An ensemble of neural networks for weather forecasting

This study presents the applicability of an ensemble of artificial neural networks (ANNs) and learning paradigms for weather forecasting in southern Saskatchewan, Canada. The proposed ensemble method for weather forecasting has advantages over other techniques like linear combination. Generally, the output of an ensemble is a weighted sum, which are weight-fixed, with the weights being determined from the training or validation data. In the proposed approach, weights are determined dynamically from the respective certainties of the network outputs. The more certain a network seems to be of its decision, the higher the weight. The proposed ensemble model performance is contrasted with multi-layered perceptron network (MLPN), Elman recurrent neural network (ERNN), radial basis function network (RBFN), Hopfield model (HFM) predictive models and regression techniques. The data of temperature, wind speed and relative humidity are used to train and test the different models. With each model, 24-h-ahead forecasts are made for the winter, spring, summer and fall seasons. Moreover, the performance and reliability of the seven models are then evaluated by a number of statistical measures. Among the direct approaches employed, empirical results indicate that HFM is relatively less accurate and RBFN is relatively more reliable for the weather forecasting problem. In comparison, the ensemble of neural networks produced the most accurate forecasts.     

Weather analysis using ensemble of connectionist learning paradigms

This paper presents a comparative analysis of different connectionist and statistical models for forecasting the weather of Vancouver, Canada. For developing the models, one year's data comprising of daily temperature and wind speed were used. A multi-layered perceptron network (MLPN) and an Elman recurrent neural network (ERNN) were trained using the one-step-secant and Levenberg–Marquardt algorithm. Radial basis function network (RBFN) was employed as an alternative to examine its applicability for weather forecasting. To ensure the effectiveness of neurocomputing techniques, the connectionist models were trained and tested using different datasets. Moreover, ensembles of the neural networks were generated by combining the MLPN, ERNN and RBFN using arithmetic mean and weighted average methods. Subsequently, performance of the connectionist models and their ensembles were compared with a well-established statistical technique. Experimental results obtained have shown RBFN produced the most accurate forecast model compared to ERNN and MLPN. Overall, the proposed ensemble approach produced the most accurate forecast, while the statistical model was relatively less accurate for the weather forecasting problem considered.     

Static, Dynamic, and Hybrid Neural Networks in Forecasting Inflation

The back-propagation neural network (BPN) model has been the most popular form of artificial neural network model used for forecasting, particularly in economics and finance. It is a static (feed-forward) model which has a learning process in both hidden and output layers. In this paper we compare the performance of the BPN model with that of two other neural network models, viz., the radial basis function network (RBFN) model and the recurrent neural network (RNN) model, in the context of forecasting inflation. The RBFN model is a hybrid model with a learning process that is much faster than the BPN model and that is able to generate almost the same results as the BPN model. The RNN model is a dynamic model which allows feedback from other layers to the input layer, enabling it to capture the dynamic behavior of the series. The results of the ANN models are also compared with those of the econometric time series models.     

Nowcasting lightning flash rate and peak wind gusts associated with severe thunderstorms using remotely sensed TRMM-LIS data

The momentous weather hazards during the pre-monsoon season (April–May) over Kolkata (22° 32′ N, 88° 20′ E), India, is mostly due to lightning flashes and surface wind gusts associated with severe thunderstorms. A multi-layer perceptron (MLP) model is developed to forecast the lightning flash rate and peak wind gusts which accompany severe thunderstorms. Meteorological parameters derived from radiosonde weather observations from 1998 to 2009 are taken as input whereas lightning data from the Lightning Imaging Sensor (LIS) and wind gusts from a ground-based observatory are taken as the target output parameters. The skill of the MLP model is compared with the multiple linear regression (MLR) analysis method, and it is observed that the MLP model provides better and more accurate forecasts than the MLR analysis method. The results also reveal that the forecast accuracy is more for surface wind gusts than for the lightning flash rate, both during training and validation of the model. The MLP model forecast is validated with the India Meteorological Department (IMD) weather observations as well as Doppler weather radar and satellite imagery of 2008 and 2009 thunderstorms.     

Fault detection and isolation in robotic manipulators via neural networks: A comparison among three architectures for residual analysis

In this article we discuss artificial neural networks‐based fault detection and isolation (FDI) applications for robotic manipulators. The artificial neural networks (ANNs) are used for both residual generation and residual analysis. A multilayer perceptron (MLP) is employed to reproduce the dynamics of the robotic manipulator. Its outputs are compared with actual position and velocity measurements, generating the so‐called residual vector. The residuals, when properly analyzed, provides an indication of the status of the robot (normal or faulty operation). Three ANNs architectures are employed in the residual analysis. The first is a radial basis function network (RBFN) which uses the residuals of position and velocity to perform fault identification. The second is again an RBFN, except that it uses only the velocity residuals. The third is an MLP which also performs fault identification utilizing only the velocity residuals. The MLP is trained with the classical back‐propagation algorithm and the RBFN is trained with a Kohonen self‐organizing map (KSOM). We validate the concepts discussed in a thorough simulation study of a Puma 560 and with experimental results with a 3‐joint planar manipulator. © 2001 John Wiley & Sons, Inc.     

Importance of satellite-retrieved rain/no-rain information on short-range weather predictions

The impact of assimilating rain (satellite-retrieved rainfall is greater than zero) and no-rain (satellite-retrieved rainfall is equal to zero) information retrieved from the Tropical Rainfall Measuring Mission (TRMM) 3B42 precipitation is assessed during Indian summer monsoon 2013 using the weather research and forecasting (WRF) model. Daily three parallel experiments are performed with and without satellite rainfall assimilation for short-range weather forecasts. Additional two experiments are performed daily to evaluate the sensitivity of cumulus parameterization on the WRF model predictions when precipitations are used for assimilation. Precipitation assimilation improves the 48 h low-level temperature, moisture, and winds predictions. Rainfall prediction is also improved over central India when satellite-retrieved rainfall information are assimilated compared to without rainfall assimilation (CNT) experiments. More improvements are seen in moisture forecasts when the Kain–Fritsch (KF) cumulus convection parameterization scheme is used against the Grell–Devenyi ensemble (GD) scheme, whereas for temperature and wind speed forecasts the Grell convection parameterization scheme performed better over the Indian region. Overall, precipitation assimilation improved the WRF model analysis and subsequent model forecasts compared with without precipitation assimilation experiments. Results show that no-rain observations also have a significant positive impact on short-range weather forecasts.     

Plane-Gaussian artificial neural network

Multilayer perceptrons (MLPs) and radial basis functions networks (RBFNs) have been widely concerned in recent years. In this paper, based on k-plane clustering (kPC) algorithm, we propose a novel artificial network model termed as Plane-Gaussian network to enlarge the arsenal of the neural networks. This network adopts a so-called Plane-Gaussian activation function (PGF) in hidden neurons. Replacing traditional central point of Gaussian radial basis function (RBF) with central hyperplane, PGF forms a band-shaped rather than spheral-shaped receptive field in RBF, which makes PGF able to express its peculiar geometrical characteristics: locality and globality. Importantly, it is also proved that PGF network (PGFN) having one hidden layer is capable of universal approximation. As a universal approximator, PGFN gives an informal way of bridging the gap between MLP and RBFN. The experiments report comparison between training time and classification accuracies on some artificial and UCI datasets and conclude that (1) PGFN runs significantly faster than MLP and (2) PGFN has comparable or better classification performance than MLP and RBFN, especially in subspace-distributed datasets.     

Immunological Algorithm-based Neural Network Learning for Sales Forecasting

This study proposes a variation immunological system (VIS) algorithm with radial basis function neural network (RBFN) learning for function approximation and the exercise of industrial computer (IC) sales forecasting. The proposed VIS algorithm was applied to the RBFN to execute the learning process for adjusting the network parameters involved. To compare the performance of relevant algorithms, three benchmark problems were used to justify the results of the experiment. With better accuracy in forecasting, the trained RBFN can be practically utilized in the IC sales forecasting exercise to make predictions and could enhance business profit.     

Knowledge acquisition using a neural network for a weather forecasting knowledge-based system

Neural network technology is experiencing rapid growth and is receiving considerable attention from almost every field of science and engineering. The attraction is due to the successful application of neural network techniques to several real world problems. Neural networks have not yet found widespread application in weather forecasting. The reason for this has been the difficulty in obtaining suitable weather forecasting data sets. In this paper we describe our experience in applying neural network techniques for acquiring the necessary knowledge to predict the weather conditions of Melbourne City and its suburbs in Australia during a 24 hour period beginning at 9 am local time. The accuracy of forecasts produced by a given forecasting procedure typically varies with factors such as geographical location, season, categories of weather, quality of input data, lead time and validity time. Two types of weather data sets assembled from the archives of the Australian Commonwealth Bureau of Meteorology are used for training the neural network. The results of the experiments are competitive and are discussed.     

Neural network models for group behavior prediction: a case of soccer match attendance

Soccer match attendance is an example of group behavior with noisy context that can only be approximated by a limited set of quantifiable factors. However, match attendance is representative of a wider spectrum of context-based behaviors for which only the aggregate effect of otherwise individual decisions is observable. Modeling of such behaviors is desirable from the perspective of economics, psychology, and other social studies with prospective use in simulators, games, product planning, and advertising. In this paper, we evaluate the efficiency of different neural network architectures as models of context in attendance behavior by comparing the achieved prediction accuracy of a multilayer perceptron (MLP), an Elman recurrent neural network (RNN), a time-lagged feedforward neural network (TLFN), and a radial basis function network (RBFN) against a multiple linear regression model, an autoregressive moving average model with exogenous inputs, and a naive cumulative mean model. We show that the MLP, TLFN, and RNN are superior to the RBFN and achieve comparable prediction accuracy on datasets of three teams from the English Football League Championship, which indicates weak importance of context transition modeled by the TLFN and the RNN. The experiments demonstrate that all neural network models outperform linear predictors by a significant margin. We show that neural models built on individual datasets achieve better performance than a generalized neural model constructed from pooled data. We analyze the input parameter influences extracted from trained networks and show that there is an agreement between nonlinear and linear measures about the most significant attributes.

     

A comparative study of vibrational response based impact force localization and quantification using radial basis function network and multilayer perceptron

Impact force identification from response sensors is important especially when force measurement using force sensor is not possible due to the installation or dynamic characteristic altering problems. For example, the bump-excited impact force acting on vehicle wheel or ship collision on an offshore structure. Among various existing impact identification approaches, neural network based force identification method has received great attention because one does not need to have a system model. Thus, it is less likely to be affected by ill-posed problem that often occurs during the inversion process. So far, previous studies focused on solving the impact force identification problem using only the conventional Multilayer Perceptron (MLP). Thus, there is a room for improvement to find an alternate algorithm that has great advantage over MLP. For this reason, this study proposes Radial Basis Function Network (RBFN) for possible further improvement in impact identification task. A comparative study between these two algorithms was conducted via experimental approach. Impact forces were made on a Perspex plate structure which was designed to produce similar dynamic behavior of a typical vehicle. Impact locations were fixed at four edges of the test rig to simulate impact events at a vehicle's wheels. Time-domain peak-to-peak and peak arrival time features were extracted from accelerometer data to use as network inputs. Few training data were taken in the way that they represent the entire range of magnitudes of all trial impacts made throughout the experiment. In overall, RBFN improved the impact localization and quantification accuracies by decreasing 32.98% and 40.91% error respectively compared to MLP. The improvement was mainly due to the RBFN's strong approximation ability and its superior tolerance to experimental noises/uncertainties.     

A new structure adaptation algorithm for RBF networks and its application

An adaptation algorithm is developed for radial basis function network (RBFN) in this paper. The RBFN is adapted on-line for both model structure and parameters with measurement data. When the RBFN is used to model a non-linear dynamic system, the structure is adapted to model abrupt change of system operating region, while the weights are adapted to model the incipient time varying parameters. Two new algorithms are proposed for adding new centres while the redundant centres are pruned, which is particularly useful for model-based control. The developed algorithm is evaluated by modelling a numerical example and a chemical reactor rig. The performance is compared with a non-adaptive model.     

A Support Vector Machine Based Approach for Forecasting of Network Weather Services

We present forecasting related results using a recently introduced technique called Support Vector Machines (SVM) for measurements of processing, memory, disk space, communication latency and bandwidth derived from Network Weather Services (NWS). We then compare the performance of support vector machines with the forecasting techniques existing in network weather services using a set of metrics like mean absolute error, mean square error among others. The models are used to make predictions for several future time steps as against the present network weather services method of just the immediate future time step. The number of future time steps for which the prediction is done is referred to as the depth of prediction set. The support vector machines forecasts are found to be more accurate and outperform the existing methods. The performance improvement using support vector machines becomes more pronounced as the depth of the prediction set increases. The data gathered is from a production environment (i.e., non-experimental).     

Application of temporal difference learning rules in short‐term traffic flow prediction

Studying dynamic behaviours of a transportation system requires the use of the system mathematical models as well as prediction of traffic flow in the system. Therefore, traffic flow prediction plays an important role in today's intelligent transportation systems. This article introduces a new approach to short‐term daily traffic flow prediction based on artificial neural networks. Among the family of neural networks, multi‐layer perceptron (MLP), radial basis function (RBF) neural network and wavenets have been selected as the three best candidates for performing traffic flow prediction. Moreover, back‐propagation (BP) has been adapted as the most efficient learning scheme in all the cases. It is shown that the coefficients produced by temporal signals improve the performance of the BP learning (BPL) algorithm. Temporal signals provide researchers with a new model of temporal difference BP learning algorithm (TDBPL). The capability and performance of TDBPL algorithm are examined by means of simulation in order to prove that the wavelet theory, with its multi‐resolution ability in comparison to RBF neural networks, is a suitable algorithm in traffic flow forecasting. It is also concluded that despite MLP applications, RBF neural networks do not provide negative forecasts. In addition, the local minimum problems are inevitable in MLP algorithms, while RBF neural networks and wavenet networks do not encounter them.     

Neural-based electricity load forecasting using hybrid of GA and ACO for feature selection

Due to deregulation of electricity industry, accurate load forecasting and predicting the future electricity demand play an important role in the regional and national power system strategy management. Electricity load forecasting is a challenging task because electric load has complex and nonlinear relationships with several factors. In this paper, two hybrid models are developed for short-term load forecasting (STLF). These models use “ant colony optimization (ACO)” and “combination of genetic algorithm (GA) and ACO (GA-ACO)” for feature selection and multi-layer perceptron (MLP) for hourly load prediction. Weather and climatic conditions, month, season, day of the week, and time of the day are considered as load-influencing factors in this study. Using load time-series of a regional power system, the performance of ACO?+?MLP and GA-ACO?+?MLP hybrid models is compared with principal component analysis (PCA)?+?MLP hybrid model and also with the case of no-feature selection (NFS) when using MLP and radial basis function (RBF) neural models. Experimental results and the performance comparison with similar recent researches in this field show that the proposed GA-ACO?+?MLP hybrid model performs better in load prediction of 24-h ahead in terms of mean absolute percentage error (MAPE).     

Predicting daily streamflow using rainfall forecasts,a simple loss module and unit hydrographs: Two Brazilian catchments

The performance of a simple, spatially-lumped, rainfall–streamflow model is compared with that of a more complex, spatially-distributed model. In terms of two model-fit statistics it is shown that for two catchments in Brazil (about 30,000 km² and 34,000 km²) with different flow regimes, the simpler catchment models, which are unit hydrograph-based and require only rainfall, streamflow and air temperature data for calibration, perform about as well as more complex catchment models that require additional information from satellite images and digitized maps of elevation, land-use and soils. Simple catchment models are applied in forecasting mode, using daily rainfall forecasts from a regional weather forecasting model. The value of the rainfall forecasts, relative to the case where rainfall is known, is assessed for both catchments. The results are discussed in the context of on-going work to compare different modelling approaches for many other Brazilian catchments, and to apply improved forecasting algorithms based on the simple modelling approach to the same, and other, catchments.     

Application of three satellite techniques in support of precipitation forecasts of a NWP model

In this study, the efficiency of an integrated operational system, based on three satellite infrared techniques, to support nowcasting and very short range forecasting of high precipitation rates provided by a numerical weather prediction (NWP) model is examined. Three algorithms, one for the detection and tracking of convective cloud cells and another two for rainfall estimation, are applied on satellite sensor data in order to qualitatively and quantitatively verify the precipitation forecasts provided by a NWP model. The application of the detection and tracking algorithm aims at verifying qualitatively, in real time, the precipitation forecasts by monitoring the detected convective cloud cells, at the same time intervals that the model forecasts are given. The application of the rainfall estimation techniques on satellite sensor data is needed for both quantitative and qualitative cross‐comparisons with the model outputs. The developed tool is applied in a case of intense precipitation over Greece. The results of the application are promising and show the potential for the implementation of the integrated system as a support tool for nowcasting and very short range forecasting by performing real‐time validation of NWP precipitation forecasts.     

APPLICATION OF ARTIFICIAL NEURAL NETWORKS IN CONTROLLED DRUG DELIVERY SYSTEMS

Estimation of release profiles of drugs normally requires time-consuming trial-and-error experiments. Feed-forward neural networks including multilayer perceptron (MLP), radial basis function network (RBFN), and generalized regression neural network (GRNN) are used to predict the release profile of betamethasone (BTM) and betamethasone acetate (BTMA) where in situ forming systems consist of poly (lactide-co-glycolide), N-methyl-1-2-pyrolidon, and ethyl heptanoat as a polymer, solvent, and additive, respectively. The input vectors of the artificial neural networks (ANNs) include drug concentration, gamma irradiation, additive substance, and type of drug. As the outputs of the ANNs, three features are extracted using the nonlinear principal component analysis technique. Leave-one-out cross-validation approach is used to train each ANN. We show that for estimation of BTM and BTMA release profiles, MLP outperforms GRNN and RBF networks in terms of reliability and efficiency.     

混合模型神经网络在短期负荷预测中的应用

提出了可应用于电力系统负荷预测的混合模型神经网络方法,该方法同时具有电力系统负荷预测的传统方法的优点及人工视网络方法的优点,该方法中,不同的负荷分量采用不同类型的预测方法,并采用基本绵谐振分量作神经网络的输入,神经网络的训练采用快速的学习算法进行,该方法具有很强的实时性和适应性,适用于没有气象资料的应用场合,仿真计算的结果表明,预测精度较传统来得高。     

Financial distress prediction by a radial basis function network with logit analysis learning

This paper presents a financial distress prediction model that combines the approaches of neural network learning and logit analysis. This combination can retain the advantages and avoid the disadvantages of the two kinds of approaches in solving such a problem. The radial basis function network (RBFN) is adopted to construct the prediction model. The architecture of RBFN allows the grouping of similar firms in the hidden layer of the network and then performs a logit analysis on these groups instead of directly on the firms. Such a manner can remedy the problem of nominal variables in the input space. The performance of the proposed RBFN is compared to the traditional logit analysis and a backpropagation neural network and demonstrates superior results to both the counterparts in predictive accuracy for unseen data.