Data-driven Artificial Neural Network Research Articles

Data-driven prediction of flame temperature and pollutant emission in distributed combustion

The flame temperature and pollutant emission (of NO and CO) characteristics in distributed combustion were examined using data-driven artificial neural network (ANN) approach. Experimental results collected from swirl-assisted distributed combustion using methane fuel at an equivalence ratio 0.9 were utilized for dataset preparation. The distributed combustion condition was created in a swirl-assisted burner (at thermal intensity of 5.72 MW/m3-atm) by diluting the main airstream with carbon dioxide. Experimental results of exhaust NO and CO concentrations and adiabatic flame temperature (AFT) derived from Chemkin-Pro® simulation were selected as the target output. The ANN model was developed with inlet airflow rates and O2 concentrations as the input, and pollutant emission and AFT as the output variables. The ANN possessed one hidden layer with variable number of neurons (10, 15, and 20). Tangent sigmoid and Log sigmoid transfer functions were tested along with feed forwards and cascade forward network schemes having Levenberg–Marquardt backpropagation training algorithms. Different learning rates of model training were investigated to determine the optimized training model. The best model (with learning rate 0.2) was selected based on the mean square error (MSE) and the strength of correlation (R) predicted for individual outputs. The model predicted very well the target outputs in both conventional swirl combustion and distributed combustion regime for wider applicability in a range of applications. A very strong correlation was predicted by the current ANN model for the overall data as indicated by the R value of 0.9842. Furthermore, the results obtained with Bayesian Regularization backpropagation method demonstrated better prediction than those from Levenberg–Marquardt algorithm.

Hybrid Data-Driven Modeling Methodology for Fast and Accurate Transient Simulation of SiC MOSFETs

To enable fast and accurate models of SiC MOSFETs for transient simulation, a hybrid data-driven modeling methodology of SiC MOSFETs is proposed. Unlike conventional modeling methods that are based on complex nonlinear equations, data-driven artificial neural networks (ANNs) are used in this article. For model accuracy, the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$I$</tex-math></inline-formula> – <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$V$</tex-math></inline-formula> characteristics are measured in the whole operation region to train the ANN. The ANN model is then combined with behavior-based equations to model the cutoff region and to avoid overfitting the ANN. In addition, the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$C$</tex-math></inline-formula> – <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$V$</tex-math></inline-formula> characteristics are modeled by ANNs with a logarithmic scale for accuracy. The proposed model is implemented and simulated in SPICE simulator SIMetrix. The simulation results are compared with experimental results from a double pulse tester to validate the proposed modeling methodology. The model is also compared with the Angelov model created by the Keysight MOSFET modeling software. The comparison results show that the proposed model is more accurate than the Angelov model. Besides, when compared to the Angelov model, the proposed model requires 30% less computation time when simulating a double pulse tester. In addition, the proposed modeling method also has better adaptability to model different types of SiC MOSFETs.

Seismic Velocity Model Building Using Recurrent Neural Networks: A Frequency-Stepping Approach

Data-driven artificial neural networks (ANNs) demonstrably offer numerous advantages over conventional deterministic methods in a wide range of geophysical problems. For seismic velocity model building, judiciously trained ANNs offer the possibility of estimating subsurface velocity models; however, there are substantial challenges with effective and efficient network training. Motivated by the multi-scale approach commonly used to address FWI non-linearity challenges, we develop a frequency-stepping velocity model building approach that uses a sequence-to-sequence recurrent neural network (RNN) with built-in long short-term memory (LSTM). The input sequences to the LSTM-RNN consist of the frequency-domain seismic data ordered by frequency from lowest available to highest usable or chosen, while the corresponding output sequences are frequency-dependent smoothed velocity models. We compare models recovered using the trained RNN to the true models qualitatively and quantitatively. The normalized root-mean-square (NRMS) misfit between the true and predicted models has a mean of 6%, which confirms that the network recovers highly accurate models.

Artificial neural network model for single-phase real gas ejectors

Ejector is a passive gas dynamic device used ubiquitously in engineering applications. Applications of ejectors have transited from traditional steam extraction to heat-driven gas compression in supercritical and trans-critical CO 2 power and refrigeration cycles. The non-ideal gas behavior in such applications adds to the existing lack of fundamental understanding of the complex gas dynamics within the ejector. A number of simplified 1-D physics-based low-fidelity models have been proposed in the literature. These models offer reasonably good predictions for most ideal gases such as air and superheated steam. However, significant deviations as high as 20% between the experimental measurements and the model are observed for real gases. On the other hand, CFD simulations for an ejector are not only computationally expensive but also require careful selection and tweaking of the turbulence model. The paper presents a data-driven artificial neural network (ANN) model to predict the critical parameters of a supersonic ejector. The model is trained using a comprehensive experimental database comprising of various working fluids and operating conditions. The framework uses a topology of seven dimensionless input parameters capturing geometry, operating conditions, and working fluid characteristics to predict two output parameters; (a) entrainment ratio and (b) operational regime of the ejector (critical or mixed mode of operation). The model relies on a network of 84 neurons in the hidden layer and uses a split ratio of 0.15 for training the dataset. The trained ANN model is able to predict the entrainment ratio within 10% accuracy, with the ability to classify the operational mode with an accuracy of 91%. • An ANN model is developed to predict the ejector performance and operating regime. • The working fluid is modeled with its molecular weight and isentropic exponent. • ANN predicts ER with a 10% deviation with experimental ER for all working fluids. • The ANN model classifies the operating regime of ejectors with 91% accuracy. • It will be effective in predicting the global performance of ejector-based systems.

Multi-objective optimisation with hybrid machine learning strategy for complex catalytic processes

Catalytic chemical processes such as hydrocracking, gasification and pyrolysis play a vital role in the renewable energy and net zero transition. Due to the complex and non-linear behaviours during operation, catalytic chemical processes require a powerful modelling tool for prediction and optimisation for smart operation, speedy green process routes discovery and rapid process design. However, challenges remain due to the lack of an effective modelling and optimisation toolbox, which requires not only a precise analysis but also a fast optimisation. Here, we propose a hybrid machine learning strategy by embedding the physics-based continuum lumping kinetic model into the data-driven artificial neural network framework. This hybrid model is adopted as the surrogate model in the multi-objective optimisation and demonstrated in the benchmarking of a hydrocracking process. The results show that the novel hybrid surrogate model exhibits the mean square error less than 0.01 by comparing with the physics-based simulation results. This well-trained hybrid model was then integrated with non-dominated-sort genetic algorithm (NSGA-II) as the surrogate model to evaluate and optimise the yield and selectivity of the hydrocracking process. The Pareto front from the multi-objective optimisation was able to identify the trade-off curve between the objective functions which is essential for the decision-making during process design. Our work indicates that adopting the hybrid machine learning strategy as the surrogate model in the multi-objective optimisation is a promising approach in various complex catalytic chemical processes to enable an accurate computation as well as a rapid optimisation.

Data-driven artificial and spiking neural networks for inverse kinematics in neurorobotics

SummaryInverse kinematics is fundamental for computational motion planning. It is used to derive an appropriate state in a robot's configuration space, given a target position in task space. In this work, we investigate the performance of fully connected and residual artificial neural networks as well as recurrent, learning-based, and deep spiking neural networks for conventional and geometrically constrained inverse kinematics. We show that while highly parameterized data-driven neural networks with tens to hundreds of thousands of parameters exhibit sub-ms inference time and sub-mm accuracy, learning-based spiking architectures can provide reasonably good results with merely a few thousand neurons. Moreover, we show that spiking neural networks can perform well in geometrically constrained task space, even when configured to an energy-conserved spiking rate, demonstrating their robustness. Neural networks were evaluated on NVIDIA's Xavier and Intel's neuromorphic Loihi chip.

Development of artificial neural networks software for arsenic adsorption from an aqueous environment

Arsenic contamination is a global problem, as it affects the health of millions of people. For this study, data-driven artificial neural network (ANN) software was developed to predict and validate the removal of As(V) from an aqueous solution using graphene oxide (GO) under various experimental conditions. A reliable model for wastewater treatment is essential in order to predict its overall performance and to provide an idea of how to control its operation. This model considered the adsorption process parameters (initial concentration, adsorbent dosage, pH, and residence time) as the input variables and arsenic removal as the only output. The ANN model predicted the adsorption efficiency with high accuracy for both training and testing datasets, when compared with the available response surface methodology (RSM) model. Based on the best model synaptic weights, user-friendly ANN software was created to predict and analyze arsenic removal as a function of adsorption process parameters. We developed various graphical user interfaces (GUI) for easy use of the developed model. Thus, a researcher can efficiently operate the software without an understanding of programming or artificial neural networks. Sensitivity analysis and quantitative estimation were carried out to study the function of adsorption process parameter variables on As(V) removal efficiency, using the GUI of the model. The model prediction shows that the adsorbent dosages, initial concentration, and pH are the most influential parameters. The efficiency was increased as the adsorbent dosages increased, decreasing with initial concentration and pH. The result show that the pH 2.0–5.0 is optimal for adsorbent efficiency (%).

A probabilistic fatigue life prediction for adhesively bonded joints via ANNs-based hybrid model

The paper is aimed at developing an efficient and robust probabilistic fatigue life prediction framework for adhesively bonded joints. This framework calibrates the fatigue life model by quantifying uncertainty in the fatigue damage evolution relation using a set of experimental fatigue life data. Probabilistic assessment of fatigue life is simulated through damage evolution along the bondline and Bayesian inference via the Markov chain Monte Carlo (MCMC) sampling method for inverse uncertainty quantification (UQ). To expedite the fatigue life simulation, a hybrid model composed of physics-based fatigue damage evolution relation and a data-driven artificial neural networks (ANNs) model is employed. The degradation of the adhesive is evaluated by the fatigue damage evolution relation which is then mapped to the strain redistribution along the bondline using the ANNs model. Once the mapping is learned by the ANNs, through data from FEA simulations, the probabilistic fatigue life prediction framework involves three successive modules: (I) fatigue damage growth (FDG) simulator, (II) uncertainty quantification (UQ), and (III) confidence bounds for fatigue life prediction. The FDG simulator can be used for simulating fatigue degradation rapidly for a given geometric configuration under any arbitrary fatigue loading spectra. The quantified uncertainties from the framework correspond to the intrinsic statistical material properties that can be used for probabilistic fatigue life prediction in any joint configuration with the same adhesive material. The probabilistic framework is verified using a single lap joint (SLJ) by quantifying uncertainties which are then used for probabilistic fatigue life prediction in laminated doublers in the bending (LDB) joint, that uses the same adhesive material as SLJ, and successfully compared with experimental data. The framework is also tested and validated by estimating probabilistic fatigue life in other joint configurations under constant and variable amplitude fatigue loading spectra.

Control of a hybrid upper-limb orthosis device based on a data-driven artificial neural network classifier of electromyography signals

This study presents the design of a functional electrical stimulation system (FES) for upper limb, including the construction and evaluation of an integral self-carrying rehabilitation device. The proposed device incorporated a hybrid system that performs the active mobilization control for the articulations or the assisted movement for the upper limb using electrical stimulation, as well as an on-line characterization of the electromyographic (EMG) signals captured in the trapezius and deltoid muscles. The orthosis was manufactured using a three-dimensional printer. The constructed device was electronically instrumented as a fully actuated robot, and it was controlled in a decentralized form by a set of state feedback (proportional-derivative) algorithms. This study proposed an interpolation method based on sigmoidal functions to solve the trajectory tracking for each actuated articulation. These algorithms used the estimated time-derivative of the tracking error provided by several explicit discretized super-twisting differentiators. The EMG signals were classified by a static multilayered artificial neural network trained with the Levenberg-Marquardt method, defining the movement intention triggered by the user. The device was tested in simulation software including the integration of several evaluation scenarios depending on the classified EMG signal and the tracking trajectory performance developed by the suggested state feedback controllers. The constructed orthosis was successfully evaluated with four volunteers showing the expected performance according to the proposed evaluation scenarios.

A data-driven artificial neural network model for predicting wind load of buildings using GSM-CFD solver

Wind load prediction is critical to the design of buildings, especially skyscrapers, because wind-induced dynamic loads and vibrations can often be a major concern that dominates the design. This paper presents a real-time wind load predictor for buildings by training an artificial neural network, with big data generated from a well-established software package for computational fluid dynamics (CFD), known as GSM-CFD. The cross-section of a building is idealized as a rectangle with various aspect ratios. The attack angles and velocity of the wind are also treated as variables in a wide range for building design considerations. We firstly establish a large number of computational models of major possible situations, and intensive computations using the GSM-CFD solver are conducted to generate detailed flow fields. Due to the high Reynold numbers of the flow fields, large eddy simulation (LES) model has been adopted in the calculation process. Both drag and lift coefficients are then computed based on the pressure distribution around the surface of the rectangular blocks. The GSM-CFD enables us to generate a mass of data considering different aspect ratios, wind velocities and attacking angles. Finally, an artificial neural network is trained to predict the drag coefficients for any given aspect ratios of the rectangular cross-sections of building subject to any given attack angles of wind with various velocities. The trained neural network is verified by comparing the results of the neural network predictions and that of the numerical simulations. The accuracy based on the test data set is found about 3.17%, which is sufficient for engineering design purposes. Our trained neural network model is applicable to other problems with a flow around rectangles with aspect ratios from 0.2 to 1, air-flow velocity of 34 m/s, and attack angles from 0° to 90°. The prediction is practically in real-time.

Modeling cyclic volatile methylsiloxanes removal efficiency from wastewater by ZnO-coated aluminum anode using artificial neural networks

Usage of cyclic volatile methyl siloxanes (cVMSs) in the industrial process is unavoidable due to their superior properties; however, it is hazardous to human health. Photocatalytic zinc oxide coated aluminum anode is used to degrade the cVMSs in wastewater. In this work, we investigated the relationship among degradation process parameters such as current density (4–20 mA/cm2), initial pH (5–9), plate distance (8–24 cm), UV intensity (0–120 W), and reaction time (30–100 min) vis-a-vis cVMSs removal efficiency by using data-driven artificial neural networks(ANN) model. The ANN model was trained using a backpropagation algorithm with the sigmoid activation function between input, hidden, and the output layers. Two hidden layers with eight neurons in each layer presented the minimum average training error (0.24) and higher (0.99) correlation coefficient values (both Pearson’s r and Adj. R2) as compared with the conventional regression model. The effect and relationship between the parameters and cVMSs removal efficiency were analyzed by single, two variable sensitivity analysis, qualitative and quantitative estimation.

A real time data driven algal bloom risk forecast system for mariculture management

In eutrophic coastal waters, harmful algal blooms (HAB) often occur and present challenges to environmental and fisheries management. Despite decades of research on HAB early warning systems, the field validation of algal bloom forecast models have received scant attention. We propose a daily algal bloom risk forecast system based on: (i) a vertical stability theory verified against 191 past algal bloom events; and (ii) a data-driven artificial neural network (ANN) model that assimilates high frequency data to predict sea surface temperature (SST), vertical temperature and salinity differential with an accuracy of 0.35oC, 0.51oC, and 0.58 psu respectively. The model does not rely on past chlorophyll measurements and has been validated against extensive field data. Operational forecasts are illustrated for representative algal bloom events at a marine fish farm in Tolo Harbour, Hong Kong. The robust model can assist with traditional onsite monitoring as well as artificial-intelligence (AI) based methods.

A spatial-and-temporal-based method for rapid particle concentration estimations in an urban environment

The increasing construction of buildings and infrastructure in cities heavily influences pollutant dispersion and causes a spread of increased particle concentrations. Real-time data and information on local pollution levels are highly desired by residents, urban planners and policy-makers. Such information is scarce due to the high cost of real-time measurement. To fill the gap, the aim of this research is to develop a model that can rapidly estimate particulate pollution based on a data-driven artificial neural network modelling approach. The key influential factors such as background pollution level, weather conditions and urban morphology are embedded in the model in association with local emission sources of pollution relating to construction activities and traffic flows. The data for urban spatial-variables (building and road) and traffic information is processed with the aid of the Geographic Information System using self-developed Python scripts. The geographic dataset containing the required information for each grid is integrated with the artificial neural network model to perform forecasting of particle concentrations. The model has been verified with measurements from a case study with 20 sample locations in Chongqing, China, showing that the average relative error of particle concentration estimation compared to measurement is 17.56% for PM10 and 16.04% for PM2.5. A map of a time-specific spatial interpolation of particle concentrations which visualises real-time pollution is consequently produced based on the method. The newly proposed method is novel and holistic which integrates spatial and time information covering aspects of urban form, weather file, traffic, construction sites. The rigorously validated model has been transformed to a robust tool for a fast estimation of real-time particle concentration in an urban environment.

Prediction of groundwater level in seashore reclaimed land using wavelet and artificial neural network-based hybrid model

It is challenging to predict groundwater level (GWL) accurately and reliably in reclaimed coastal areas, the dynamic of which has been greatly altered by the land reclamation activities, due to its diverse engineering boundaries. In this study, the data-driven Artificial Neural Network (ANN) models (nonlinear input-output network (NIO), nonlinear autoregressive network with exogenous inputs (NARX), and wavelet-NARX (WA-NARX)) are used to predict the GWL in a newly reclaimed land on Zhoushan Island, China. The models use semi-diurnal tide (SDT) and precipitation as input variables, and the wavelet coherence (WTC) is applied to analyze the response of GWL to SDT. The results show that the WA-NARX hybrid model provides better prediction performance, especially for short-term periods. Meanwhile, we also explore the correlations between GWL and SDT in the filled and clay layers by using the WTC and the global coherence coefficient (GCC). The results show strong correlations at 0.5-, 1-, and 15-day time scales (resonance periodicities), which are then used as prediction periods for ANN models. The correlation is stronger for the filled layer than that for the clay layer indicates GWL in the filled layer is more sensitive to SDT, especially at 0.5-day time scale. The predicted results also confirm that SDT and precipitation have great influences on GWL with better prediction in the filled layer.

A comparison between knowledge-driven fuzzy and data-driven artificial neural network approaches for prospecting porphyry Cu mineralization; a case study of Shahr-e-Babak area, Kerman Province, SE Iran

The study area, located in the southern section of the Central Iranian volcano–sedimentary complex, contains a large number of mineral deposits and occurrences which is currently facing a shortage of resources. Therefore, the prospecting potential areas in the deeper and peripheral spaces has become a high priority in this region. Different direct and indirect methods try to predict promising areas for future explorations, most of which are very time-consuming and costly. The main goal of mineral prospecting is applying a transparent and robust approach for identifying high potential areas to be explored further in the future. This work presents the procedure taken to create two different Cu-mineralization prospectivity maps. The first map is created using a knowledge-driven fuzzy technique and the second one by a data-driven Artificial Neural Network (ANN) approach. In this study aim is to investigate the results of applying the ANN technique and to compare them with the outputs of applying the fuzzy logic method. The geo-datasets employed for creating evidential maps of porphyry Cu mineralization include the solid geology map, alteration map, faults, dykes, airborne total magnetic intensity, airborne gamma-ray spectrometry data (U, Th, K and total count), and known Cu occurrences. Based on this study, the ANN technique is a better predictor of Cu mineralization compared to the fuzzy logic method. The ANN technique, due to capabilities such as classification, pattern matching, optimization, and prediction, is useful in identifying the anomalies associated with the Cu mineralization.

Forecasting of Horizontal Gas Well Production Decline in Unconventional Reservoirs using Productivity, Soft Computing and Swarm Intelligence Models

The most widely used production decline forecasting tools are numerical reservoir simulation, material balance estimates and advanced methods of production decline analysis. Besides, these existing production decline evaluation techniques for unconventional reserves estimations have underlying limitations and assumptions incorporated into their formulation which can result into under- and overestimation. This further raises the debate about which decline curve analysis (DCA) is better than the other for unconventional reservoirs predictions. Currently, data-driven artificial neural network (ANN) has emerged as a new paradigm capable of mapping complex functional relationships. In this present study, ANN technology was used to calibrate tight gas carbonate field historical declining trends which exhibit high early peak production rate and quick decline. Hereafter, making reliable predictions posed as a challenge for the complex target field and hard computing protocols. Therefore, this research applied and tested the capability of backpropagation artificial neural network (BPANN), radial basis function neural network (RBNN) and generalized regression neural network (GRNN) as DCA techniques for predicting the historical production decline trends of an ultra-low porosity and permeability tight gas reservoir. The optimum trained ANN DCA models developed were validated with another surrounding well’s dataset, producing acceptable results in agreement with the actual field data. The GRNN DCA model’s testing performance was poor, and its smoothing parameter was optimized with particle swarm optimization (PSO) algorithm which gave satisfactory results comparable to standalone BPANN and RBFNN DCA models. Furthermore, the optimum ANN DCA models’ generalization strength across the entire field dataset revealed that the developed models’ predictions were robust as compared to the data-driven Arps hyperbolic and power law exponential DCA models. This was evident from the statistical performance criteria employed which indicated that BPANN, RBFNN and PSO-GRNN DCA models are plausible better fit models for matching the target field’s historical decline performances. Also, a novel non-Darcy flow horizontal well productivity evaluation model for the target field was developed based on stress sensitivity coefficient (SSC) of permeability, tortuosity factor, Klinkenberg effect, near-wellbore turbulence effect and threshold pressure gradient (TPG) for validating the ANN DCA models predictions. The productivity model was validated with published horizontal well model with closely matched results. For inflow performances, the horizontal well model with turbulence minimizes negative effects on non-Darcy flow rates than without turbulence. Additionally, pressure drawdown influences the tight gas well productivity such that the lower the pressure drawdown, the smaller the tight gas well productivity. The operating points of the tight gas well were determined through inflow performance relation and tubing performance relation at different SSC and TPG for $$ 3{\raise0.5ex\hbox{$\scriptstyle 1$} \kern-0.1em/\kern-0.15em \lower0.25ex\hbox{$\scriptstyle 2$}} $$ in tubing size. In another case, synthetic unconventional simulation data with long production history were used for future forecast of 5, 24 and 70 months of production rate and cumulative production which gave pretty close results for all the DCA models, unlike the real field datasets, where the empirical rate-time models under and overestimate. In all, these ANN DCA models and the horizontal well productivity model derived will serve as new computational tools for complementing existing DCA techniques for better understanding of unconventional reservoirs’ production decline performance.

Study of the spatial distribution of groundwater quality using soft computing and geostatistical models

Groundwater quality monitoring and simulation is of great importance in arid and semi-arid regions due to water scarcity in such regions. Geostatistical methods are appropriate techniques for interpolating and spatial mapping of groundwater qualitative parameters. In the present study, geostatistic-based Kriging and Co-Kriging methods were used and compared with data-driven artificial neural network (ANN) and ANFIS models for predicting spatial distribution of groundwater electrical conductivity (EC), then the best model was selected for further sampling in the studied region. Data from 24 wells in the Keshit, Bam Normashir, and Rhmtabad plains (Kerman province, Iran) during the period of 2002–2011 were used for analysis. Root mean square error (RMSE), R2, and mean absolute error (MAE) statistical indices were used for assessing the applied models. Results showed that for both the models, i.e. triple input model (including longitude, latitude, and the number of months) and the quadruple-input model (including longitude, latitude, number of months, and Cl), ANN model presented the most accurate results with the lowest RMSE and MAE and the highest R2 values. Attending to the geostatistical methods, Co-Kriging method outperformed the Kriging method. Finally, using the tested geostatistical model, the EC data were produced for the regions without observational values.

Novel Formulation of Adaptive MPC as EKF Using ANN Model: Multiproduct Semibatch Polymerization Reactor Case Study

In this paper, a novel formulation for nonlinear model predictive control (MPC) has been proposed incorporating the extended Kalman filter (EKF) control concept using a purely data-driven artificial neural network (ANN) model based on measurements for supervisory control. The proposed scheme consists of two modules focusing on online parameter estimation based on past measurements and control estimation over control horizon based on minimizing the deviation of model output predictions from set points along the prediction horizon. An industrial case study for temperature control of a multiproduct semibatch polymerization reactor posed as a challenge problem has been considered as a test bed to apply the proposed ANN-EKFMPC strategy at supervisory level as a cascade control configuration along with proportional integral controller [ANN-EKFMPC with PI (ANN-EKFMPC-PI)]. The proposed approach is formulated incorporating all aspects of MPC including move suppression factor for control effort minimization and constraint-handling capability including terminal constraints. The nominal stability analysis and offset-free tracking capabilities of the proposed controller are proved. Its performance is evaluated by comparison with a standard MPC-based cascade control approach using the same adaptive ANN model. The ANN-EKFMPC-PI control configuration has shown better controller performance in terms of temperature tracking, smoother input profiles, as well as constraint-handling ability compared with the ANN-MPC with PI approach for two products in summer and winter. The proposed scheme is found to be versatile although it is based on a purely data-driven model with online parameter estimation.

Forecasting intraday call arrivals using the seasonal moving average method

Research into time series forecasting for call center management suggests that a forecast based on the simple Seasonal Moving Average (SMA) method outperforms more sophisticated approaches at long horizons where capacity planning decisions are made. However in the short to medium term where decisions concerning the scheduling of agents are required, the SMA method is usually outperformed. This study is the first systematic evaluation of the SMA method across averages of different lengths using call arrival data sampled at different frequencies from 5min to 1h. A hybrid method which combines the strengths of the SMA method and nonlinear data-driven artificial neural networks (ANNs) is proposed to improve short-term accuracy without deteriorating long-term performance. Results of forecasting the intraday call arrivals to banks in the US, UK and Israel indicate that the proposed method outperforms standard benchmarks, and leads to improvements in forecasting accuracy across all horizons.

Parameterized data-driven fuzzy model based optimal control of a semi-batch reactor

A parameterized data-driven fuzzy (PDDF) model structure is proposed for semi-batch processes, and its application for optimal control is illustrated. The orthonormally parameterized input trajectories, initial states and process parameters are the inputs to the model, which predicts the output trajectories in terms of Fourier coefficients. Fuzzy rules are formulated based on the signs of a linear data-driven model, while the defuzzification step incorporates a linear regression model to shift the domain from input to output domain. The fuzzy model is employed to formulate an optimal control problem for single rate as well as multi-rate systems. Simulation study on a multivariable semi-batch reactor system reveals that the proposed PDDF modeling approach is capable of capturing the nonlinear and time-varying behavior inherent in the semi-batch system fairly accurately, and the results of operating trajectory optimization using the proposed model are found to be comparable to the results obtained using the exact first principles model, and are also found to be comparable to or better than parameterized data-driven artificial neural network model based optimization results.