Model-based Predictive Approach Research Articles

Adaptive SARIMA modelling for continuous chamber temperature tracking in ultra-low temperature freezers

Ultra-low temperature (ULT) freezers are vital for storing perishable biological materials, requiring continuous monitoring of chamber thermal conditions to ensure sample integrity. A reliable dynamic data-driven model is key for smart proactive surveillance of the ULT freezers through real-time status tracking. However, the dynamic nature of ULT freezers, influenced by external disturbances, demands adaptive models to sustain performance. Complex models thus face challenges due to time-consuming parameter updates. To ensure model prediction performance while minimising model updating time, this study proposes a Seasonal Autoregressive Integrated Moving Average (SARIMA) model-based approach for short-term ULT freezer chamber temperature prediction. This method solely relies on historical chamber temperature data, eliminating the need for additional measurements. Moreover, two adaptive duty cycle length (DCL) selection strategies and a one-step ahead DCL estimation method are developed to address auto-correlation issues caused by recurring duty cycles. Comparative analyses with reference models demonstrate that the SARIMA model, incorporating future DCL effects, excels in one-step prediction and multi-step forecast under stable and variant DCL periods. Moreover, it outperforms the reference models in auto-correlation representation, affirming its reliability in predicting future states. The proposed modelling approach promises an effective method for freezing chamber temperature prediction, enabling continuous status tracking and early fault detection. Moreover, it holds great potential for modelling other critical temperatures in refrigeration systems and facilitating flexible energy operations.

Application of deep learning with bivariate models for genomic prediction of sow lifetime productivity-related traits.

Pig breeders cannot obtain phenotypic information at the time of selection for sow lifetime productivity (SLP). They would benefit from obtaining genetic information of candidate sows. Genomic data interpreted using deep learning (DL) techniques could contribute to the genetic improvement of SLP to maximize farm profitability because DL models capture nonlinear genetic effects such as dominance and epistasis more efficiently than conventional genomic prediction methods based on linear models. This study aimed to investigate the usefulness of DL for the genomic prediction of two SLP-related traits; lifetime number of litters (LNL) and lifetime pig production (LPP). Two bivariate DL models, convolutional neural network (CNN) and local convolutional neural network (LCNN), were compared with conventional bivariate linear models (i.e., genomic best linear unbiased prediction, Bayesian ridge regression, Bayes A, and Bayes B). Phenotype and pedigree data were collected from 40,011 sows that had husbandry records. Among these, 3,652 pigs were genotyped using the PorcineSNP60K BeadChip. The best predictive correlation for LNL was obtained with CNN (0.28), followed by LCNN (0.26) and conventional linear models (approximately 0.21). For LPP, the best predictive correlation was also obtained with CNN (0.29), followed by LCNN (0.27) and conventional linear models (approximately 0.25). A similar trend was observed with the mean squared error of prediction for the SLP traits. This study provides an example of a CNN that can outperform against the linear model-based genomic prediction approaches when the nonlinear interaction components are important because LNL and LPP exhibited strong epistatic interaction components. Additionally, our results suggest that applying bivariate DL models could also contribute to the prediction accuracy by utilizing the genetic correlation between LNL and LPP.

Prediction of Electric Vehicle Mileage According to Optimal Energy Consumption Criterion

In the field of electric vehicle usage, an inherent challenge lies in the restricted mileage capacity prior to requiring a recharge, hindering broader acceptance of electric vehicles. To alleviate this concern, enhancing the comprehension of vehicle energy consumption and range plays a pivotal role in easing the anxieties of electric vehicle drivers. Within this context, a novel model-based predictive approach is introduced for estimating electric vehicle energy consumption. This method considers the vehicle's specific parameters, the road network's topology, and actual traffic conditions. Through the macro model of electric vehicle energy consumption, real-time summary data can be extracted using conventional map-based web services. By representing the road network as a weighted directed graph tailored to the energy consumption model, an algorithm aids in mileage optimization by determining the optimal path for immediate use. The resultant motion range from this approach offers improved precision and dependability in contrast to conventional strategies based on average consumption and distance.

Service-oriented model-based fault prediction and localization for service compositions testing using deep learning techniques

As service-oriented computing systems become more buoyant and complex, the occurrence of faults dramatically increases. Fault prediction plays a crucial role in the service-oriented computing paradigm, aiming to reduce testing cost while maximizing testing quality to utilize testing resources effectively and increase the reliability of service compositions. Although various fault prediction techniques were considered in software testing, service-oriented systems were less fortunate, in which most of the studies have focused on single web services testing rather than service compositions. Moreover, mainly the detection of faulty/non-faulty services was addressed, ignoring the estimate of faults count, their severity, as well as predicting when and where such faults would occur. In this paper, a multilateral model-based fault prediction and localization approach is proposed using deep learning techniques for web service compositions testing rather than single web service testing, which uniquely predicts not only faulty services, but also their count and severity level, location of faults, and time at which faults would occur. Three deep learning models are investigated: Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) and a proposed hybrid model based on both CNN and RNN. The proposed approach is language-independent, as it adopts process metrics rather than code metrics to overcome the code unavailability concern of services. The experimental analysis adopted main performance metrics on multiple public datasets to evaluate its efficiency and effectiveness. The results indicated that the hybrid CNN_RNN model achieves an average accuracy range of 84%–95.7%, where the RNN and CNN models individually achieve 75%–90% and 70%-79.3% respectively. Thus, the hybrid model increases the accuracy level by 5%–10% and 15%–20%, while achieving the least mean square error of 30% and 60% compared to the RNN and CNN models respectively. In terms of time, the RNN model consumes less average time as of 30–50 ms for the different datasets of variant sizes compared to the CNN and hybrid CNN_RNN models that consume 79–102 and 177–224 ms respectively. Thus, RNN model consumes around 50%–80% less time than those of the CNN and hybrid models respectively.

A Novel Model-Assisted Decentralized Multi-Agent Reinforcement Learning for Joint Optimization of Hybrid Beamforming in Massive MIMO mmWave Systems

To meet the explosive increasing demands for high data rates, hybrid beamforming has become a promising paradigm for massive MIMO millimeter wave (mmWave) systems. In this paper, we study the joint optimization of hybrid beamforming for maximizing the achievable sum rate in networked dynamic massive MIMO mmWave systems. To address this problem, we propose a novel Model-Assisted Decentralized Multi-Agent Reinforcement Learning (MAD-MARL) algorithm, which incorporates the following three designs: 1) a model-based prediction approach, which can improve the policy learning speed greatly and also predict each agent's potential future so as to alleviate the impact of delayed information sharing in a networked system, 2) an attentional prediction sharing approach, which is to effectively reduce the non-stationarity of the environment by combining the proposed model-based prediction and self-attention strategy for reaching global consensus and cooperative inter-agent coordination in a distributed manner, and 3) a decentralized model-free learning approach, which trains the agents with the assistance of model-based predictions. Extensive simulations are conducted and the numerical results demonstrate that MAD-MARL can significantly increase the learning speed and also greatly improve the overall performance compared with existing work.

Augmented model-based framework for battery remaining useful life prediction

Traditional, model-based approaches for predicting the remaining useful life (RUL) of a rechargeable battery cell simply update and extrapolate a mathematical model which describes the evolution of the cell’s capacity fade trend. These approaches are straightforward but tend to break down when the capacity fade trend changes over the cell’s lifetime. To retain the desirable properties of model-based prediction approaches (uncertainty quantification, long-term accuracy, limited physical meaning) and improve their overall accuracy in RUL prediction, we augment empirical model-based prediction with data-driven error correction. Our approach decomposes the task of RUL prediction into two steps: 1) Offline training of data-driven models for RUL error correction and 2) Online data-driven correction of model-based RUL prediction. The approach is evaluated on five datasets consisting of 237 cells: 1) three open-source datasets, 2) one proprietary dataset, and 3) a simulated out-of-distribution dataset. Results show that data-driven error correction effectively reduces root-mean-square-error by 40% and mean uncertainty calibration error by 34% compared to a model-based approach alone. The proposed approach is also shown to be more conservative in its uncertainty estimates than a purely data-driven RUL prediction approach. Special attention is given to ensure the initial model-based uncertainty estimates are propagated through the data-driven error correction model and considered in the final RUL prediction. The enhanced uncertainty quantification of our approach makes it suitable for deployment in an online predictive maintenance scheduling framework.

Deep Learning model-based approach for preoperative prediction of Ki67 labeling index status in a noninvasive way using magnetic resonance images: A single-center study

ObjectivesKi67 is an important biomarker of pituitary adenoma (PA) aggressiveness. In this study, PA invasion of surrounding structures is investigated and deep learning (DL) models are established for preoperative prediction of Ki67 labeling index (Ki67LI) status using conventional magnetic resonance (MR) images. MethodsWe reviewed 362 consecutive patients with PAs who underwent endoscopic transsphenoidal surgery, of which 246 patients with primary PA are selected for PA invasion analysis. MRI data from 234 of these PA patients are collected to develop DL models to predict Ki67LI status, and DL models were tested on 27 PA patients in the clinical setting. ResultsPA invasion is observed in 46.8% of cases in the Ki67 ≥ 3% group and 33.3% of cases in the Ki67 < 3% group. Three deep-learning models are developed using contrast-enhanced T1-weighted images (ceT1WI), T2-weighted images (T2WI), and multimodal images (ceT1WI+T2WI), respectively. On the validation dataset, the prediction accuracy of the ceT1WI model, T2WI model, and multimodal model were 87.4%, 89.4%, and 89.2%, respectively. In the clinical test, 27 MR slices with the largest tumors from 27 PA patients were tested using the ceT1WI model, T2WI model, and multimodal model, the average accuracy of Ki67LI status prediction was 63%, 77.8%, and 70.4%, respectively. ConclusionPreoperative prediction of PA Ki67LI status in a noninvasive way was realized with the DL model by using MRI. T2WI model outperformed the ceT1WI model and multimodal model. This end-to-end model-based approach only requires a single slice of T2WI to predict Ki67LI status and provides a new tool to help clinicians make better PA treatment decisions.

Weight optimized unscented Kalman filter for degradation trend prediction of lithium-ion battery with error compensation strategy

Due to the inevitable degradation of Lithium-ion batteries (LIBs) during its lifetime, remaining useful life (RUL) prediction methods are adopted for ensuring the stable and safety operation of electrical equipment. To make up the deficiencies of single model-based or data-driven prediction approach, this paper proposes a new hybrid framework using a weight optimization unscented Kalman filter (WOUKF) and attention based bi-directional long short-term memory (BiLSTM-AM). To be specific, firstly, a Fourier model is proposed to describe the degradation of LIBs instead of the traditional double exponential model. Then, a WOUKF algorithm is designed to identify the model parameters efficiently even in the case of poor initialization. Next, the trends of prediction residual between the WOUKF and the true capacity is established by BiLSTM-AM, which is fed back to the WOUKF for updating model parameters during the prediction period. In addition, an error compensation scheme is developed to further improve prediction performance. Finally, the effectiveness of the proposed prognosis framework is verified on two battery datasets. The simulation results show that the proposed method has better prediction accuracy. The maximum root mean squared error (RMSE) and mean absolute error (MAE) of the proposed hybrid framework are 3.5% and 3%, respectively.

Adaptive optics control with multi-agent model-free reinforcement learning.

We present a novel formulation of closed-loop adaptive optics (AO) control as a multi-agent reinforcement learning (MARL) problem in which the controller is able to learn a non-linear policy and does not need a priori information on the dynamics of the atmosphere. We identify the different challenges of applying a reinforcement learning (RL) method to AO and, to solve them, propose the combination of model-free MARL for control with an autoencoder neural network to mitigate the effect of noise. Moreover, we extend current existing methods of error budget analysis to include a RL controller. The experimental results for an 8m telescope equipped with a 40x40 Shack-Hartmann system show a significant increase in performance over the integrator baseline and comparable performance to a model-based predictive approach, a linear quadratic Gaussian controller with perfect knowledge of atmospheric conditions. Finally, the error budget analysis provides evidence that the RL controller is partially compensating for bandwidth error and is helping to mitigate the propagation of aliasing.

Rigorous Predictive Noise Modeling Approach for Model-Based Onset Detection and Enhanced Picking of P-Waves in Seismic Signals

Detecting and picking the onset of P-waves in seismic signals has a fairly rich literature, among which model-based (predictive) approaches hold immense promise. A majority of these models are usually built on certain critical assumptions, namely, stationarity, linearity, and Gaussianity. Despite their criticality, very little reported literature exists on validating these assumptions on real seismic data. Furthermore, the predictive capabilities of these models are not utilized to their best potential in detection. Therefore, there exists a strong need for a rigorous approach to model seismic noise and develop a method that builds on the statistical properties of residuals resulting from the noise models. The objectives of this work are (i) to critically study certain long-held assumptions in seismic noise modeling, (ii) to develop rigorous time-series models for background noise that are commensurate with the noise properties, so as to (iii) devise a residual-based method for detection and enhanced picking of P-waves. An important finding of this work, arising from our study on 185 historical data sets, is that these standard assumptions do not hold for most of the data sets under study; rather, they exhibit additional special features such as heteroskedasticity and integrating effects. Consequent to these novel discoveries, we develop auto-regressive integrated moving average- generalized auto-regressive conditionally heteroskedastic (ARIMA-GARCH) models for seismic noise. The proposed residual-based detector and picker is found to be highly effective with a 90&#x0025; detection rate while picking 91&#x0025; of the events with an accuracy of &#x2264; 0.625 seconds based on tests with 100 historical data sets. Further, when the noise model is used in combination with the existing AIC-based pickers, the number of events picked with an accuracy of &#x2264; 0.625 seconds is 50&#x0025; more than the existing AR-AIC picker. Therefore, the proposed method can be considered highly competitive and effective in P-wave detection and picking.

Trajectory Prediction of Preceding Target Vehicles Based on Lane Crossing and Final Points Generation Model Considering Driving Styles

Reliable trajectory prediction of preceding target vehicles (PTVs) is crucial for the planning and decision making of automated vehicles. However, the future trajectory is affected by the driver's intention and diverse driving styles, which can hardly be predicted precisely, especially when the vehicle performs a lane change maneuver. In this study, we propose a lane crossing and final points generation (CFPG) model-based trajectory prediction approach for PTVs, in which the key influence factors such as the driver's intention and the mixed driving style are included. Firstly, we build a maneuver and stage recognition model upon the long short term memory (LSTM) to infer the current maneuver of the preceding target vehicle. Furthermore, the approach predicts the lane crossing point using a physics-based model combining with a deep conditional generative model trained by a deep neural network. Moreover, a maneuver-based model is adopted to predict the final point according to the prediction interval. In order to avoid the possible cumulative error caused by iteratively generating trajectories in traditional methods, we use a curve fitting method to obtain the predicted trajectory. Lane changing data collected from naturalistic driving scenarios are used to verify the proposed approach, and the results suggest more accurate and reliable prediction trajectories compared with conventional methods.

Model-Based Predictive Rotor Field-Oriented Angle Compensation for Induction Machine Drives

In this paper, a model-based predictive rotor field-oriented angle compensation approach is proposed for induction machine drives. Indirect rotor field-oriented control is widely used in induction machine drives for its simple implementation and low cost. However, the accuracy of the rotor field-oriented angle is affected by variable parameters such as the rotor resistance and inductance. An inaccurate rotor field-oriented angle leads to a degradation of the torque and dynamic performance, especially in the high-speed flux-weakening region. Therefore, the d-axis and q-axis currents in the rotation reference frame are predicted based on the model and compared with the feedback current to correct the rotor field-oriented angle. To improve the stability and robustness, the proposed predictive algorithm is based on the storage current, voltage, and velocity data. The algorithm can be easily realized in real-time. Finally, the simulated and experimental results verify the algorithm’s effectiveness on a 7.5 kW induction machine setup.

A Probability Distribution Model-Based Approach for Foot Placement Prediction in the Early Swing Phase With a Wearable IMU Sensor.

Predicting the next foot placement of humans during walking can help improve compliant interactions between humans and walking aid robots. Previous studies have focused on foot placement estimation with wearable inertial sensors after heel-strike, but few have predicted foot placements in advance during the early swing phase. In this study, a Bayesian inference-based foot placement prediction approach was proposed. Possible foot placements were modeled as a probability distribution grid map. With selected foot motion feature events detected sequentially in the early swing phase, the foot placement probability map could be updated iteratively using the feature models we built. The weighted center of the probability distribution was regarded as the predicted foot placement. Prediction errors were evaluated with collected walking data sets. When testing with the data from inertial measurement units, the prediction errors were (5.46 cm ± 10.89 cm, -0.83 cm ± 10.56 cm) for cross-velocity walking data and (-4.99 cm ± 12.31 cm, -11.27 cm ± 7.74 cm) for cross-subject-cross-velocity walking data. The results were comparable to previous works yet the prediction could be made earlier. For the subject who walked with more stable gaits, the prediction error can be further decreased. The proposed foot placement prediction approach can be utilized to help walking aid robots adjust their pose before each heel-strike event during walking, which will make human-robot interactions more compliant. This study is also expected to inspire additional probabilistic gait analysis works.

Evaluation of Non-Invasive Ankle Joint Effort Prediction Methods for Use in Neurorehabilitation Using Electromyography and Ultrasound Imaging.

Reliable measurement of voluntary human effort is essential for effective and safe interaction between the wearer and an assistive robot. Existing voluntary effort prediction methods that use surface electromyography (sEMG) are susceptible to prediction inaccuracies due to non-selectivity in measuring muscle responses. This technical challenge motivates an investigation into alternative non-invasive effort prediction methods that directly visualize the muscle response and improve effort prediction accuracy. The paper is a comparative study of ultrasound imaging (US)-derived neuromuscular signals and sEMG signals for their use in predicting isometric ankle dorsiflexion moment. Furthermore, the study evaluates the prediction accuracy of model-based and model-free voluntary effort prediction approaches that use these signals. The study evaluates sEMG signals and three US imaging-derived signals: pennation angle, muscle fascicle length, and echogenicity and three voluntary effort prediction methods: linear regression (LR), feedforward neural network (FFNN), and Hill-type neuromuscular model (HNM). In all the prediction methods, pennation angle and fascicle length significantly improve the prediction accuracy of dorsiflexion moment, when compared to echogenicity. Also, compared to LR, both FFNN and HNM improve dorsiflexion moment prediction accuracy. The findings indicate FFNN or HNM approach and using pennation angle or fascicle length predict human ankle movement intent with higher accuracy. The accurate ankle effort prediction will pave the path to safe and reliable robotic assistance in patients with drop foot.

Supervised Machine Learning Model-Based Approach for Performance Prediction of Students

Predicting students’ performance is one of the crucial issue for learning contexts, since it helps to develop alternative recommendation systems for academically weak students. Thus, several methods and practices have been applied for educational improvement. However, most of the existing methods do not determine the performance of the students. In this study, we have studied the execution of six machine learning models (Decision tree, Random Forest, Support Vector Machine, Logistic Regression, Ada Boost, Stochastic Gradient Descent) to analyze and evaluate the students’ achievements. The performance is evaluated in term of accuracy, precision, sensitivity and f-measure. Among the selected models, the results validate that the efficiency of Stochastic Gradient Descent is better in training small dataset. In addition, it also produces the higher accuracy as compared with other models. This contribution aims to develop the best model which may derive the conclusion on students' academic achievement.

Comparative Study of a Biomechanical Model-based and Black-box Approach for Subject-Specific Movement Prediction.

The performance and safety of human robot interaction (HRI) can be improved by using subject-specific movement prediction. Typical models include biomechanical (parametric) or black-box (non-parametric) models. The current work aims to investigate the benefits and drawbacks of these approaches by comparing elbow-joint torque predictions based on electromyography signals of the elbow flexors and extensors. To this end, a parameterized biomechanical model is compared to a non-parametric (Gaussian-process) approach. Both models showed adequate results in predicting the elbow-joint torques. While the non-parametric model requires minimal modeling effort, the parameterized biomechanical model can lead to deeper insight of the underlying subject specific musculoskeletal system.

Predictive control of a permanent magnet synchronous generator connected to an MMC converter in an oscillating water column based power plant

Modular Multilevel Converters (MMC) are traditionally used in high-power applications although their field of application has lately widened to lower powers. In medium and low-power applications such as electric drives, a fast control system is usually required and model-based predictive control approaches (MPC) are highly suitable to obtain superior performance. Unfortunately, although MPC is widely used with two-level three-phase power converters, typically in motor and generator control applications, it has been barely developed for MMC. Recent studies discuss MPC approaches for MMC but they are primarily focused on improving the internal operation of MMC. In contrast, this paper presents an MPC which is designed to control the torque and the magnetic field of a surface permanent magnet synchronous generator (SPMSG) by means of an MMC. The proposed approach features a new and fast method to find the location of the reference voltage vector among the large number of multilevel vectors present in typical MMCs. The performance of the new control system has been demonstrated for an oscillating water column (OWC) based power plant simulation. The control system stands out for its fast response, low THD currents and reasonably low computing time which enables its easy implementation in microcontrollers (MCU).

Reduction of common‐mode voltage using a simplified FSC‐MPC for a five‐phase induction motor drive

This paper presents a cascaded closed-loop control of a five-phase induction motor (IM) drive system with common-mode voltage reduction capability. The outer loop uses standard PI regulators for speed reference tracking, while the inner control loop is implemented using a reduced order finite-state model-based current predictive approach. Establishing certain constraints, different combinations of switching state vectors are obtained, leading to the generation of different magnitudes of common-mode voltages. As a result, precise speed and current control is achieved in addition to the reduction of common-mode voltage and computational burden. Experimental results are presented to validate the concept.

Inferring Pedestrian Motions at Urban Crosswalks

Robust prediction of pedestrian behavior is one of the most challenging problems for autonomous driving. Particularly, predicting pedestrian crossings at crosswalks is of considerable importance for avoiding accidents on the one hand and not unnecessarily slowing down traffic on the other hand. Traditional model-based motion tracking and prediction approaches have difficulties in capturing abrupt changes in motions, as humans can perform them. In this paper, an approach for predicting pedestrian motions that combines established motion tracking algorithms with data-driven methods is presented. The approach is built upon a hierarchical structure, where first, the intent of each pedestrian is classified. Then, the approach computes several qualitative metrics, such as time-to-cross, for the pedestrians classified as crossing. The approach is evaluated on a challenging urban data set collected for different types of crosswalks such as roundabouts and straight roads. The evaluation also provides a thorough analysis of the generalization performance of the proposed approach.

A hidden-Gamma model-based filtering and prediction approach for monotonic health factors in manufacturing

In the context of Smart Monitoring and Fault Detection and Isolation in industrial systems, the aim of Predictive Maintenance technologies is to predict the happening of process or equipment faults. In order for a Predictive Maintenance technology to be effective, its predictions have to be both accurate and timely for taking strategic decisions on maintenance scheduling, in a cost-minimization perspective. A number of Predictive Maintenance technologies are based on the use of “health factors”, quantitative indicators associated with the equipment wear that exhibit a monotone evolution. In real industrial environment, such indicators are usually affected by measurement noise and non-uniform sampling time. In this work we present a methodology, formulated as a stochastic filtering problem, to optimally predict the evolution of the aforementioned health factors based on noisy and irregularly sampled observations. In particular, a hidden Gamma process model is proposed to capture the nonnegativity and nonnegativity of the derivative of the health factor. As such filtering problem is not amenable to a closed form solution, a numerical Monte Carlo approach based on particle filtering is here employed. An adaptive parameter identification procedure is proposed to achieve the best trade-off between promptness and low noise sensitivity. Furthermore, a methodology to identify the risk function associated to the observed equipment based on previous maintenance data is proposed. The present study is motivated and tested on a real industrial Predictive Maintenance problem in semiconductor manufacturing, with reference to a dry etching equipment.