Feedforward Component Research Articles

Imputation Strategies in Time Series Based on Language Models

Incomplete time series present a significant challenge for downstream analysis. In the field of time series, Large Language Models are already being used for prediction, classification, and, in rare cases, imputation. This study thoroughly examines the imputation of time series using Large Language Models. Within a defined experimental setup, current state-of-the-art time series imputation methods are compared with the performance of Large Language Models. Parameter-efficient fine-tuning methods are applied to adapt the Large Language Models to the imputation task. The results indicate that the models are suitable for time series imputation. The performance of these models depends on the number of parameters and the type of pre-training. Small specialized models, such as BERT, compete with models like Llama2 and outperform them on selected datasets. Furthermore, it becomes clear that the attention and feedforward network components of Large Language Models are particularly well-suited for adaptation to imputation, and parameter-efficient methods are also performance-enhancing.

Feedback–feedforward signal control with exogenous demand estimation in congested urban road networks

To cope with uncertain traffic patterns and traffic models, traffic-responsive signal control strategies in the literature are designed to be robust to these uncertainties. These robust strategies still require sensing infrastructure to implement traffic-responsiveness. In this paper, we take a novel perspective and show that it is possible to use the already necessary sensing infrastructure to estimate the uncertain quantities in real time. Specifically, resorting to the store-and-forward model, we design a novel network-wide traffic-responsive strategy that estimates the occupancy and exogenous demand in each link, i.e., entering (exiting) vehicle flows at the origins (destinations) of the network or within links, in real time. Borrowing from optimal control theory, we design an optimal linear quadratic control scheme, consisting of a linear feedback term, of the occupancy of the road links, and a feedforward component, which accounts for the varying exogenous vehicle load on the network. Thereby, the resulting control scheme is a simple feedback–feedforward controller, which is fed with occupancy and exogenous demand estimates, and is suitable for real-time implementation. Numerical simulations for the urban traffic network of Chania, Greece, show that, for realistic surges in the exogenous demand, the proposed solution significantly outperforms tried-and-tested solutions that ignore the exogenous demand.

Model-based hybrid control of combined active–passive vertical zero-gravity system

Ground-based full-scale physical experiments on spacecraft have become an essential part of ground simulations. However, achieving a vertical translational degree of freedom for zero-gravity simulation has become a key technological bottleneck for six-degree-of-freedom ground simulations. The combined active–passive vertical zero-gravity system discussed in this paper can achieve zero-gravity simulation in the vertical direction; however, it exhibits deficiencies such as inaccurate modeling, low control precision, and significant control delays. Addressing the issue of inaccurate modeling, this study uses interpretable ODE neural networks to identify the dynamic model of the system. For resolving the problems of poor control precision and significant delays, this study proposes a hybrid control algorithm based on the ODE dynamic model. The controller includes nominal model feedforward, PID feedback, and learning-based feedforward components. Finally, comparative experiments are designed to validate the proposed model-based hybrid control algorithm. Compared to PID-Feedforward controller, the hybrid controller exhibits a higher response speed, while compared to model-based feedforward controller, it exhibits a better control accuracy. Therefore, the proposed hybrid control algorithm based on the ODE dynamic model effectively enhances the control response speed and improves the control precision of the system.

Structural Learning Benefits in a Visuomotor Adaptation Task Generalize to a Contralateral Effector

Structural learning is characterized by facilitated adaptation following training on a set of sensory perturbations all belonging to the same structure (e.g., ‘visuomotor rotations’). This generalization of learning is a core feature of the motor system and is often studied in the context of interlimb transfer. However, such transfer has only been demonstrated when participants learn to counter a specific perturbation in the sensory feedback of their movements; we determined whether structural learning in one limb generalized to the contralateral limb. We trained 13 participants to counter random visual feedback rotations between +/-90 degrees with the right hand and subsequently tested the left hand on a fixed rotation. The structural training group showed faster adaptation in the left hand in both feedforward and feedback components of reaching compared to 13 participants who trained with veridical reaching, with lower initial reaching error, and straighter, faster, and smoother movements than in the control group. The transfer was ephemeral – benefits were confined to roughly the first 20 trials. The results demonstrate that the motor system can extract invariant properties of seemingly random environments in one limb, and that this information can be accessed by the contralateral limb.

Development of reciprocal connections between the dorsal lateral geniculate nucleus and the thalamic reticular nucleus

The thalamic reticular nucleus (TRN) serves as an important node between the thalamus and neocortex, regulating thalamocortical rhythms and sensory processing in a state dependent manner. Disruptions in TRN circuitry also figures prominently in several neurodevelopmental disorders including epilepsy, autism, and attentional defects. An understanding of how and when connections between TRN and 1st order thalamic nuclei, such as the dorsal lateral geniculate nucleus (dLGN), develop is lacking. We used the mouse visual thalamus as a model system to study the organization, pattern of innervation and functional responses between TRN and the dLGN. Genetically modified mouse lines were used to visualize and target the feedforward and feedback components of these intra-thalamic circuits and to understand how peripheral input from the retina impacts their development.Retrograde tracing of thalamocortical (TC) afferents through TRN revealed that the modality-specific organization seen in the adult, is present at perinatal ages and seems impervious to the loss of peripheral input. To examine the formation and functional maturation of intrathalamic circuits between the visual sector of TRN and dLGN, we examined when projections from each nuclei arrive, and used an acute thalamic slice preparation along with optogenetic stimulation to assess the maturation of functional synaptic responses. Although thalamocortical projections passed through TRN at birth, feedforward axon collaterals determined by vGluT2 labeling, emerged during the second postnatal week, increasing in density through the third week. Optogenetic stimulation of TC axon collaterals in TRN showed infrequent, weak excitatory responses near the end of week 1. During weeks 2–4, responses became more prevalent, grew larger in amplitude and exhibited synaptic depression during repetitive stimulation. Feedback projections from visual TRN to dLGN began to innervate dLGN as early as postnatal day 2 with weak inhibitory responses emerging during week 1. During week 2–4, inhibitory responses continued to grow larger, showing synaptic depression during repetitive stimulation. During this time TRN inhibition started to suppress TC spiking, having its greatest impact by week 4–6. Using a mutant mouse that lacks retinofugal projections revealed that the absence of retinal input led to an acceleration of TRN innervation of dLGN but had little impact on the development of feedforward projections from dLGN to TRN. Together, these experiments reveal how and when intrathalamic connections emerge during early postnatal ages and provide foundational knowledge to understand the development of thalamocortical network dynamics as well as neurodevelopmental diseases that involve TRN circuitry.

Glycolytic lactate in diabetic kidney disease.

Lactate elevation is a well-characterized biomarker of mitochondrial dysfunction, but its role in diabetic kidney disease (DKD) is not well defined. Urine lactate was measured in patients with type 2 diabetes (T2D) in 3 cohorts (HUNT3, SMART2D, CRIC). Urine and plasma lactate were measured during euglycemic and hyperglycemic clamps in participants with type 1 diabetes (T1D). Patients in the HUNT3 cohort with DKD had elevated urine lactate levels compared with age- and sex-matched controls. In patients in the SMART2D and CRIC cohorts, the third tertile of urine lactate/creatinine was associated with more rapid estimated glomerular filtration rate decline, relative to first tertile. Patients with T1D demonstrated a strong association between glucose and lactate in both plasma and urine. Glucose-stimulated lactate likely derives in part from proximal tubular cells, since lactate production was attenuated with sodium-glucose cotransporter-2 (SGLT2) inhibition in kidney sections and in SGLT2-deficient mice. Several glycolytic genes were elevated in human diabetic proximal tubules. Lactate levels above 2.5 mM potently inhibited mitochondrial oxidative phosphorylation in human proximal tubule (HK2) cells. We conclude that increased lactate production under diabetic conditions can contribute to mitochondrial dysfunction and become a feed-forward component to DKD pathogenesis.

STVANet: A spatio-temporal visual attention framework with large kernel attention mechanism for citywide traffic dynamics prediction

Enhancing the efficiency and safety of the Intelligent Transportation System requires effective modeling and prediction of citywide traffic dynamics. Most studies employ convolutional neural networks (CNNs) with a 3D convolutional structure or spatio-temporal models with self-attention mechanisms to capture the spatio-temporal information of traffic distribution. Although 3D CNNs excel at capturing local contextual information, they are computationally complex due to the large number of parameters and cannot capture long-range dependence. By contrast, although self-attention mechanisms originally designed to address challenges in natural language processing can capture long-range dependence, their application to 2D image structures requires breaking down the inherent 2D context into a 1D sequence, increasing the computational complexity and neglecting the adaptability between local contextual information and channels. Accordingly, we propose a spatio-temporal visual attention neural network (STVANet), a novel spatio-temporal visual attention 2D CNN, which integrates a unique visual attention module with a large kernel attention (LKA) mechanism, a squeeze-and-excitation (SE) mechanism and a feedforward component to capture long-range dependence and channel information in urban traffic data while preserving the 2D image structure. LKA-based spatio-temporal attention networks extract spatial and temporal features from weekly, daily, and recent hourly periods, and aggregate them with weighted consideration of external features to make predictions. Evaluation of real-world datasets demonstrates STVANet’s superiority over baseline models, showcasing its potential in citywide traffic prediction.

A Transformer and LSTM-Based Approach for Blind Well Lithology Prediction

Petrographic prediction is crucial in identifying target areas and understanding reservoir lithology in oil and gas exploration. Traditional logging methods often rely on manual interpretation and experiential judgment, which can introduce subjectivity and constraints due to data quality and geological variability. To enhance the precision and efficacy of lithology prediction, this study employed a Savitzky–Golay filter with a symmetric window for anomaly data processing, coupled with a residual temporal convolutional network (ResTCN) model tasked with completing missing logging data segments. A comparative analysis against the support vector regression and random forest regression model revealed that the ResTCN achieves the smallest MAE, at 0.030, and the highest coefficient of determination, at 0.716, which are indicative of its proximity to the ground truth. These methodologies significantly enhance the quality of the training data. Subsequently, a Transformer–long short-term memory (T-LS) model was applied to identify and classify the lithology of unexplored wells. The input layer of the Transformer model follows an embedding-like principle for data preprocessing, while the encoding block encompasses multi-head attention, Add & Norm, and feedforward components, integrating the multi-head attention mechanism. The output layer interfaces with the LSTM layer through dropout. A performance evaluation of the T-LS model against established rocky prediction techniques such as logistic regression, k-nearest neighbor, and random forest demonstrated its superior identification and classification capabilities. Specifically, the T-LS model achieved a precision of 0.88 and a recall of 0.89 across nine distinct lithology features. A Shapley analysis of the T-LS model underscored the utility of amalgamating multiple logging data sources for lithology classification predictions. This advancement partially addresses the challenges associated with imprecise predictions and limited generalization abilities inherent in traditional machine learning and deep learning models applied to lithology identification, and it also helps to optimize oil and gas exploration and development strategies and improve the efficiency of resource extraction.

USV Trajectory Tracking Control Based on Receding Horizon Reinforcement Learning.

We present a novel approach for achieving high-precision trajectory tracking control in an unmanned surface vehicle (USV) through utilization of receding horizon reinforcement learning (RHRL). The control architecture for the USV involves a composite of feedforward and feedback components. The feedforward control component is derived directly from the curvature of the reference path and the dynamic model. Feedback control is acquired through application of the RHRL algorithm, effectively addressing the problem of achieving optimal tracking control. The methodology introduced in this paper synergizes with the rolling time domain optimization mechanism, converting the perpetual time domain optimal control predicament into a succession of finite time domain control problems amenable to resolution. In contrast to Lyapunov model predictive control (LMPC) and sliding mode control (SMC), our proposed method employs the RHRL controller, which yields an explicit state feedback control law. This characteristic endows the controller with the dual capabilities of direct offline and online learning deployment. Within each prediction time domain, we employ a time-independent executive-evaluator network structure to glean insights into the optimal value function and control strategy. Furthermore, we substantiate the convergence of the RHRL algorithm in each prediction time domain through rigorous theoretical proof, with concurrent analysis to verify the stability of the closed-loop system. To conclude, USV trajectory control tests are carried out within a simulated environment.

An Industrial Control System for Cement Sulfates Content Using a Feedforward and Feedback Mechanism

This study examines the design and long-term implementation of a feedforward and feedback (FF–FB) mechanism in a control system for cement sulfates applied to all types of cement produced in two mills at a production facility. We compared the results with those of a previous controller (SC) that operated in the same unit. The Shewhart charts of the annual SO3 mean values and the nonparametric Mann–Whitney test demonstrate that, for the FF–FB controller, the mean values more effectively approach the SO3 target than the older controller in two out of the three cement types. The s-charts for the annual standard deviation of all cement types and mills indicate that the ratio of the central lines of FF–FB to SC ranges from 0.39 to 0.59, representing a significant improvement. The application of the error propagation technique validates and explains these improvements. The effectiveness of the installed system is due to two main factors. The feedforward (FF) component tracks the set point of SO3 when the mill begins grinding a different type of cement, while the feedback (FB) component effectively attenuates the fluctuations in the sulfates of the raw materials.

Ripple force compensation based on ILC–ASMC for linear servo system

The ripple force, which displays intricate dynamic characteristics and is fundamentally nonlinear in linear servo systems, poses challenges in modeling and compensation. To achieve superior suppression of the ripple force, this paper proposes a new scheme called the iterative learning control (ILC)–adaptive sliding-mode control (ASMC), which combines ASMC and ILC. Initially, the servo system model is established, followed by an analysis of the ripple force mechanism and obtaining a simplified model. Then, an adaptive sliding-mode structure with a feed-forward component is proposed to enhance the servo dynamic response based on the design of a sliding-mode surface. In addition, a new ripple force compensator based on the ILC is designed, allowing for iterative learning from the previous period to accurately estimate the ripple force online. It is worth noting that the process is divided into the 0th iteration and the ith ( i ≥ 1) iteration to avoid the selection for initial values. The 0th iteration ensures initial system stability and provides the initial iteration values, while subsequent iterations approximate the real ripple force through iterative learning. Importantly, the controller and the compensator are combined through the same sliding-mode surface, obviously simplifying the system structure. Finally, experimental results validate the effectiveness of the proposed scheme.

Improving the Feedforward Component for Recent Variants of Predictive Functional Control

A recent study demonstrated that the use of feedforward information with conventional Predictive Functional Control (PFC) leads to unexpected inconsistencies, with subsequent negative impacts on tuning and behaviour. A proposal was made to define the coincident point differently and shown to reduce the lag in the closed-loop PFC responses and applied to some systems with benign dynamics. Other recent work has looked at parameterisations of the future input to deal with challenging open-loop dynamics and significantly extended the range of problems for which PFC can be effective. This paper combines the two concepts, and thus proposes an algorithm that has both more effective and simple tuning than original PFC as well as being applicable to a range of challenging dynamics.

Achieving Sales Forecasting with Higher Accuracy and Efficiency: A New Model Based on Modified Transformer

With the exponential expansion of e-commerce, an immense volume of historical sales data has been generated and amassed. This influx of data has created an opportunity for more accurate sales forecasting. While various sales forecasting methods and models have been applied in practice, existing ones often struggle to fully harness sales data and manage significant fluctuations. As a result, they frequently fail to make accurate predictions, falling short of meeting enterprise needs. Therefore, it is imperative to explore new models to enhance the accuracy and efficiency of sales forecasting. In this paper, we introduce a model tailored for sales forecasting based on a Transformer with encoder–decoder architecture and multi-head attention mechanisms. We have made specific modifications to the standard Transformer model, such as removing the Softmax layer in the last layer and adapting input embedding, position encoding, and feedforward network components to align with the unique characteristics of sales forecast data and the specific requirements of sales forecasting. The multi-head attention mechanism in our proposed model can directly compute the dot product results in a single step, addressing long-term time-dependent computation challenges while maintaining lower time complexity and greater interpretability. This enhancement significantly contributes to improving the model’s accuracy and efficiency. Furthermore, we provide a comprehensive formula representation of the model for the first time, facilitating better understanding and implementation. We conducted experiments using sales datasets that incorporate various factors influencing sales forecasts, such as seasons, holidays, and promotions. The results demonstrate that our proposed model significantly outperforms seven selected benchmark methods, reducing RMSLE, RMSWLE, NWRMSLE, and RMALE by approximately 48.2%, 48.5%, 45.2, and 63.0%, respectively. Additionally, ablation experiments on the multi-head attention and the number of encoder–decoders validate the rationality of our chosen model parameters.

Effects of model fidelity and uncertainty on a model-based attitude controller for satellites with flexible appendages

This paper investigates the effects of model fidelity and parameter uncertainty on the performance of a hybrid model-based feedback-feedforward control scheme for attitude tracking of a satellite with flexible appendages. The feedforward component is an inverse model-based term produced through a computational approach known as inverse simulation (InvSim), which works by iteratively solving a discretised reference trajectory. The hybrid controller’s feedback is proportional-derivative (PD) based, using body attitude and rate feedback to provide stability and robustness. Furthermore, to ensure that the flexible modes do not trigger instability, the PD control gains are tuned to give a closed-loop response that is significantly slower than the flexible modes. Additionally, excitation of the flexible modes is reduced by minimising jerk through polynomial rest-to-rest manoeuvres, following the shortest quaternion path using spherical–linear-interpolation (SLERP). The effects of the appendage flexing on attitude tracking are then compensated through the feedforward element of the hybrid controller, with performance being compared to a traditional PD tracking law. The effect of the model fidelity on the performance of the hybrid controller is investigated through the use of both rigid body and multiple-fidelity finite-element mathematical models. Additionally, the effect of uncertainties in the model parameters is investigated to determine the accuracy of the model required to obtain significant improvement in attitude tracking. It is found that in the absence of any model parameter uncertainty, the hybrid controller outperforms the PD tracking control law by at least one order of magnitude when the finite-element model is used. Increasing the number of finite elements was found to provide no significant improvement in performance, with one element being sufficient and favourable with its lower computational overhead. It was also found that to ensure good performance compared to the PD tracking controller, the uncertainty in the inertia tensor should be <1%. Similarly, uncertainty in the first flexible modal frequency should be <0.5 rad/s.

Event-Triggered Control with Intermittent Communications over Erasure Channels for Leader–Follower Problems with the Combined-Slip Effect

In this article, we investigate the vehicle path-following problem for a vehicle-to-vehicle (V2V)–enabled leader–follower scenario and propose an integrated control policy for the following vehicle to accurately follow the leader’s path. We propose a control strategy for the follower vehicle to maintain a velocity-dependent distance relative to the leader vehicle while stabilizing its longitudinal and lateral dynamics considering the combined-slip effect and tire force saturation. In light of reducing wireless communication errors and efficient usage of battery power and resources, we propose an intermittent V2V communication in which transmissions are scheduled based on an event-triggered law. An event is triggered and a transmission is scheduled in subsequent sample time if some of the well-defined path-following error functions (relative distance error and lateral error) exceed given tolerance bounds. Considering that the V2V communication channel might be erroneous or a transmission fails due to, e.g., vehicles’ distance or low battery power, we consider data loss in the V2V channel. Our proposed control law consists of two components: a receding horizon feedback controller with state constraints based on a safe operation envelop and a feedforward controller that generates complementary control inputs when the leader’s states are successfully communicated to the follower. To mitigate the effects of data loss on the follower’s path-following performance, we design a remote estimator for the follower to predict the leader’s state using its on-board sensor equipment when an event is triggered but the corresponding state information is not received by the follower due to a packet loss. Incorporating this estimator allows the follower to apply cautionary control inputs knowing that the path-following error had exceeded a tolerance bound. We show that while the feedback controller stabilizes the follower’s dynamics, the feedforward component improves the safety margins and reduces the path-following errors even in the presence of data loss. High-fidelity simulations are performed using CarSim to validate the effectiveness of our proposed control architecture specifically in harsh maneuvers and high-slip scenarios on various road surface conditions.

Wet clutch pressure hysteresis compensation control under variable oil temperatures for electro-hydraulic actuators

The hysteresis characteristics and oil temperatures of wet clutch electro-hydraulic actuators have a significant impact on clutch pressure tracking performance, thereby influencing the launch, shift, and mode transition performance of both fuel and hybrid electric vehicles. In this study, we introduce a novel compensator for wet clutch pressure hysteresis that effectively adapts to variations in oil temperature. Experimental investigations were conducted to examine the hysteresis characteristic of the electro-hydraulic actuator and its susceptibility to changes in oil temperature. Subsequently, a temperature-dependent modified generalized Prandtl–Ishlinskii (TDMGPI) hysteresis model that accounts for the influence of oil temperature was established. To mitigate the effects of oil temperature fluctuations on the controller, we developed a fuzzy proportional-integral-derivative (FPID) controller, which dynamically adjusts controller gain. Furthermore, a feedforward-feedback controller for clutch pressure was established, incorporating the inverse solution of the TDMGPI model as the feedforward component and the FPID controller as the feedback element. The experiments were conducted at various oil temperatures, revealing that the proposed controller significantly enhances response speed and pressure tracking accuracy, maintaining good control performance across the tested range of oil temperatures.

Improved genomic prediction using machine learning with Variational Bayesian sparsity

BackgroundGenomic prediction has become a powerful modelling tool for assessing line performance in plant and livestock breeding programmes. Among the genomic prediction modelling approaches, linear based models have proven to provide accurate predictions even when the number of genetic markers exceeds the number of data samples. However, breeding programmes are now compiling data from large numbers of lines and test environments for analyses, rendering these approaches computationally prohibitive. Machine learning (ML) now offers a solution to this problem through the construction of fully connected deep learning architectures and high parallelisation of the predictive task. However, the fully connected nature of these architectures immediately generates an over-parameterisation of the network that needs addressing for efficient and accurate predictions.ResultsIn this research we explore the use of an ML architecture governed by variational Bayesian sparsity in its initial layers that we have called VBS-ML. The use of VBS-ML provides a mechanism for feature selection of important markers linked to the trait, immediately reducing the network over-parameterisation. Selected markers then propagate to the remaining fully connected feed-forward components of the ML network to form the final genomic prediction. We illustrated the approach with four large Australian wheat breeding data sets that range from 2665 lines to 10375 lines genotyped across a large set of markers. For all data sets, the use of the VBS-ML architecture improved genomic prediction accuracy over legacy linear based modelling approaches.ConclusionsAn ML architecture governed under a variational Bayesian paradigm was shown to improve genomic prediction accuracy over legacy modelling approaches. This VBS-ML approach can be used to dramatically decrease the parameter burden on the network and provide a computationally feasible approach for improving genomic prediction conducted with large breeding population numbers and genetic markers.

Integrated thrust ripple identification and compensation for linear servo system using an MP algorithm

Generally there are two methods for the thrust ripple modeling: one is to identify the eigenvalues and the model parameters of thrust ripple sequentially under rigorous operating conditions; another is to treat thrust ripple as a general and inaccurate disturbance. To get rid of the constraints and further enhance the tracking accuracy of the linear servo system, a novel integrated identification and compensation scheme using an improved matching pursuit algorithm is proposed in this paper. First, the dynamics of the linear motor is formulated, and the cause of thrust ripple is analyzed. Then, a time–frequency atomic library of thrust ripple is constructed with the consideration of the frequency characteristics of position command, and meanwhile, the two optimal atoms are obtained successively by maximizing the inner product for the thrust ripple reconstruction. Based on the above, the eigenvalues and model parameters of thrust ripple can be simultaneously identified in an online manner. Finally, the feedforward compensation component is devised by the identification results to suppress thrust ripple. In order to verify the superiority of the proposed scheme, simulations and experiments are conducted compared with the conventional methods.

Field-Oriented Predictive Control Structure for Synchronous Reluctance Motors

This paper presents a cascade predictive control structure based on field-oriented control (FOC) in the dq rotor reference frame for the synchronous reluctance machine (SynRM). The constant d-axis current control strategy was used, and thus, the electromagnetic torque was directly controlled by the q-axis current. Because the model of the two axes of currents from the inner loop is a coupled non-linear multivariable one, to control in a non-interaction and linear way the two currents, their decoupling was achieved through feedforward components. Following the decoupling, two independent monovariable linear systems resulted for the two current dynamics that were controlled using model predictive control (MPC) algorithms, considering their ability to automatically handle the state bounds. The most important bounds for SynRM are the limits imposed on currents and voltages, which in the dq plane correspond to a circular limit. To avoid computational effort, linear limitations were adopted through polygonal approximations, resulting in rectangular regions in the dq plane. For the outer loop that controls the angular speed with a constrained MPC algorithm, the q-axis current closed-loop dynamics and the torque linear equation were considered. To evaluate the performance of the proposed cascade predictive control structure, a simulation study using MPC controllers versus PI ones was conducted.

Online Learning-Informed Feedforward-Feedback Controller Synthesis for Path Tracking of Autonomous Vehicles

High-performance path tracking is a key technology for autonomous vehicles. Feedforward-feedback control architectures are suitable for accurate path tracking with adequate margins of stability. For system modelling in the feedforward component, the learning-based method has been proven to be a promising approach owing to its model-free framework. However, the offline-learned data model trained with collection data is confined by its feature space, resulting in insufficient generalization. As a solution, in this study, we introduce an online learning network – the recurrent high-order neural network (RHONN) – to characterize vehicle behaviors. The RHONN is used to feature vehicle behaviors in a timely manner with a high fidelity and flexible form. The equilibrium at the preview point on the desired path is found based on the online-identified RHONN model, and its induced steering angle is taken as the feedforward command. For the feedback steering controller, the preview point position-based control law incorporating the steady vehicle sideslip angle is adopted to enhance the stability performance. Finally, in the CarSim/Simulink environment, the performance of the designed RHONN-informed feedforward and feedback controller is validated in two typical scenarios – double-lane change and single-turn. The validation results reveal that the proposed approach offers better tracking accuracy in linear and nonlinear regions than other techniques. More notably, the average execution time (3.55 ms) is less than the sampling frequency of the controller (50 ms), which further confirms the applicability and efficiency of the proposed approach.