Application of an EWMA combining technique to the prediction of currency exchange rates

Financial forecasting is an important and challenging task for both academic researchers and business practitioners. The recent trend to improve the prediction accuracy is to combine individual forecasts using a simple average or weighted average where the weight reflects the inverse of the prediction error. In the existing combining methods, however, the errors between actual and predicted values are equally reflected in the weights regardless of the time order in a forecasting horizon. In this paper, we propose a new approach where the forecasting results of Generalized AutoRegressive Conditional Heteroskedastic (GARCH), neural network, and random walk models are combined based on a weight that reflects the inverse of the exponentially weighted moving average of the Mean Absolute Percentage Error (MAPE) of each individual prediction model. The results of an empirical study indicate that the proposed method has a better accuracy than the GARCH, neural network, and random walk models, and also combining methods based on using the MAPE for the weight.     

Prediction of BRIC Stock Price Using ARIMA,SutteARIMA, and Holt-Winters

The novel coronavirus has played a disastrous role in many countries worldwide. The outbreak became a major epidemic, engulfing the entire world in lockdown and it is now speculated that its economic impact might be worse than economic deceleration and decline. This paper identifies two different models to capture the trend of closing stock prices in Brazil (BVSP), Russia (IMOEX.ME), India (BSESN), and China (SSE), i.e., (BRIC) countries. We predict the stock prices for three daily time periods, so appropriate preparations can be undertaken to solve these issues. First, we compared the ARIMA, SutteARIMA and Holt-Winters (H-W) methods to determine the most effective model for predicting data. The stock closing price of BRIC country data was obtained from Yahoo Finance. That data dates from 01 November 2019 to 11 December 2020, then divided into two categories--training data and test data. Training data covers 01 November 2019 to 02 December 2020. Seven days (03 December 2020 to 11 December 2020) of data was tested to determine the accuracy of the models using training data as a reference. To measure the accuracy of the models, we obtained the means absolute percentage error (MAPE) and mean square error (MSE). Prediction model Holt-Winters was found to be the most suitable for forecasting the Brazil stock price (BVSP) while MAPE (0.50) and MSE (579272.65) with Holt-Winters (smaller than ARIMA and SutteARIMA), model SutteARIMA was found most appropriate to predict the stock prices of Russia (IMOEX.ME), India (BSESN), and China (SSE) when compared to ARIMA and Holt-Winters. MAPE and MSE with SutteARIMA: Russia (MAPE:0.7; MSE:940.20), India (MAPE:0.90; MSE:207271.16), and China (MAPE: 0.72; MSE: 786.28). Finally, Holt-Winters predicted the daily forecast values for the Brazil stock price (BVSP) (12 December to 14 December 2020 i.e., 115757.6, 116150.9 and 116544.1), while SutteARIMA predicted the daily forecast values of Russia stock prices (IMOEX.ME) (12 December to 14 December 2020 i.e., 3238.06, 3241.54 and 3245.01), India stock price (BSESN) (12 December to 14 December 2020 i.e.,. 45709.38, 45828.71 and 45948.05), and China stock price (SSE) (11 December to 13 December 2020 i.e., 3397.56, 3390.59 and 3383.61) for the three time periods.     

旋转机械故障诊断中的主从混合神经网络模型研究

针对ＢＰ网络在旋转机械故障诊断应用中的不足，借助Ｈｏｐｆｉｅｌｄ网络的优良特性，建立了以反馈式Ｈｏｐｆｉｅｌｄ网络为主控网络、前馈式ＢＰ网络为从网络的主从混合神经网络模型。通过这个网络模型的设计、动力学行为分析、学习算法的描述和测试以及它在旋转机械故障诊断中的应用，结果表明：该网络模型具有收敛速度快、稳定性好、最小系统误差等优点，是一种实现旋转机械故障诊断的优良网络模型。     

Prediction and control of surface roughness for the milling of Al/SiC metal matrix composites based on neural networks

In recent years, there has been a significant increase in the utilization of Al/SiC particulate composite materials in engineering fields, and the demand for accurate machining of such composite materials has grown accordingly. In this paper, a feed-forward multi-layered artificial neural network (ANN) roughness prediction model, using the Levenberg-Marquardt backpropagation training algorithm, is proposed to investigate the mathematical relationship between cutting parameters and average surface roughness during milling Al/SiC particulate composite materials. Milling experiments were conducted on a computer numerical control (CNC) milling machine with polycrystalline diamond (PCD) tools to acquire data for training the ANN roughness prediction model. Four cutting parameters were considered in these experiments: cutting speed, depth of cut, feed rate, and volume fraction of SiC. These parameters were also used as inputs for the ANN roughness prediction model. The output of the model was the average surface roughness of the machined workpiece. A successfully trained ANN roughness prediction model could predict the corresponding average surface roughness based on given cutting parameters, with a 2.08% mean relative error. Moreover, a roughness control model that could accurately determine the corresponding cutting parameters for a specific desired roughness with a 2.91% mean relative error was developed based on the ANN roughness prediction model. Finally, a more reliable and readable analysis of the influence of each parameter on roughness or the interaction between different parameters was conducted with the help of the ANN prediction model.The full text can be downloaded at https://link.springer.com/article/10.1007/s40436-020-00326-x     

膨润土塑性混凝土渗透系数的组合预测模型

在分析膨润土塑性混凝土各组分对渗透系数影响规律的基础上,将人工神经网络（ANN）与支持向量机（SVM）等方法融合使用,建立膨润土塑性混凝土渗透系数与各组分掺量之间的映射关系组合预测模型;并应用均值绝对百分率误差（MAPE）、均方根误差（RMSE）和有效系数（Er）等性能指标对预测禊趋的准确性进行对比与评价;结果表明,融合后的组合预测模型对膨润土塑性混凝土渗透系数的预测精度较高,为塑性混凝土渗透系数的预测提供了理论依据,在膨润土塑性混凝土的配合比设计及应用塑性混凝土施工等方面具有一定的参考价值。     

Magnetic hysteresis modeling via feed-forward neural networks

A general neural approach to magnetic hysteresis modeling is proposed. The general memory storage properties of systems with rate independent hysteresis are outlined. Thus, it is shown how it is possible to build a neural hysteresis model based on feed-forward neural networks (NN's) which fulfills these properties. The identification of the model consists in training the NN's by usual training algorithms such as backpropagation. Finally, the proposed neural model has been tested by comparing its predictions with experimental data     

Artificial neural network (ANN) approach for modeling of Pb(II) adsorption from aqueous solution by Antep pistachio (Pistacia Vera L.) shells

A three-layer artificial neural network (ANN) model was developed to predict the efficiency of Pb(II) ions removal from aqueous solution by Antep pistachio (Pistacia Vera L.) shells based on 66 experimental sets obtained in a laboratory batch study. The effect of operational parameters such as adsorbent dosage, initial concentration of Pb(II) ions, initial pH, operating temperature, and contact time were studied to optimise the conditions for maximum removal of Pb(II) ions. On the basis of batch test results, optimal operating conditions were determined to be an initial pH of 5.5, an adsorbent dosage of 1.0 g, an initial Pb(II) concentration of 30 ppm, and a temperature of 30 degrees C. Experimental results showed that a contact time of 45 min was generally sufficient to achieve equilibrium. After backpropagation (BP) training combined with principal component analysis (PCA), the ANN model was able to predict adsorption efficiency with a tangent sigmoid transfer function (tansig) at hidden layer with 11 neurons and a linear transfer function (purelin) at output layer. The Levenberg-Marquardt algorithm (LMA) was found as the best of 11 BP algorithms with a minimum mean squared error (MSE) of 0.000227875. The linear regression between the network outputs and the corresponding targets were proven to be satisfactory with a correlation coefficient of about 0.936 for five model variables used in this study.     

基于神经网络的电弧增材制造铝合金力学性能预测

目的 预测不同工艺参数下电弧增材制造铝合金的力学性能。方法 通过实验建立了电弧增材制造6061铝合金及Ti C增强6061铝合金力学性能的数据集,并建立了一种以焊接电流、焊接速度、脉冲频率、TiC颗粒含量为输入,以屈服强度和抗拉强度为输出的神经网预测模型,对比了反向传播神经网络(BP)、粒子群算法优化BP神经网络(PSO-BP)、遗传算法优化BP神经网络(GA-BP)3种预测模型的精度。结果 与BP模型和PSO-BP模型相比,GA-BP预测模型具有更好的预测精度。其中,GA-BP模型预测6061铝合金屈服强度最佳结果的相关系数(R)为0.965,决定系数(R²)为0.93,平均绝对误差(Mean Absolute Error,MAE)为2.35,均方根误差(Root Mean Square Error,RMSE)为2.67;预测Ti C增强的6061铝合金抗拉强度最佳结果的R=1,R²高达0.99,MAE为0.46,RMSE为0.49,GA-BP具有良好的预测精度。结论 BP、PSO-BP、GA-BP 3种神经网络模型可以用来预测电弧增材制造...     

Novel Computing for the Delay Differential Two-Prey and One-Predator System

The aim of these investigations is to find the numerical performances of the delay differential two-prey and one-predator system. The delay differential models are very significant and always difficult to solve the dynamical kind of ecological nonlinear two-prey and one-predator system. Therefore, a stochastic numerical paradigm based artificial neural network (ANN) along with the Levenberg-Marquardt backpropagation (L-MB) neural networks (NNs), i.e., L-MBNNs is proposed to solve the dynamical two-prey and one-predator model. Three different cases based on the dynamical two-prey and one-predator system have been discussed to check the correctness of the L-MBNNs. The statistic measures of these outcomes of the dynamical two-prey and one-predator model are chosen as 13% for testing, 12% for authorization and 75% for training. The exactness of the proposed results of L-MBNNs approach for solving the dynamical two-prey and one-predator model is observed with the comparison of the Runge-Kutta method with absolute error ranges between 10⁻⁰⁵ to 10⁻⁰⁷. To check the validation, constancy, validity, exactness, competence of the L-MBNNs, the obtained state transitions (STs), regression actions, correlation presentations, MSE and error histograms (EHs) are also provided.     

Optimization of Short Load Forecasting in Electricity Market of Iran Using Artificial Neural Networks

Accurate short-term load forecasting (STLF) is one of the essential requirements for power systems. In this paper, two different seasonal artificial neural networks (ANNs) are designed and compared in terms of model complexity, robustness, and forecasting accuracy. Furthermore, the performance of ANN partitioning is evaluated. The first model is a daily forecasting model which is used for forecasting hourly load of the next day. The second model is composed of 24 sub-networks which are used for forecasting hourly load of the next day. In fact, the second model is partitioning of the first model. Time, temperature, and historical loads are taken as inputs for ANN models. The neural network models are based on feed-forward back propagation which are trained and tested using data from electricity market of Iran during 2003 to 2005. Results show a good correlation between actual data and ANN outcomes. Moreover, it is shown that the first designed model consisting of single ANN is more appropriate than the second model consisting of 24 distinct ANNs. Finally ANN results are compared to conventional regression models. It is observed that in most cases ANN models are superior to regression models in terms of mean absolute percentage error (MAPE).     

Modular artificial neural network for prediction of petrophysicalproperties from well log data

An application of Kohonen's self-organizing map (SOM), learning-vector quantization (LVQ) algorithms, and commonly used backpropagation neural network (BPNN) to predict petrophysical properties obtained from well-log data are presented. A modular, artificial neural network (ANN) comprising a complex network made up from a number of subnetworks is introduced. In this approach, the SOM algorithm is applied first to classify the well-log data into a predefined number of classes, This gives an indication of the lithology in the well. The classes obtained from SOM are then appended back to the training input logs for the training of supervised LVQ. After training, LVQ can be used to classify any unknown input logs. A set of BPNN that corresponds to different classes is then trained. Once the network is trained, it is then used as the classification and prediction model for subsequent input data. Results obtained from example studies using the proposed method have shown to be fast and accurate as compared to a single BPNN network     

多硝基含能化合物5 s爆发点的定量构效关系

为精确预测含能材料的5 s爆发点,解决大量新型含能材料实验测试难度大、安全数据不全等问题,基于定量构效关系(QSPR)原理,研究多硝基含能材料分子结构与5 s爆发点(ln T_E)间的内在定量关系。应用集成学习算法随机森林(RF)筛选出8个对5 s爆发点具有显著影响的分子描述符;采用人工神经网络(ANN)建立90种多硝基含能材料5 s爆发点的预测模型。73种训练集的复决定系数为0.918,均方根误差为0.036,平均绝对误差为0.027。17个检验样本的复决定系数为0.903,均方根误差为0.061,平均绝对误差为0.053。对模型进行了验证以及应用域评价。结果表明：模型具备较好的预测性和泛化性能,可用于对多硝基含能材料的5 s爆发点进行精度较高的预测,有效解决现有含能材料的爆发点数据不够全面的问题,为相关产品研制与生产安全提供参考。     

阀门内漏识别及内漏速率量化技术研究

阀门作为天然气管线的关键部件,若发生内漏会带来经济损失及生产安全隐患。因此,阀门内漏的有效诊断及内漏速率的准确量化具有重大意义。针对复杂背景噪声下内漏诊断效率不高的问题,以内漏信号和非泄漏噪声信号的功率谱密度图作为输入,构建了阀门内漏卷积神经网络(convolutional neural network, CNN)识别模型;针对物理理论及浅层网络模型在多工况阀门内漏数据上存在量化误差大的问题,构建了阀门内漏速率深度信念网络(deep belief network, DBN)量化回归模型,并与支持向量回归机、BP神经网络等模型进行了对比研究。研究结果表明:所构建模型的内漏识别准确率及内漏速率量化平均绝对百分比误差分别为99%和9.101 2,证实了所构建模型的高效性,为阀门内漏诊断与评价开拓了新的研究方向。     

阀门内漏识别及内漏速率量化技术研究

阀门作为天然气管线的关键部件,若发生内漏会带来经济损失及生产安全隐患。因此,阀门内漏的有效诊断及内漏速率的准确量化具有重大意义。针对复杂背景噪声下内漏诊断效率不高的问题,以内漏信号和非泄漏噪声信号的功率谱密度图作为输入,构建了阀门内漏卷积神经网络(convolutional neural network, CNN)识别模型;针对物理理论及浅层网络模型在多工况阀门内漏数据上存在量化误差大的问题,构建了阀门内漏速率深度信念网络(deep belief network, DBN)量化回归模型,并与支持向量回归机、BP神经网络等模型进行了对比研究。研究结果表明:所构建模型的内漏识别准确率及内漏速率量化平均绝对百分比误差分别为99%和9.101 2,证实了所构建模型的高效性,为阀门内漏诊断与评价开拓了新的研究方向。     

Fabrication and modelling of the macro-mechanical properties of cross-ply laminated fibre-reinforced polymer composites using artificial neural network

The mechanical behaviour of fibre-reinforced polymer composites (FRPCs) is considered very complex due to many factors such as composition, material type, manufacturing process and end user applications. This article presents the mechanical properties and artificial neural network (ANN) modelling results of cross-ply laminated FRPCs. Twenty composite samples were fabricated by varying the number of layers of carbon fibre and glass fibre as reinforcement and polyphenylene sulphide and high-density polyethylene as matrix. Mechanical properties were measured in terms of flexural modulus, hardness, impact and transverse rupture strength. Multilayer feed-forward backpropagation ANN approach was used to predict the mechanical properties by using material type, composition and number of reinforcement and matrix layers as input variables. From 20 data patterns, 16 were used for network training and remaining 4 were used to test the models. Furthermore, trend analysis was also performed to understand the influence of inputs on developed models. It is evident from the ANN prediction results that there is good correlation between predicted and actual values within acceptable mean absolute error. The outcomes of this research will help to reduce cost and time by eliminating tedious composite property measurements and to fabricate tailored composites meeting application requirements.     

Artificial neural network-based prediction technique for coating thickness in Fe-Al coatings fabricated by mechanical milling

The objective of this study was to evaluate the effect of milling time, milling speed and particle size of initial powders on the coating thickness of Fe-Al intermetallic coating by using artificial neural network (ANN). Coating morphology and cross-section microstructures were evaluated using a scanning electron microscope (SEM). It was found that an increase in the milling time provided an increase in the coating-layer thickness due to the cold welding process between particles and the steel substrate. The microstructure of the coating surface was refined by ball impacts in the milling process. As a result of this study, the ANN was found to be successful for predicting the coating thickness of Fe-Al intermetallic coatings. The correlation between the predicted values and the experimental data of the feed-forward back-propagation ANN was quite adequate. The mean absolute percentage error (MAPE) for the predicted values didn’t exceed 7.46%. The ANN model can be used for predicting the coating thickness of Fe-Al intermetallic coating produced for different milling time, milling speed and particle size.     

Enhancement of mechanical properties of 3D printed hydroxyapatite by combined low and high molecular weight polycaprolactone sequential infiltration.

A new infiltration technique using a combination of low and high molecular weight polycaprolactone (PCL) in sequence was developed as a mean to improve the mechanical properties of three dimensional printed hydroxyapatite (HA). It was observed that using either high (M _n~80,000) or low (M _n~10,000) molecular weight infiltration could only increase the flexural modulus compared to non-infiltrated HA, but did not affect strength, strain at break and energy at break. In contrast, a combination of low and high molecular infiltration in sequence increased the flexural modulus, strength and energy at break compared to those of non-infiltrated HA or infiltrated by high or low molecular weight PCL alone. This overall enhancement was found to be attributed to the densification of low molecular weight PCL and the reinforcement of high molecular PCL concurrently. The combined low and high molecular weight infiltration in sequence also maintained high osteoblast proliferation and differentiation of the composites at the similar level of the HA. Densification was a dominant mechanism for the change in modulus with porosity and density of the infiltrated HA/PCL composites. However, both densification and the reinforcing performance of the infiltration phase were crucial for strength and toughening enhancement of the composites possibly by the defect healing and stress shielding mechanisms. The sequence of using low molecular weight infiltration and followed by high molecular infiltration was seen to provide the greatest flexural properties and highest cells proliferation and differentiation capabilities.