Proposed Neural Network Model Research Articles

Predicting Solar Power Generation Utilized in Iraq Power Grid using Neural Network

The prediction of a photovoltaic (PV) energy production over time is considered the major challenge to integrate it with the utilize grid. This is because it is affected by many factors, including geographical locations and weather conditions. Hence, accurately forecasting PV generation is a crucial stage for ensuring a grid stability. In this paper, several studies are discussed between 2015 to 2023 based on various term forecasting conditions. Then, a neural network (NN) is employed to forecast a medium-term PV power generation by gathering meteorological data specific to the southern region of Iraq. The proposed NN model based on Python language is trained to find the PV power prediction for various seasons of the year using solar radiation and surrounding temperature of weather condition tests. While, the MATLAB Simulink model is designed to address the PV actual power for the same tests. The results show that the root mean square error of a PV generation test at autumn season provides the lowest percentage of 92.2119 W when compared with the other three seasons

A Neural Network Approach to Estimate the Frequency of a Cantilever Beam with Random Multiple Damages.

Natural frequency is an important parameter in the structural health monitoring (SHM) system. Any changes in this parameter indicate structural alteration due to damage. This study provides a neural network (NN) solution as an alternative to the finite element (FE) method to measure the natural frequencies of a cantilever beam with random multiple damage. It is based on a statistical dataset of a free vibration test obtained from the APDL (Ansys parametric design language) simulation using a MATLAB (matrix laboratory) script. The script can generate an unlimited number of possible damage combinations for any given parameters with the help of the Monte Carlo (MC) technique. MC helps to generate a random number of damages in random locations at each simulation. Damage conditions are controlled by three parameters including damage severity and damage size (in terms of the mean and standard deviation of damage). Moreover, the method proposes a curve-fitting equation to validate the predicted natural frequency for the first three modes obtained from the neural network model. Both methods are in good agreement with each other, having minimal errors in the range of 0.2-3% for each mode. The frequency result shows that the beam frequency is 8.6486 Hz if the area reduction is 10%, whereas it comes down to 7.2338 Hz if there is a 30% area reduction. A two-level factorial test shows that damage severity is the most impactful factor compared to the damage sizes on the frequency shift event. This indicates that damage alters the composition of the beam and has an impact on its frequency change with the assumed damage parameters. Therefore, the proposed NN model can estimate the frequency shift for various damage scenarios. It can be utilized in the vibration-based damage identification process to predict the frequency changes of the damaged beam without any computational burden.

An artificial neural network model in predicting VTEC over central Anatolia in Turkey

This research investigates the capability of artificial neural networks to predict vertical total electron content (VTEC) over central Anatolia in Turkey. The VTEC dataset was derived from the 19 permanent Global Positioning System (GPS) stations belonging to the Turkish National Permanent GPS Network-Active (TUSAGA-Aktif) and International Global Navigation Satellite System Service (IGS) networks. The study area is located at 32.6°E-37.5°E and 36.0°N-42.0°N. Considering the factors inducing VTEC variations in the ionosphere, an artificial neural network (NN) with seven input neurons in a multi-layer perceptron model is proposed. The KURU and ANMU GPS stations from the TUSAGA-Aktif network are selected to implement the proposed neural network model. Based on the root mean square error (RMSE) results from 50 simulation tests, the hidden layer in the NN model is designed with 41 neurons since the lowest RMSE is achieved in this attempt. According to the correlation coefficients, absolute and relative errors, the NN VTEC provides better predictions for hourly and quarterly GPS VTEC. In addition, this paper demonstrates that the NN VTEC model shows better performance than the global IRI2016 model. Regarding the spatial contribution of the GPS network to TEC prediction, the KURU station performs better than ANMU station in fitting with the proposed NN model in the station-based comparison.

Artificial neural networks approach for prediction of axial loading capacity of circular normal strength concrete columns confined by both transverse steel reinforcement and carbon fiber reinforced polymer wraps

A few empirical models for the axial loading capacity (ALC) of circular normal strength concrete (NSC) columns wrapped by carbon fiber reinforced polymer (CFRP) sheets and interior transverse steel reinforcement (TSR) (CSC columns) are available in the literature. The deficiency of those models is that they were proposed based on a small number of tests by considering limited parameters of CSC columns. Therefore, the main aim of the current investigation is to propose the improved empirical models for the ALC of CSC columns by including the interaction mechanism between TSR and FRP confining behavior. To secure this aim, a general regression analysis technique and artificial neural networks (NNs) on the experimental outcomes of 76 CSC columns collected from the previous investigations were employed. The proposed NN model was adjusted for the different number of hidden layers and neurons to achieve an optimized model. The suggested NN and empirical models portrayed a close agreement with the testing database with R2 = 0.998 and R2 = 0.892, respectively. The NN model reported a higher accuracy than the theoretical model. The comparative investigation solidly authenticated the superiority and accuracy of the anticipated strength models for CSC columns.

Deep learning for component fault detection in electricity transmission lines

Component fault detection and inventory are one of the most significant bottlenecks facing the electricity transmission and distribution utility establishments especially in developing countries for delivery of efficient services to the customers and to ensure proper asset audit and management for network optimization and load forecasting. For lack of technology and data, insecurity, the complexity associated with traditional methods, untimeliness, and general human cost, electricity assets monitoring, and management have remained a big problem in many developing countries. In view of this, we explored the use of oblique UAV imagery with high spatial resolution and fine-tuned deep Convolutional Neural Networks (CNNs) for automatic faulty component inspection and inventory in an Electric power transmission network (EPTN). This study investigated the capability of the Single Shot Multibox Detector (SSD), a one-stage object detection model on the electric transmission power line imagery to localize, detect and classify faults. Our proposed neural network model is a CNN based on a multiscale layer feature pyramid network (FPN) using aerial image patches and ground truth to localise and detect faults through a one-phase procedure. The SSD Rest50 architecture variation performed the best with a mean Average Precision (mAP) of 89.61%. All the developed SSD-based models achieve a high precision rate and low recall rate in detecting faulty components, thus achieving acceptable balance levels of F1-score and representation. We have established in this paper that combined use of UAV imagery and computer vision presents a low-cost method for easy and timely electricity asset inventory, especially in developing countries. This study also provides the guide to various considerations when adopting this technology in terms of the choice of deep learning architecture, adequate training samples over multiple fault characteristics, effects of data augmentation, and balancing of intra-class heterogeneity.

Short-term prediction of metro passenger flow with multi-source data: A neural network model fusing spatial and temporal features

Accurate short-term prediction of metro passenger flow can ensure the efficient operation of metro system and comfort and safety of passengers. Most existing studies use a single data source such as smart card data to predict metro passenger flow, which only consider the internal passenger flows of the metro system and fails to consider the potential passenger flows outside metro stations. However, the potential passenger flow outside metro stations actually has an impact on metro passenger flow in the short term. Using multiple sources of data including smart card data, mobile phone data and metro network data, this paper presents a neural network (NN) model for short-term prediction of metro passenger flow. In the proposed NN model, various information are taken into account by extracting spatial and temporal features inside and outside the metro system. The proposed model structure includes long short-term memory layers to extract historical pattern of the extracted features and fully connected layers to grasp potential connection and interaction among the features. The advantages of the proposed model are demonstrated by using multi-source data from Suzhou, China. The results show that the proposed NN model outperforms various baseline models in terms of accuracy and stability. It can also be found that the inclusion of mobile phone data improves the accuracy of prediction.

Single Probe Imitation of Multi-Depth Capacitive Soil Moisture Sensor Using Bidirectional Recurrent Neural Network

This work proposes a single probe imitation of multidepth capacitive soil moisture sensor for low-cost and energy-efficient implementation of IoT-assisted wireless sensor network (IoWSN) farm monitoring infrastructure. A conditioning circuit (CC) is devised to capture the behavior of soil water movement and its impact on soil moisture around the probe at different depths. The captured correlation is used to train the proposed neural network (NN) models to estimate the soil water content (SWC) at different soil depths based on a single measurement taken at the reference depth. To adjust the weight of the neurons, training and test dataset are collected through a measurement campaign. The data are collected during the cropping period of paddy vegetation by deploying 150 sensor nodes at different depths in bare land before sowing. Two NN models—artificial neural network (ANN) and bidirectional long short term memory network (BLSTM)—are proposed and compared based on the accuracy of SWC estimation. To demonstrate the efficacy of the concept, the proposed sensor design is compared with relevant soil moisture sensors reported over the past five years. The root mean square error (RMSE), <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${R^{2}}$ </tex-math></inline-formula> and mean absolute percentage error (MAPE)-based analysis validate the significance of the proposed NN models.

Prediction of Total Electron Content (TEC) using Neural Network over Anomaly Crest Region Bhopal

In the present work neural network has been developed for the prediction of Vertical Total Electron Content (VTEC) over Bhopal (23.2°N, 77.4°E & MLAT 14.2°N), an equatorial anomaly crest region. The study is based on the VTEC data recorded by a multi-frequency multi-constellation GNSS receiver installed at Bhopal during May 2016 to August 2017. Neural Network (NN) consists of artificial neurons that are used to learn data patterns adjusting weights between input, hidden and output layers by computational modelling. In order to obtain the optimal number of neurons in the hidden layer, the performance of four systems of neural network have been tested. Out of which, the fourth neural network has been used to predict VTEC with 7 input neurons in the present study. The number of hidden layer neurons is a factor that affects the performance of the trained networks. The input layer of the network includes the year, day of the year, hour of the day, geographic latitude, geographic longitude, SSN and IRI NmF2. We have compared the TEC obtained from the neural network prediction (NN TEC), GPS derived TEC (GPS TEC) and TEC from IRI-2016 model (IRI TEC) and analysis has been made. We observed that the Root Mean Square Error (RMSE) decreases with increase in the number of hidden layer neurons. The RMSE comes out to be lowest (1.96 TECU), when the number of hidden layer neurons are 41.The TEC predicted from the NN model and estimated from GPS receiver have same trend at 01:00 and 07:00 UT. The modelled NN TEC values are in good agreement with GPS TEC values for all seasons and 0.99 correlation coefficient was calculated for summer months, but IRI TEC always underestimates GPS TEC in all seasons. The NN model excellently predicted the TEC values of August 2017 by putting the credentials of August 2016 and only 0.55–32.16% relative errors were estimated between them whereas it reaches up to 93.86% with IRI TEC. Thus, the result shows that the proposed NN model can predict the diurnal and seasonal GPS TEC more accurate than IRI over anomaly crest region Bhopal. However, the inclusion of IRI-NmF2 as input layer neuron increases the network performance.

Neural Network aided PMSM multi-objective design and optimization for more-electric aircraft applications

This study uses the Neural Network (NN) technique to optimize design of surface-mounted Permanent Magnet Synchronous Motors (PMSMs) for More-Electric Aircraft (MEA) applications. The key role of NN is to provide dedicated correction factors for the analytical PMSM mass and loss estimation within the entire design space. Based on that, a globally optimal design can be quickly obtained. Matching the analytical estimation with Finite-Element Analysis (FEA) is the main research target of training the NN. Conventional analytical formulae serve as the basis of this study, but they are prone to loss accuracy (especially for a large design space) due to their assumptions and simplifications. With the help of the trained NNs, the analytical motor model can give an estimation as accurate as the FEA but with super less time during the optimization process. The Average Correction Factor (ACF) approach is regarded as the comparison method to demonstrate the excellent performance of the proposed NN model. Furthermore, a NN aided three-stage-seven-step optimization methodology is proposed. Finally, a Pole-10-Slot-12 PMSM case study is given to demonstrate the feasibility and gain of the NN aided multi-objective optimization approach. In this case, the NN aided analytical model can generate one motor design in 0.04 s while it takes more than 1 min for the used FEA model.

Predicting Rate Constants of Hydroxyl Radical Reactions with Alkanes Using Machine Learning.

The hydrogen abstraction reactions of the hydroxyl radical with alkanes play an important role in combustion chemistry and atmospheric chemistry. However, site-specific reaction constants are difficult to obtain experimentally and theoretically. Recently, machine learning has proved its ability to predict chemical properties. In this work, a machine learning approach is developed to predict the temperature-dependent site-specific rate constants of the title reactions. Multilayered neural network (NN) models are developed by training the site-specific rate constants of 11 reactions, and several schemes are designed to improve the prediction accuracy. The results show that the proposed NN models are robust in predicting the site-specific and overall rate constants.

Predicting the Local Response of Metastatic Brain Tumor to Gamma Knife Radiosurgery by Radiomics With a Machine Learning Method.

PurposeThe current study proposed a model to predict the response of brain metastases (BMs) treated by Gamma knife radiosurgery (GKRS) using a machine learning (ML) method with radiomics features. The model can be used as a decision tool by clinicians for the most desirable treatment outcome.Methods and MaterialUsing MR image data taken by a FLASH (3D fast, low-angle shot) scanning protocol with gadolinium (Gd) contrast-enhanced T1-weighting, the local response (LR) of 157 metastatic brain tumors was categorized into two groups (Group I: responder and Group II: non-responder). We performed a radiomics analysis of those tumors, resulting in more than 700 features. To build a machine learning model, first, we used the least absolute shrinkage and selection operator (LASSO) regression to reduce the number of radiomics features to the minimum number of features useful for the prediction. Then, a prediction model was constructed by using a neural network (NN) classifier with 10 hidden layers and rectified linear unit activation. The training model was evaluated with five-fold cross-validation. For the final evaluation, the NN model was applied to a set of data not used for model creation. The accuracy and sensitivity and the area under the receiver operating characteristic curve (AUC) of the prediction model of LR were analyzed. The performance of the ML model was compared with a visual evaluation method, for which the LR of tumors was predicted by examining the image enhancement pattern of the tumor on MR images.ResultsBy the LASSO analysis of the training data, we found seven radiomics features useful for the classification. The accuracy and sensitivity of the visual evaluation method were 44 and 54%. On the other hand, the accuracy and sensitivity of the proposed NN model were 78 and 87%, and the AUC was 0.87.ConclusionsThe proposed NN model using the radiomics features can help physicians to gain a more realistic expectation of the treatment outcome than the traditional method.

Neural Network Model for Heart Disease Diagnosis

The heart disease diagnosis system is proposed inthis study. This kind of diagnosis systems enhance medical careand helps doctors. In this paper, heart disease dataset fromkaggle web site is used. Neural Network is examined andanalyzed for different structures as an optimizer, loss function,and batch size. The simulation results show that the proposedneural network model has 90,16% accuracy.

Neural network‐based energy management of multi‐source (battery/UC/FC) powered electric vehicle

Due to increased environmental pollution and global warming concerns, the use of energy storage units that can be supported by renewable energy resources in transportation becomes more of an issue and plays a vital role in terms of clean energy solutions. However, utilization of multiple energy storage units together in an electric vehicle makes the powertrain system more complex and difficult to control. For this reason, the present study proposes an advanced energy management strategy (EMS) for range extended battery electric vehicles (BEVs) with complex powertrain structure. Hybrid energy storage system (HESS) consists of battery, ultra-capacitor (UC), fuel cell (FC) and the vehicle is propelled with two complementary propulsion machines. To increase powertrain efficiency, traction power is simultaneously shared at different rates by propulsion machines. Propulsion powers are shared by HESS units according to following objectives: extending battery lifetime, utilizing UC and FC effectively. Primarily, to optimize the power split in HESS, a convex optimization problem is formulated to meet given objectives that results 5 years prolonged battery lifetime. However, convex optimization of complex systems can be arduous due to the excessive number of parameters that has to be taken into consideration and not all systems are suitable for linearization. Therefore, a neural network (NN)-based machine learning (ML) algorithm is proposed to solve multi-objective energy management problem. Proposed NN model is trained with convex optimization outputs and according to the simulation results the trained NN model solves the optimization problem within 92.5% of the convex optimization one.

Efficient pothole detection using smartphone sensors

Road safety remains a casualty in India, with potholes wrecking asphalt pavements by the dozens. A study in 2017 recorded that potholes caused the budget for road safety to increase by a whopping 100.4 per cent, and even doubled the death toll from that of the year prior. To address this situation, an effective solution is required that ensures the drivers’ safety and can prove beneficial for long term measures. This can be established by employing an apt pothole detection system which is simple yet functional. In this paper, the method for such a system is described which uses accelerometer and gyroscope, both built in the modern day smartphones, to sense potholes. Pothole induced vibrations can be measured on the axis reading, making them distinguishable. Our proposed Neural Network model is trained and evaluated on the data acquired from the sensors and classifies the potholes from the non-potholes. The neural network gives a classification accuracy of 94.78 per cent. It also presents a solid precision-recall trade-off with 0.71 precision and 0.81 recall, considerably high for a problem with class imbalance. The results indicate that the method is suitable for creating an accurate and sensitive supervised model for pothole detection.

A New Approach of Hybrid Bee Colony Optimized Neural Computing to Estimate the Soil Compression Coefficient for a Housing Construction Project

In the design phase of housing projects, predicting the settlement of soil layers beneath the buildings requires the estimation of the coefficient of soil compression. This study proposes a low-cost, fast, and reliable alternative for estimating this soil parameter utilizing a hybrid metaheuristic optimized neural network (NN). An integrated method of artificial bee colony (ABC) and the Levenberg–Marquardt (LM) algorithm is put forward to train the NN inference model. The model is capable of delivering the response variable of soil compression coefficient a set of physical properties of soil. A large-scale real-life urban project at Hai Phong city (Vietnam) was selected as a case study. Accordingly, a dataset of 441 samples with their corresponding testing values of the compression coefficient has been collected and prepared during the construction phase. Experimental outcomes confirm that the proposed NN model with the hybrid ABC-LM training algorithm has attained the highly accurate estimation of the soil compression coefficient with root mean square error (RMSE) = 0.008, mean absolute percentage error (MAPE) = 10.180%, and coefficient of determination (R2) = 0.864. Thus, the proposed machine learning method can be a promising tool for geotechnical engineers in the design phase of housing projects.

Spread-F prediction model for the equatorial Chumphon station, Thailand

This work proposes a range spread-F (RSF) prediction model using the neural network (NN) over the equatorial Chumphon (CPN) region in Thailand. The RSF model is constructed by using five input spaces including the diurnal variations, seasonal variations, geographic latitude, solar flux index (F10.7), and magnetic index (Ap). The RSF NN model is trained with three years of RSF data during 2013 to 2015 from Chumphon (CPN) station (Latitude = 10.7°N, Longitude = 99.4°E) and the performance of the proposed RSF NN model is validated using the dataset of 2016. As a result, the RSF NN model achieves 98.3% accuracy of all correct predictions even with the limited available data. The results show that the proposed NN model yields a lower RSF probability than the actual observation by about 7.3%, but the overestimation of the proposed NN model is 2.5% in both the equinoxes and solstices. In addition, we discover that the IRI-2016 model mostly overestimates the RSF probability when compared with the actual observation for all seasons in 2016, particularly, in equinoctial months over Chumphon station.

An efficient neural network for solving convex optimization problems with a nonlinear complementarity problem function

In this paper, we present a one-layer recurrent neural network (NN) for solving convex optimization problems by using the Mangasarian and Solodov (MS) implicit Lagrangian function. In this paper by using Krush–Kuhn–Tucker conditions and MS function the NN model was derived from an unconstrained minimization problem. The proposed NN model is one layer and compared to the available NNs for solving convex optimization problems, which has a better performance in convergence time. The proposed NN model is stable in the sense of Lyapunov and globally convergent to optimal solution of the original problem. Finally, simulation results on several numerical examples are presented and the validity of the proposed NN model is demonstrated.

Establishment of Prediction Models of Asphalt Pavement Performance based on a Novel Data Calibration Method and Neural Network

This paper aims to develop models to forecast the deterioration of pavement conditions including rutting, roughness, skid-resistance, transverse cracking, and pavement surface distress. A data quality control method was proposed to rebuild the performance data based on the idea of longest increasing or decreasing subsequences. Neural network (NN) was used to develop the five models, and principal component analysis (PCA) was applied to reduce the dimension of traffic variables. The influence of different input variables on the model outputs was discussed respectively by comparing their mean impact values (MIV). Results show that the proposed NN models demonstrated great potential for accurate prediction of pavement conditions, with an average testing R-square of 0.8692. The results of sensitivity analysis revealed that recent pavement conditions may influence the future pavement conditions significantly. Rutting and roughness were sensitive to pavement age and maintenance type. The materials of original pavement asphalt layer were highly relevant to the prediction of pavement roughness, skid-resistance, and pavement surface distress. Moreover, traffic loads obviously affected the pavement skid-resistance and transverse cracking. Pavement and bridge had different effect on surface distress. The material of the base has a remarkable impact on the initiation and development of transverse cracks. Disease treatment in terms of pavement cracking—such as sticking the cracks, excavating and filling the cracks—shows a high MIV in the prediction model of transverse cracking and pavement surface distress.

Synthetic retrieval of hourly net ecosystem exchange using the neural network model with combined MI and GOCI geostationary sensor datasets and ground-based measurements

ABSTRACTNet ecosystem carbon dioxide (CO2) exchange (NEE) is a key parameter for understanding the terrestrial plant ecosystems, but it is difficult to monitor or predict over large areas at fine temporal resolutions. In this research, we estimated the hourly NEE using a combination of the integrated neural network (NN) model with geostationary satellite imagery to overcome the limitations of existing daily polar orbiting satellite-derived carbon flux products. Two sets of satellite imageries (i.e. the meteorological imager (MI) and geostationary ocean colour imager (GOCI) aboard communication, ocean, and meteorological satellite (COMS)) and CO2 flux data derived from eddy covariance measurements were used to verify the feasibility of applying hourly geostationary satellite imagery with an NN-based approach for estimating NEE at high temporal resolutions. For the NN model, the optimum neuronal architecture was established using an NN with one hidden layer that was trained using the Levenberg–Marquardt back propagation algorithm. The hourly NEE values estimated in test period from the NN model using the combined COMS MI and GOCI imagery and ground measurements as model inputs were compared with the eddy covariance NEE values from the measurement tower, which yielded reliable statistical agreement. The hourly NEE results from the NN model based on COMS MI and GOCI imagery and ground measurement data had the highest accuracy (RMSE = 2.026 μmol m−2 s−2, R = 0.975), while the root mean square error (RMSE) and the regression coefficient (R) generated by the NN model based on satellite imagery as the sole input variable were relatively lower (RMSE = 3.230 μmol m−2 s−2, R = 0.952). Although the simulations for the satellite-only NEE were showed as lower accuracy than the NN model that included all input variables, the hourly variations in NEE also appeared to describe its daily growth and development pattern well, indicating the possibility of deriving hourly-based products from the proposed NN model using geostationary satellite data as inputs.

A comparison of neural network-based predictions of foF2 with the IRI-2012 model at conjugate points in Southeast Asia

This paper presents the development of Neural Network (NN) model for the prediction of the F2 layer critical frequency (foF2) at three ionosonde stations near the magnetic equator of Southeast Asia. Two of these stations including Chiang Mai (18.76°N, 98.93°E, dip angle 12.7°N) and Kototabang (0.2°S, 100.3°E, dip angle 10.1°S) are at the conjugate points while Chumphon (10.72°N, 99.37°E, dip angle 3.0°N) station is near the equator. To produce the model, the feed forward network with backpropagation algorithm is applied. The NN is trained with the daily hourly values of foF2 during 2004–2012, except 2009, and the selected input parameters, which affect the foF2 variability, include day number (DN), hour number (HR), solar zenith angle (C), geographic latitude (θ), magnetic inclination (I), magnetic declination (D) and angle of meridian (M) relative to the sub-solar point, the 7-day mean of F10.7 (F10.7_7), the 81-day mean of SSN (SSN_81) and the 2-day mean of Ap (Ap_2). The foF2 data of 2009 and 2013 are then used for testing the NN model during the foF2 interpolation and extrapolation, respectively. To examine the performance of the proposed NN, the root mean square error (RMSE) of the observed foF2, the proposed NN model and the IRI-2012 (CCIR and URSI options) model are compared. In general, the results show the same trends in foF2 variation between the models (NN and IRI-2012) and the observations in that they are higher during the day and lower at night. Besides, the results demonstrate that the proposed NN model can predict the foF2 values more closely during daytime than during nighttime as supported by the lower RMSE values during daytime (0.5≤RMSE≤1.0 for Chumphon and Kototabang, 0.7≤RMSE≤1.2 at Chiang Mai) and with the highest levels during nighttime (0.8≤RMSE≤1.5 for Chumphon and Kototabang, 1.2≤RMSE≤2.0 at Chiang Mai). Furthermore, the NN model predicts the foF2 values more accurately than the IRI model at the three sites on average, as clearly seen on the yearly RMSE averages. The RMSE values of NN model are lower than those of both CCIR and URSI options, and in terms of the yearly percentage improvements, the NN model gives improvement of around 10–15% in 2009 and 10% in 2013 for Chiang Mai, 20–25% in 2009 and 5–10% in 2013 for Chumphon, and around 18–25% in 2009 and 20–30% in 2013 at Kototabang. Although the NN model predicts the foF2 values closely to the observed data and produces more accurate prediction than the IRI models, in some cases, the IRI model performs better than the NN model. Hence, there is still room for further improvement of the proposed NN model.