Multilayer Perceptron Neural Network Model Research Articles

Assessing the protection mechanisms on Enterobacter aerogenes ATCC 13048 by potentially probiotic strain Lacticaseibacillus casei XN18: An experimental and modeling study

In this study, Lacticaseibacillus casei XN18 had a remarkable resistant to simulated gastrointestinal conditions, hydrophobicity (38.60%), auto-aggregation (29.80%), co-aggregation (21.10%), adhesion (9.50%), anti-adhesion (24.40–36.90%), antioxidant activity (46.47%), cholesterol assimilation (41.10%), and antimicrobial effect on some pathogenic microorganisms. The modified double layer method, and Enterobacter aerogenes (inhibition zone (IZ) = 9.10 mm) and Listeria monocytogenes (IZ = 14.60 mm) were the most sensitive and resistant pathogens to the probiotic strain. The Lb. casei was sensitive to ciprofloxacin (IZ = 23 mm) and nitrofurantoin (IZ = 25.10 mm), semi-sensitive to imipenem (IZ = 18.80 mm), erythromycin (IZ = 16.90 mm), and chloramphenicol (IZ = 17.90 mm), and resistant to ampicillin (IZ = 9.60 mm) and nalidixic acid (IZ = 9.90 mm). The Lb. casei showed no haemolytic and DNase properties, and it could therefore be used for health-promoting purposes. In the next section, multilayer perceptron (MLP) neural network (NN) and gaussian process regression (GPR) models with k-fold cross validation method were used for predicting the rate of probiotic viability based on three levels of pH and time. The results showed that GPR has the lowest error. The mean absolute percentage error (MAPE), root mean absolute error (RMSE) and coefficient of determination (R2) for GPR and MLP models were 1.49 ± 0.40, 0.21 ± 0.03, 0.98 ± 0.05 and 6.66 ± 0.98, 0.83 ± 0.23 0.82 ± 0.09, respectively. So, the GPR model can be reliably used as a useful method to predict the probiotic viability in similar cases.

Stock Market Prediction Using a Hybrid of Deep Learning Models

Financial markets play an essential role in developing modern society and enabling the deployment of economic resources. This study focuses on predicting stock prices using deep learning models. In particular, the daily closing prices of two different stocks from the Casablanca Stock Market Viz Bank of Africa and Itissalat Al-Maghrib (IAM) are considered. The datasets were pre-processed and passed through the Long Short-Term Memory (LSTM), Multi-Layer Perceptron (MLP), and Convolutional Neural Networks (CNN) models. The models’ performances were compared based on the performance evaluation metrics, viz: mean squared error (MSE) and root mean squared error (RMSE) and Mean Absolute Error (MAE). The paper proposes a novel hybrid model. The hybrid design of the model improves its predictive power as the results of the Hybrid network performance surpassed all the other models.

A stacked machine learning model for multi-step ahead prediction of lake surface water temperature

Lake surface water temperature is one of the most important physical and ecological indices of lakes, which has frequently been used as the indicator to evaluate the impact of climate change on lakes. Knowing the dynamics of lake surface water temperature is thus of great significance. The past decades have witnessed the development of different modeling tools to forecast lake surface water temperature, yet, simple models with fewer input variables, while maintaining high forecasting accuracy are scarce. Impact of forecast horizons on model performance has seldom been investigated. To fill the gap, in this study, a novel machine learning algorithm by stacking multilayer perceptron and random forest (MLP-RF) was employed to forecast daily lake surface water temperature using daily air temperature as the exogenous input variable, with the Bayesian Optimization procedure applied for tuning the hyperparameters. Prediction models were developed using long-term observed data from eight Polish lakes. The MLP-RF stacked model showed very good forecasting capabilities for all lakes and forecast horizons, far better than shallow multilayer perceptron neural network, a model coupling wavelet transform and multilayer perceptron neural network, non-linear regression and air2water models. A reduction in model performance was observed as the forecast horizon increased. However, the model also performs well with a forecast horizon of several days (e.g., 7 days ahead, testing stage: R2 - [0.932, 0.990], RMSE °C - [0.77, 1.83], MAE °C - [0.55, 1.38]). In addition, the MLP-RF stacked model has proven to be reliable for both intermediate temperatures and minimum and maximum peaks. The model proposed in this study will be useful to the scientific community in predicting lake surface water temperature, thus contributing to studies on such sensitive aquatic ecosystems as lakes.

Sensorless Adaptive control of VSI-Fed Induction Motor Drive with Optimized MLP-Neural Network

ABSTRACT Multilayer-Perceptron Neural Network (MLP-NN)-based sensorless speed control of adaptive Indirect Field-Oriented Control (IFOC) strategy is implemented for online parameter estimation of Induction Motor Drive (IMD) fed from Common mode voltage injection Space vector PWM (CVSVPWM) based Voltage Source Inverter. Harris Hawks Optimization (HHO) is implemented in this work, to train the MLP-NN model by choosing the optimal weight and biases for the estimation of accurate parameters and speed of IMD. The objective of optimal MLP-NN is to improve the IMD reliability and response fast during dynamic operation. The model performances are evaluated by employing statistical metrics of MSE, RMSE, MAE, MAPE, and R for training and testing. These are reported for testing to be 0.000602064, 0.0245, 0.4015, 0.25474, and 0.9997 which indicates the best-fitted prediction model and proves the minimized error. The results reveal that an optimized MLP-NN accomplishes promising performance in estimating the parameters and speed with the least errors such as rs is 3.82%, rr is 4.19%, ls is 0.41%, lr is 0.72%, lm is 0.21%, and strongly tracking of reference speed. In addition, HHO is also employed to evolve the gains of the PI-controller in adaptive-IFOC for generation of reference signals by reducing the computational effort.

Rapid discrimination of the identity of infant formula by triple-channel models

Infant formula is related to children's life and health. However, the existing identification methods for infant formula are time-consuming, costly and prone to environmental pollution. Therefore, a simple, efficient and less polluting identification method for infant formula is urgently needed. The aim of this study was to distinguish between goat and cow infant formula using HS-SPME-GC–MS and E-nose combined with triple-channel models. The results indicated that the main difference of them attributed to thirteen volatile compounds and three sensor variables. Based on this, the linear discriminant and partial least squares discriminant analyses were conducted, and a multilayer perceptron neural network model was constructed to identify the commercial samples. There was a high percentage of correct classifications (>90%) in samples. Together, our work demonstrated that the volatile compounds of infant formula combined with chemometric analysis were effective and rapid for detecting two infant formulas.

Preeclampsia Susceptibility Assessment Based on Deep Learning Modeling and Single Nucleotide Polymorphism Analysis.

The early diagnosis of preeclampsia, a key outlook in improving pregnancy outcomes, still remains elusive. The present study aimed to examine the interleukin-13 and interleukin-4 pathway potential in the early detection of preeclampsia as well as the relationship between interleukin-13 rs2069740(T/A) and rs34255686(C/A) polymorphisms and preeclampsia risk to present a combined model. This study utilized raw data from the GSE149440 microarray dataset, and an expression matrix was constructed using the RMA method and affy package. The genes related to the interleukin-13 and interleukin-4 pathway were extracted from the GSEA, and their expression levels were applied to design multilayer perceptron and PPI graph convolutional neural network models. Moreover, genotyping for the rs2069740(T/A) and rs34255686(C/A) polymorphisms of the interleukin-13 gene were tested using the amplification refractory mutation system PCR method. The outcomes revealed that the expression levels of interleukin-4 and interleukin-13 pathway genes could significantly differentiate early preeclampsia from normal pregnancy. Moreover, the present study's data suggested significant differences in the genotype distribution, the allelic frequencies and some of the risk markers of the study, in the position of rs34255686 and rs2069740 polymorphisms between the case and control groups. A combined test of two single nucleotide polymorphisms and an expression-based deep learning model could be designed for future preeclampsia diagnostic purposes.

Multilayer perceptron neural network with regression and ranking loss for patient-specific quality assurance

Patient-specific quality assurance (PSQA) of volumetric modulated arc therapy (VMAT) treatment plans is crucial to enable the plans to be validated for clinical acceptance. However, performing PSQA for clinical delivery is labor-intensive and time-consuming. The existing prediction models do not take into account the dynamic delivery process of VMAT plans. To solve the above problems and improve accuracy of PSQA, this paper presents a multilayer perceptron (MLP) neural network model with regression and ranking loss to predict the gamma passing rate (GPR). The proposed model combines a convolutional neural network with multiple MLP blocks for extracting inter-image correlation features of plan files during dynamic delivery. To focus on the similarity and specificity of multiple VMAT plans, a regression and ranking loss function with dynamic weights is proposed to optimize the training process. In addition, a clinical workflow is proposed to combine the designed model with measurement-based PSQA to screen potential risky plans better. A total of 690 VMAT plans from multiple treatment sites are collected to validate the performance. For 2%/2 mm, 3%/2 mm and 3%/3 mm, the best result of mean absolute error and max error between measured and predicted GPR are 2.17%, 1.25%, 0.74%, and 7.89%, 4.29%, 3.05%, respectively. Experimental results demonstrate that the proposed method has a state-of-the-art performance and can improve the VMAT PSQA process and reduce PSQA workloads.

Analysis of Hotel Water-Use Behavior Based on the MLP-SEM Model

As a representative service industry, the hotel industry has a complex water-use structure and high water consumption. It is of great significance to investigate the mechanisms determining hotel water-use behavior for demand analysis, as this would make it possible to enhance water-use efficiency and enact targeted water-saving measures. Using Spearman’s hierarchical correlation coefficient, the multi-layer perceptron (MLP) neural network model, and the structural equation model (SEM), in this study, we explored the mechanism determining hotel consumers’ water-use behavior from different dimensions and constructed a typical water-use behavior model based on the MLP-SEM model. In terms of individual water-use behavior, the results showed that individual characteristics, water-conservation awareness, and consumption behavior possessed significant differences regarding their influence on and correlation with various water-use behaviors. The most relevant factors influencing each behavior, namely washing up, hand washing, and drinking, were daily stay in the hotel, education, and income. Gender had the greatest impact on bathing and toilet-flushing water-use behaviors. The importance of daily stay in the hotel was 0.181, which meant that this was the most significant factor influencing the direct water-use behavior of hotel guests. The following factors were identified: hotel type, income, age, and gender. Typical individual characteristics had a significant impact on main water-use behaviors, whereas typical consumption behaviors had no effect. These results can provide a foundation for relevant research in other industries and serve as a basis for a prediction model of water consumption in hotels based on water-use behavior. Furthermore, they provide a basis for the delicate management of water-use behavior in hotels, making it possible to effectively guide the public to consciously adopt water-saving habits, thus improving water efficiency, which could alleviate the shortage of water resources in the long-term.

Estimation of water's surface elevation in compound channels with converging and diverging floodplains using soft computing techniques

Abstract In this research, water's surface elevation in compound channels with converging and diverging floodplains using soft computing models including the Multi-Layer Perceptron Neural Network (MLPNN), Group Method of Data Handling (GMDH), Neuro-Fuzzy Group Method of Data Handling (NF-GMDH) and Support Vector Machine (SVM) was modeled and predicted. For this purpose, laboratory data published in this field were used. Parameters including convergence angle (with a positive sign) and divergence angle (with a negative sign), relative depth, and relative distance were used as input variables. The results showed that all the used models have appropriate performance. However, the best performance was related to the SVM model with statistical indicators R2 = 0.998 and RMSE = 0.008 in the testing stage. The use of the adaptive fuzzy approach in developing the GMDH model led to a remarkable increase in accuracy so that its statistical indicators in the testing stage reached R2 = 0.985 and RMSE = 0.203. It was found that the best performance of the activation and kernel functions in the development of the MLPNN model and the SVM is related to the sigmoid and radial tangent functions.

Bio-hydrogen production from the photocatalytic conversion of wastewater: Parametric analysis and data-driven modelling using nonlinear autoregressive with exogeneous input and back-propagated multilayer perceptron neural networks

The quest for energy and environmental sustainability necessitates an increasing interest in the photocatalytic conversion of wastewater to biohydrogen. However, the complexity of the photocatalytic conversion and the low productivity of the biohydrogen produced has become a major concern in the scale-up of the process. This study employs a data-driven approach to model biohydrogen production from the photocatalytic conversion of wastewater. Having ascertained the influence of five different parameters namely catalyst size, reaction temperature, catalyst among, irradiation time, and radiation intensity on the biohydrogen production through parametric analysis, the data were employed to model the process using multilayer perceptron neural network (MLPNN) and nonlinear autoregressive neural network (NARX). Both the MLPNN and NARX models were trained using Levenberg-Marquardt (LM), Bayesian regularization (BR), and scaled conjugate gradient (SCG) algorithms. The performance of 20 network architectures was tested for MLPNN-LM, MLPNN-BR, MLPNN-SCG, NARX-LM, NARX-BR, and NARX-SCG. The analysis revealed that the best network architectures of 5-14-1, 5-11-1, 5-7-1, 5-14-1, 5-15-1, and 5-7-1 were obtained for the MLPNN-LM, MLPNN-BR, MLPNN-SCG, NARX-LM, NARX-BR, and NARX-SCG, respectively. All the models demonstrated a good predictability of the biohydrogen production as evidenced by the coefficient of determination (R2) > 0.9 and low root mean square error (RMSE) values. The best performance was displayed by MLPNN-BR model with R2 of 0.999 and RMSE of 0.138. The independent variable analysis shows that all the factors significantly influence the predicted biohydrogen production. The catalyst size has the most significant effect on the predicted hydrogen production as indicated by the importance value of 0.329.

Prediction of Credit Card Loan Risk Based on Multilayer Perceptron Neural Network Model

Under the background of Internet finance and the popularization of intelligent terminal equipment, the scale of data and the diversity of information have developed exponentially rapidly, while the bad debt rate of personal credit is vague and much higher than that of commercial banks in the traditional technology period. Most of the traditional credit evaluation is driven by models, and the robustness of the risk control system is poor, which cannot meet the increasingly complex demand for default risk prediction. Since commercial banks have accumulated large-scale data assets, it has great significance to combine machine learning technology to help banks extract effective information from massive data and achieve risk assessment of borrowers, thereby reducing the risk of default in borrowing. This paper selects a credit card related data set, uses principal component analysis to extract feature vectors, and obtains a risk assessment index system for borrowers. Then, based on MLP neural network, a credit card loan risk prediction model is constructed. The study reads: The MLP model based on principal component analysis has high accuracy, fast running speed and strong stability. It is an ideal model for commercial banks to evaluate the credit risk of borrowers. This study can provide a new reference for commercial banks to solve the problem of credit risk prediction.

Improvement of spatial estimation for soil organic carbon stocks in Yuksekova plain using Sentinel 2 imagery and gradient descent-boosted regression tree.

Carbon sequestration in earth surface is higher than the atmosphere, and the amount of carbon stored in wetlands is much greater than all other land surfaces. The purpose of this study was to estimate soil organic carbon stocks (SOCS) and investigate spatial distribution pattern of Yuksekova wetlands and surrounding lands in Hakkari province of Turkey using machine learning and remote sensing data. Disturbed and undisturbed soil samples were collected from 10-cm depth in 50 locations differed with land use and land cover. Vegetation, soil, and moisture indices were calculated using Sentinel 2 Multispectral Sensor Instrument (MSI) data. Significant correlations (p≤0.01) were obtained between the indices and SOCS; thus, the remote sensing indices (ARVI 0.43, BI -0.43, GSI -0.39, GNDI 0.44, NDVI 0.44, NDWI 0.38, and SRCI 0.51) were used as covariates in multi-layer perceptron neural network (MLP) and gradient descent-boosted regression tree (GBDT) machine learning models. Mean absolute error, root mean square error, and mean absolute percentage error were 3.94 (Mg C ha -1), 6.64 (Mg C ha-1), and 9.97%, respectively. The simple ratio clay index (SRCI), which represents the soil texture, was the most important factor in the SOCS estimation variance. In addition, the relationship between SRCI and Topsoil Grain Size Index revealed that topsoil clay content is a highly important parameter in spatial variation of SOCS. The spatial SOCS values obtained using the GBDT model and the mean SOCS values of the CORINE land cover classes were significantly different. The land cover has a significant effect on SOC in Yuksekova plain. The mean SOCS for continuously ponded fields was 45.58 Mg C ha-1, which was significantly different from the mean SOCS of arable lands. The mean SOCS in arable lands, with significant areas of natural vegetation, was 50.22 Mg C ha-1 and this amount was significantly higher from the SOCS of other land covers (p<0.01). The wetlands had the highest SOCS (61.46 Mg C ha-1), followed by the lands principally occupied by natural vegetation and used as rangelands around the wetland (50.22 Mg C ha-1). Environmental conditions had significant effect on SOCS in the study area. The use of remote sensing indices instead of using single bands as estimators in the GBDT algorithm minimized radiometric errors, and reliable spatial SOCS information was obtained by using the estimators. Therefore, the spatial estimation of SOCS can be successfully determined with up-to-date machine learning algorithms only using remote sensing predictor variables. Reliable estimation of SOCS in wetlands and surrounding lands can help understand policy and decision makers the importance of wetlands in mitigating the negative impacts of global warming.

Color constancy from a pure color view.

Great efforts have been made on illuminant estimation in both academia and industry, leading to the development of various statistical- and learning-based methods. Little attention, however, has been given to images that are dominated by a single color (i.e., pure color images), though they are not trivial to smartphone cameras. In this study, a pure color image dataset, "PolyU Pure Color," was developed. A lightweight feature-based multilayer perceptron (MLP) neural network model-"Pure Color Constancy (PCC)"-was also developed for estimating the illuminant of pure color images using four color features (i.e., the chromaticities of the maximal, mean, brightest, and darkest pixels) of an image. The proposed PCC method was found to have significantly better performance for pure color images in the PolyU Pure Color dataset and comparable performance for normal images in two existing image datasets, in comparison to the various state-of-the-art learning-based methods, with a good cross-sensor performance. Such good performance was achieved with a much smaller number of parameters (i.e., around 400) and a very short processing time (i.e., around 0.25 ms) for an image using an unoptimized Python package. This makes the proposed method possible for practical deployments.

Classification of the Central Effects of Transcutaneous Electroacupuncture Stimulation (TEAS) at Different Frequencies: A Deep Learning Approach Using Wavelet Packet Decomposition with an Entropy Estimator

The field of signal processing using machine and deep learning algorithms has undergone significant growth in the last few years, with a wide scope of practical applications for electroencephalography (EEG). Transcutaneous electroacupuncture stimulation (TEAS) is a well-established variant of the traditional method of acupuncture that is also receiving increasing research attention. This paper presents the results of using deep learning algorithms on EEG data to investigate the effects on the brain of different frequencies of TEAS when applied to the hands in 66 participants, before, during and immediately after 20 min of stimulation. Wavelet packet decomposition (WPD) and a hybrid Convolutional Neural Network Long Short-Term Memory (CNN-LSTM) model were used to examine the central effects of this peripheral stimulation. The classification results were analysed using confusion matrices, with kappa as a metric. Contrary to expectation, the greatest differences in EEG from baseline occurred during TEAS at 80 pulses per second (pps) or in the ‘sham’ (160 pps, zero amplitude), while the smallest differences occurred during 2.5 or 10 pps stimulation (mean kappa 0.414). The mean and CV for kappa were considerably higher for the CNN-LSTM than for the Multilayer Perceptron Neural Network (MLP-NN) model. As far as we are aware, from the published literature, no prior artificial intelligence (AI) research appears to have been conducted into the effects on EEG of different frequencies of electroacupuncture-type stimulation (whether EA or TEAS). This ground-breaking study thus offers a significant contribution to the literature. However, as with all (unsupervised) DL methods, a particular challenge is that the results are not easy to interpret, due to the complexity of the algorithms and the lack of a clear understanding of the underlying mechanisms. There is therefore scope for further research that explores the effects of the frequency of TEAS on EEG using AI methods, with the most obvious place to start being a hybrid CNN-LSTM model. This would allow for better extraction of information to understand the central effects of peripheral stimulation.

Machine learning classification of non-Markovian noise disturbing quantum dynamics

In this paper machine learning and artificial neural network models are proposed for the classification of external noise sources affecting a given quantum dynamics. For this purpose, we train and then validate support vector machine, multi-layer perceptron and recurrent neural network models with different complexity and accuracy, to solve supervised binary classification problems. As a result, we demonstrate the high efficacy of such tools in classifying noisy quantum dynamics using simulated data sets from different realizations of the quantum system dynamics. In addition, we show that for a successful classification one just needs to measure, in a sequence of discrete time instants, the probabilities that the analysed quantum system is in one of the allowed positions or energy configurations. Albeit the training of machine learning models is here performed on synthetic data, our approach is expected to find application in experimental schemes, as e.g. for the noise benchmarking of noisy intermediate-scale quantum devices.

Automatic measurement and prediction of Chinese Grown Pigs weight using multilayer perceptron neural networks

The knowledge of body size/weight is necessary for the general growth enhancement of swine as well as for making informed decisions that concern their health, productivity, and yield. Therefore, this work aims to automate the collection of pigs’ body parameters using images from Kinect V2 cameras, and the development of Multilayer Perceptron Neural Network (MLP NN) models to predict their weight. The dataset obtained using 3D light depth cameras contains 9980 pigs across the S21 and S23 breeds, and then grouped into 70:15:15 training, testing, and validation sets, respectively. Initially, two MLP models were built and evaluations revealed that model 1 outperformed model 2 in predicting pig weights, with root mean squared error (RMSE) values of 5.5 and 6.0 respectively. Moreover, employing a normalized dataset, two new models (3 and 4) were developed and trained. Subsequently, models 2, 3, and 4 performed significantly better with a RMSE value of 5.29 compared to model 1, which has a RMSE value of 6.95. Model 3 produced an intriguing discovery i.e. accurate forecasting of pig weights using just two characteristics, age and abdominal circumference, and other error values show corresponding results

Intelligent Measurement of Void Fractions in Homogeneous Regime of Two Phase Flows Independent of the Liquid Phase Density Changes

Determining the amount of void fraction of multiphase flows in pipelines of the oil, chemical and petrochemical industries is one of the most important challenges. Performance of capacitance based two phase flow meters highly depends on the fluid properties. Fluctuation of the liquid phase properties such as density, due to temperature and pressure changes, would cause massive errors in determination of the void fraction. A common approach to fix this problem is periodic recalibration of the system, which is a tedious task. The aim of this study is proposing a method based on artificial intelligence (AI), which offers the advantage of intelligent measuring of the void fraction regardless of the liquid phase changes without the need for recalibration. To train AI, a data set for different liquid phases is required. Although it is possible to obtain the required data from experiments, it is time-consuming and also incorporates its own specific safety laboratory consideration, particularly working with flammable liquids such as gasoline, oil and gasoil. So, COMSOL Multiphysics software was used to model a homogenous regime of two-phase flow with five different liquid phases and void fractions. To validate the simulation geometry, initially an experimental setup including a concave sensor to measure the capacitance by LCR meter for the case that water used as the liquid phase, was established. After validation of the simulated geometry for concave sensor, a ring sensor was also simulated to investigate the best sensor type. It was found that the concave type has a better sensitivity. Therefore, the concave type was used to measure the capacitance for different liquid phases and void fractions inside the pipe. Finally, simulated data were used to train a Multi-Layer Perceptron (MLP) neural network model in MATLAB software. The trained MLP model was able to predict the void fraction independent of the liquid phase density changes with a Mean Absolute Error (MAE) of 1.74.

Performance Comparison of Machine Learning Disruption Predictors at JET

Reliable disruption prediction (DP) and disruption mitigation systems are considered unavoidable during international thermonuclear experimental reactor (ITER) operations and in the view of the next fusion reactors such as the DEMOnstration Power Plant (DEMO) and China Fusion Engineering Test Reactor (CFETR). In the last two decades, a great number of DP systems have been developed using data-driven methods. The performance of the DP models has been improved over the years both for a more appropriate choice of diagnostics and input features and for the availability of increasingly powerful data-driven modelling techniques. However, a direct comparison among the proposals has not yet been conducted. Such a comparison is mandatory, at least for the same device, to learn lessons from all these efforts and finally choose the best set of diagnostic signals and the best modelling approach. A first effort towards this goal is made in this paper, where different DP models will be compared using the same performance indices and the same device. In particular, the performance of a conventional Multilayer Perceptron Neural Network (MLP-NN) model is compared with those of two more sophisticated models, based on Generative Topographic Mapping (GTM) and Convolutional Neural Networks (CNN), on the same real time diagnostic signals from several experiments at the JET tokamak. The most common performance indices have been used to compare the different DP models and the results are deeply discussed. The comparison confirms the soundness of all the investigated machine learning approaches and the chosen diagnostics, enables us to highlight the pros and cons of each model, and helps to consciously choose the approach that best matches with the plasma protection needs.

Retrieving rice (Oryza sativa L.) net photosynthetic rate from UAV multispectral images based on machine learning methods.

Photosynthesis is the key physiological activity in the process of crop growth and plays an irreplaceable role in carbon assimilation and yield formation. This study extracted rice (Oryza sativa L.) canopy reflectance based on the UAV multispectral images and analyzed the correlation between 25 vegetation indices (VIs), three textural indices (TIs), and net photosynthetic rate (Pn) at different growth stages. Linear regression (LR), support vector regression (SVR), gradient boosting decision tree (GBDT), random forest (RF), and multilayer perceptron neural network (MLP) models were employed for Pn estimation, and the modeling accuracy was compared under the input condition of VIs, VIs combined with TIs, and fusion of VIs and TIs with plant height (PH) and SPAD. The results showed that VIs and TIs generally had the relatively best correlation with Pn at the jointing-booting stage and the number of VIs with significant correlation (p< 0.05) was the largest. Therefore, the employed models could achieve the highest overall accuracy [coefficient of determination (R 2) of 0.383-0.938]. However, as the growth stage progressed, the correlation gradually weakened and resulted in accuracy decrease (R 2 of 0.258-0.928 and 0.125-0.863 at the heading-flowering and ripening stages, respectively). Among the tested models, GBDT and RF models could attain the best performance based on only VIs input (with R 2 ranging from 0.863 to 0.938 and from 0.815 to 0.872, respectively). Furthermore, the fusion input of VIs, TIs with PH, and SPAD could more effectively improve the model accuracy (R 2 increased by 0.049-0.249, 0.063-0.470, and 0.113-0.471, respectively, for three growth stages) compared with the input combination of VIs and TIs (R 2 increased by 0.015-0.090, 0.001-0.139, and 0.023-0.114). Therefore, the GBDT and RF model with fused input could be highly recommended for rice Pn estimation and the methods could also provide reference for Pn monitoring and further yield prediction at field scale.

Characterization of Relative Movements between Blocks Observed in a Concrete Dam and Definition of Thresholds for Novelty Identification Based on Machine Learning Models

Dam surveillance activities are based on observing the structural behaviour and interpreting the past behaviour supported by the knowledge of the main loads. For day-to-day activities, data-driven models are usually adopted. Most applications consider regression models for the analysis of horizontal displacements recorded in pendulums. Traditional regression models are not commonly applied to the analysis of relative movements between blocks due to the non-linearities related to the simultaneity of hydrostatic and thermal effects. A new application of a multilayer perceptron neural network model is proposed to interpret the relative movements between blocks measured hourly in a concrete dam under exploitation. A new methodology is proposed for threshold definition related to novelty identification, taking into account the evolution of the records over time and the simultaneity of the structural responses measured in the dam under study. The results obtained through the case study showed the ability of the methodology presented in this work to characterize the relative movement between blocks and for the identification of novelties in the dam behaviour.