Accuracy Of Dynamic Prediction Research Articles

Dynamic performance prediction of PEM electrolyzers combining an electrochemical neural network model and a lumped thermal capacitance model

ABSTRACT The proton exchange membrane (PEM) electrolyzer electrochemical models based on conventional analytical and empirical approaches require numerous parameters, which are difficult to obtain accurately, leading to lower dynamic performance prediction accuracy while the parameters of the lumped thermal capacitance model are relatively simple and easily identifiable. This study proposes an integrated modeling framework, which combines a backpropagation (BP) neural network model for electrochemical performance and a lumped thermal capacitance model for the assessment of thermal performance. The neural network model is trained on polarization data collected from 44°C to 76°C. The BP neural network, optimized through iterative hyperparameter tuning, achieves high predictive accuracy with three hidden layer nodes and tansig and purelin activation functions, exhibiting less than 1% error in polarization curve predictions at four additional temperatures and current densities from 0.04 to 2.6 A/cm2. The thermal model, calibrated with experimental data, predicts thermal-balance temperatures with root mean square error (RMSE) of 0.08°C to 0.52°C under four different operating conditions. The dynamic performance of the integrated model is validated under three different operating conditions. This comprehensive model enables detailed analysis of critical operating variables, providing actionable insights for selecting and optimizing operating conditions in practical applications.

Structural Online Damage Identification and Dynamic Reliability Prediction Method Based on Unscented Kalman Filter.

As sensor monitoring technology continues to evolve, structural online monitoring and health management have found numerous applications across various fields. However, challenges remain concerning the real-time diagnosis of structural damage and the accuracy of dynamic reliability predictions. In this paper, a structural online damage identification and dynamic reliability prediction method based on Unscented Kalman Filter (UKF) is presented. Specifically, in the Wiener degradation process with random effects on structural performance, the structural damage identification is initially realized using UKF. Following that, the EM algorithm is employed for estimating the performance model parameters. Eventually, dynamic reliability prediction is realized based on conditional probability. The simulation results indicate that the method effectively estimates the damage state during the structure's use while providing accurate, real-time, and dynamic reliability predictions for the system.

Aridity-Driven Change in Microbial Carbon Use Efficiency and Its Linkage to Soil Carbon Storage.

Global warming is generally predicted to increase aridity in drylands, while the effects of aridity changes on microbial carbon use efficiency (CUE) and its linkage to soil organic carbon (SOC) storage remain unresolved, limiting the accuracy of soil carbon dynamic predictions under changing climates. Here, by employing large-scale soil sampling from 50 sites along an ~6000 km aridity gradient in northern China, we report a significant decreasing trend in microbial CUE (ranging from approximately 0.07 to 0.59 across the aridity gradient) with increasing aridity. The negative effect of aridity on microbial CUE was further verified by an independent moisture manipulation experiment, which revealed that CUE was lower under lower moisture levels than under higher moisture levels. Aridity-induced increases in physicochemical protection or decreases in microbial diversity primarily mediated the decrease in CUE with increasing aridity. Moreover, we found a highly positive microbial CUE-SOC relationship, and incorporating CUE improved the explanatory power of SOC variations along the aridity gradient. Our findings provide empirical evidence for aridity-induced reductions in microbial CUE over a broad geographic scale and highlight that increasing aridity may be a crucial mechanism underlying SOC loss by suppressing the ability of soil microorganisms to sequester carbon.

Nonlinear hydrodynamic assessment of a floating solar double-hull substructure using viscous numerical wave tank

This paper assesses a viscous numerical solver for hydrodynamic simulations of floating double-hull substructures supporting solar panels, critical for advancing floating solar technologies. Using the OpenFOAM repository, the study develops a model based on the Finite Volume method and Volume of Fluid approach, focusing on a range of wave frequencies, from weakly to strongly nonlinear, at intermediate depths. A free decay test calibrated the mesh morphology of the control volume, enabling comparative analysis of single and double-cylinder structures exposed to Stokes second-order nonlinear waves, alongside nonlinear potential models and experimental data. Spectral analysis of these solutions quantifies the nonlinearities' influence on structural responses and high-order dynamic prediction accuracy. The key findings demonstrate that the viscous model accurately predicted heave and surge responses, outperforming the potential model under steep waves. The gap distance between the cylinder in the double-cylinder platforms significantly affects fluid flow and stability, especially under shorter waves. Accurate wave-body simulations require appropriate mesh resolution and tailored wavemaker functions for nonlinear waves. These findings highlight the need to incorporate viscous effects in floating solar platform design to enhance stability and performance in real-world environments.

Improving dynamic prognostication of patients with hepatocellular carcinoma by node fusion algorithm equipped survival path mapping.

e13594 Background: Patients with hepatocellular carcinoma (HCC) always require routine surveillance and repeated treatment. Our previous work (1) developed a machine learning methodology called “Survival Path Mapping”, which converted the time series data into a cascading survival map to facilitate dynamic prognosis prediction and treatment planning. However, in previous work, the exponential growth of paths raised challenges of exhausting limited cohort and over-fitting risk. Novel algorithm enhancing data utilization in the Survival Path Mapping was in urgent need. Methods: A node fusion algorithm was designed. First, feature selection algorithm bifurcated each parent-nodes into temporary child-nodes, which constitutes temporary child-nodes set. Then, clustering was performed within temporary child-nodes set according to key features of each child-node. Finally, in each cluster, multiple child-nodes would be merged into one if there was no significant difference in survival. For nodes with bifurcation, a treatment recommendation algorithm was developed. A total of 19446 patients with newly diagnosed primary HCC at Sun Yat-sen University Cancer Center (SYSUCC) between 2008 and 2023 were retrospectively included as internal cohort (divided into internal train and holdout validation cohort); meanwhile 1135 patients from five hospitals were included as multicenter external validation cohort. Dedicated natural language processing program was developed to extract structured features from medical imaging reports. The time series data of patients were converted into data of time slices, with an interval of 3 months. Results: The original survival path mapping (raw-SP) consists of 463 different nodes, while the node fusion survival path mapping (fusion-SP) consisted of 84 nodes (81.8% decrease from raw-SP). The 3-year c-index of fusion-SP were 0.745 (95% CI 0.717-0.772), significantly better (p<0.001) than that of raw-SP: 0.725 (95% CI 0.696-0.755), in the internal holdout validation. Both raw-SP and fusion-SP significantly (p < 0.001) outperformed conventional BCLC and CNLC staging. In the multicenter external validation, the fusion-SP also demonstrated better performance than raw-SP and BCLC staging system. The recommended therapy calculated based fusion-SP system for patients receiving second-line or subsequent treatment had superior or non-inferior efficacy compared to BCLC guideline’s recommendation. Conclusions: The proposed node fusion algorithm solved exponential growth of paths in previous survival path algorithm, improving the accuracy and external validity of dynamic prognosis prediction for HCC patients. The fusion-SP model also retained its transparency and could be used to guide dynamic treatment planning, with its effectiveness in treatment guidance evaluated in prospective trials in the future. 1. Nat Commun 9, 2230; 2018.

Multiscale simulation and experimental studies on modal damping of ultra‐thick CFRP laminates

AbstractThe vibration characterization of ultra‐thick laminates widely used in aerospace and marine industries was different from that of thin laminates. However, modal testing and simulation methods were generally limited to relatively thin laminates. To explore the vibration behavior of ultra‐thick laminates, a novel multiscale simulation technology based on the representative volume modal strain energy (RVMSE) method is presented for predicting modal damping in this paper. The adaptive Greedy‐Based Generation (GBG) algorithm is used to create a 3D random distribution of fibers for composites. Modal shape information of the structure is obtained according to the Lanczos algorithm, which is used as an input parameter for the microscopic model. This is followed by homogenization based on the volumetric analysis performed on representative volume elements (RVEs) to determine the strain energy in six directions. The calculation of modal damping is performed by using the modal strain energy (MSE) method and the damping properties of the six directions. The fast rate of convergence and high accuracy of the method are demonstrated through different examples. The vibration response predictions produced by novel RVMSE methodology are shown to closely match experimental measurements, providing scope to expand the application of this approach to more complex, ultra‐thick laminate components.Highlights The RVMSE method uses a higher level of homogenization to overcome the limitations of the traditional MSE method for calculating the damping of ultra‐thick CFRP laminates. The modal parameter error is controlled to less than 10%, and the experimental results fill the gap of the damping data of ultra‐thick CFRP structure. The proposed RVMSE model improves the accuracy of dynamic response prediction for ultra‐thick laminates.

Finite element model updating and response prediction of a frame structure based on optimal sensor placement

This paper proposes an approach for finite element (FE) model updating and response prediction of frame structures based on optimal sensor placement (OSP), which integrates sensor placement optimization, mode expansion, and model updating techniques. Firstly, sensor optimization layout and modal testing analysis are conducted on a bolted laboratory frame structure. The covariance-driven stochastic subspace identification (SSI-COV) method identifies the real-world structure’s natural frequencies, modal damping, and mode shapes. Secondly, the complete mode shapes are expanded using the measured incomplete modal data from the limited number of sensors. Thirdly, a multi-objective function based on frequency and mode shapes is established to adjust the parameters of the FE model. This ensures that the updated model accurately represents the dynamic properties of the actual structure within a specific frequency range. Finally, the Rayleigh damping of the frame structure is estimated, and the damping matrix is assembled to enhance the accuracy of dynamic response prediction in the updated model. By comparing the response prediction results of the updated FE model with and without considering the updated damping effects to the measurement data of the real-world structure, it is demonstrated that the proposed method considering updated damping effects can more effectively predict the structural response.

Examining the impact of critical attributes on hard drive failure times: Multi‐state models for left‐truncated and right‐censored semi‐competing risks data

AbstractThe ability to predict failures in hard disk drives (HDDs) is a major objective of HDD manufacturers since avoiding unexpected failures may prevent data loss, improve service reliability, and reduce data center downtime. Most HDDs are equipped with a threshold‐based monitoring system named self‐monitoring, analysis and reporting technology (SMART). The system collects several performance metrics, called SMART attributes, and detects anomalies that may indicate incipient failures. SMART works as a nascent failure detection method and does not estimate the HDDs' remaining useful life. We define critical attributes and critical states for hard drives using SMART attributes and fit multi‐state models to the resulting semi‐competing risks data. The multi‐state models provide a coherent and novel way to model the failure time of a hard drive and allow us to examine the impact of critical attributes on the failure time of a hard drive. We derive dynamic predictions of conditional survival probabilities, which are adaptive to the state of the drive. Using a dataset of HDDs equipped with SMART, we find that drives are more likely to fail after entering critical states. We evaluate the predictive accuracy of the proposed models with a case study of HDDs equipped with SMART, using the time‐dependent area under the receiver operating characteristic curve (AUC) and the expected prediction error (PE). The results suggest that accounting for changes in the critical attributes improves the accuracy of dynamic predictions.

Model optimization in system equivalent model mixing using analytical modal analysis

This study pertains to the application of system equivalent model mixing to an experimental and analytical model. System equivalent model mixing is applied to a suspension subframe model, and virtual points at the suspension mount points are set as target points. Subsequently, the six degree-of-freedom frequency response functions at the virtual points are predicted. To improve the accuracy of dynamic characteristic prediction at the target points, system equivalent model mixing with extended compatibility is applied. Points in the test or analysis that can be used to apply system equivalent model mixing are not limited. The accuracy of the system equivalent model mixing is analyzed based on changes in the points in the analysis and test model. The mixing points for the system equivalent model are optimized via preliminary analysis in the pre-test stage.

Entropy-Aware Time-Varying Graph Neural Networks with Generalized Temporal Hawkes Process: Dynamic Link Prediction in the Presence of Node Addition and Deletion

This paper addresses the problem of learning temporal graph representations, which capture the changing nature of complex evolving networks. Existing approaches mainly focus on adding new nodes and edges to capture dynamic graph structures. However, to achieve more accurate representation of graph evolution, we consider both the addition and deletion of nodes and edges as events. These events occur at irregular time scales and are modeled using temporal point processes. Our goal is to learn the conditional intensity function of the temporal point process to investigate the influence of deletion events on node representation learning for link-level prediction. We incorporate network entropy, a measure of node and edge significance, to capture the effect of node deletion and edge removal in our framework. Additionally, we leveraged the characteristics of a generalized temporal Hawkes process, which considers the inhibitory effects of events where past occurrences can reduce future intensity. This framework enables dynamic representation learning by effectively modeling both addition and deletion events in the temporal graph. To evaluate our approach, we utilize autonomous system graphs, a family of inhomogeneous sparse graphs with instances of node and edge additions and deletions, in a link prediction task. By integrating these enhancements into our framework, we improve the accuracy of dynamic link prediction and enable better understanding of the dynamic evolution of complex networks.

Monitoring and Prediction of Horizontal Displacement of Underground Enclosure Piles in Subway Foundation Pits.

Urban rail transportation and underground space development and utilization are initiatives to solve urban traffic congestion. The monitoring and prediction of the stability of underground enclosure piles in foundation pits is a key index for the dynamic evaluation of the stability of underground space engineering. In this paper, we focused on the problem that the dynamic prediction accuracy and stability of foundation pit retaining piles in the Qingdao area are not high. Based on the analysis of various time function curves and the physical meaning of the parameters, we proposed the Adjusted-Logistic time function model, which introduces three physical parameters to adjust the deformation velocity and deformation acceleration in different stages for better accuracy. It could predict the deformation process of underground enclosure piles under different geological engineering conditions. It was verified in the field that the root-mean-square error (RMSE) of the Adjusted-Logistic function was 0.5316, MAE was 0.3752, and R2 was 0.9937, which were better than those of Gompertz, Weibull, and Knothe time function models. Meanwhile, it showed that with the increase of excavation depth, the maximum horizontal displacement of the underground enclosure piles gradually decreased and finally stabilized at 0.62-0.71 H. We established a catastrophe model of the horizontal displacement cusp at the observation point of the underground enclosure piles by using the time series of the measured data. It could determine the weak location of the underground enclosure pile stability and realize the multipoint warning of the foundation pit stability, which would ensure safe construction.

Long-run forecasting surface and groundwater dynamics from intermittent observation data: An evaluation for 50 years

The accurate prediction of water dynamics is critical for operational water resource management. In this study, we propose a novel approach to perform long-term forecasts of daily water dynamics, including river levels, river discharges, and groundwater levels, with a lead time of 7–30 days. The approach is based on the state-of-the-art neural network, bidirectional long short-term memory (BiLSTM), to enhance the accuracy and consistency of dynamic predictions. The operation of this forecasting system relies on an in-situ database observed for over 50 years with records gauging in 19 rivers, the karst aquifer, the English Channel, and the meteorological network in Normandy, France. To address the problem of missing measurements and gauge installations over time, we developed an adaptive scheme in which the neural network is regularly adjusted and re-trained in response to changing inputs during a long operation. Advances in BiLSTM with extensive learning past-to-future and future-to-past further help to avoid time-lag calibration that simplifies data processing.The proposed approach provides high accuracy and consistent prediction for the three water dynamics within a similar accuracy range as an on-site observation, with approximately 3 % error in the measurement range for the 7 day-ahead predictions and 6 % error for the 30 d-ahead predictions. The system also effectively fills the gap in actual measurements and detects anomalies at gauges that can last for years. Working with multiple dynamics not only proves that the data-driven model is a unified approach but also reveals the impact of the physical background of the dynamics on the performance of their predictions. Groundwater undergoes a slow filtration process following a low-frequency fluctuation, favoring long-term prediction, which differs from other higher-frequency river dynamics. The physical nature drives the predictive performance even when using a data-driven model.

Dynamic Target Tracking Control of Autonomous Underwater Vehicle Based on Trajectory Prediction.

Underwater dynamic target tracking technology has a wide application prospect in marine resource exploration, underwater engineering operations, naval battlefield monitoring, and underwater precision guidance. Aiming at the underwater dynamic target tracking problem, an autonomous underwater vehicle tracking control method based on trajectory prediction is studied. First, a deep learning-based target detection algorithm is developed. For the image collected by the multibeam forward-looking sonar image, this algorithm uses the YOLO v3 network to determine the target in a sonar image and obtain the position of the target. Then, a time profit Elman neural network (TPENN) is constructed to predict the trajectory information of the dynamic target. Compared with an ordinary Elman neural network, its accuracy of dynamic target prediction is increased. Finally, underwater tracking of the dynamic target is realized using the model predictive controller (MPC), and the tracking result is stable and reliable. Through simulations and experiment, the proposed underwater dynamic target tracking control method is demonstrated to be effective and feasible.

Multivariate linear-regression variable parameter spatio-temporal zoning model for temperature prediction in steel rolling reheating furnace

In the steel industry, temperature control and prediction of billets during heating in steel rolling reheating furnace (SRRF) can guarantee product performance and save energy consumption. However, the nonlinear time-varying (NTV) and distributed of SRRF are a great challenge for the accurate temperature control of billets. It is important to predict and regulate the temperature of the SRRF online and to identify the causal variables that may lead to temperature variations. In this paper, a new multivariate linear regression variable parameter spatio-temporal (MLR-VPST) zoning model approach is proposed to predict the temperature. Firstly, a MLR-VPST zoning model method is proposed to predict the temperature of the SRRF, which can handle the distribution characteristics of the temperature field and the NTV characteristics. Secondly, a new least squares matrix block (LSMB) algorithm is proposed to obtain batch parameter solutions, which can improve the dynamic and temperature prediction accuracy. Thirdly, a zoning model approach is proposed to describe the degree of fitting of the SRRF model in each zone, and the linear correlation is gradually increasing from the reheating zone to the soaking zone. It makes the temperature of the soaking zone more accurate. Simulation results of real-time data from steel company show that it meets the requirements of the rolling process.

Trajectories of Change in Acute Dynamic Risk Ratings and Associated Risk for Recidivism in Paroled New Zealanders: A Joint Latent Class Modelling Approach

ObjectivesPrior studies indicate risk for recidivism declines with time spent in the community post-incarceration. The current study tested whether declines in risk scores occurred uniformly for all individuals in a community corrections sample or whether distinct groups could be identified on the basis of similar trajectories of change in acute risk and time to recidivism. We additionally tested whether accounting for group heterogeneity improved prospective prediction of recidivism.MethodsThis study used longitudinal, multiple-reassessment data gathered from 3,421 individuals supervised on parole in New Zealand (N = 92,104 assessments of theoretically dynamic risk factors conducted by community corrections supervision officers). We applied joint latent class modelling (JLCM) to model group trajectories of change in acute risk following re-entry while accounting for data missing due to recidivism (i.e., missing not at random). We compared accuracy of dynamic predictions based on the selected joint latent class model to an equivalent joint model with no latent class structure.ResultsWe identified four trajectory groups of acute dynamic risk. Groups were consistently estimated across a split sample. Trajectories differed in direction and degree of change but using the latent class structure did not improve discrimination when predicting recidivism.ConclusionsThere may be significant heterogeneity in how individuals’ assessed level of acute risk changes following re-entry, but determining risk for recidivism should not be based on probable group membership. JLCM revealed heterogeneity in early re-entry unlikely to be observed using traditional analytic approaches.

Ship Navigation Behavior Prediction Based on AIS Data

Real-time and accurate ship navigation dynamic prediction can effectively improve maritime supervision’s intelligence and precision level and ensure ship navigation safety. To further enhance the accuracy of ship navigation dynamic prediction, this paper uses ship AIS as the data source and proposes an improved LSTM navigation dynamic prediction model based on the attention mechanism. Firstly, a set of pre-processing means, including navigation data extraction, abnormal data processing, and missing data interpolation, is proposed to solve the problems of information loss and inaccuracy in AIS data and incomplete retention of dynamic navigation features; secondly, combining the dynamic characteristics of ship navigation in AIS data with time series, using longitude, latitude, heading, speed, a ship heading and time increment as input to establish a dynamic forecasting model for navigation based on LSTM; The existing navigation sequence coding distortion and space-time data incoherence problem, an optimized Attention-LSTM neural network navigation dynamic prediction method is proposed, and the accuracy and robustness of the model are verified by simulation analysis. The results show that this method can achieve high-precision prediction of the ship’s longitude, latitude, heading, and speed.

Mobile Interface CMS Method for Vibration Characteristics Prediction of Mistuned Bladed Disk with Different Coupling Degrees

The prediction of vibration characteristics was studied in the mistuned bladed disk by the mobile interface prestressed component mode synthesis (CMS) superelement method. When the strongly, generally, and weakly coupling in the mistuned bladed disk, according to the results of the direct FEM method, the prediction accuracy of this method was verified and compared with the fixed-interface CMS method by using the relative error of dynamic frequency, vibration mode matching function, and dimensionless root mean square error of vibration amplitudes. It is pointed that for mistuned bladed disk in the strong coupling, the prediction accuracy of dynamic frequency and vibration amplitudes are higher by the mobile interface CMS method and the vibration modes are matched with the direct method. In weak coupling, the results of dynamic frequency and vibration modes predicted by the mobile interface CMS method and the fixed-interface CMS method are consistent with the direct method, but the vibration amplitudes’ prediction error of the mobile interface CMS method is lower than that of the fixed-interface CMS method. In general coupling, the mobile interface CMS method has higher dynamic frequency prediction accuracy at low order, and the two methods have comparable dynamic frequency prediction accuracy at high order. The vibration modes predicted by the two methods are matched with the direct FEM method, and the prediction accuracy of vibration amplitude by the mobile interface CMS method is better than that of the fixed-interface CMS method. The results indicate that the mobile interface CMS method could more accurately predict vibration characteristics of the mistuned bladed disk with different coupling degrees and could be an effective measurement for studying the vibration characteristics of the mistuned bladed disk system.

Integrating the temperature vegetation dryness index and meteorology parameters to dynamically predict crop yield with fixed date intervals using an integral regression model

This paper describes integrating remote sensing and meteorological data for dynamic winter wheat yield prediction at fixed date intervals with an integral regression model. In most cases, both rainfall and irrigation supply the water needs of crops. We postulated that meteorological precipitation could not fully represent the actual amount of water available to crops, whereas the temperature vegetation dryness index (TVDI) can represent actual water, which monitors soil moisture. According to this, meteorological and remote-sensing data based on sunshine hours, temperature, and TVDI formed a light–temperature–water data combination, which are the three most fundamental and critical factors related to crop growth. The new data combination was used as the input variables of the integral regression model, which could predict the crop yield at a fixed date interval of any length. The dynamic results were then compared with that of the meteorology data combination (sunshine hours, temperature, and precipitation). We selected a total of seven county-level areas in semi-humid and arid regions in China. The proposed dynamic prediction approach achieved good results at the county level from March 21 to maturity at 16-day intervals with predicted R2 > 0.94 and root mean squared error range of 15.54–121.01 kg ha−1 between predicted and actual yield with leave-one-year-out cross-validation for each prediction date. The trend of dynamic prediction accuracy had a relatively stable state. In addition, TVDI contained more crop water information, and prediction results using TVDI and meteorology were generally superior to that using only meteorology for semi-humid and arid areas, with the best results for arid regions. Thus, integrating meteorology and TVDI could effectively dynamically predict yield at county level for most agroclimatic regions, and particularly for arid regions. The work has implications for dynamic yield prediction and improving prediction accuracy and can be extended to other areas.

Model predictive current control with adaptive-adjusting timescales for PMSMs

A model predictive current control (MPCC) with adaptive-adjusting method of timescales for permanent magnet synchronous motors (PMSMs) is proposed in this paper to improve the dynamic response and prediction accuracy in transient-state, while lessening the computational burden and improving the control performance in steady-state. The timescale characteristics of different parts of MPCC, such as signal sampling, prediction calculation, control output, model error correction, are analyzed, and the algorithm architecture of MPCC with multi-timescale is proposed. The difference between reference and actual speed, and the change rate of actual speed are utilized to discriminate the transient state of speed change and load change, respectively. Adaptive-adjusting method of control period and prediction stepsize are illustrated in detail after operation condition discrimination. Experimental results of a PMSM are presented to validate the effectiveness of proposed MPCC. In addition, comparative evaluation of single-step MPCC with fixed timescale and proposed MPCC is conducted, which demonstrates the superiority of proposed control strategy.

Accuracy of machine learning-based prediction of medication adherence in clinical research

Medication non-adherence represents a significant barrier to treatment efficacy. Remote, real-time measurement of medication dosing can facilitate dynamic prediction of risk for medication non-adherence, which in-turn allows for proactive clinical intervention to optimize health outcomes. We examine the accuracy of dynamic prediction of non-adherence using data from remote real-time measurements of medication dosing. Participants across a large set of clinical trials (n = 4,182) were observed via a smartphone application that video records patients taking their prescribed medication. The patients’ primary diagnosis, demographics, and prior indication of observed adherence/non-adherence were utilized to predict (1) adherence rates ≥ 80% across the clinical trial, (2) adherence ≥ 80% for the subsequent week, and (3) adherence the subsequent day using machine learning-based classification models. Empirically observed adherence was demonstrated to be the strongest predictor of future adherence/non-adherence. Collectively, the classification models accurately predicted adherence across the trial (AUC = 0.83), the subsequent week (AUC = 0.87) and the subsequent day (AUC = 0.87). Real-time measurement of dosing can be utilized to dynamically predict medication adherence with high accuracy.