Current Step Research Articles

Feedback-linearization decoupling based coordinated control of air supply and thermal management for vehicular fuel cell system

The coordinated control of the coupled system of air supply and thermal management is one of the important ways to improve the efficiency of proton exchange membrane fuel cell (PEMFC) system. The novelty of this paper is to design a feedback-linearization decoupling based coordinated controller combining with active disturbance rejection control (ADRC) method. Aiming at strong coupling of air supply and thermal management system, we have proposed a three-dimensional particle swarm coordinated optimization algorithm to calculate the steady-state reference, then, a feedback-linearization decoupling controller is designed to achieve independent control of air flow, cathode pressure and temperature. Besides, aiming at time-varying, nonlinear interference, we have designed an ADRC based on the decoupled method to achieve better dynamic response and stability. The current step load test reveals that the overshoot and the average tracking error are much smaller under the air flow-pressure-temperature decoupling coordinated controller compared that under the proportional-integral-derivative (PID) controller and the air flow-pressure decoupling controller. The test results under different ambient temperatures indicate that the proposed coordinated strategy reduces the tracking error by over 10 %, the efficiency of fuel cell system is increased by 3.5%–4.5 %.

Mitigating Ill-posedness in Parameter Estimation under Sparse measurement for Linear time-varying systems employing Virtual Sensor Responses

Monitoring linear time-varying (LTV) systems using model-based approaches typically requires dense instrumentation and high-fidelity support models, resulting in significant computational and financial costs. Bayesian filtering-based methods adopt a joint state-parameter estimation approach for LTV system monitoring wherein states/parameters are observed as well as inferred from the measurements. Eventually, sparsity in measurement while dealing with high-fidelity models can aggravate the ill-posedness in the estimation diverging the estimation to impractical or no solutions. To enhance estimation resolution and precision, an alternative approach can be supporting estimation with virtual sensor measurements sampled from future time steps. Virtual measurements are in fact measurements sampled from a future time step relative to the time at which estimation is sought. To leverage virtual measurements, the estimation needs to be time-delayed, enabling the observation of states/parameters through measurements taken at both current and subsequent time steps, thereby alleviating the ill-posedness. This method has been explored for an LTV spring-mass-dashpot system, where the numerical investigation utilizes an interacting filtering environment to estimate states and parameters in the presence of sparse measurements. The study has shown how incorporating additional information can stabilize the estimation process, leading to improved estimates for both states and parameters.

Geometrically nonlinear bending analysis of laminated thin plates based on classical laminated plate theory and deep energy method

This paper establishes a geometrically nonlinear bending analysis framework using the deep energy method and the classical laminated plate theory (CLPT) for laminated plates. Inspired by the transfer learning technique, a load applied to a laminated plate can be divided into multiple load steps. The network parameters for the current load step, with the exception of the initial step, are initialized by inheriting values from their preceding steps. Including both von Kármán and Green-Lagrange strains, the plate strains are computed using the automatic differentiation and integrated along the thickness direction per laminate plate based on the constitutive theory. By combining the outputs of neural network, the external potential energy can be obtained, and the optimized network parameters are given by minimizing the total system potential energy of the laminated plate. In order to validate the proposed approach, several numerical examples are calculated, and the present solutions are compared with those given by the literature and the Finite Element Analysis (FEA). The results show that the proposed approach is indeed feasible, can reach high levels of precision under varying loads while offering a simplified calculation strategy.

A U-Net architecture as a surrogate model combined with a geostatistical spectral algorithm for transient groundwater flow inverse problems

Characterizing groundwater flow parameters is crucial for understanding complex aquifer systems, and inverse techniques play a fundamental role in modeling hydrogeological parameters and assessing their uncertainties. Nonetheless, the use of a forward model in these methods can be highly time-consuming, especially with an increasing number of model parameters. To address this issue, we propose a surrogate model based on a U-Net architecture that replaces the transient groundwater flow model, reducing runtime and enabling a fast quantification of uncertainties related to key parameters, including heterogeneous hydraulic conductivity, boundary conditions, specific storage, and pumping rate. The surrogate is trained using limited evaluations of the forward model to learn the physical relationship between hydraulic conductivity fields and transient hydraulic heads measured on-site. The physical principles of the studied problem, including boundary conditions, specific storage, and source terms, are also mapped and introduced as inputs to the model to enhance its understanding of the governing equation of transient groundwater flow. To speed up learning using image–image regression, the previously predicted transient hydraulic heads also serve as an input to predict the transient heads at the current time step. Once the model is trained, we use a spectral geostatistical method to solve the inverse problem, a pumping test of 12 h, using the surrogate model in place of the forward model. Our study demonstrates that the trained U-Net accurately reproduces the state variables corresponding to a specific parameter field, and in terms of computational demand, using U-Net as a surrogate model reduces the required computational time by approximately an order of magnitude for the defined problem. The proposed approach offers an efficient and accurate method for groundwater flow parameter characterization and uncertainty quantification in complex aquifer systems.

STBGRN: A Traffic Prediction Model Based on Spatiotemporal Bidirectional Gated Recurrent Units and Graph Convolutional Residual Networks

With the development of society and the advancement of urbanization, the development of intelligent transportation system has attracted much attention. Efficient and accurate traffic flow prediction is one of the core tasks in the research of intelligent transportation system. Most of the existing spatiotemporal traffic flow prediction models do not make full use of various periodic spatiotemporal dynamic characteristics of traffic flow, and it is difficult to effectively capture the complex spatiotemporal changes of traffic flow. To accurately predict the traffic flow of complex urban network, this paper proposes a traffic prediction model (STBGRN) based on spatiotemporal bidirectional gated cycle unit (ST-BiGRU) and graph convolution residual network (GCN-ResNet). The spatiotemporal information in traffic data is learned using the spatiotemporal bidirectional GRU coupled with the forward and backward information of the current time step, and the topology structure of the traffic network is captured by the graph convolution residual network, which is combined to complete the prediction task of the urban traffic network. In this paper, STBGRN was tested on two real data sets, SZ-taxi and Los-loop, and compared with other baseline methods, the RMSE index reached 3.857 and 5.461, and the MAE index reached 2.702 and 3.781, respectively. Accuracy index is 72.66% and 91.35%, R2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$${R}^{2}$$\\end{document} index is 0.858 and 0.845, var index is 0.858 and 0.849. The experimental results show that STBGRN model can obtain spatiotemporal dependence from complex traffic data, and can be used in spatiotemporal traffic data processing.

Getting slapped and kicked: the experience of judicial bullying for lawyers providing publicly funded criminal defence

Judicial bullying has received increased attention in Australia in the past decade. Traditional attitudes that dismissed judicial bullying as a rite of passage have been superseded by recognition about the negative impact it has on lawyers’ mental health and wellbeing. In interviews about wellbeing and burnout with lawyers in five Australian jurisdictions, judicial bullying was identified as having a significant role in lawyers’ mental health and careers. This research affirms current formal steps being undertaken in Victoria to address judicial bullying, but also emphasises the informal mechanisms employed by lawyers to mitigate the risk and reduce the impact of judicial bullying.

Interleukin-6 Modulates Action Potential Generation in Freshly Isolated Rat Dorsal Root Ganglia Neurons And its Stimulating Effect Is Exacerbated in Polycystic Kidney Disease

Our laboratory has demonstrated a central role for increased afferent nerve renal activity (ARNA) in the progression of cystogenesis in the PCK rat model of polycystic kidney disease (PKD). The underlying cause for the pathological increase in ARNA is unknown. Inflammatory cell infiltration and increased proinflammatory cytokine content, such as interleukin 6 (IL-6), have been reported in PKD. We hypothesized inflammatory cytokines directly increase sensory nerve activity, and this response is exacerbated in PKD. We measured action potential generation (AP) in freshly isolated dorsal root ganglion (DRG) neurons in PCK rats and non-cystic controls, and their response to IL-6. DRGs (T10-L1) were isolated from Sprague-Dawley (SD) or PCK rat (Age: 10-26 weeks) spinal cords, enzymatically digested and mechanically dissociated. Neurons were plated on coverslips, and AP were recorded within 24-hours (whole-cell patch-clamp). We tested the effects of IL-6 (0.04 μg/ml) and bradykinin (BK; 40μg/ml) on AP under 100, 200 pA, and 300 pA current clamp steps. IL-6 elicited either an excitatory response or inhibitory response in AP. In excitatory cells (EC), AP was normalized to BK peak response at +300 pA. In inhibited cells (IC), AP was normalized to baseline at +300 pA. PCK and IL-6 effects were assessed by two-way ANOVA (Bonferroni post hoc test;*p<.05); data as mean ± SEM. Proportion of ECs and ICs did not differ between SD and PCK rats. IL-6 increased AP at 300 pA from 0.48±0.28 (baseline) to 0.95±0.05 (p<.05) in SD ECs cells. BK further increased AP, showing no cell fatigue by IL-6 exposure. In PCK ECs, IL-6 increased AP at 100 pA (0.07±0.05 vs. 0.52±0.25), at 200 pA (0.47±0.27 vs. 1.39±0.81), and 300 pA (0.54±0.25 vs. 1.11±0.47). BK further increased AP at 100 pA to 0.70± 0.28 but decreased AP at 200 pA to 1.21±0.40 and at +300 pA to 1.00±0.00, indicating that neurons activity was likely plateaued by IL-6 at +200 and +300 pA. In both SD and PCK ICs, there was a main effect of IL-6 and BK treatment across all current steps (p<0.05). In SD ICs, IL-6 decreased AP at 200 pA (1.03±0.67 vs. 0.28±0.24) and 300 pA (1.00±0.00 vs. 0.28±0.24). BK had no effect at +200 pA and partially reversed the effect of IL-6 at +300 pA (AP increase to 0.53±0.27). In PCK ICs, IL-6 decreased AP at +200 pA (0.21±0.08 vs. 0.11±0.09) and +300 pA (1.00±0.00 vs. 0.42±0.24). BK increased AP at +200 pA (0.67±0.37) and +300 pA (0.71±0.33). These data partially support our hypothesis that IL-6 increases sensory nerve discharge. This effect was not observed uniformly across cells, as there were inhibitory and excitatory responses in both PCK and SD rats. We conclude that pro-inflammatory cytokine IL-6 can increase a subset of DRG neuron activity and responsiveness, an effect potentiated in PKD. This neuro-immune interaction might contribute to the underlying increase in ARNA and cystogenesis in PKD. Further experimentation is underway to test neuronal sensitivity to IL-6 and other inflammatory cytokines, as well as the posited potentiated effects in PKD. Supported in part by the PKD Foundation Research Award 1022534 (CB); NIDDK/PKD RRC Pilot and Feasibility Award U24DK126110 (CB). This is the full abstract presented at the American Physiology Summit 2024 meeting and is only available in HTML format. There are no additional versions or additional content available for this abstract. Physiology was not involved in the peer review process.

Dual Engine Boosts Organic Photoelectrochemical Transistor for Enhanced Modulation and Bioanalysis.

Organic photoelectrochemical transistor (OPECT) has shown substantial potential in the development of next-generation bioanalysis yet is limited by the either-or situation between the photoelectrode types and the channel types. Inspired by the dual-photoelectrode systems, we propose a new architecture of dual-engine OPECT for enhanced signal modulation and its biosensing application. Exemplified by incorporating the CdS/Bi2S3 photoanode and Cu2O photocathode within the gate-source circuit of Ag/AgCl-gated poly(3,4-ethylenedioxythiophene):poly(styrenesulfonate) (PEDOT:PSS) channel, the device shows enhanced modulation capability and larger transconductance (gm) against the single-photoelectrode ones. Moreover, the light irritation upon the device effectively shifts the peak value of gm to zero gate voltage without degradation and generates larger current steps that are advantageous for the sensitive bioanalysis. Based on the as-developed dual-photoelectrode OPECT, target-mediated recycling and etching reactions are designed upon the CdS/Bi2S3, which could result in dual signal amplification and realize the sensitive microRNA-155 biodetection with a linear range from 1 fM to 100 pM and a lower detection limit of 0.12 fM.

An improved constant current step-based grey wolf optimization algorithm for photovoltaic systems

In this paper, an improved constant current step based on the grey wolf optimization (CCS-GWO) algorithm for photovoltaic systems is investigated. The development of grey wolf optimization has been widely spread over photovoltaic applications. This method is one of the metaheuristic swarm optimization algorithm groups inspired by an optimum means of chasing prey by grey wolves. The proposed technique applies constant current steps to the pack of wolves (alpha, beta, and omega) by monitoring the average of the internal current step and external current step in order to target the leader alpha wolf position. Moreover, the proposed technique solves the convergence process issues, low convergence speed, and premature local optima problems of the traditional GWO algorithm. This CCS-GWO algorithm accurately tracks the maximum power point from the photovoltaic systems for load charging in different partial shading conditions (PSCs). A number of standard benchmark functions are presented with low average cost functions and their corresponding standard deviation values. The simulation results revealed that the proposed CCS-GWO approach outperforms the existing GWO and GA algorithms in terms of efficiency (98.55%) and tracking time (0.3 s).

Anomalous Current Steps in 3D Graphene Electrochemical Systems at Room Temperature.

In this work, we report a new phenomenon in electrochemical systems whereby uniform current steps of 1 mA per 0.5 × 0.5 × 0.1 cm3 (width × width × depth) of electrode volume occurred during the electrodeposition of gold and silver nanoparticles onto 3D microporous graphene on nickel layers (GF/Ni) at room temperature. The effect was exhibited only at specific applied electrical potentials. The experiments (magnetic interference, temperature dependence, and surface area dependence) were repeated, and the results were reproducible. Finally, we proposed classical electrochemical theory using the Butler-Volmer equation and quantum theory using the Landauer formalism to describe this new effect. Both theories could be used to explain the experimental results: temperature dependence, surface area dependence, blocking effects, and external magnetic field dependence. In addition, the stepwise current presented in this work facilitates the trapping and supplying of a large amount of electric charge via an inherent magnetic field in a sharp time step (∼1 s). A video clip of the recorded effect can be found at https://youtu.be/pPJh45w1sUQ.

Current Fertility Preservation Steps in Young Women Suffering from Cancer and Future Perspectives.

Childhood cancer incidence, especially in high-income countries, has led to a focus on preserving fertility in this vulnerable population. The common treatments, such as radiation and certain chemotherapeutic agents, though effective, pose a risk to fertility. For adult women, established techniques like embryo and egg freezing are standard, requiring ovarian stimulation. However, for prepubescent girls, ovarian tissue freezing has become the primary option, eliminating the need for hormonal preparation. This review describes the beginning, evolution, and current situation of the fertility preservation options for this young population. A total of 75 studies were included, covering the steps in the current fertility preservation protocols: (i) ovarian tissue extraction, (ii) the freezing method, and (iii) thawing and transplantation. Cryopreservation and the subsequent transplantation of ovarian tissue have resulted in successful fertility restoration, with over 200 recorded live births, including cases involving ovarian tissue cryopreserved from prepubescent girls. Despite promising results, challenges persist, such as follicular loss during transplantation, which is attributed to ischemic and oxidative damage. Optimizing ovarian tissue-freezing processes and exploring alternatives to transplantation, like in vitro systems for follicles to establish maturation, are essential to mitigating associated risks. Further research is required in fertility preservation techniques to enhance clinical outcomes in the future. Ovarian tissue cryopreservation appears to be a method with specific benefits, indications, and risks, which can be an important tool in terms of preserving fertility in younger women.

BDCore: Bidirectional Decoding with Co-graph Representation for Joint Entity and Relation Extraction

Relation extraction has become a crucial step for the automatic construction of Knowledge Graph (KG). Recently, researchers leverage Sequence-to-Sequence (Seq2Seq) models for Joint Entity and Relation Extraction (JERE). Nevertheless, traditional decoding methods entail the generation of the target sequence incrementally from left to right by the decoder, without the ability to revise earlier predictions when errors occur. This limitation becomes evident when decoding errors manifest prior to the current decoding step. Furthermore, the interrelations among triplets originating from the same sentence exhibit a robust correlation, which has been overlooked. In this paper, we propose Bidirectional Decoding with Co-graph representation (BDCore) to address the issues mentioned above. Specifically, we first introduce a backward decoder to decode the target sequence in a reverse order. Then, the forward decoder introduces two attention mechanisms to simultaneously considering the hidden states of the encoder and the backward decoder. Thus, the backward decoding information helps to alleviate the negative impact of the forward decoding errors. Besides, we construct a relation co-occurrence graph (Co-graph) and exploit Graph Convolutional Network (GCN) to capture the relation correlation. The extensive experiments demonstrate the benefits of the proposed bidirectional decoding and co-graph representation for relation extraction. Compared to the previous methods, our approach significantly outperforms the baselines on the NYT benchmark.

Solving a System of One-Dimensional Hyperbolic Delay Differential Equations Using the Method of Lines and Runge-Kutta Methods

In this paper, we consider a system of one-dimensional hyperbolic delay differential equations (HDDEs) and their corresponding initial conditions. HDDEs are a class of differential equations that involve a delay term, which represents the effect of past states on the present state. The delay term poses a challenge for the application of standard numerical methods, which usually require the evaluation of the differential equation at the current step. To overcome this challenge, various numerical methods and analytical techniques have been developed specifically for solving a system of first-order HDDEs. In this study, we investigate these challenges and present some analytical results, such as the maximum principle and stability conditions. Moreover, we examine the propagation of discontinuities in the solution, which provides a comprehensive framework for understanding its behavior. To solve this problem, we employ the method of lines, which is a technique that converts a partial differential equation into a system of ordinary differential equations (ODEs). We then use the Runge–Kutta method, which is a numerical scheme that solves ODEs with high accuracy and stability. We prove the stability and convergence of our method, and we show that the error of our solution is of the order O(Δt+h¯4), where Δt is the time step and h¯ is the average spatial step. We also conduct numerical experiments to validate and evaluate the performance of our method.

One-Step Forward and Backtrack: Overcoming Zig-Zagging in Loss-Aware Quantization Training

Weight quantization is an effective technique to compress deep neural networks for their deployment on edge devices with limited resources. Traditional loss-aware quantization methods commonly use the quantized gradient to replace the full-precision gradient. However, we discover that the gradient error will lead to an unexpected zig-zagging-like issue in the gradient descent learning procedures, where the gradient directions rapidly oscillate or zig-zag, and such issue seriously slows down the model convergence. Accordingly, this paper proposes a one-step forward and backtrack way for loss-aware quantization to get more accurate and stable gradient direction to defy this issue. During the gradient descent learning, a one-step forward search is designed to find the trial gradient of the next-step, which is adopted to adjust the gradient of current step towards the direction of fast convergence. After that, we backtrack the current step to update the full-precision and quantized weights through the current-step gradient and the trial gradient. A series of theoretical analysis and experiments on benchmark deep models have demonstrated the effectiveness and competitiveness of the proposed method, and our method especially outperforms others on the convergence performance.

STOCK PRICE PREDICTION OF NIFTY 50

In this project a forecasting system is presented which can predict the change in stock market value of Nifty Fifty index using the historical data of various factors influencing the stock value. Traditional stock market prediction models only factor in the trends of the historical data of the targeted stock value which is then fed to a time series algorithm to get a forecast. Since the real-world stock market is constantly being influenced by various factors, it is not possible to take all those factors into account and make the forecast, let alone using the target stock by itself. So, the objective of this project is to analyse and use the trends in the factors that majorly influence Nifty Fifty index like international stock indices and commodities that affect the economy addon Recurrent Neural Networks (RNN) which has the architecture that can retain the past information and use it in the prediction along with the observations at the current time step. This will allow the predictions to be more accurate and make it possible to make a better decision when investing in the Nifty Fifty index.

Spectrometer as a quantitative sensor for predicting the weld depth in laser welding

Spectrometers have been used for quality assessment or as auxiliary sensors in multisensor systems to record emissions at different wavelengths. This study developed a single-spectrometer sensor to quantitatively predict the laser penetration depth in Cu/Al overlap and dual-phase (DP) steel bead-on-plate welding joints. The spectrometer signals were sampled at 100 Hz, and an optical coherence tomography (OCT) sensor was used to estimate the in situ penetration depth. A one-dimensional (1D) convolution neural network (CNN) model was developed using the current step of the spectrometer signal. Furthermore, a two-dimensional (2D) CNN model was developed using 10 recent steps of the spectrometer signals. Consequently, 1732 and 3000 data points for DP steel and Al/Cu overlap welding were used for the penetration-depth modeling. Through model training, mean absolute errors of 22.67 and 52.96 µm were achieved for the Al/Cu overlap and DP steel welding, respectively, corresponding to approximately 1.5 % of the substrate thickness. The 2D-CNN models outperformed the 1D-CNN models for both material combinations. However, a slight delay was observed in the response of the models when confronted with abrupt output changes. This study confirmed that spectroscopy can serve as the sole quantitative sensing method for laser welding monitoring and control.

Learning to walk with logical embedding for knowledge reasoning

The path-based model has remarkably succeeded in the knowledge graph (KG) multi-hop reasoning task. It employs all available resources to accomplish various complex path reasoning tasks and continuously explores new graph paths. However, existing multi-hop reasoning methods rely heavily on the high reward, which is fed back to the model when the agent searches for the target. In contrast, most previous methods focused on efficiently querying the correct answer while disregarding the logic and validity of the entire reasoning chain. It contradicts the intention of various complex reasoning tasks in real-world scenarios. Unreasonable paths will cause the selection to deviate from normal cognition, resulting in invalid path resource information. Therefore, we must be able to complete specific tasks via these reliable paths. In order to address these issues, we proposed a Reinforcement Learning-Based Knowledge Reasoning Model with Logical Embedding (RKLE) to enhance the interpretability of the reasoning chain. RKLE assembles the logical structure (query structure) with additional nodes in the current step and develops the logical reward shaping to assist the agent in selecting a more reasonable path. Finally, experimental results on several benchmarks demonstrate that our approach can search for the correct answers more efficiently than existing path-based methods and that the corresponding reasoning chain is interpretable.

Attentional bias for hands: Cascade dual‐decoder transformer for sign language production

AbstractSign Language Production (SLP) refers to the task of translating textural forms of spoken language into corresponding sign language expressions. Sign languages convey meaning by means of multiple asynchronous articulators, including manual and non‐manual information channels. Recent deep learning‐based SLP models directly generate the full‐articulatory sign sequence from the text input in an end‐to‐end manner. However, these models largely down weight the importance of subtle differences in the manual articulation due to the effect of regression to the mean. To explore these neglected aspects, an efficient cascade dual‐decoder Transformer (CasDual‐Transformer) for SLP is proposed to learn, successively, two mappings SLPhand: Text → Hand pose and SLPsign: Text → Sign pose, utilising an attention‐based alignment module that fuses the hand and sign features from previous time steps to predict more expressive sign pose at the current time step. In addition, to provide more efficacious guidance, a novel spatio‐temporal loss to penalise shape dissimilarity and temporal distortions of produced sequences is introduced. Experimental studies are performed on two benchmark sign language datasets from distinct cultures to verify the performance of the proposed model. Both quantitative and qualitative results show that the authors’ model demonstrates competitive performance compared to state‐of‐the‐art models, and in some cases, achieves considerable improvements over them.

Dynamic Co-Embedding Model for Temporal Attributed Networks.

In this article, we study the problem of embedding temporal attributed networks, with the goal of which is to learn dynamic low-dimensional representations over time for temporal attributed networks. Existing temporal network embedding methods only learn the representations for nodes, which are unable to capture the dynamic affinities between nodes and attributes. Moreover, existing co-embedding methods that learn the static embeddings of both nodes and attributes cannot be naturally utilized to obtain their dynamic embeddings for temporal attributed networks. To address these issues, we propose the dynamic co-embedding model for temporal attributed networks (DCTANs) based on the dynamic stochastic state-space framework. Our model captures the dynamics of a temporal attributed network by modeling the abstract belief states representing the condition of the nodes and attributes of current time step, and predicting the transitions between temporal abstract states of two successive time steps. Our model is able to learn embeddings for both nodes and attributes based on their belief states at each time step of the temporal attributed network, while the state transition tendency for predicting the future network can be tracked and the affinities between nodes and attributes can be preserved. Experimental results on real-world networks demonstrate that our model achieves substantial performance gains in several static and dynamic graph mining applications compared with the state-of-the-art static and dynamic models.

State of Health Estimation for Lithium-Ion Batteries Under Arbitrary Usage Using Data-Driven Multimodel Fusion

Accurately estimating the state of health (SoH) of batteries is indispensable for the safety, reliability, and optimal energy and power management of electric vehicles. However, from a data-driven perspective, complications, such as dynamic vehicle operating conditions, stochastic user behaviors, and cell-to-cell variations, make the estimation task challenging. This work develops a data-driven multi-model fusion method for SoH estimation under arbitrary usage profiles. All possible operating conditions are categorized into six scenarios. For each scenario, an appropriate feature set is extracted to indicate the SoH. Based on the obtained features, four machine learning algorithms are applied individually to train SoH estimation models using time-series data. In addition to the estimates at the current time step, a histogram data-based and online adaptive model is taken from previous work for predicting the next-step SoH. Then, a Kalman filter is applied to systematically fuse the results of all the estimation and prediction models. Experimental data collected from different types of batteries operated under diverse profiles verify the effectiveness and practicability of the developed method, as well as its superiority over individual models.