General Artificial Neural Network Research Articles

Transferrable Model-Agnostic Meta-learning for Short-Term Household Load Forecasting With Limited Training Data

This letter proposes a transferrable model-agnostic meta-learning (T-MAML) approach for short-term load forecasting for single households. The proposed approach enables multiple households to collaboratively train a generic artificial neural network (ANN) model. The generic ANN model is then further trained at each target household node for the STLF purpose. The proposed T-MAML based STLF approach is featured by: (1) significant reduction of computation and communication costs on the household side; and (2) superior STLF performance, especially when there is limited load data for training in a target household. Experiments based on a real Australian residential dataset are conducted to validate the effectiveness of the proposed approach.

A Novel Vertical Wire-Bonding Compensation Structure Adaptively Modeled and Optimized With GRNN and GA Methods for System in Package

In this article, a novelvertical wire-bonding interconnect structure is intelligently modeled and optimized with general artificial neural network (GRNN) and genetic algorithm (GA) for multilayered system in package. A compensation structure is constructed with a hybrid inductive and capacitive technique, while a capacitive stripline with series inductive short-end via is designed underneath the 50-Ω transmission line. The GRNN algorithm is employed to build the electromagnetic model databases during the procedure of GA optimization. In comparison with the conventional optimization algorithm, the output performances can be directly achieved with collaboratively combined methods, which can significantly reduce the calculation time. From the measurement results, the return loss is improved significantly while the parasitic inductive behavior of the bonding wire is eliminated with the presented design. Moreover, compared with the traditional compensation techniques, no additional area is occupied on the surface plane of the wire-bonding interconnection, which is more suitable for the high-integrated circuit system.

2D Gaze Estimation Based on Pupil-Glint Vector Using an Artificial Neural Network

Gaze estimation methods play an important role in a gaze tracking system. A novel 2D gaze estimation method based on the pupil-glint vector is proposed in this paper. First, the circular ring rays location (CRRL) method and Gaussian fitting are utilized for pupil and glint detection, respectively. Then the pupil-glint vector is calculated through subtraction of pupil and glint center fitting. Second, a mapping function is established according to the corresponding relationship between pupil-glint vectors and actual gaze calibration points. In order to solve the mapping function, an improved artificial neural network (DLSR-ANN) based on direct least squares regression is proposed. When the mapping function is determined, gaze estimation can be actualized through calculating gaze point coordinates. Finally, error compensation is implemented to further enhance accuracy of gaze estimation. The proposed method can achieve a corresponding accuracy of 1.29°, 0.89°, 0.52°, and 0.39° when a model with four, six, nine, or 16 calibration markers is utilized for calibration, respectively. Considering error compensation, gaze estimation accuracy can reach 0.36°. The experimental results show that gaze estimation accuracy of the proposed method in this paper is better than that of linear regression (direct least squares regression) and nonlinear regression (generic artificial neural network). The proposed method contributes to enhancing the total accuracy of a gaze tracking system.

BRNeural ‑ Artificial Neural Networks Simulator with Topology Multilayer Perceptron Using the Encog Framework

The Artificial Neural Network (ANN) approach has had been applied in solutions for several problems such as classification, prediction and pattern recognition. Mostly, these problems belong to diversified areas of knowledge, not necessarily related to Computer Science, e.g. in agriculture, specifically in precision agriculture, recognition of diseases in plants, among others. Therefore, it is possible to visualize the multidisciplinary feature that ANNs have, motivating the development of tools for simulating more generic ANN, making its use possible by users whose knowledge of this type of methodology isn't deep. Inside this context, the just described work has had the goal of developing a tool that is part of the BRNeural research project, whose main goal was the development of a simulator named BRNeural (Brasil Neural) which is focused on the scientific and professional community since one of its proposals is to offer a generic environment to aid the creation, train and operate with ANNs. This tool consists in a module for training and testing ANNs with the Multilayer Perceptron (MLP) topology and it was developed using the ENCOG framework. For validating this module and all implemented training algorithms, the simulator was submitted to pattern recognition tests using agrometeorological data such as temperature, relative humidity, solar radiation, atmospheric pressure, wind direction, wind speed and temperature of leaf wetness, comparing the results obtained with another simulator which was developed using the MLP framework (Medeiros, 2003) with the specific purpose of classifying diseases in severity indexes. After validating, the simulator was submitted to other pattern recognition tests whose goal was the classification of diseases in the wheat like Blight, Spot blotch and Mildew in severity indexes through agrometeorological variables, showing its features such as output normalizations and the training algorithms Backpropagation, Resilient propagation and Quick propagation. The best results were obtained with the tests for the classification of diseases using severity indexes, 83.63% correct answers for Blight using the Quick propagation training algorithm and Equilateral normalization, 86.95% of correct answers for Mildew with Resilient Propagation algorithm and OneOf normalization and, for Spot blotch, 91.96% correct answers with Quick and Resilient Propagation algorithms, both using Equilateral normalization.

A neural network model to predict knee adduction moment during walking based on ground reaction force and anthropometric measurements

The external knee adduction moment (KAM) is a major variable for the evaluation of knee loading during walking, specifically in patients with knee osteoarthritis. However, assessment of the KAM is limited to locations where full motion laboratories are available. The purpose of this study was to develop and test a simple method to predict the KAM using only force plate and anthropometric measurements. Three groups of 28 knees (asymptomatic, mild osteoarthritis, and severe osteoarthritis) were studied. Walking trials were collected at different speeds using a motion capture system and a force plate. The reference KAM was calculated by inverse dynamics. For the prediction, inter-subject artificial neural networks were designed using 11 inputs coming from the ground reaction force and the mechanical axis alignment. The predicted KAM curves were similar to the reference curves with median mean absolute deviation (MAD) of 0.36%BW⁎Ht and median correlation coefficient of 0.966 over 756 individual trials. When comparing mean group curves, the median MAD was 0.09%BW⁎Ht and the median correlation coefficient 0.998. The peak values and the angular impulses extracted from the predicted and reference curves were significantly correlated, and the same significant differences were obtained among the three groups when the predicted or when the reference curves were used for 95% of the comparisons. In conclusion, this study demonstrated that a simple method using a generic artificial neural network can predict the KAM curve during walking with a high level of significance and provides a practical option for a broader evaluation of the KAM.

Calibrating artificial neural networks by global optimization

Artificial neural networks (ANNs) are used extensively to model unknown or unspecified functional relationships between the input and output of a “black box” system. In order to apply the generic ANN concept to actual system model fitting problems, a key requirement is the training of the chosen (postulated) ANN structure. Such training serves to select the ANN parameters in order to minimize the discrepancy between modeled system output and the training set of observations. We consider the parameterization of ANNs as a potentially multi-modal optimization problem, and then introduce a corresponding global optimization (GO) framework. The practical viability of the GO based ANN training approach is illustrated by finding close numerical approximations of one-dimensional, yet visibly challenging functions. For this purpose, we have implemented a flexible ANN framework and an easily expandable set of test functions in the technical computing system Mathematica. The MathOptimizer Professional global–local optimization software has been used to solve the induced (multi-dimensional) ANN calibration problems.

An overview on constitutive modelling to predict elevated temperature flow behaviour of fast reactor structural materials

This overview emphasized the aspects of formulation and application of various constitutive models developed by us in recent past viz. Johnson- Cook (JC), modified Zerilli-Armstrong (MZA), strain compensated Arrhenius type model and artificial neural network (ANN) model to predict elevated temperature flow behaviour of fast reactor structural materials. It has been shown that the JC model is not able to represent the high temperature flow behaviour of both alloy D9 and the modified 9Cr-1Mo as it does not incorporate the coupled effect of strain and temperature, and of strain rate and temperature. The new materials model based on Zerilli-Armstrong (ZA) equation considers the coupled effects of temperature and strain and of strain rate and temperature on the flow stress and hence has the capability to predict flow stress over a wider domain temperature and strain rate in comparison with JC model. The formulation and application of strain compensated Arrhenius type constitutive model to predict high temperature flow behaviour of alloy D9 and modified 9Cr-1Mo has been discussed. Development and application of a generic ANN based constitutive model to predict high temperature deformation behaviour of austenitic stainless steels has been highlighted. Finally, a comparative analysis of the merits and shortcomings of these models has been made.

Rainfall‐runoff modeling through hybrid intelligent system

This study explores the potential of integrating two different artificial intelligence techniques, namely neural network and fuzzy logic, effectively to model the rainfall‐runoff process from rainfall and runoff information. The integration is achieved through representing fuzzy system computations in a generic artificial neural network (ANN) architecture, which is functionally equivalent to a fuzzy inference system. The model is initialized by a hyperellipsoidal fuzzy clustering (HEC) procedure, which identifies suitable numbers of fuzzy if‐then rules through proper partition of the input space. The parameters of the membership functions are optimized using a nonlinear optimization procedure. The consequent functions are chosen to be linear in their parameters, and a standard least squares error method is employed for parameter estimation. The proposed model is tested on two case studies: Narmada basin in India and Kentucky basin in the United States. The results are highly encouraging as the model is able to explain more than 92% of the variance. The performance of the proposed model is found to be comparable to that of an adaptive neural based fuzzy inference system (ANFIS) developed for both the basins. The number of parameters in the proposed model is fewer compared to ANFIS, and the former can be trained in lesser time. It is also observed that the proposed model simulates the peak flow better than ANFIS. Overall, the study suggests that the proposed model can potentially be a viable alternative to ANFIS for use as an operational tool for rainfall runoff modeling purposes.

Towards a generic artificial neural network model for dynamic predictions of algal abundance in freshwater lakes

A generalised architecture of a feedforward ANN for the prediction of algal abundance is suggested. It simplifies practical model applications, rationalises data collection and preprocessing, improves model validity, and enables meaningful comparison of ANN models between lakes. The generic ANN model considers the key driving variables of algal growth such as phosphorous, nitrogen, underwater light and water temperature as input nodes and predicts algal species abundance or biomass as output. Two model structures were used; one for same-day and one for 30-days ahead predictions of algal abundance. ANN models with and without hidden layers were compared to determine the impact of the addition of non-linear processing capabilities on model performance. A bootstrap aggregation method was found to reduce test set prediction error and to mitigate the effects of overfitting. The model was validated by means of time-series data from six different freshwater lakes.

On-line learning control of manipulators based on artificial neural network models

This paper addresses the tracking control problem of robotic manipulators with unknown and changing dynamics. In this study, nonlinear dynamics of the robotic manipulator is assumed to be unknown and a control scheme is developed to adaptively estimate the unknown manipulator dynamics utilizing generic artificial neural network models to approximate the underlying dynamics. Based on the error dynamics of the controller, a parameter update equation is derived for the adaptive ANN models and local stability properties of the controller are discussed. The proposed scheme is simulated and successfully tested for trajectory following tasks. The controller also demonstrates remarkable performance in adaptation to changes in manipulator dynamics.

Conguring ANN for Inundation Areas Identication based on Relevant Thematic Layers

Flooding is one of the major natural disasters that can cause not only disruptions to daily lives but also damages to our properties. The study of flood models to determine inundation areas has therefore become increasingly more important for decision makers and the authorities. Being readily accessible, the Artificial Neural Network (ANN) scheme was frequently adopted for hydrology and flood modeling. Most ANN techniques primarily take into account rainfall data and then predict runoff consequences. Despite convincing successes, they usually neglect other causative factors. This study thus focuses on configuring and improving generic ANN for inundation areas identification using various flood deterministic attributes. Accordingly, an ANN was developed using nine flood causative factors, derived from relevant thematic layers. They consist of flood plain in the past, height above sea level, water density, water blockage, sub basin areas, soil drainage capability, land uses and monthly rainfall, whose prognostic values were previously reported and assessed against the full scale census and comprehensive GIS survey with satisfying cogency. The guidelines and precautions suggested in this paper may be applied to various ubiquitous ANN based frameworks for flood forecasting and related risks assessment.