Feedforward Neural Network Approach Research Articles

Direct and inverse simulations of hydrodynamic and thermal characteristics in a room with random boundary conditions by feedforward neural network modelling

Artificial Intelligence (AI) or Neural Network (NN) approaches are starting to be used in built environment areas. This study demonstrates that feedforward neural network (FNN) modelling is capable of simulating the hydrodynamic and thermal characteristics of a room with a radiant heating and cooling system. The FNN model was trained and tested to simulate air velocity and temperature profiles based on boundary conditions as direct simulation and to restore boundary conditions as inverse simulation, using a scaled and normalised database generated by a Computational Fluid Dynamics (CFD) model. In the CFD model, the non-dimensional form of the governing equations, including the radiative transfer equation (RTE) and vorticity equation, were solved using the discrete ordinate method (DOM) and finite difference method (FDM) respectively. The impacts of training database size, hidden layers, and the number of neurons in each layer on the accuracy of the FNN results in the direct and inverse simulations were evaluated in terms of the average root mean square error (RMSE) over the testing data. In the direct simulation, using a smaller CFD database with more hidden layers and neurons achieved comparable accuracy to an FNN trained with a larger CFD database and a less complex FNN. Furthermore, in the inverse simulation, where the FNN model was trained to use a subdomain of temperature profiles to simulate or restore the boundary conditions, the effect of the size of the subdomain on the accuracy in terms of RMSE was studied. It demonstrated that the FNN approach could conduct inverse simulations that are typically beyond the capability of conventional CFD modelling approaches. The results show that the boundary conditions could be accurately restored using subdomains of the temperature profile, covering approximately 60% of the main domain.

Machine learning approach for prediction of ionospheric irregularities on ROTI index over the Northern anomaly crest in Egypt during solar cycle 24

This paper reports the prediction results of the ionospheric irregularities represented by the ROTI index close to the equatorial ionization anomaly's Northern peak. A feedforward backpropagation Neural Network (NN) approach is implemented using a time series Nonlinear Autoregressive approach with Exogenous inputs (NARX). We used the data from a dual frequency GPS-SCINDA station located at Helwan (geographic coordinates of 29.86°N, 31.32°E, and MLAT of 29.94°N) in Egypt. To allow the model to be independently tested during varying levels of solar activities, we collected a 5-minute resolution of ROTI data for the solar cycle 24 from 2009 to 2017. The factors that influence the development of ionospheric irregularities are involved in the established model representing the diurnal and seasonal variations as well as solar and geomagnetic activity parameters. To support learning, we included IRI-foF2 and IRI-hmF2 parameters in the input layer neurons to improve the model learning about the behavior of the ionospheric F layer. The results show that the NN-ROTI values precisely match the GPS-ROTI values with an RMSE of 0.106 TECU/min and a prediction efficiency of 95%. The predicted values are highly correlated with the originals, with a regression of 0.89. Furthermore, the irregularities were more prevalent during the equinox months than during the solstice months. It is also observed that predicted NN-ROTI values in all seasons have RMSEs less than 0.03 and 0.05 TECU/min for the low and high solar activity, respectively. The results also showed a clear correlation between solar activity and the occurrence percentage of ionospheric irregularities indicating a solar cycle dependency. The prediction values of the NN-ROTI occurrence % demonstrated a remarkable correlation with the observed GPS-ROTI occurrence % across SC24.

Systemic risk prediction using machine learning: Does network connectedness help prediction?

The global financial crisis not only highlights the important role of financial network connectedness, but also urges regulators pay more attention to systemic risk. We study whether network connectedness helps systemic risk prediction by using machine learning techniques, including k-nearest neighbor (KNN), support vector regression (SVR), extreme gradient boosting (XGB), and feedforward neural network (FNN). Based on prediction results, we use the fingerprint model to deliver interpretability evaluations of financial variables and network connectedness. We find that (1) the linear and nonlinear effects of stock market volatility and network connectedness effectively interpret systemic risk; (2) the interaction effects between stock market volatility and other drivers have a stronger ability to predict system risk in SVR, KNN, and XGB approaches, while network connectedness performs more profoundly in FNN approach; and (3) the interaction effect between stock market volatility and network connectedness is evident in all approaches.

A pruned feed-forward neural network (pruned-FNN) approach to measure air pollution exposure.

Environmental epidemiology studies require accurate estimations of exposure intensities to air pollution. The process from air pollutant emission to individual exposure is however complex and nonlinear, which poses significant modeling challenges. This study aims to develop an exposure assessment model that can strike a balance between accuracy, complexity, and usability. In this regard, neural networks offer one possible approach. This study employed a custom-designed pruned feed-forward neural network (pruned-FNN) approach to calculate the air pollution exposure index based on emission time and rates, terrain factors, meteorological conditions, and proximity measurements. The model's performance was evaluated by cross-validating the estimated exposure indexes with ground-based monitoring records. The pruned FNN can predict pollution exposure indexes (PEIs) that are highly and stably correlated with the monitored air pollutant concentrations (Spearman's rank correlation coefficients for tenfold cross-validation (mean ± standard deviation: 0.906 ± 0.028) and for random cross-validation (0.913 ± 0.024)). The predicted values are also close to the ground truth in most cases (95.5% of the predicted PEIs have relative errors smaller than 10%) when the training datasets are sufficiently large and well-covered. The pruned-FNN method can make accurate exposure estimations using a flexible number of variables and less extensive data in a less money/time-consuming manner. Compared to other exposure assessment models, the pruned FNN is an appropriate and effective approach for exposure assessment that covers a large geographic area over a long period of time.

Lithium-Ion Battery State of Health Prediction With a Robust Collaborative Augmented Hidden Layer Feedforward Neural Network Approach

Lithium-ion (Li-ion) batteries play an important role in providing necessary energy when acting as a main or backup source of electricity. Indeed, the unavailability of battery aging discharge data in most real-world applications makes the State of Health (SoH) assessment very challenging. Alternatively, accelerated aging is therefore adopted to emulate the degradation process and to achieve an SoH estimate. However, accelerated aging generates limited deterioration patterns suffering from a higher level of complexity due to the non-linearity and non-stationarity imposed by harsh conditions. In this context, this paper aims to provide a predictive model capable of solving incomplete data problems by providing two main solutions for each of the problems of complexity and missing patterns, respectively. First, to overcome the problem of lack of patterns, a robust collaborative feature extractor (RCFE) is designed by collaborating between a set of improved restricted Boltzmann machines (I-RBMs) to be able to share learning knowledge among different locally trained I-RBMs to create a more generalized global extraction model. Second, a set of RCFEs is then evolved through a neural network with an augmented hidden layer (NAHL) to enhance the predictive ability by further exploring representation learning to overcome pattern complexity issues. The designed RCFE-NAHL is trained to predict SoH using constant current (CC) discharge characteristics by implying multiple characteristics recorded through the constant voltage (CV) charging process as indicators of health. The proposed SoH prediction approach performances are evaluated on a set of battery life cycles from the well-known NASA database. In this context, the achieved results clearly highlight the higher accuracy and robustness of the proposed learning model.

Maternal exposure to metal components of PM2.5 and low birth weight in New Mexico, USA.

Infants with low birth weight (LBW) are more likely to have health problems than normal weight infants. In studies examining the associations between particulate matter (PM) exposures and LBW, there is a tendency to focus on PM2.5 as a whole. However, insufficient information is available regarding the effects of different components of PM2.5 on birth weight. This study identified the associations between maternal exposure to 10 metal components of PM2.5 and LBW in offspring based on small area (divided by population size) level data in New Mexico, USA, from 2012 to 2016. This study used a pruned feed-forward neural network (pruned-FNN) approach to estimate the annual average exposure index to each metal component in each small area. The linear regression model was employed to examine the association between maternal PM2.5 metal exposures and LBW rate in small areas, adjusting for the female percentage and race/ethnicity compositions, marriage status, and educational level in the population. An interquartile range increase in maternal exposure to mercury and chromium of PM2.5 increased LBW rate by 0.43% (95% confidence interval (CI): 0.18-0.68%) and 0.63% (95% CI: 0.15-1.12%), respectively. These findings suggest that maternal exposure to metal components of air pollutants may increase the risk of LBW in offspring. With no similar studies in New Mexico, this study also posed great importance because of a higher LBW rate in New Mexico than the national average. These findings provide critical information to inform further epidemiological, biological, and toxicological studies.

Fatigue & Fracture of Engineering Materials & Structures Virtual Special Issue: Data science and machine learning for fatigue and fracture assessment

<scp>Fatigue &amp; Fracture of Engineering Materials &amp; Structures Virtual Special Issue</scp>: Data science and machine learning for fatigue and fracture assessment

Lithium-Ion Battery State of Charge Estimation Using Deep Neural Network

In an electric vehicle (EV), the battery management system (BMS) is crucial for managing the health and safety of the battery. The accurate estimation of battery state of charge (SOC) offers critical information about the battery’s remaining capacity. The SOC of the battery mainly depends on its non-linear internal parameters, battery chemistry, ambient temperature, aging factor etc. So, accurate SOC estimation is still a significant challenge. Many researchers have developed several model-based methods that are more complex to develop. Another approach is a data-driven based SOC estimation algorithm, which is less complex but requires more data and it may be inaccurate. In this context, this paper presents a robust and accurate SOC estimation algorithm for a Lithium-ion battery using a deep learning feed-forward neural network (DLFFNN) approach. The proposed algorithm accurately characterizes the battery’s non-linear behavior. To develop a robust SOC estimation algorithm, data is collected at different temperatures with 5% error in data (4 mV-voltage, 110 mA-current, 5∘C temperature) is added to battery datasets. The obtained results demonstrated that the performance of the proposed DLFNN is robust and accurate on different drive cycles with 1.14% Root mean squared error (RMSE), 0.66% mean absolute error (MAE), and 6.65% maximum error (MAX).

Evaluation of Deep Learning CNN Model for Recognition of Devanagari Digit

Devanagari character and digit recognition are a difficult undertaking because writing style depends on a person’s traits and differs from person to person. We get more precise results in digit recognition, thanks to deep learning convolutional neural networks (CNNs), which function similarly to the human brain. In this study, the CNN method was put into practice and contrasted with the feed-forward neural network and random forest approaches. In comparison to previous methods, CNN has reportedly provided an accuracy rating of up to 99.2%. CNN is effective with both organized and unstructured data, including pictures, video, and audio.

Design of motor cable artificial muscle (MC-AM) with tendon sheath–pulley system (TSPS) for musculoskeletal robot

AbstractIn an unstructured environment, the arm can perform complicated tasks with rapidity, flexibility, and robustness. It is difficult to configure multiple artificial muscles similar to an arm in the compact space of a robotic arm. When muscle tension is transferred, mechanisms like tendon-sheath/tendon-pulley may be installed in a compact space to develop musculoskeletal robots that are closer to the arm. However, handling variable frictional nonlinearity and elastic cable deformation is necessary for transmission stability. In this study, the modular artificial muscle system (MAMS), including motor cable artificial muscle and tendon sheath–pulley system (TSPS), that can be installed remotely and transmit muscle tension in narrow paths, is designed. The feed-forward multi-layer neural network (FF-MNN) approach is utilized to discuss the relationship between the measurable input tension of TSPS and the unmeasurable output tension and cable elongation. Subsequently, the lightweight musculoskeletal arm (LM-Arm) is built to verify the validity of MAMS. Through trials, the experiments of MAMS after friction compensating and the LM-Arm’s end-point 3D trajectory tracking are investigated. The results show that average errors of the active and passive muscles tension are 3.87 N and 3.51 N, respectively, under conditions of larger load and higher contraction velocity. The average muscle length error of trajectory tracking is 0.00078 m (0.72%). The suggested MAMS may successfully build a musculoskeletal robot that has similar flexibility and morphology to the arm. It can also be utilized to power various pieces of machinery, such as rescue robot, invasive surgical robots, dexterous hands, and wearable exoskeletons.

A Feedforward Neural Network Approach for the Detection of Optically Thin Cirrus From IASI-NG

A Feedforward Neural Network Approach for the Detection of Optically Thin Cirrus From IASI-NG

Artificial neural network approach for monthly air temperature estimations and maps

Accurate air temperature estimation plays a crucial role in weather planning, and hydrological modeling techniques including water supply, drought analysis, and demand problems of a country. In this study, monthly maximum air temperature (Tmax), minimum air temperature (Tmin), and mean air temperature (T) values for Türkiye were modeled using the Feed-forward neural network (FNN) and Elman neural network (ENN) approaches. In the developed models, monthly Tmax, Tmin, and T values at 81 measurement stations over the period of 1927 through 2021 recorded by the Turkish State Meteorological Service (known as MGM in Türkiye) were used. While the periodicity (month of the year) and geographical variables (latitude, longitude, and altitude) were used as inputs in the developed models, the Tmax, Tmin, and T variables were defined as the output. First of all, long-term monthly air temperature maps were created with the help of the ArcGIS software program using the observed values. Then, maps of long-term air temperatures were created with the ArcGIS software program using the predicted values of the developed artificial neural network models. To prove the accuracy of the models, the predicted long-term air temperature maps were compared with those actual maps. The obtained results showed that the long-term monthly air temperature maps of the studied regions could be created with a small and acceptable error as a function of geographical information and periodicity component. By creating artificial intelligence maps in this way, it will be possible to accurately predict the long-term monthly air temperatures as a function of latitude, longitude, altitude, and the number of months at any point where no measurement is made.

Mineral potential modelling of orogenic gold systems in the granites-tanami Orogen, Northern Territory, Australia: A multi-technique approach

In this study we adopted fuzzy inference system (FIS), transformed predictor map-based random forest (RF) and feed-forward deep neural network (FF-DNN) approaches to mineral potential modelling (MPM) of the Granites-Tanami Orogen (GTO), Northern Territory, Australia, to (i) predict gold prospectivity, and (ii) compare the performance of these models to those previously generated in the study area. The current study, which utilized 19 predictor maps previously developed by Roshanravan et al. (2020) in the framework of a mineral systems approach, took into consideration the learnings gained and results obtained from previous approaches to modelling the gold potential of the GTO (Roshanravan et al., 2020: continuous fuzzy gamma, geometric average, data-driven index overlay and non-transformed predictor map-based RF; Roshanravan et al., 2021: cuckoo optimization algorithm). Importantly, our study area extending over > 46,000 km2 in the GTO, represents one of the few, if not the only, areas worldwide that has been subjected to eight different modelling techniques, an approach that delivered unique insights that cannot be achieved by utilizing a single technique. For example, our multi-technique approach enabled us to study what effect predictor maps may have on model performance and resulting mineral potential maps. In the case of data-driven MPM with RF, arguably-one of the most popular machine learning techniques, the effect of non-transformed (i.e., original) and transformed (i.e., derivative) predictor maps has never been evaluated simultaneously. To achieve this, we generated a transformed predictor map-based RF potential model with the aim of comparing it to the non-transformed predictor map-based RF model of Roshanravan et al. (2020). The FF-DNN and Mamdani-type FIS approaches to MPM were carried out specifically for the purpose of comparing and further investigating the modelling results. With respect to the Mamdani-type FIS, the concentration-area (C-A) fractal technique served to more objectively determine thresholds of the underlying mathematical functions, a novel approach that aids in reducing bias resulting from human input and expert opinion. The thresholded results demonstrate that the RF potential map generated with transformed predictor maps outperformed all other data-driven and continuous models. Overall, our multi-technique approach to MPM and the comparing and contrasting of a large set of resulting gold potential models, not only offered insights that helped to develop and calibrate new tools and techniques but also delivered what we believe are more robust gold exploration targets. As a result, we strongly recommend to using a multi-technique approach to MPM, which we hope will soon replace the currently accepted and widely utilized single-technique approach.

Reconstructing ocean subsurface salinity at high resolution using a machine learning approach

Abstract. A gridded ocean subsurface salinity dataset with global coverage is useful for research on climate change and its variability. Here, we explore the feed-forward neural network (FFNN) approach to reconstruct a high-resolution (0.25∘ × 0.25∘) ocean subsurface (1–2000 m) salinity dataset for the period 1993–2018 by merging in situ salinity profile observations with high-resolution (0.25∘ × 0.25∘) satellite remote-sensing altimetry absolute dynamic topography (ADT), sea surface temperature (SST), sea surface wind (SSW) field data, and a coarse-resolution (1∘ × 1∘) gridded salinity product. We show that the FFNN can effectively transfer small-scale spatial variations in ADT, SST, and SSW fields into the 0.25∘ × 0.25∘ salinity field. The root-mean-square error (RMSE) can be reduced by ∼11 % on a global-average basis compared with the 1∘ × 1∘ salinity gridded field. The reduction in RMSE is much larger in the upper ocean than the deep ocean because of stronger mesoscale variations in the upper layers. In addition, the new 0.25∘ × 0.25∘ reconstruction shows more realistic spatial signals in the regions with strong mesoscale variations, e.g., the Gulf Stream, Kuroshio, and Antarctic Circumpolar Current regions, than the 1∘ × 1∘ resolution product, indicating the efficiency of the machine learning approach in bringing satellite observations together with in situ observations. The large-scale salinity patterns from 0.25∘ × 0.25∘ data are consistent with the 1∘ × 1∘ gridded salinity field, suggesting the persistence of the large-scale signals in the high-resolution reconstruction. The successful application of machine learning in this study provides an alternative approach for ocean and climate data reconstruction that can complement the existing data assimilation and objective analysis methods. The reconstructed IAP0.25∘ dataset is freely available at https://doi.org/10.57760/sciencedb.o00122.00001 (Tian et al., 2022).

A Deep-Learning Approach to Load Modeling in Modern Power Distribution System

Modern Power Distribution Networks (MPDNs) are no longer passive because Distributed Generations (DGs) are integrated with them to enhance system reliability and power quality. For this reason, load modeling has to be updated to capture the new dynamics of active DNs. This paper presents a composite load modeling for a grid-connected photovoltaic (PV) distribution network using the Levenberg-Marquardt algorithm in the deep learning feed-forward neural network approach. Load modeling is constructing a relationship between input excitation(s) and output response(s); it can be used for simulation studies, stability analysis, and control/protection design. A grid-connected PV distribution network was modeled in Matlab/Simulink and generates data for training and model estimation. The estimated model was tested and validated using a laboratory experimental test bed. Results of the model exhibit a good fitness of 99.8% and 97.2% in active and reactive power models respectively during training. While 97.84% and 94.65% respectively were obtained during testing. The estimation errors were found to be 0.0025 and 0.0049 for active and reactive powers respectively with 0.0473 and 0.0701 corresponding errors in testing.

Validation of machine learning approaches for estimating wheel fatigue loads at the front suspension of a race car during track driving

AbstractThe wheel loads of a race car have been estimated in view of structural durability assessments. First, the front left double‐wishbone suspension of a rear‐wheel‐drive race vehicle has been instrumented; then, wheel loads have been estimated by means of four approaches: (i) a geometric matrix (GM) method, (ii) a feedforward neural network (FNN) approach applied to the fully instrumented suspension (FIS), (iii) a FNN approach involving a reduced number of sensors (the partially instrumented suspension (PIS)) and an inertial measurement unit (IMU), and (iv) a linear modeling approach (LM). After having trained the FNNs by using suspension signals acquired in a racetrack as inputs and related tire forces measured with the GM method as targets, the FNN‐based methods have been validated on three different racetracks by comparing the estimated loads with those estimated by means of the GM method. According to the results achieved, the FNN approaches are effective for the estimation of the wheel forces.

Estimation of wind turbine output power using soft computing models

ABSTRACT Among renewable energy generation technologies, wind energy has become one of the most outstanding issues, especially in the last decade. Wind speed is the most critical parameter influencing the power obtained from a wind turbine. The unstable structure of the wind causes an impossibility to receive a direct theoretical relation between wind power and speed. Accordingly, obtaining a simulation of the generated turbine power concerning the approaching wind speed has become vitally essential, nowadays. In the current study, generated wind turbine power (P) has been predicted using three forecasting methods. The computer was trained using wind speed (V) and the turbine rotor rotational speed (ṅ) as the inputs of the forecasting methods. The methods used for this purpose were to include adaptive neuro-fuzzy inference system (ANFIS), Elmanneural network (ENN), and feed-forward neural network (FNN) approaches. In the training of the programs, among the cumulative of 43,800 wind speed, turbine rotor rotational speed, and wind power data, 80% and 20% of the total data were used for training and testing stages of the algorithms, respectively. The statistical results of the computations demonstrated that among three methods, ANFIS gave better outcomes when compared to ENN and FNN, in both the training and testing stages of the algorithms. The proposed models of the current study have revealed low and acceptable mean absolute error (MAE) and root mean square error (RMSE) statistical error results of ANFIS tool; corresponding to 52.448 kW and 87.204 kW error, respectively, were obtained at the training stage, whereas 48.675 kW and 78.453 kW error, respectively, were obtained at the testing stage. For the estimation of wind power, while the coefficient of determination (R 2) was detected as R2 = 0.9948 and 0.9961 in training and testing stages, respectively, with ANFIS model, it was found out as R2 = 0.9942 and 0.9957 with ENN model and R2 = 0.9943 and 0.9956 with FNN model. Namely, it has been reported that ANFIS tool can be successfully applied in wind output power estimations, as long as wind speed and turbine rotor rotational speed values are provided without the need for numerous experimental measurements, which induces additional time, labor, and measurement expenses.

Accuracy Analysis of Data Fraud Detection for Company Transactions Using Two-Layered Feed Forward Neural Network Approach Compared with Random Forest

Aim: To find accuracy of data fraud detection for company transactions using two-layered feed forward Neural Network approaches compared with Random Forest. Materials and methods: Number of groups is 2. Group 1 - Two-layered feed forward Neural Network, and Group 2 - Random Forest. Results: The accuracy of the innovative feed forward Neural Network method is 83.4% and Random Forest is 85.5%. The accuracy of the Random Forest algorithm appears to be a slight increase. Conclusion: Random Forest algorithm finds the fraud detection appearance to be better accuracy than the feed forward Neural Network.

Inversion of 2D cross-hole electrical resistivity tomography data using artificial neural network.

Geophysical inversion is often ill-posed because of its nonlinearity and the ordinary measured data of measured data. To deal with these problems, an artificial neural network (ANN) has been introduced with the capability of a nonlinear and complex problem for geophysical inversion. This study aims to invert 2D cross-hole electrical resistivity tomography data using a feedforward back-propagation neural network (FBNN) approach. To generate the synthetic data to train the model, eighteen forward models (100 to 600 Ω.m homogeneous medium and three different locations of 10 Ω.m of the grouted bulb) with a dipole-dipole array configuration were adopted. The effect of the hyperparameter on the performance of the proposed FBNN model was examined. Various datasets from the laboratory testing result were also tested using the suggested FBNN model and then the error between the actual and predicted area in each model was determined. The results show that our suggested FBNN model, with the trainrp training function, 4 hidden layers, 75 neurons in each hidden layer, 0.8 learning rate, 1 of momentum coefficient, and 54,000 training data points, has higher performance and better accuracy than other models. It was found that the error value of the FBNN model was about 15% to 18% lower compared to the conventional inversion model.

Predicting the near field underwater explosion response of coated composite cylinders using multiscale simulations, experiments, and machine learning

Prediction of underwater explosion response of coated composite cylinders using machine learning (ML) requires a large, consistent, accurate, and representative dataset. However, such reliable large experimental dataset is not readily available. Besides, the ML algorithms need to abide by the fundamental laws of physics to avoid non-physical predictions. To address these challenges, this paper synergistically integrates ML with high-throughput multiscale finite element (FE) simulations to predict the response of coated composite cylinders subjected to nearfield underwater explosion. The simulated responses from the multiscale approach correlate very well with the experimental observations. After validation of the multiscale approach, a representative and consistent dataset containing more than 3800 combinations is developed using high-throughput multiscale simulation by varying the fiber/matrix/coating material properties, coating thickness as well as experimental variables such as explosive energy and stand-off distance. The dataset is leveraged to predict the response of coated composite cylinders subjected to nearfield underwater explosion using a feed-forward multilayer perceptron-based neural network (NN) approach which shows excellent predictions. Overall, the synergistic approach powered by physics-based simulations presented here can potentially enable materials scientists and engineers to make intelligent, informed decisions in the purview of innovative design strategies for underwater explosion mitigation in composite structures.