Soft sensing technique has been extensively used to predict key quality variables in industrial systems. However, due to the difficulty of quality variable acquisition, only limited labeled data samples are available, and a large number of unlabeled ones are discarded. This raises a big challenge to build a high-quality soft sensor model. In order to furthest exploit information contained in both the labeled and unlabeled data, this paper proposes a principal component-based semi-supervised extreme learning machine (referred to as PCSELM) model. Through this model, extracting latent features and learning nonlinear input-output relationship can be simultaneously performed. In this way, unlabeled samples are utilized efficiently for feature representation and model accuracy improvement. Moreover, mixed regularizations are employed to work in conjunction with the PCSELM to obtain high generality and flexibility. We also derive an efficient parameter learning algorithm with theoretically guaranteed convergence. Comprehensive experiments are conducted via an industrial process. Comparison results illustrate that the proposed PCSELM outperforms other representative semi-supervised algorithms. Note to Practitioners —Industrial processes in general incorporate unlabeled samples which are ubiquitous in real world applications. The focus of this paper is to develop a semi-supervised soft sensor model (PCSELM) that is capable to learn the nonlinear features and regression relationship efficiently with both the labeled and unlabeled samples. The proposed model can automatically implement the feature representation and the input-output relationship description. In addition, we introduce mixed norms for the model objective function to improve the final prediction performance and generalization. A feasible model optimization technique with proved convergence is also derived. Experimental results based on a real industrial dataset manifest that PCSELM achieves better prediction accuracy than its peers.

Great efforts have been devoted to the intelligent control of autonomous systems. Yet, most of existing methods fail to effectively handle the uncertainty of their environment and models. Uncertain locations of dynamic obstacles pose a major challenge for their optimal control and safety, while their linearization or simplified system models reduce their actual performance. To address them, this paper presents a new model predictive control framework with finite samples and a Gaussian model, resulting in a chance-constrained program. Its nominal model is combined with a Gaussian process. Its residual model uncertainty is learned. The resulting method addresses an efficiently solvable approximate formulation of a stochastic optimal control problem by using approximations for efficient computation. There is no perfect distribution knowledge of a dynamic obstacle's location uncertainty. Only finite samples from sensors or past data are available for moment estimation. We use the uncertainty propagation of a system's state and obstacles' locations to derive a general collision avoidance condition under tight concentration bounds on the error of the estimated moments. Thus, this condition is suitable for different obstacles, e.g., bounding box and ellipsoid obstacles. We provide proved guarantees on the satisfaction of the chance-constraints corresponding to the nominal yet unknown moments. Simulation examples of a vehicle's control are used to show that the proposed method can well realize autonomous control and obstacle avoidance of a vehicle, when it operates in an uncertain environment with moving obstacles. It outperforms the existing moment methods in both performance and computational time.

Autonomous driving is producing various novel achievements in integrating different control technologies with machine learning to control intelligent vehicles, i.e., learning-based control. Yet, few work formally consider the confidence of learning-based models and their residual model uncertainties. As a result, existing methods lack safety guarantees in real-world applications. To address this issue, this work proposes a Linear-Bayesian-Regression-based Model Predictive Control (MPC) method for vehicles with unknown dynamics and subject to possibly unbounded disturbances. Model parameters can be obtained by using posterior calculations via prior estimations and systems’ historical data. The expected value of resulting posterior distribution is chosen as linear model's parameters for controller design. Furthermore, the corresponding posterior distribution can be used to derive an upper bound for the variance of residual model uncertainties. In analogy to the concept of Tube-MPC, this work introduces a dynamic feedback gain which is combined with the standard nominal MPC method. It derives provable security guarantees with respect to state and input chance constraints for complex unknown dynamical systems in the form of probabilities. A convex-reformulation-based optimization method is proposed to enable efficient computation of control laws and ensure the recursive feasibility. Simulations of a Vander Pol's oscillator's control and a vehicle's tracking control illustrate the design procedure and effectiveness of the proposed method.

A crane system often works in a complex environment. It is difficult to model or learn its true dynamics by traditional system identification approaches. If a dynamics model is created by minimizing its prediction error, its use tends to introduce inaccuracies and thus lead to suboptimal performance. Is it possible to learn the dynamics model of a crane that can achieve the best performance, instead of learning its true dynamics? This work answers the question by presenting a performance-driven model predictive control (P-MPC) algorithm for a two-dimensional underactuated bridge crane. In the proposed dual-layer control architecture, an inner-loop controller uses a proportional–integral–derivative controller to achieve anti-sway rapidly. An outer-loop controller uses MPC to ensure accurate trolley positioning under control constraints. Compared with classical MPC, this work proposes a data-driven method for plant modeling and controller parameter updating. By considering the control target at the learning stage, the method can avoid adjusting the controller to deal with uncertainty. We use Bayesian optimization in an active learning framework where a locally linear dynamics model is learned with the intent of maximizing control performance and then used in conjunction with optimal control schemes to efficiently design a controller for a given task. The model is updated directly based on the performance observed in experiments on the physical system in an iterative manner till a desired performance is achieved. The controller parameters and prediction models of the best closed-loop performance can be found through continuous experiments and iterative optimization. Simulation and experiment results show that we can explicitly find the dynamics model that produces the best performance for an actual system, and the method can quickly suppress swing and realize accurate trolley positioning. The results verified its effectiveness, feasibility, and superior performance on comparing it with state-of-the-art methods.

Maximizing the performance of a system without reference over an infinite horizon is a challenging problem for iterative control tasks. This paper introduces a Value Approximator-based Learning Model Predictive Control (VALMPC) framework that aims to enhance the system's performance by learning from previous trajectories. We introduce a value approximator to recursively reconstruct a terminal cost function and reformulate an infinite time optimization problem to a finite time one. This work proposes a novel controller design approach, and shows its recursive feasibility and stability. Moreover, the convergence of closed-loop trajectory and the optimality of steady trajectory as iterations proceed to the infinity are proven for general nonlinear systems. Simulation and comparison results show the lower storage requirement of the proposed control method than two state-of-the-art methods. Its resulting trajectory is validated to achieve the optimality.

Soft sensing is a promising solution to predict key quality variables in various industries. One of the major obstacles to building an accurate data-driven soft sensor is the scarcity of labeled data and the challenge of extracting useful information from unlabeled data. To mitigate this issue, this work proposes a semi-supervised soft sensing method called dual attention-aided cooperative deep spatiotemporal-feature-extraction network. It leverages an encoder-decoder structure to explicitly exploit the spatial and temporal information in both labeled and unlabeled data, enabling efficient utilization of the latter for prediction performance improvement. The encoder is used to realize more detailed spatiotemporal dependencies capture while a dual attention mechanism is developed for feature extraction. In addition, a gated neuron is added between the encoder and decoder to boost model accuracy by quantifying the contributions of extracted features and adaptively fusing them. To optimize our model while incorporating both labeled and unlabeled data, a mixed form loss is employed in the decoder. Experiments are carried out on a real-life industrial process. The results demonstrate that our proposed model achieves state-of-the-art performance.