Wyniki wyszukiwania - BazTech

1

Predicting sea surface salinity in a tidal estuary with machine learning

Guillou Nicolas, Chapalain Georges, Petton Sébastien

Oceanologia

|

2023

|

Vol. 65 (2)

318--332

EN

As an indicator of exchanges between watersheds, rivers and coastal seas, salinity may provide valuable information about the exposure, ecological health and robustness of marine ecosystems, including especially estuaries. The temporal variations of salinity are traditionally approached with numerical models based on a physical description of hydrodynamic and hydrological processes. However, as these models require large computational resources, such an approach is, in practice, rarely considered for rapid turnaround predictions as requested by engineering and operational applications dealing with the ecological monitoring of estuaries. As an alternative efficient and rapid solution, we investigated here the potential of machine learning algorithms to mimic the non-linear complex relationships between salinity and a series of input parameters (such as tide-induced free-surface elevation, river discharges and wind velocity). Beyond regression methods, the attention was dedicated to popular machine learning approaches including MultiLayer Perceptron, Support Vector Regression and Random Forest. These algorithms were applied to six-year observations of sea surface salinity at the mouth of the Elorn estuary (bay of Brest, western Brittany, France) and compared to predictions from an advanced ecological numerical model. In spite of simple input data, machine learning algorithms reproduced the seasonal and semi-diurnal variations of sea surface salinity characterised by noticeable tide-induced modulations and low-salinity events during the winter period. Support Vector Regression provided the best estimations of surface salinity, improving especially predictions from the advanced numerical model during low-salinity events. This promotes the exploitation of machine learning algorithms as a complementary tool to process-based physical models.

2

Fault location of distribution network with distributed generation based on Karrenbauer transform and support vector machine regression

Wang Siming, Zhao Kaikai

Archives of Electrical Engineering

|

2023

|

Vol. 72, nr 2

461--481

EN

As the capacity and scale of distribution networks continue to expand, and distributed generation technology is increasingly mature, the traditional fault location is no longer applicable to an active distribution network and "two-way" power flow structure. In this paper, a fault location method based on Karrenbauer transform and support vector machine regression (SVR) is proposed. Firstly, according to the influence of Karrenbauer transformation on phase angle difference before and after section fault in a low-voltage active distribution network, the fault regions and types are inferred preliminarily. Then, in the feature extraction stage, combined with the characteristics of distribution network fault mechanism, the fault feature sample set is established by using the phase angle difference of the Karrenbauer current. Finally, the fault category prediction model based on SVR was established to solve the problem of a single-phase mode transformation modulus and the indistinct identification of two-phase short circuits, then more accurate fault segments and categories were obtained. The proposed fault location method is simulated and verified by building a distribution network system model. The results show that compared with other methods in the field of fault detection, the fault location accuracy of the proposed method can reach 98.56%, which can enhance the robustness of rapid fault location.

3

Prediction of aerosol optical depth over Pakistan using novel hybrid machine learning model

Zaheer Komal, Saeed Sana, Tariq Salman

Acta Geophysica

|

2023

|

Vol. 71, no. 4

2009--2029

EN

Impact of aerosols on health includes both long-term chronic irritation and inflammation of the respiratory tract. Aerosol optical depth (AOD), a crucial optical parameter that assesses the extinction effect of atmospheric aerosols, is frequently used to estimate the extent of air pollution on large scales. So, the better prediction of AOD is crucial for understanding the health impacts of aerosols. The accurate prediction of AOD is difficult due to its nonlinear relationships with other climatic variables, uncertainties, and time series variable characteristics. In this paper, a machine learning (ML) model such as support vector regression (SVR), novel hybrid SVR-GWO model (SVR integrated with gray wolf optimizer (GWO)), and statistical model multi-linear regression (MLR) are used to predict AOD. Also, for SVR-GWO model, SVR hyper-parameters are optimized using meta-heuristic GWO algorithm. Satellite-based data of Pakistan is used for the prediction of AOD on monthly bases. In addition, preprocessing techniques of forward feature selection (FFS) is utilized to select the optimal input features for the SVR-GWO, SVR and MLR models to predict AOD. The performance of the novel hybrid SVR-GWO, SVR, and MLR model is analyzed using RMSE, MAE, RRMSE, R2 and Taylor diagram, and it is found that hybrid SVR-GWO model (RMSE = 0.07, MAE = 0.06, RRMSE = 0.22 and R2=0.60) is better than ordinary SVR model (RMSE = 0.10, MAE = 0.07, RRMSE = 0.29 and R2=0.18) and MLR model (RMSE = 0.11, MAE = 0.07, RRMSE = 0.32 and R2=0.03). Keynotes: (a) The study demonstrates the potential of ML models such as SVR-GWO for accurate prediction of AOD, which can aid in better understanding of the health impacts of aerosols. (b) The use of preprocessing techniques like FFS and optimization algorithms like GWO can significantly improve the performance of the ML (SVR-GWO) model in predicting AOD. (c) The findings of this study can be useful for policymakers and healthcare professionals in identifying regions and populations at risk of aerosol-induced respiratory health issues and designing effective interventions to mitigate them.

4

Porosity prediction using Fuzzy SVR and FCM SVR from well logs of an oil field in south of Iran

Moosavi Nastaran, Bagheri Majid, Nabi-Bidhendi Majid, Heidari Reza

Acta Geophysica

|

2023

|

Vol. 71, no. 2

769--782

EN

Identification of petrophysical parameters including porosity plays an important role to evaluate hydrocarbon reservoirs. A precise prediction of porosity in oil and gas reservoirs may prevent lots of costs before drilling operations. Porosity obtained from core analysis in laboratory is the most reliable one, while they are very expensive and not always accessible. Inappropriate or missing data in under-survey locations are a key challenge for reservoir engineers. In this paper, support vector regression (SVR) is used to estimate porosity in one of the oil fields in south of Iran. SVR creates models due to structural risk minimization methods which help us to produce models with better generalization and less risk of overfitting. Definitely, measured data are always contaminated with noise. One of the common methods to reduce noise and outliers in data is to process them before using them to train the algorithm; during processing, outliers and some noisy data can be suppressed from data, while it is not always easy to distinguish real data from noise. In this paper, we modified SVR to Fuzzy SVR and Fuzzy C Means (FCM) SVR, which are used to decrease effect of noise on model, and then by adding artificial noise including random noise and outliers to data we investigated how these two methods respond to presence of noise. The results show the presence of noise and outliers in data can alter the center locations and distribution of data points in clusters in FCM SVR. Similarly, it can change the variance of Gaussian membership function we used for Fuzzy SVR, but overall, the results show Fuzzy SVR model is notably more robust against noise compared to FCM SVR. Correlation coefficient (CC) calculated between model and core data decreased from 78 to 67% after noise added to data in FCM SVR model, however, calculated CC for Fuzzy SVR remained almost steady altering from 87 to 86%. Subsequently, calculated root mean square error (RMSE) between models and core data increased from 0.0376 to 0.03827 for Fuzzy SVR, while RMSE jumped from 0.0448 to 0.0517 for FCM SVR.

5

Evaluation of the performance of data‑driven approaches for filling monthly precipitation gaps in a semi‑arid climate conditions

Katipoğlu Okan Mert

Acta Geophysica

|

2023

|

Vol. 71, no. 5

2265--2285

EN

Missing data cause problems in meteorological, hydrological, and climate analysis. The observation data should be complete and cover long periods to make the research more accurate and reliable. Artificial intelligence techniques have attracted interest for completing incomplete meteorological data in recent years. In this study the abilities of machine learning models, artificial neural networks, the nonlinear autoregressive with exogenous input (NARX) model, support vector regression, Gaussian processes regression, boosted tree, bagged tree (BAT), and linear regression to fill in missing precipitation data were investigated. In developing the machine learning model, 70% of the dataset was used for training, 15% for testing, and 15% for validation. The Bayburt, Tercan, and Zara precipitation stations, which are closest to the Erzincan station and have the highest correlation coefficients, were used to fill the data gaps. The accuracy of the constructed models was tested using various statistical criteria, such as root-mean-square error (RMSE), mean absolute error (MAE), Nash–Sutcliffe model efficiency coefficient (NSE), and determination coefficient (R2) and graphical approaches such as scattering, box plots, violin plots, and Taylor diagrams. Based on the comparison of model results, it was concluded that the BAT model with R2: 0.79 and NSE: 0.79 and error (RMSE: 11.42, and MAE: 7.93) was the most successful in the completion of missing monthly precipitation data. The contribution of this research is assist in the choice of the best and most accurate method for estimating precipitation data in semi-arid regions like Erzincan.

6

Bivariate simulation of river fow using hybrid intelligent models in sub basins of Lake Urmia, Iran

Eslamitabar Vahed, Ahmadi Farshad, Sharafati Ahmad, Rezaverdinejad Vahid

Acta Geophysica

|

2023

|

Vol. 71, no. 2

873--892

EN

In this study, the performance of continuous autoregressive moving average (CARMA), CARMA-generalized autoregressive conditional heteroscedasticity (CARMA-GARCH), random forest, support vector regression and ant colony optimization (SVR-ACO), and support vector regression and ant lion optimizer (SVR-ALO) models in bivariate simulating of discharge based on the rainfall variables in monthly time scale was evaluated over four sub-basins of Lake Urmia, located in northwestern Iran. The models were assessed in two stages: train and test. The results showed that the CARMA-GARCH hybrid model offered better performance in all cases than the stand-alone CARMA. The improvement percentages of the error rate in the CARMA model compared to the CARMA-GARCH hybrid model in the Mahabad Chai, Nazlu Chai, Siminehrood, and Zola Chai sub-basins were 9, 20, 17, and 6.4%, respectively, in the training phase. Among the models, the hybrid SVR models integrated with ACO and ALO optimization algorithms presented the best performance based on the Taylor diagram and evaluation criteria. Considering the use of ant colony and ant lion optimization algorithms to optimize the support vector regression model’s parameters, these models offered the best performance in the study area to simulate the discharge. The improvement percentages of the error rate in the SVR-ACO model compared to the CARMA-GARCH hybrid model in the Mahabad Chai, Nazlu Chai, Siminehrood, and Zola Chai sub-basins were 11, 10, 19, and 21%, respectively, in the training phase. In contrast, the random forest model provided the lowest accuracy and the highest error in discharge simulation.

7

Impact of air pollution on maize and wheat production

Chen Haorui, Wenzhi Zeng, Li Jiuying, Ma Tao, Liu Shenzhou, Lei Guoqing, Gaiser Thomas, Srivastava Amit Kumar

Ecological Chemistry and Engineering. S

|

2022

|

Vol. 29, nr 2

237--256

EN

To determine the effects of air pollution on crop yields, weather, air pollution, and maize and winter wheat yield data from 331 cities in China from 2014 to 2016 were collected and analysed. Furthermore, support vector regression and the crop growth model were applied to extrapolate the air pollution data of Beijing and Hetian and verify the relationship between air pollution and yield. Precisely, heavy air pollution usually occurred in North China, but less than moderate air pollution levels affected crop yields statistically insignificantly. Moreover, both the winter wheat and maize yields increased in moderate air pollution periods but decreased in heavy air pollution periods in 2014, 2015 and 2016. Importantly, a threshold value was necessary for the heavy air pollution periods to trigger a yield decrease. The threshold values of maize in 2015 and 2016 were 7 days and 5 days, respectively, while that of winter wheat was 10 days in both 2015 and 2016. Once the heavy air pollution periods exceeded the threshold value, both the winter wheat and maize yields decreased linearly with the periods. PM2.5 was the main air pollutant in Beijing in 2014, while PM2.5 and PM10 were the main air pollutants in Hetian in both 2015 and 2016. Regardless of whether the main air pollutant was PM2.5 or PM10, the simulated potential winter wheat yields by the crop growth model with moderate air pollution for the whole growth period were all higher than the yields under observed and heavy air pollution conditions.

8

Hybrid support vector regression models with algorithm of innovative gunner for the simulation of groundwater level

Roshni Thendiyath, Mirzania Ehsan, Hasanpour Kashani Mahsa, Bui Quynh-Anh Thi, Shamshirband Shahab

Acta Geophysica

|

2022

|

Vol. 70, no 4

1885--1898

EN

Groundwater level time series is a prime factor for variety of groundwater studies and is of great significance for the management of groundwater resources. Quality control of groundwater level observations is essential for hydrological applications. Artificial Intelligent techniques deal with highly nonlinear interactions and complex hydrological process and hence can be a better alternative for groundwater level prediction. In this research, the performances of Support Vector Regression (SVR) and SVR ensembled with metaheuristic Algorithm of Innovative Gunner (AIG) models were evaluated in simulating the monthly groundwater level of the Shabestar plain during the period 2001–2019. The 80 and 20% of the monthly dataset were used for training and testing the developed models. The efficiency of the developed models was compared using different statistical indices including correlation coefficient (R), Nash–Sutcliffe Efficiency (NSE) coefficient, Root-Mean-Square Error (RMSE), RMSE-observation standard deviation ratio (RSR) and Legates & McCabe’s Index (ELM). The results showed that the hybrid model (SVR-AIG) generates accurate estimations in combinatory patterns. Moreover, among the SVR and SVR-AIG models with different input scenarios, the SVR-AIG model showed best results for scenario 6 (M6) in both the training stage (R=0.995, NSE=0.99, RMSE=0.151 (m), RSR=0.096 and ELM =0.916) and the testing stage (R=0.941, NSE=0.879, RMSE=0.146 (m), RSR=0.346 and ELM =0.660). The hybrid SVR-AIG model is shown to be more accurate and robust than the SVR models, providing a novel capability to capture unknown time-varying dependencies. In general, the results of the proposed model are promising and it provides a reliable insight for water resources planners in conducting future research of groundwater resources.

9

Fuzzy support vector regression for permeability estimation of petroleum reservoir using well logs

Moosavi Nastaran, Bagheri Majid, Majid Nabi-Bidhendi, Heidari Reza

Acta Geophysica

|

2022

|

Vol. 70, no 1

161--172

EN

In this paper we propose a three-step approach to predict permeability. First, by using Electrofacies Analysis (EA), data are classified into several clusters. We take advantage of EA to overcome abrupt changes of permeability which its unpredictability prevents a machine to be learned. EA is also helpful for wells that suffer from core data. Second, fuzzy membership functions are applied on data points in each Electrofacies Log (EL). Third, Support Vector Regression (SVR) is employed to predict permeability using fuzzy clustered data for areas with core missing data. To perform this process, we applied the proposed technique on four well sets of a gas field located in South of Iran; three wells devoted to training and the fourth remained for testing operation. Seven ELs derived using Multi Regression Graph-Based Clustering (MRGC) method. MRGC is able to estimate more appropriate number of clusters without prior knowledge compared to other three algorithms for our case-study area. Then, fuzzy membership functions applied to data. Thereafter, SVR applied to both fuzzy and not-fuzzy ELs. Consequently, the predicted permeability log for both fuzzy and not-fuzzy inputs correlated to real permeability (core data obtained from plugs in laboratory) in the test well. Finally, predicted permeability for each face merged together to make an estimated permeability for the whole test well. The results show that predicted permeability obtained from application of SVR on fuzzy data (FSVR) has a notably better correlation with core data for both clusters individually and the whole data compared to SVR.

10

Weather drought index prediction using the support vector regression in the Ansegmir Watershed, Upper Moulouya, Morocco

Bekri My Hachem, El Hmaidi Abdellah, Ousmana Habiba, El Faleh El Mati, Berrada Mohamed, El Aissaoui Kamal, Essahlaoui Ali, El Ouali Abdelhadi

Journal of Water and Land Development

|

2021

|

no. 50

187--194

EN

The purpose of this study is to develop mathematical models based on artificial intelligence: Models based on the support vectors regression (SVR) for drought forecast in the Ansegmir watershed (Upper Moulouya, Morocco). This study focuses on the prediction of the temporal aspect of the two drought indices (standardized precipitation index – SPI and standardized precipitation-evapotranspiration index – SPEI) using six hydro-climatic variables relating to the period 1979–2013. The model SVR3-SPI: RBF, ε = 0.004, C = 20 and γ = 1.7 for the index SPI, and the model SVR3-SPEI: RBF ε = 0.004, C = 40 and γ = 0.167 for the SPEI index are significantly better in comparison to other models SVR1, SVR2 and SVR4. The SVR model for the SPI index gave a correlation coefficient of R = 0.92, MSE = 0.17 and MAE = 0.329 for the learning phase and R = 0.90, MSE = 0.18 and MAE = 0.313 for the testing phase. As for the SPEI index, the overlay is slightly poorer only in the case of the SPI index between the observed values and the predicted ones by the SVR model. It shows a very small gap between the observed and predicted values. The correlation coefficients R = 0.88 for the learning, R = 0.86 for testing remain higher and corresponding to a quadratic error average MSE = 0.21 and MAE = 0.351 for the learning and MSE = 0.21 and MAE = 0.350 for the testing phase. The prediction of drought by SVR model remain useful and would be extremely important for drought risk management.

11

Shallow, Deep, Ensemble models for Network Device Workload Forecasting

Liu Cenru

Annals of Computer Science and Information Systems

|

2020

|

Vol. 21

101--104

EN

Reliable prediction of workload-related characteristics of monitored devices is important and helpful for management of infrastructure capacity. This paper presents 3 machine learning models (shallow, deep, ensemble) with different complexity for network device workload forecasting. The performance of these models have been compared using the data provided in FedCSIS'20 Challenge. The R2 scores achieved from the cascade Support Vector Regression (SVR) based shallow model, Long short-term memory (LSTM) based deep model, and hierarchical linear weighted ensemble model are 0.2506, 0.2831, and 0.3059, respectively, and was ranked 3rd place in the preliminary stage of the challenges.

12

Kernel function based regression approaches for estimating the oxygen transfer performance of plunging hollow jet aerator

Kumar M., Tiwari N. K., Ranjan S.

Journal of Achievements in Materials and Manufacturing Engineering

|

2019

|

Vol. 95, nr 2

74--84

EN

Purpose: To evaluate the capability of various kernels employed with support vector regression (SVR) and Gaussian process regression (GPR) techniques in estimating the volumetric oxygen transfer coefficient of plunging hollow jets. Design/methodology/approach: In this study, a data set of 81 observations is acquired from laboratory experiments of hollow jets plunging on the surface of water in the tank. The jet variables: jet velocity, jet thickness, jet length, and water depth are varied accordingly and the values of volumetric oxygen transfer coefficient is computed. An empirical relationship expressing the oxygenation performance of plunging hollow jet aerator in terms of jet variables is formulated using multiple nonlinear regression. The performance of this nonlinear relationship is compared with various kernel function based SVR and GPR models. Models developed with the training data set (51 observations) are checked on testing data set (24 observations) for performance comparison. Sensitivity analysis is carried out to examine the influence of jet variables in effecting the oxygen transfer capabilities of plunging hollow jet aerator. Findings: The overall comparison of kernels yielded good estimation performance of Radial Basis Function kernel (RBF) and Pearson VII Function kernel (PUK) using the SVR technique which is followed by nonlinear regression, and other kernel function based regression models. Research limitations/implications: The results of the study pertaining to the performance of kernels are based on the current experimental conditions and the estimation potential of the regression models may fluctuate beyond the selection of current data range due to datadependant learning of the soft computing models. Practical implications: Volumetric oxygen transfer coefficient of plunging hollow jets can be predicted precisely using SVR model by employing RBF as kernel function as compared to empirical correlation and other kernel function based regression models. Originality/value: The comparative analysis of kernel functions is conducted in this study. In previous studies, the predictive modelling approaches are implemented in simulating the aeration properties of cylindrical solid jets only, while this paper simulates the volumetric oxygen transfer coefficient of diverging hollow jets with the jet variables by utilizing polynomial, normalized polynomial, PUK, and RBF kernels in SVR and GPR.

13

Prognozowanie wytrzymałości na ściskanie betonu natryskowego przy zastosowaniu inteligentnych metod obliczeniowych: sztucznej sieci neuronowej i regresji wektorów wspierających

Kalhori Hamid, Bagherpour Raheb

Cement Wapno Beton

|

2019

|

R. 22/84, nr 2

126--136

PL

Wytrzymałość na ściskanie jest jedną z najważniejszych właściwości betonu. Wynik badania 28-dniowej wytrzymałości na ściskanie betonu lub betonu natryskowego jest wskaźnikiem ich jakości, który bardzo zależy od składu mieszanki betonowej. Niektóre czynniki takie jak rodzaj cementu, stosunek woda/cement, suma drobnego i grubego kruszywa w mieszance, domieszki i in. wpływają na wytrzymałość betonu. Bardzo trudno jest prognozować wytrzymałość betonu z powodu dużej liczby takich parametrów. Obecnie, dysponując inteligentnymi metodami obliczeniowymi, modelowanie odgrywa szczególną rolę w naukach inżynierskich i prognozowaniu zachowania się materiału. Dlatego w artykule przedstawiono wyniki badań różnych mieszanek betonu natryskiwanego zawierającego pył krzemionkowy i zbiory ich wytrzymałości na ściskanie po 28 dniach. Modele ANN oraz SVR zastosowano do prognozowania wytrzymałości na ściskanie betonu natryskowego biorąc pod uwagę parametry mieszanek betonu natryskiwanego jako dane wejściowe. Współczynnik korelacji (R), średni bezwzględny błąd procentowy (MAPE) i pierwiastek błędu średniokwadratowego (RMSE) są statystykami użytymi jako miary efektywności proponowanych modeli prognostycznych. Porównanie wszystkich wyników obliczeń z wynikami doświadczeń wskazuje na zadawalającą dokładność wyników uzyskiwanych przy użyciu proponowanych inteligentnych metod obliczeniowych. Wyniki przeprowadzonych badań wskazują, że zarówno model ANN jak i model SVR są dogodnymi narzędziami do oszacowania wytrzymałości na ściskanie betonu natryskiwanego.

EN

Compressive strength is one of the most important mechanical properties of concrete. 28-day compressive strength test is the acceptance measure of concrete or shotcrete, which is highly affected by the mix design. Some parameters like water/cement ratio, amount of fine and coarse aggregates in mix, admixtures and so on affect shotcrete strength. Due to the large number of such parameters, it is very difficult to predict the shotcrete strength. Today, owing to intelligent methods, modeling has a particular role in engineering sciences and predicting material behavior. Therefore, this paper examines different mix designs of shotcrete containing microsilica and records their 28-day compressive strength. Regarding shotcrete mix design parameters as inputs, ANN and SVR models were used to predict compressive strength of shotcretes. The correlation coefficient (R), mean absolute percentage error (MAPE) and the root mean square error (RMSE) statics are used for performance evaluation of proposed predictive models. All of the results showed that the accuracy of the proposed soft computing methods is quite satisfactory as compared to experimental results. The finding of this study indicated that the both ANN and SVM models are sufficient tools for estimating the compressive strength of shotcrete.

14

Training subset selection for support vector regression

Liu Cenru, Cen Jiahao

Annals of Computer Science and Information Systems

|

2019

|

Vol. 18

11--14

EN

As more and more data are available, training a machine learning model can be extremely intractable, especially for complex models like Support Vector Regression (SVR) train- ing of which requires solving a large quadratic programming optimization problem. Selecting a small data subset that can effectively represent the characteristic features of training data and preserve their distribution is an efficient way to solve this problem. This paper proposes a systematic approach to select the best representative data for SVR training. The distribution of both predictor and response variables are preserved in the selected subset via a 2-layer data clustering strategy. A 2-layer step-wise greedy algorithm is introduced to select best data points for constructing a reduced training set. The proposed method has been applied for predicting deck's win rates in the Clash Royale Challenge, in which 10 subsets containing hundreds of data examples were selected from 100k for training 10 SVR models to maximize their prediction performance evaluated using R-squared metric. Our final submission having a R2 score of 0.225682 won the 3rd place among over 1200 solutions submitted by 115 teams.

15

Greedy incremental support vector regression

Ruta Dymitr, Cen Ling, Vu Quang Hieu

Annals of Computer Science and Information Systems

|

2019

|

Vol. 18

7--9

EN

Support Vector Regression (SVR) is a powerful supervised machine learning model especially well suited to the normalized or binarized data. However, its quadratic complexity in the number of training examples eliminates it from training on large datasets, especially high dimensional with frequent retraining requirement. We propose a simple two-stage greedy selection of training data for SVR to maximize its validation set accuracy at the minimum number of training examples and illustrate the performance of such strategy in the context of Clash Royale Challenge 2019, concerned with efficient decks' win rate prediction. Hundreds of thousands of labelled data examples were reduced to hundreds, optimized SVR was trained on to maximize the validation R2 score. The proposed model scored the first place in the Cash Royale 2019 challenge, outperforming over hundred of competitive teams from around the world.

16

Efficient support vector regression with reduced training data

Cen Ling, Vu Quang Hieu, Ruta Dymitr

Annals of Computer Science and Information Systems

|

2019

|

Vol. 18

15--18

EN

Support Vector Regression (SVR) as a supervised machine learning algorithm have gained popularity in various fields. However, the quadratic complexity of the SVR in the number of training examples prevents it from many practical applications with large training datasets. This paper aims to explore efficient ways that maximize prediction accuracy of the SVR at the minimum number of training examples. For this purpose, a clustered greedy strategy and a Genetic Algorithm (GA) based approach are proposed for optimal subset selection. The performance of the developed methods has been illustrated in the context of Clash Royale Challenge 2019, concerned with decks' win rate prediction. The training dataset with 100,000 examples were reduced to hundreds, which were fed to SVR training to maximize model prediction performance measured in validation R2 score. Our approach achieved the second highest score among over hundred participating teams in this challenge.

17

Soft-Sensing in Batch Annealing Based on Finite Differential Method and Support Vector Regression

Kačur Ján, Durdán Milan, Laciak Marek, Flegner Patrik

Advances in Science and Technology. Research Journal

|

2019

|

Vol. 13, no 4

70--86

EN

The temperature of annealed steel coils is a determining variable of the future steel sheets quality. This variable also determines the energy consumption in operation. Unfortunately, the monitoring of coil inner temperature is problematic due to the furnace environment with high temperature, coil structure, and annealing principle. Currently, there are no measuring principles that can measure the temperature inside the heat-treated product in a non-destructive manner. In this paper, the soft sensing of inner temperature based on the theory of non-stationary heat conduction and approach based on Support Vector Regression (SVR) was presented. The results showed that a black-box approach based on the SVR could replace an analytic approach, though with lesser performance. Several annealing experiments were performed to create a training data set and model performance improvement in the estimation of inner coil temperatures. The proposed software based on non-stationary heat conduction can calculate the behavior of inner coil temperature from the measured boundary temperatures that are measured by thermocouples. The soft-sensing principles presented in this paper were verified under laboratory conditions and on the data obtained from a real annealing plant.

18

A fuzzy KNN-based model for significant wave height prediction in large lakes

Nikoo M. R., Kerachian R., Alizadeh M. R.

Oceanologia

|

2018

|

No. 60 (2)

153--168

EN

Some algorithms based on fuzzy set theory (FST) such as fuzzy inference system (FIS) and adaptive-network-based fuzzy inference system (ANFIS) have been successfully applied to significant wave height (SWH) prediction. In this paper, perhaps for the first time, the fuzzy K-nearest neighbor (FKNN) algorithm is utilized to develop a fuzzy wave height prediction model for large lakes, where the fetch length depends on the wind direction. As fetch length (or wind direction) can affect the wave height in lakes, this variable is also considered as one of the inputs of the prediction model. The results of the FKNN model are compared with those of some soft computing techniques such as Bayesian networks (BNs), regression tree induction (named M5P), and support vector regression (SVR). The developed FKNN model is used for SWH prediction in the western part of Lake Superior in North America. The results show that the FKNN and M5P model can outperform the other soft computing techniques.

19

Accurate prediction of continuous blood glucose based on support vector regression and differential evolution algorithm

Hamdi T., Ben Ali J., Di Costanzo V., Fnaiech F., Moreau E., Ginoux J. M.

Biocybernetics and Biomedical Engineering

|

2018

|

Vol. 38, no. 2

362--372

EN

Type 1 diabetes (T1D) is a chronic disease requiring patients to know their blood glucose values in order to ensure blood glucose levels as close to normal as possible. Hence, the ability to predict blood glucose levels is of a great interest for clinical researchers. In this sense, the literature is rich with several solutions that can predict blood glucose levels. Unfortunately, these methods require the patient to specific their daily activities: meal intake, insulin injection and emotional factors, which can be error prone. To reduce this burden on the patent, this work proposes to use only continuous glucose monitoring (CGM) data to predict blood glucose levels independently of other factors. To support this, support vector regression (SVR) and differential evolution (DE) algorithms were investigated. The proposed method is validated using real CGM data of 12 patients. The obtained average of root mean square error (RMSE) was 9.44, 10.78, 11.82 and 12.95 mg/dL for prediction horizon (PH) respectively equal to 15, 30, 45 and 60 min. The results of the present study and comparison with some previous works show that the proposed method holds promise. The SVR based on DE algorithm achieved high prediction accuracy while being robustness, automatic, and requiring no human intervention.

20

Intermittent demand forecasting using data mining techniques

Kaya Gamze Ogcu, Turkyilmaz Ali

Applied Computer Science

|

2018

|

Vol. 14, no 2

38--47

EN

Intermittent demand occurs randomly with changing values and a lot of periods having zero demand. Ad hoc intermittent demand forecasting techniques have been developed which take special intermittent demand characteristics into account. Besides traditional techniques and specialized methods, data mining offers a better alternative for intermittent demand forecasting since data mining methods are powerful techniques. This study contributes to the current literature by showing the benefit of using data mining methods for intermittent demand forecasting purpose by comprising mostly used data mining methods.