of the proposed models were found to be very close to each other. However, the results indicated that the artificial neural network, which has the R value of 0.92 ± 0.03, achieved the best classification performance overall accuracy on the 10-fold holdout sample. The predictions of all machine learning models were well correlated with measurement data. [ABSTRACT FROM AUTHOR]

A fast and accurate classification method for sewage sludge biological activity classification is of great significance for wastewater treatment. However, the data are often imbalanced and the accuracy of traditional classification algorithms applied to imbalanced small classes of data is very low. Such small classes are crucial application data. Therefore, based on the analysis of eight microorganisms, a novel method is proposed in this paper for the classification of activated sludge known as balanced support-vector-based back-propagation (SV-BP) neural network. It first splits the multiclass classification problem into a plurality of pairwise classification problems and uses a support vector machine (SVM) to achieve equalization. Second, the new dataset is produced, following which back-propagation neural network (BPNN) is used for training and classification. To examine the efficiency of the model, 1731 real data points are collected from a wastewater treatment factory and divide the data into four classes with the help of wastewater experts. Based on the new model, data redundancy and noise are greatly reduced. With area under the curve (AUC) measurements, we find that the AUC of SV-BP is 6.9% higher than classical BPNN. In addition, the small-class recognition rate of SV-BP is far better than that by classical BPNN and SVM algorithms. [ABSTRACT FROM AUTHOR]

Due to limited water resources and the increasing demand for agricultural products, it is significantly important to operate surface water reservoirs optimally, especially those located in arid and semi-arid regions. This paper investigates uncertainty-based optimal operation of a multi-purpose water reservoir system by using four optimization models. The models include dynamic programming (DP), stochastic DP (SDP) with inflow classification (SDP/Class), SDP with inflow scenarios (SDP/Scenario), and sampling SDP (SSDP) with historical scenarios (SSDP/Hist). The performance of the models was tested in Zayandeh-Rud Reservoir system in Iran by evaluating how their release policies perform in a simulation phase. While the SDP approaches were better than the DP approach, the SSDP/Hist model outperformed the other SDP models. We also assessed the effect of ensemble streamflow predictions (ESPs) that were generated by artificial neural networks on the performance of SSDP/Hist. Application of the models to the Zayandeh-Rud case study demonstrated that SSDP in combination with ESPs and the K-means technique, which was used to cluster a large number of ESPs, could be a promising approach for real-time reservoir operation. [ABSTRACT FROM AUTHOR]

Water distribution systems, and other infrastructures, are increasingly being pervaded by sensing technologies, collecting a growing volume of data aimed at supporting operational and investment decisions. These sensors monitor system characteristics, i.e. flows, pressures and water quality, such as in pipes. This paper presents the application of pattern matching techniques and binary associative neural networks for novelty detection in such data. A protocol for applying pattern matching to automatically recognise specific waveforms in time series based on their shapes is described together with a system called Advanced Uncertain Reasoning Architecture (AURA) Alert for autonomous determination of novelty. AURA is a class of binary neural network that has a number of advantages over standard artificial neural network techniques for condition monitoring including a sound theoretical basis to determine the bounds of the system operation. Results from application to several case studies are provided including both hydraulic and water quality data. In the case of pattern matching, the results demonstrated some transferability of burst patterns across District Metered Areas; however limitations in performance and difficulties with assembling pattern libraries were found. Results for the AURA system demonstrate the potential for robust event detection across multiple parameters providing valuable information for diagnosis; one example also demonstrates the potential for detection of precursor information, vital for proactive management. [ABSTRACT FROM AUTHOR]

In rainfall-runoff modeling, the wavelet-ANN model, which includes a wavelet transform to capture multi-scale features of the process, as well as an artificial neural network (ANN) to predict the runoff discharge, is a beneficial approach. One of the essential steps in any ANN-based development process is determination of dominant input variables. This paper presents a two-stage procedure to model the rainfall-runoff process of the Delaney Creek and Payne Creek Basins, Florida, USA. The two-stage procedure includes data pre-processing and model building stages. In the data preprocessing stage, a wavelet transform is used to decompose the rainfall and runoff time series into several sub-series at different scales. Subsequently, independent sub-series are chosen via a self-organizing map (SOM). In the model building stage, selected sub-series are imposed as input data to a feed-forward neural network (FFNN) to forecast runoff discharge. To make a better interpretation of the model efficiency, the proposed model is compared with the Auto Regressive Integrated Moving Average with eXogenous input (ARIMAX) and with the ad hoc FFNN methods, without any data pre-processing. The results proved that the proposed model leads to better outcome especially in term of determination coefficient for detecting peak points (DCpeak). [ABSTRACT FROM AUTHOR]

Accurate and reliable forecasts of key hydrological variables such as stream flow are of importance due to their profound impacts on real world water resources applications. Data-driven methods have proven their applicability to modeling complex and non-linear hydrological processes. This paper presents a novel visual modeling system that has been developed to overcome the problems involved in implementation of data-driven models for hydrological forecasts using conventional programming languages: problems such as the effort and skill needed to program the models, the lack of reusability of existing models, and the lack of shared tools to perform tedious tasks such as preprocessing data. The system provides an integrated visual modeling environment within which users are able to graphically design and verify specific forecasting models for particular problems without writing code. A set of popular data-driven models are offered by the system. Plug-in models created by wrapping existing code are also allowed to run within the system due to the system's open architecture. The system's feasibility and capability is demonstrated through a case study of forecasting 1-day ahead flow in a river basin located in China. The encouraging simulation results show that the system can simplify the process of implementing hydrological forecast. [ABSTRACT FROM AUTHOR]

There are several statistical downscaling methods available for generating local-scale meteorological variables from large-scale model outputs. There is still no universal single method, or group of methods, that is clearly superior, particularly for downscaling daily precipitation. This paper compares different statistical methods for downscaling daily precipitation from numerical weather prediction model output. Three different methods are considered: (i) hybrids; (ii) neural networks; and (iii) nearest neighbor-based approaches. These methods are implemented in the Saguenay watershed in northeastern Canada. Suites of standard diagnostic measures are computed to evaluate and inter-compare the performances of the downscaling models. Although results of the downscaling experiment show mixed performances, clear patterns emerge with respect to the reproduction of variation in daily precipitation and skill values. Artificial neural network-logistic regression (ANN-Logst), partial least squares (PLS) regression and recurrent multilayer perceptron (RMLP) models yield greater skill values, and conditional resampling method (SDSM) and K-nearest neighbor (KNN)-based models show the potential to capture the variability in daily precipitation. [ABSTRACT FROM AUTHOR]

Prediction of bridge pier scour depth is essential for safe and economical bridge design. Keeping in mind the complex nature of bridge scour phenomenon, there is a need to properly address the methods and techniques used to predict bridge pier scour. Up to the present, extensive research has been carried out for pier scour depth prediction. Different modeling techniques have been applied to achieve better prediction. This paper presents a new soft computing technique called gene-expression programming (GEP) for pier scour depth prediction using laboratory data. A functional relationship has been established using GEP and its performance is compared with other artificial intelligence (AI)-based techniques such as artificial neural networks (ANNs) and conventional regression-based techniques. Laboratory data containing 529 datasets was divided into calibration and validation sets. The performance of GEP was found to be highly satisfactory and encouraging when compared to regression equations but was slightly inferior to ANN. This slightly inferior performance of GEP compared to ANN is offset by its capability to provide compact and explicit mathematical expression for bridge scour. This advantage of GEP over ANN is the main motivation for this work. The resulting GEP models will add to the existing literature of AI-based inductive models for bridge scour modeling. [ABSTRACT FROM AUTHOR]

The maximum depth of scouring around spur dikes plays an important role in the hydraulic design process. There have been many studies on the maximum depth of scouring, but there is little information available on the time variation of scour depth. In this paper, the time variation of scouring around the first spur dike in a series was investigated experimentally. Experiments were carried out in four different bed materials under different flow intensities (U/Ucr ). To achieve a time development of scouring around the first spur dike, more than 750 sets of experimental data were collected. The results showed that 70-90% of the equilibrium scour depths were occurring during the initial 20% of the overall time of scouring. Based on the data analysis, a regression model and artificial neural networks (ANNs) were developed. The models were compared with other empirical equations in the literature. However, the results showed that the developed regression model is quite accurate and more practical, but the ANN models by feed forward back propagation and radial basis function provide a better prediction of observation. Finally, by sensitivity analysis, the most and the least effective parameters, which affected time variation of scouring, were determined. [ABSTRACT FROM AUTHOR]

The prediction of groundwater levels in a well has immense importance in the management of groundwater resources, especially in arid regions. This paper investigates the abilities of neuro-fuzzy (NF) and artificial neural network (ANN) techniques to predict the groundwater levels. Two different NF and ANN models comprise various combinations of monthly variablities, that is, air temperature, rainfall and groundwater levels in neighboring wells. The result suggests that the NF and ANN techniques are a good choice for the prediction of groundwater levels in individual wells. Also based on comparisons, it is found that the NF computing techniques have better performance than the ANN models in this case. [ABSTRACT FROM AUTHOR]

Missing values are a common problem faced in the analysis of hydrometric data. The need for complete hydrological data, especially hydrometric data for planning, development and designing hydraulic structures, has become increasingly important. Reasonably estimating these missing values is significant for the complete analysis and modeling of the hydrological cycle. The major objective of this paper is to estimate the missing annual maximum hydrometric data by using artificial neural networks (ANN). Sixteen stations, with 28 years of measurements, in the catchment area of the Sefdroud watershed in the north of Iran were selected for this investigation. Comparison between the results of ANN and the nonlinear regression method (NLR) illustrated the efficiency of artificial neural networks and their ability to rebuild the missing data. According to the coefficient of determination (R²) and the root mean squared value of error (RMSE), it was concluded that ANN provides a better estimation of the missing data. [ABSTRACT FROM AUTHOR]

A promising new approach for eco-environmental modelling, such as algal growth prediction, is the data-driven modeling using machine learning techniques: an artificial neural network (ANN) being a typical method. Another method growing in popularity, based on the M5 model tree (MT) algorithm, is the use of piecewise linear regression models at the leaf nodes of the tree. M5 MTs using partial least-squares regression (PLSR) proposed in this paper were tested on a particular dataset and then compared to M5 MTs, MLF- and RBF-ANN and k nearest neighbours (kNN). With the dataset partitioned to periods of algal growth and no growth, M5 MTs using PLSR showed better results for algal growth prediction in the reservoir than using the annual dataset and other algorithms. This gives the idea that the M5-PLSR MTs, in spite of the lack of data, more effectively seeks latent vectors between the closely correlated multivariate dataset partitioned using clustering techniques. M5-PLSR MTs is a promising approach when there is a shortage of data required to build a more transparent learning process model, and a combination with clustering is recommended. [ABSTRACT FROM AUTHOR]

The present paper deals with exploring the use of Artificial Neural Networks (ANN) for forecasting the Blue Nile river flows in Sudan. Four ANN rainfall-runoff models based on the structure of the well-known multi-layer perceptron are developed. These models use the rainfall index as a common external input, with the rainfall index being a weighted sum of the recent and current rainfall. These models differ in terms of the additional external inputs being used by the model. The additional inputs are basically the seasonal expectations of both the rainfall index and the observed discharge. The results show that the model, which uses both the seasonal expectation of the observed discharge and the rainfall index as additional inputs, has the best performance. The estimated discharges of this model are further updated using a non-linear Auto-Regressive Exogenous-input model (NARXM)-ANN river flow forecasting output-updating procedure. In this way, a real-time river flow forecasting model is developed. The results show that the forecast updating has significantly enhanced the quality of the discharge forecasts. The results also indicate that the ANN has considerable potential to be used for river flow forecasting in developing countries. [ABSTRACT FROM AUTHOR]

After a brief review of the mutation from modern to postmodern conditions of society, that is, from societies of knowers to societies of consumers of knowledge, the position of the knowledge provider in hydroinformatics is introduced. The provider's changing role in a society with an ever-increasing emphasis on communication is introduced. The sociotechnical forces that have driven numerical modelling, first from second to third generation, then to the fourth generation, and now into a fifth generation, are explicated. These forces are exemplified by an increasingly active stakeholder participation in projects, joint policy formulation and on-line management. The applications to the increasingly problematic developments in the so-called 'Third World' are then considered. The overall consequences for the future of numerical modelling practice are then explored. The role of open-source software developments interacting with proprietorial software, together with the development and introduction of service-oriented architectures are explained. The concept of the knowledge supply chain is then advanced to introduce the concept of the extended halo of the global knowledge provider. This paper concludes with an analysis of appropriate and less appropriate business models. [ABSTRACT FROM AUTHOR]

The increasing demand for uncertainty assessment in stream-flow forecasts has drawn the hydrological community's interest toward ensemble forecasting techniques. The widespread deterministic hydrological forecasting point of view focuses to a great extent on the search for a hydrological model that would come as close as possible to "perfection" (i.e. the aim is to implement a model that produces a point forecast that is as close as possible as the observed outcome). On the other hand, ensemble forecasting departs from the deterministic point of view by avoiding the assumption that the "perfect" model exists and instead focuses on issuing a type of forecast that accounts explicitly for the uncertainty inherent to the forecasting process as a whole. In this paper, one-day-ahead hydrological ensemble forecasts obtained by stacked neural networks are presented and analysed. To do so, three simple performance assessment criteria are presented. Those criteria were originally developed in the meteorological and statistical communities to accommodate the need for a quality assessment methodology that is coherent with the probabilistic nature of ensemble weather forecasts. It will be shown that, even though the ensemble forecasts suffer from underdispersion, they outperform point forecasts. [ABSTRACT FROM AUTHOR]

This paper provides a general overview about the use of fuzzy inference systems in the important field of river flow forecasting. It discusses the overall operation of the main two types of fuzzy inference systems, namely Mamdani and Takagi-Sugeno-Kang fuzzy inference systems, and the critical issues related to their application. A literature review of existing studies dealing with the use of fuzzy inference systems in river flow forecasting models is presented, followed by some recommendations for future research areas. This review shows that fuzzy inference systems can be used as effective tools for river flow forecasting, even though their application is rather limited in comparison to the popularity of neural networks models. In addition to this, it was found that there are several unresolved issues requiring further attention before more clear guidelines for the application of fuzzy inference systems can be given. [ABSTRACT FROM AUTHOR]

Heuristic search techniques are highly flexible, though they represent computationally intensive optimization methods that may require thousands of evaluations of expensive objective functions. This paper integrates MODSIM, a generalized river basin network flow model, a particle swarm optimization (PSO) algorithm and artificial neural networks into a modeling framework for optimum water allocations at basin scale. MODSIM is called in the PSO model to simulate a river basin system operation and to evaluate the fitness of each set of selected design and operational variables with respect to the model's objective function, which is the minimization of the system's design and operational cost. Since the direct incorporation of MODSIM into a PSO algorithm is computationally prohibitive, an ANN model as a meta-model is trained to approximate the MODSIM modeling tool. The resulting model is used in the problem of optimal design and operation of the upstream Sirvan river basin in Iran as a case study. The computational efficiency of the model makes it possible to analyze the model performance through changing its parameters so that better solutions are obtained compared to those of the original PSO-MODSIM model. [ABSTRACT FROM AUTHOR]

Watershed runoff in areas with heavy seasonal snow cover is usually estimated using physically based conceptual hydrologic models. Such simulation models normally require a snowmelt algorithm consisting of a surface energy balance and some accounting of internal snowpack processes to be part of the modeling system. On the other hand, artificial neural networks are flexible mathematical structures that are capable of identifying such complex nonlinear relationships between input and output datasets from historical precipitation, temperature and streamflow records. This paper presents the findings of a study on using a form of time-delayed neural network, namely time-lagged feedforward neural network (TLFN), that implicitly accounts for snow accumulation and snowmelt processes through the use of logical values and tapped delay lines. The logical values (in the form of symbolic inputs) are used to implicitly include seasonal information in the TLFN model. The proposed method has been successfully applied for improved precipitation-runoff modeling of both the Chute-du-Diable reservoir inflows and the Serpent River flows in northeastern Canada where river flows and reservoir inflows are highly influenced by seasonal snowmelt effects. The study demonstrates that the TLFN with logical values is capable of modeling the precipitation-runoff process in a cold and snowy climate by relying on 'logical input values' and tapped delay lines to implicitly recognize the temporal input- output patterns in the historical data. The study results also show that, once the appropriate input patterns are identified, the time-lagged neural network based models performed quite well, especially for spring peak flows, and demonstrated comparable performance in simulating the precipitation-runoff processes to that of a physically based hydrological model, namely HBV. [ABSTRACT FROM AUTHOR]

Physically based (process) models based on mathematical descriptions of water motion are widely used in river basin management. During the last decade the so-called data-driven models are becoming more and more common. These models rely upon the methods of computational intelligence and machine learning, and thus assume the presence of a considerable amount of data describing the modelled system's physics (i.e. hydraulic and/or hydrologic phenomena). This paper is a preface to the special issue on Data Driven Modelling and Evolutionary Optimization for River Basin Management, and presents a brief overview of the most popular techniques and some of the experiences of the authors in data-driven modelling relevant to river basin management. It also identifies the current trends and common pitfalls, provides some examples of successful applications and mentions the research challenges. [ABSTRACT FROM AUTHOR]

Haifa-A is the first of two case studies relating to the POWADIMA research project. It comprises about 20% of the city's water-distribution network and serves a population of some 60,000 from two sources. The hydraulic simulation model of the network has 126 pipes, 112 nodes, 9 storage tanks, 1 operating valve and 17 pumps in 5 discrete pumping stations. The complex energy tariff structure changes with hours of the day and days of the year. For a dynamically rolling operational horizon of 24h ahead, the real-time, near-optimal control strategy is calculated by a software package that combines a genetic algorithm (GA) optimizer with an artificial neural network (ANN) predictor, the latter having replaced a conventional hydraulic simulation model to achieve the computational efficiency required for real-time use. This paper describes the Haifa-A hydraulic network, the ANN predictor, the GA optimizer and the demand- forecasting model that were used. Thereafter, it presents and analyses the results obtained for a full (simulated) year of operation in which an energy cost saving of some 25% was achieved in comparison to the corresponding cost of current practice. Conclusions are drawn regarding the achievement of aims and future prospects. [ABSTRACT FROM AUTHOR]

This paper is intended to serve as an introduction to the POWADIMA research project, whose objective was to determine the feasibility and efficacy of introducing real-time, near-optimal control for water-distribution networks. With that in mind, its contents include the current state-of-the-art and some of the difficulties that would need to be addressed if the goal of near-optimal control was to be achieved, Subsequently, the approach adopted is outlined, together with the reasons for the choice. Since it would be somewhat impractical to use a conventional hydraulic simulation model for real-time, near-optimal control, the methodology includes replicating the model by an artificial neural network which, computationally, is far more efficient. Thereafter, the latter is embedded in a dynamic genetic algorithm, designed specifically for real-time use. In this way, the near-optimal control settings to meet the current demands and minimize the overall pumping costs up to the operating horizon can be derived. The programme of work undertaken in achieving this end is then described. By way of conclusion, the potential benefits arising from implementing the control system developed are briefly reviewed, as are the possibilities of using the same approach for other application areas. [ABSTRACT FROM AUTHOR]

This paper describes a simple mathematical technique that uses a genetic algorithm and least squares optimization to obtain a functional approximation (or computer program) for a given data set. Such an optimal functional form is derived from a pre-defined general functional formulation by selecting optimal coefficients, decision variable functions, and mathematical operators. In the past, functional approximations have routinely been obtained through the use of linear and non-linear regression analysis. More recent methods include the use of genetic algorithms and genetic programming. An example application based on a data set extracted from the commonly used Moody diagram has been used to demonstrate the utility of the proposed method. The purpose of the application was to determine an explicit expression for friction factor and to compare its performance to other available techniques. The example application results in the development of closed form expressions that can be used for evaluating the friction factor for turbulent pipe flow. These expressions compete well in accuracy with other known methods, validating the promise of the proposed method in identifying useful functions for physical processes in a very effective manner. The proposed method is simple to implement and has the ability to generate simple and compact explicit expressions for a given response function. [ABSTRACT FROM AUTHOR]

Support Vector Machines are kernel machines useful for classification and regression problems. In this paper, they are used for non-linear regression of environmental data. From a structural point of view, Support Vector Machines are particular Artificial Neural Networks and their training paradigm has some positive implications. In fact, the original training approach is useful to overcome the curse of dimensionality and too strict assumptions on statistics of the errors in data. Support Vector Machines and Radial Basis Function Regularised Networks are presented within a common structural framework for non-linear regression in order to emphasise the training strategy for support vector machines and to better explain the multi-objective approach in support vector machines' construction. A support vector machine's performance depends on the kernel parameter, input selection and -tube optimal dimension. These will be used as decision variables for the evolutionary strategy based on a Genetic Algorithm, which exhibits the number of support vectors, for the capacity of machine, and the fitness to a validation subset, for the model accuracy in mapping the underlying physical phenomena, as objective functions. The strategy is tested on a case study dealing with groundwater modelling, based on time series (past measured rainfalls and levels) for level predictions at variable time horizons. [ABSTRACT FROM AUTHOR]

A step that should be considered when developing artificial neural network (ANN) models for water resources applications is the selection of an appropriate transformation of the data. In general, the primary motivations for data transformation are: (1) to scale the data so as to be commensurate with the transfer function in the output layer; (2) to standardise each of the variables; (3) to provide a suitable initialization of the ANN; and (4) to modify the distribution of the input variables to provide a better mapping to the outputs. In this paper, five different transformations are investigated in an attempt to improve the ANN's forecasting ability. These are: linear transformation, logarithmic transformation, histogram equalization, seasonal transformation and a transformation to normality. A case study is presented in which each of the ANN models developed using the different transformation techniques is used to forecast salinity in the River Murray at Murray Bridge (South Australia) 14 days in advance. When tested on a validation set from July 1992 to March 1998, the model developed using the linear transformation resulted in the lowest root mean squared forecasting error. This finding further strengthens the claim that the probability distribution of the data does not need to be known to develop effective ANN models. No improvement in the ANN model's forecasting ability was made using the logarithmic, seasonal and normality transformations. The model developed using histogram equalization produced good results for data within the training domain but was not robust on new patterns outside of the calibration range. [ABSTRACT FROM AUTHOR]