Advances in Artificial Neural Networks – Methodological Development and Application

Graph neural networks-enhanced relation prediction for ecotoxicology (GRAPE)

Exposure to toxic chemicals threatens species and ecosystems. This study introduces a novel approach using Graph Neural Networks (GNNs) to integrate aquatic toxicity data, providing an alternative to complement traditional in vivo ecotoxicity testing. This study pioneers the application of GNN in ecotoxicology by formulating the problem as a relation prediction task. GRAPE's key innovation lies in simultaneously modelling 444 aquatic species and 2826 chemicals within a graph, leveraging relations from existing datasets where informative species and chemical features are augmented to make informed predictions. Extensive evaluations demonstrate the superiority of GRAPE over Logistic Regression (LR) and Multi-Layer Perceptron (MLP) models, achieving remarkable improvements of up to a 30% increase in recall values. GRAPE consistently outperforms LR and MLP in predicting novel chemicals and new species. In particular, GRAPE showcases substantial enhancements in recall values, with improvements of ≥ 100% for novel chemicals and up to 13% for new species. Specifically, GRAPE correctly predicts the effects of novel chemicals (104 out of 126) and effects on new species (7 out of 8). Moreover, the study highlights the effectiveness of the proposed chemical features and induced network topology through GNN for accurately predicting metallic (74 out of 86) and organic (612 out of 674) chemicals, showcasing the broad applicability and robustness of the GRAPE model in ecotoxicological investigations. The code/data are provided at https://github.com/csiro-robotics/GRAPE.

SDBA: Score Domain-Based Attention for DNA N4-Methylcytosine Site Prediction from Multiperspectives

In tasks related to DNA sequence classification, choosing the appropriate encoding methods is challenging. Some of the methods encode sequences based on prior knowledge that limits the ability of the model to obtain multiperspective information from the sequences. We introduced a new trainable ensemble method based on the attention mechanism SDBA, which stands for Score Domain-Based Attention. Unlike other methods, we fed the task-independent encoding results into the models and dynamically ensembled features from different perspectives using the SDBA mechanism. This approach allows the model to acquire and weight sequence features voluntarily. SDBA is conceptually general and empirically powerful. It has achieved new state-of-the-art results on the benchmark data sets associated with DNA N4-methylcytosine site prediction.

Research on safety evaluation of collapse risk in highway tunnel construction based on intelligent fusion

To solve the problems of untimely and low accuracy of tunnel project collapse risk prediction, this study proposes a method of multi-source information fusion. The method uses the PSO-SVM model to predict the surrounding rock displacement. With the prediction index as the benchmark, the Cloud Model (CM) is used to calculate the basic probability assignment value. At the same time, the improved D-S theory is used to fuse the monitoring data, the advanced geological forecast, and the tripartite information indicators of site inspection patrol. This method is applied to the risk assessment of Jinzhupa Tunnel, and the decision-makers adjust the risk factors in time according to the prediction level. In the end, the tunnel did not collapse on a large scale.

Application of Machine Learning in Material Synthesis and Property Prediction

Material innovation plays a very important role in technological progress and industrial development. Traditional experimental exploration and numerical simulation often require considerable time and resources. A new approach is urgently needed to accelerate the discovery and exploration of new materials. Machine learning can greatly reduce computational costs, shorten the development cycle, and improve computational accuracy. It has become one of the most promising research approaches in the process of novel material screening and material property prediction. In recent years, machine learning has been widely used in many fields of research, such as superconductivity, thermoelectrics, photovoltaics, catalysis, and high-entropy alloys. In this review, the basic principles of machine learning are briefly outlined. Several commonly used algorithms in machine learning models and their primary applications are then introduced. The research progress of machine learning in predicting material properties and guiding material synthesis is discussed. Finally, a future outlook on machine learning in the materials science field is presented.

Nondestructive Estimation of Hazelnut (Corylus avellana L.) Terminal Velocity and Drag Coefficient Based on Some Fruit Physical Properties Using Machine Learning Algorithms

Hazelnut culture originated in Turkey, which has the highest volume and area of hazelnut production in the world. For the design and sizing of equipment and structures in agricultural operations for the hazelnut industry, especially harvesting operations and post-harvest operations, it is essential that an understanding of hazelnuts' aerodynamic properties, i.e., terminal velocity and drag coefficient, is acquired. In this study, the moisture, mass, density, projected area, surface area, and geometric diameter were used as independent variables in the data set, and the dependent variables terminal velocity and drag coefficient estimation were determined. In this study, logistic regression (LR), support vector regression (SVR), and artificial neural networks (ANNs) were used based on machine learning methods. When the results were evaluated according to R2 (determination coefficient), MSE (mean squared error), and MAE (mean absolute error) metrics, it was seen that the most successful models were the ANN, SVR, and LR, respectively. According to the R2 metric, the ANN method achieved 91.5% for the terminal velocity of hazelnuts and 85.9% for the drag coefficient of hazelnuts. Using the independent variables in the study, it was seen that the terminal velocity and drag coefficient value of hazelnuts could be successfully estimated.

Machine Learning Methods for Small Data Challenges in Molecular Science

Small data are often used in scientific and engineering research due to the presence of various constraints, such as time, cost, ethics, privacy, security, and technical limitations in data acquisition. However, big data have been the focus for the past decade, small data and their challenges have received little attention, even though they are technically more severe in machine learning (ML) and deep learning (DL) studies. Overall, the small data challenge is often compounded by issues, such as data diversity, imputation, noise, imbalance, and high-dimensionality. Fortunately, the current big data era is characterized by technological breakthroughs in ML, DL, and artificial intelligence (AI), which enable data-driven scientific discovery, and many advanced ML and DL technologies developed for big data have inadvertently provided solutions for small data problems. As a result, significant progress has been made in ML and DL for small data challenges in the past decade. In this review, we summarize and analyze several emerging potential solutions to small data challenges in molecular science, including chemical and biological sciences. We review both basic machine learning algorithms, such as linear regression, logistic regression (LR), k-nearest neighbor (KNN), support vector machine (SVM), kernel learning (KL), random forest (RF), and gradient boosting trees (GBT), and more advanced techniques, including artificial neural network (ANN), convolutional neural network (CNN), U-Net, graph neural network (GNN), Generative Adversarial Network (GAN), long short-term memory (LSTM), autoencoder, transformer, transfer learning, active learning, graph-based semi-supervised learning, combining deep learning with traditional machine learning, and physical model-based data augmentation. We also briefly discuss the latest advances in these methods. Finally, we conclude the survey with a discussion of promising trends in small data challenges in molecular science.

Exploring the influence of personal factors on physiological responses to mental imagery in sport

Imagery is a well-known technique in mental training which improves performance efficiency and influences physiological arousal. One of the biomarkers indicating the amount of physiological arousal is skin conductance level (SCL). The aim of our study is to understand how individual differences in personality (e.g. neuroticism), general imagery and situational sport anxiety are linked to arousal measuring with SCL in situational imagery. Thirty participants aged between 14 and 42 years (M = 22.93; SD = 5.24), with sport experience ranging between 2 and 20 years (M = 10.15; SD = 4.75), took part in our study. Participants listened to each previously recorded script and then were asked to imagine the scene for a minute. During the task SCL was monitored using the Biofeedback Expert 2000. Machine learning predictive models based on artificial neural networks have been trained for prediction of physiological response, as a function of selected psychological tests. We found an association among neuroticism, prestart anxiety, and general tendency to use imagery with SCL. From a practical point of view our results may help athletes, coaches, and psychologists to be more aware of the role of individual differences in sport.

Application of artificial neural network model in predicting profitability of Indian banks

The aim of this study is to predict the profitability of Indian banks. Several factors both internal and external, affecting bank profitability were derived from extensive review of literature. We used Artificial Neural Network (ANN) with cross-validation technique to perform predictive analysis. ANN was chosen due to its flexibility and non-linear modelling capability. Several structures of ANN with a single and two hidden layers along with varying hidden neurons were implemented. Further, a comparison was made with the multiple linear regression (MLR) model. We found the models based on ANN to offer very accurate results in prediction and are marginally better as compared to the regression model. Higher accuracy of the model makes a significant difference due to the astronomically large size of the balance sheet of banks. This article is unique in the approach of handling the panel data for predictive analysis wherein the training of the model was done on a single bank’s data, thus, reducing the panel data to a time series data. This approach shows the ability to work with large panel data and make accurate predictions.

Artificial Neural Network Training Using Structural Learning with Forgetting for Parameter Analysis of Injection Molding Quality Prediction

The analysis of influential machine parameters can be useful to plan and design a plastic injection molding process. However, current research in parameter analysis is mostly based on computer-aided engineering (CAE) or simulation which have been demonstrated to be inadequate for analyzing complex behavioral changes in the real injection molding process. More advanced approaches using machine learning technology specifically with artificial neural networks (ANNs) brought promising results in terms of prediction accuracy. Nevertheless, the black box and distributed representation of ANN prevent humans from gaining an insight into which process parameters give a significant influence on the final prediction output. Therefore, in this paper, we develop a simpler ANN model by using structural learning with forgetting (SLF) as the algorithm for the training process. Instead of typical backpropagation which generated a fully connected layer of the ANN model, SLF only reveals the important neurons and connections. Hence, the training process of SLF leaves only influential connections and neurons. Since each of the neurons specifically on the input layer represent each of the injection molding parameters, the ANN-SLF model can be further investigated to determine the influential process parameters. By applying SLF to the ANN training process, this experiment has successfully extracted a set of significant injection molding process parameters.

Prediction Models for Evaluating Resilient Modulus of Stabilized Aggregate Bases in Wet and Dry Alternating Environments: ANN and GEP Approaches

Stabilized aggregate bases are vital for the long-term service life of pavements. Their stiffness is comparatively higher; therefore, the inclusion of stabilized materials in the construction of bases prevents the cracking of the asphalt layer. The effect of wet−dry cycles (WDCs) on the resilient modulus (Mr) of subgrade materials stabilized with CaO and cementitious materials, modelled using artificial neural network (ANN) and gene expression programming (GEP) has been studied here. For this purpose, a number of wet−dry cycles (WDC), calcium oxide to SAF (silica, alumina, and ferric oxide compounds in the cementitious materials) ratio (CSAFRs), ratio of maximum dry density to the optimum moisture content (DMR), confining pressure (σ3), and deviator stress (σ4) were considered input variables, and Mr was treated as the target variable. Different ANN and GEP prediction models were developed, validated, and tested using 30% of the experimental data. Additionally, they were evaluated using statistical indices, such as the slope of the regression line between experimental and predicted results and the relative error analysis. The slope of the regression line for the ANN and GEP models was observed as (0.96, 0.99, and 0.94) and (0.72, 0.72, and 0.76) for the training, validation, and test data, respectively. The parametric analysis of the ANN and GEP models showed that Mr increased with the DMR, σ3, and σ4. An increase in the number of WDCs reduced the Mr value. The sensitivity analysis showed the sequences of importance as: DMR > CSAFR > WDC > σ4 > σ3, (ANN model) and DMR > WDC > CSAFR > σ4 > σ3 (GEP model). Both the ANN and GEP models reflected close agreement between experimental and predicted results; however, the ANN model depicted superior accuracy in predicting the Mr value.

Hyperspectral plant sensing for differentiating glyphosate-resistant and glyphosate-susceptible johnsongrass through machine learning algorithms

BACKGROUND

Johnsongrass (Sorghum halepense) is one of the weeds that evolves resistance to glyphosate [N-(phosphonomethyl)-glycine], the most widely used herbicide, and the weed may cause agronomic troublesome in the southern USA. This paper reports a study on developing a hyperspectral plant sensing approach to explore the spectral features of glyphosate-resistant (GR) and glyphosate-sensitive (GS) plants to evaluate this approach using machine learning algorithms to differentiate between GR and GS plants.

RESULTS

On average, GR plants have higher spectral reflectance compared with GS plants. The sensitive spectral bands were optimally selected using the successive projections algorithm respectively wrapped with the machine learning algorithms of k-nearest neighbors (KNN), random forest (RF), and support vector machine (SVM) with Fisher linear discriminant analysis (FLDA) to classify between GS and GS plants. At 3 weeks after transplanting (WAT) KNN and SVM could not acceptably classify the GR and GS plants but they improved significantly with the stages to have their overall accuracies reaching 73% and 77%, respectively, at 5 WAT. RF and FLDA had a better ability to classify the plants at 3 WAT but RF was low in accuracy at 2 WAT while FLDA dropped accuracy to 50% at 4 WAT from 57% at 3 WAT and raised it to 73% at 5 WAT.

CONCLUSIONS

Previous studies were conducted developing the hyperspectral imaging approach to differentiate GR Palmer amaranth from GS Palmer amaranth and GR Italian ryegrass from GS Italian ryegrass with classification accuracies of 90% and 80%, respectively. This study demonstrated that the hyperspectral plant sensing approach could be developed to differentiate GR johnsongrass from glyphosate-sensitive GS johnsongrass with the highest classification accuracy of 77%. The comparison with our previous studies indicated that the similar hyperspectral approach could be used and transferred from classification across different GR and GS weed biotypes, such as Palmer amaranth, Italian ryegrass and johnsongrass, so it is highly possible for classification of more other GR and GS weed biotypes as well. On the basis of classic pattern recognition approaches the process of plant classification can be enhanced by modeling using machine learning algorithms. © 2022 Society of Chemical Industry. This article has been contributed to by U.S. Government employees and their work is in the public domain in the USA.

A Study on the Impact of Digital Finance on Regional Productivity Growth Based on Artificial Neural Networks

The relationship between financial development and economic growth has become a hot topic in recent years and for China, which is undergoing financial liberalisation and policy reform, the efficiency of the use of digital finance and the deepening of the balance between quality and quantity in financial development are particularly important for economic growth. This paper investigates the utility of digital finance and financial development on total factor productivity in China using interprovincial panel data decomposing financial development into financial scale and financial efficiency; an interprovincial panel data model is used to explore the utility of digital finance on total factor productivity. This involves the collection and preprocessing of financial data, including feature engineering, and the development of an optimised predictive model. We preprocess the original dataset to remove anomalous information and improve data quality. This work uses feature engineering to select relevant features for fitting and training the model. In this process, the random forest algorithm is used to effectively avoid overfitting problems and to facilitate the dimensionality reduction of the relevant features. In determining the model to be used, the random forest regression model was chosen for training. The empirical results show that digital finance has contributed to productivity growth but is not efficiently utilised; China should give high priority to improving financial efficiency while promoting financial expansion; rapid expansion of finance without a focus on financial efficiency will not be conducive to productivity growth.

Predictive Models of Phytosterol Degradation in Rapeseeds Stored in Bulk Based on Artificial Neural Networks and Response Surface Regression

The need to maintain the highest possible levels of bioactive components contained in raw materials requires the elaboration of tools supporting their processing operations, starting from the first stages of the food production chain. In this study, artificial neural networks (ANNs) and response surface regression (RSR) were used to develop models of phytosterol degradation in bulks of rapeseed stored under various temperatures and water activity conditions (T = 12-30 °C and aw = 0.75-0.90). Among ANNs, networks based on a multilayer perceptron (MLP) and a radial basis function (RBF) were tested. The model input constituted aw, temperature and storage time, whilst the model output was the phytosterol level in seeds. The ANN-based modeling turned out to be more effective in estimating phytosterol levels than the RSR, while MLP-ANNs proved to be more satisfactory than RBF-ANNs. The approximation quality of the ANNs models depended on the number of neurons and the type of activation functions in the hidden layer. The best model was provided by the MLP-ANN containing nine neurons in the hidden layer equipped with the logistic activation function. The model performance evaluation showed its high prediction accuracy and generalization capability (R2 = 0.978; RMSE = 0.140). Its accuracy was also confirmed by the elliptical joint confidence region (EJCR) test. The results show the high usefulness of ANNs in predictive modeling of phytosterol degradation in rapeseeds. The elaborated MLP-ANN model may be used as a support tool in modern postharvest management systems.

A multi-source information fusion approach in tunnel collapse risk analysis based on improved Dempster-Shafer evidence theory

The tunneling collapse is the main engineering hazard in the construction of the drilling-and-blasting method. The accurate assessment of the tunneling collapse risk has become a key issue in tunnel construction. As for assessing the tunneling collapse risk and providing basic risk controlling strategies, this research proposes a novel multi-source information fusion approach that combines Bayesian network (BN), cloud model (CM), support vector machine (SVM), Dempster–Shafer (D–S) evidence theory, and Monte Carlo (MC) simulation technique. Those methods (CM, BN, SVM) are used to analyze multi-source information (i.e. statistical data, physical sensors, and expert judgment provided by humans) respectively and construct basic probability assignments (BPAs) of input factors under different risk states. Then, these BPAs will be merged at the decision level to achieve an overall risk evaluation, using an improved D–S evidence theory. The MC technology is proposed to simulate the uncertainty and randomness of data. The novel approach has been successfully applied in the case of the Jinzhupa tunnel of the Pu-Yan Highway (Fujian, China). The results indicate that the developed new multi-source information fusion method is feasible for (a) Fusing multi-source information effectively from different models with a high-risk assessment accuracy of 98.1%; (b) Performing strong robustness to bias, which can achieve acceptable risk assessment accuracy even under a 20% bias; and (c) Exhibiting a more outstanding risk assessment performance (97.9% accuracy) than the single-information model (78.8% accuracy) under a high bias (20%). Since the proposed reliable risk analysis method can efficiently integrate multi-source information with conflicts, uncertainties, and bias, it provides an in-depth analysis of the tunnel collapse and the most critical risk factors, and then appropriate remedial measures can be taken at an early stage.

Impacts of Future Sea-Level Rise under Global Warming Assessed from Tide Gauge Records: A Case Study of the East Coast Economic Region of Peninsular Malaysia

The effects of global warming are putting the world’s coasts at risk. Coastal planners need relatively accurate projections of the rate of sea-level rise and its possible consequences, such as extreme sea-level changes, flooding, and coastal erosion. The east coast of Peninsular Malaysia is vulnerable to sea-level change. The purpose of this study is to present an Artificial Neural Network (ANN) model to analyse sea-level change based on observed data of tide gauge, rainfall, sea level pressure, sea surface temperature, and wind. A Feed-forward Neural Network (FNN) approach was used on observed data from 1991 to 2012 to simulate and predict the sea level change until 2020 from five tide gauge stations in Kuala Terengganu along the East Coast of Malaysia. From 1991 to 2020, predictions estimate that sea level would increase at a pace of roughly 4.60 mm/year on average, with a rate of 2.05 ± 7.16 mm on the East Coast of Peninsular Malaysia. This study shows that Peninsular Malaysia’s East Coast is vulnerable to sea-level rise, particularly at Kula Terengganu, Terengganu state, with a rate of 1.38 ± 7.59 mm/year, and Tanjung Gelang, Pahang state, with a rate of 1.87 ± 7.33 mm/year. As a result, strategies and planning for long-term adaptation are needed to control potential consequences. Our research provides crucial information for decision-makers seeking to protect coastal cities from the risks of rising sea levels.

Effects of Meteorological Parameters on Surface Water Loss in Burdur Lake, Turkey over 34 Years Landsat Google Earth Engine Time-Series

The current work aims to examine the effect of meteorological parameters as well as the temporal variation on the Burdur Lake surface water body areas in Turkey. The data for the surface area of Burdur Lake was collected over 34 years between 1984 and 2019 by Google Earth Engine. The seasonal variation in the water bodies area was collected using our proposed extraction method and 570 Landsat images. The reduction in the total area of the lake was observed between 206.6 km2 in 1984 to 125.5 km2 in 2019. The vegetation cover and meteorological parameters collected that affect the observed variation of surface water bodies for the same area include precipitation, evapotranspiration, albedo, radiation, and temperature. The selected meteorological variables influence the reduction in lake area directly during various seasons. Correlations showed a strong positive or negative significant relationship between the reduction and the selected meteorological variables. A factor analysis provided three components that explain 82.15% of the total variation in the data. The data provide valuable references for decision makers to develop sustainable agriculture and industrial water use policies to preserve water resources as well as cope with potential climate changes.

Mathematical Modeling of Growth and Paclitaxel Biosynthesis in Corylus avellana Cell Culture Responding to Fungal Elicitors Using Multilayer Perceptron-Genetic Algorithm

Paclitaxel is the top-selling anticancer medicine in the world. In vitro culture of Corylus avellana has been made known as a promising and inexpensive strategy for producing paclitaxel. Fungal elicitors have been named as the most efficient strategy for enhancing the biosynthesis of secondary metabolites in plant cell culture. In this study, endophytic fungal strain HEF₁₇ was isolated from C. avellana and identified as Camarosporomyces flavigenus. C. avellana cell suspension culture (CSC) elicited with cell extract (CE) and culture filtrate (CF) derived from strain HEF₁₇, either individually or combined treatment, in mid and late log phase was processed for modeling and optimizing growth and paclitaxel biosynthesis regarding CE and CF concentration levels, elicitor adding day, and CSC harvesting time using multilayer perceptron-genetic algorithm (MLP-GA). The results displayed higher accuracy of MLP-GA models (0.89-0.95) than regression models (0.56-0.85). The great accordance between the predicted and observed values of output variables (dry weight, intracellular, extracellular and total yield of paclitaxel, and also extracellular paclitaxel portion) for both training and testing subsets supported the excellent performance of developed MLP-GA models. MLP-GA method presented a promising tool for selecting the optimal conditions for maximum paclitaxel biosynthesis. An Excel^® estimator, HCC-paclitaxel, was designed based on MLP-GA model as an easy-to-use tool for predicting paclitaxel biosynthesis in C. avellana CSC responding to fungal elicitors.

Mapping the Distribution of Shallow Groundwater Occurrences Using Remote Sensing-Based Statistical Modeling over Southwest Saudi Arabia

Identifying shallow (near-surface) groundwater in arid and hyper-arid areas has significant societal benefits, yet it is a costly operation when traditional methods (geophysics and drilling) are applied over large domains. In this study, we developed and successfully applied methodologies that rely heavily on readily available temporal, visible, and near-infrared radar and thermal remote sensing data sets and field data, as well as statistical approaches to map the distribution of shallow (1–5 m deep) groundwater occurrences in Al Qunfudah Province, Saudi Arabia, and to identify the factors controlling their development. A four-fold approach was adopted: (1) constructing a digital database to host relevant geologic, hydrogeologic, topographic, land use, climatic, and remote sensing data sets, (2) identifying the distribution of areas characterized by shallow groundwater levels, (3) developing conceptual and statistical models to map the distribution of shallow groundwater occurrences, and (4) constructing an artificial neural network (ANN) and multivariate regression (MR) models to map the distribution of shallow groundwater, test the models over areas of known depth to groundwater (area of Al Qunfudah city and surroundings: 294 km2), and apply the better of the two models to map the shallow groundwater occurrences across the entire Al Qunfudah Province (area: 4680 km2). Findings include: (1) high performance for the ANN (92%) and MR (88%) models in predicting the distribution of shallow groundwater using temporal-derived remote sensing products (e.g., normalized difference vegetation index (NDVI), radar backscatter coefficient, precipitation, and brightness temperature) and field data (depth to water table), (2) areas witnessing shallow groundwater levels show high NDVI (mean and standard deviation (STD)), radar backscatter coefficient values (mean and STD), and low brightness temperature (mean and STD) compared to their surroundings, (3) correlations of temporal groundwater levels and satellite-based precipitation suggest that the observed (2017–2019) rise in groundwater levels is related to an increase in precipitation in these years compared to the previous three years (2014–2016), and (4) the adopted methodologies are reliable, cost-effective, and could potentially be applied to identify shallow groundwater along the Red Sea Hills and in similar settings worldwide.

A Novel Autonomous Perceptron Model for Pattern Classification Applications

Pattern classification represents a challenging problem in machine learning and data science research domains, especially when there is a limited availability of training samples. In recent years, artificial neural network (ANN) algorithms have demonstrated astonishing performance when compared to traditional generative and discriminative classification algorithms. However, due to the complexity of classical ANN architectures, ANNs are sometimes incapable of providing efficient solutions when addressing complex distribution problems. Motivated by the mathematical definition of a quantum bit (qubit), we propose a novel autonomous perceptron model (APM) that can solve the problem of the architecture complexity of traditional ANNs. APM is a nonlinear classification model that has a simple and fixed architecture inspired by the computational superposition power of the qubit. The proposed perceptron is able to construct the activation operators autonomously after a limited number of iterations. Several experiments using various datasets are conducted, where all the empirical results show the superiority of the proposed model as a classifier in terms of accuracy and computational time when it is compared with baseline classification models.

Designing, Developing, and Implementing a Forecasting Method for the Produced and Consumed Electricity in the Case of Small Wind Farms Situated on Quite Complex Hilly Terrain

Accurate forecasting of the produced and consumed electricity from wind farms is an essential aspect for wind power plant operators. In this context, our research addresses small-scale wind farms situated on hilly terrain, with the main purpose of overcoming the low accuracy limitations arising from the wind deflection, caused by the quite complex hilly terrain. A specific aspect of our devised forecasting method consists of incorporating advantages of recurrent long short-term memory (LSTM) neural networks, benefiting from their long-term dependencies, learning capabilities, and the advantages of feed-forward function fitting neural networks (FITNETs) that have the ability to map between a dataset of numeric inputs and a set of numeric targets. Another specific element of our approach consists of improving forecasting accuracy by means of refining the accuracy of the weather data input parameters within the same weather forecast resolution area. The developed method has power plant operators as main beneficiaries, but it can also be successfully applied in order to assess the energy potential of hilly areas with deflected wind, being useful for potential investors who want to build this type of wind farms. The method can be compiled and incorporated in the development of a wide range of customized applications targeting electricity forecasting for small wind farms situated on hilly terrain with deflected wind. The experimental results, the implementation of the developed method in a real production environment, its validation, and the comparison between our proposed method and other ones from the literature, confirm that the developed forecasting method represents an accurate, useful, and viable tool that addresses a gap in the current state of knowledge regarding the necessity for an accurate forecasting method that is able to predict with a high degree of accuracy both the produced and consumed electricity for small wind power plants situated on quite complex hilly terrain with deflected wind.

Research on Degeneration Model of Neural Network for Deep Groove Ball Bearing Based on Feature Fusion

Aiming at the pitting fault of deep groove ball bearing during service, this paper uses the vibration signal of five different states of deep groove ball bearing and extracts the relevant features, then uses a neural network to model the degradation for identifying and classifying the fault type. By comparing the effects of training samples with different capacities through performance indexes such as the accuracy and convergence speed, it is proven that an increase in the sample size can improve the performance of the model. Based on the polynomial fitting principle and Pearson correlation coefficient, fusion features based on the skewness index are proposed, and the performance improvement of the model after incorporating the fusion features is also validated. A comparison of the performance of the support vector machine (SVM) model and the neural network model on this dataset is given. The research shows that neural networks have more potential for complex and high-volume datasets.

A Regularizer Approach for RBF Networks Under the Concurrent Weight Failure Situation

Many existing results on fault-tolerant algorithms focus on the single fault source situation, where a trained network is affected by one kind of weight failure. In fact, a trained network may be affected by multiple kinds of weight failure. This paper first studies how the open weight fault and the multiplicative weight noise degrade the performance of radial basis function (RBF) networks. Afterward, we define the objective function for training fault-tolerant RBF networks. Based on the objective function, we then develop two learning algorithms, one batch mode and one online mode. Besides, the convergent conditions of our online algorithm are investigated. Finally, we develop a formula to estimate the test set error of faulty networks trained from our approach. This formula helps us to optimize some tuning parameters, such as RBF width.

Artificial Intelligence versus Statistical Modeling and Optimization of Cholesterol Oxidase Production by using Streptomyces Sp

Cholesterol oxidase (COD) is a bi-functional FAD-containing oxidoreductase which catalyzes the oxidation of cholesterol into 4-cholesten-3-one. The wider biological functions and clinical applications of COD have urged the screening, isolation and characterization of newer microbes from diverse habitats as a source of COD and optimization and over-production of COD for various uses. The practicability of statistical/ artificial intelligence techniques, such as response surface methodology (RSM), artificial neural network (ANN) and genetic algorithm (GA) have been tested to optimize the medium composition for the production of COD from novel strain Streptomyces sp. NCIM 5500. All experiments were performed according to the five factor central composite design (CCD) and the generated data was analysed using RSM and ANN. GA was employed to optimize the models generated by RSM and ANN. Based upon the predicted COD concentration, the model developed with ANN was found to be superior to the model developed with RSM. The RSM-GA approach predicted maximum of 6.283 U/mL COD production, whereas the ANN-GA approach predicted a maximum of 9.93 U/mL COD concentration. The optimum concentrations of the medium variables predicted through ANN-GA approach were: 1.431 g/50 mL soybean, 1.389 g/50 mL maltose, 0.029 g/50 mL MgSO₄, 0.45 g/50 mL NaCl and 2.235 ml/50 mL glycerol. The experimental COD concentration was concurrent with the GA predicted yield and led to 9.75 U/mL COD production, which was nearly two times higher than the yield (4.2 U/mL) obtained with the un-optimized medium. This is the very first time we are reporting the statistical versus artificial intelligence based modeling and optimization of COD production by Streptomyces sp. NCIM 5500.

Lung cancer prediction using neural network ensemble with histogram of oriented gradient genomic features

This paper reports an experimental comparison of artificial neural network (ANN) and support vector machine (SVM) ensembles and their "nonensemble" variants for lung cancer prediction. These machine learning classifiers were trained to predict lung cancer using samples of patient nucleotides with mutations in the epidermal growth factor receptor, Kirsten rat sarcoma viral oncogene, and tumor suppressor p53 genomes collected as biomarkers from the IGDB.NSCLC corpus. The Voss DNA encoding was used to map the nucleotide sequences of mutated and normal genomes to obtain the equivalent numerical genomic sequences for training the selected classifiers. The histogram of oriented gradient (HOG) and local binary pattern (LBP) state-of-the-art feature extraction schemes were applied to extract representative genomic features from the encoded sequences of nucleotides. The ANN ensemble and HOG best fit the training dataset of this study with an accuracy of 95.90% and mean square error of 0.0159. The result of the ANN ensemble and HOG genomic features is promising for automated screening and early detection of lung cancer. This will hopefully assist pathologists in administering targeted molecular therapy and offering counsel to early stage lung cancer patients and persons in at risk populations.

Computational Resources for MHC Ligand Identification

Advances in the high-throughput determination of functional modulators of major histocompatibility complex (MHC) and improved computational predictions of MHC ligands have rendered the rational design of immunomodulatory peptides feasible. Proteome-derived peptides and 'reverse vaccinology' by computational means will play a driving role in future vaccine design. Here we review the molecular mechanisms of the MHC mediated immune response, present the computational approaches that have emerged in this area of biotechnology, and provide an overview of publicly available computational resources for predicting and designing new peptidic MHC ligands.

Smart Sensor for Online Detection of Multiple-Combined Faults in VSD-Fed Induction Motors

Induction motors fed through variable speed drives (VSD) are widely used in different industrial processes. Nowadays, the industry demands the integration of smart sensors to improve the fault detection in order to reduce cost, maintenance and power consumption. Induction motors can develop one or more faults at the same time that can be produce severe damages. The combined fault identification in induction motors is a demanding task, but it has been rarely considered in spite of being a common situation, because it is difficult to identify two or more faults simultaneously. This work presents a smart sensor for online detection of simple and multiple-combined faults in induction motors fed through a VSD in a wide frequency range covering low frequencies from 3 Hz and high frequencies up to 60 Hz based on a primary sensor being a commercially available current clamp or a hall-effect sensor. The proposed smart sensor implements a methodology based on the fast Fourier transform (FFT), RMS calculation and artificial neural networks (ANN), which are processed online using digital hardware signal processing based on field programmable gate array (FPGA).

Alternative sensor system and MLP neural network for vehicle pedal activity estimation

It is accepted that the activity of the vehicle pedals (i.e., throttle, brake, clutch) reflects the driver’s behavior, which is at least partially related to the fuel consumption and vehicle pollutant emissions. This paper presents a solution to estimate the driver activity regardless of the type, model, and year of fabrication of the vehicle. The solution is based on an alternative sensor system (regime engine, vehicle speed, frontal inclination and linear acceleration) that reflects the activity of the pedals in an indirect way, to estimate that activity by means of a multilayer perceptron neural network with a single hidden layer.