In modern process industries, data-driven soft sensors have become indispensable for monitoring critical process variables that are inaccessible to direct measurement. Nevertheless, the accurate modeling of industrial processes remains challenging due to their intrinsic complexities, such as time-series behaviors, measurement outliers, redundant variables, and potential concept drift. Existing approaches can address subsets of these challenges but generally lack a unified mechanism that integrates robust offline modeling, variable-importance analysis, and efficient online adaptation. To address these issues, this study proposes an online-updating robust soft sensor framework based on bidirectional long short-term memory (BiLSTM) with integrated gradients (IG) and smoothed quantile loss (SQLoss). During offline modeling, a soft-sensing model is constructed using a BiLSTM, and the proposed SQLoss is introduced to reduce the influence of outliers; the IG method is then employed to evaluate the importance of input variables, enabling input variable selection. During online operation, model parameters associated with significant variables are selectively updated based on IG-derived variable importance, thereby addressing concept drift. Finally, experimental results on an industrial desulfurization process demonstrate that, compared with the best-performing competing basic learner, the proposed SQLoss-BiLSTM-IG reduces the average root mean squared error (RMSE) and mean absolute percentage error by 5.26% and 1.87%, respectively, while increasing the average correlation coefficient by 1.94%; in the online evaluation, the proposed updating strategy achieves a mean RMSE of 2.441, demonstrating its effectiveness in handling concept drift. Moreover, the analysis of key variable importance is consistent with field experience, offering valuable insights for optimizing the desulfurization control system.

Sound zone control enables independent listening regions within a shared acoustic environment. The variable span trade-off (VAST) filter balances acoustic contrast and signal distortion via a subspace dimension parameter; however, this parameter typically requires manual tuning, and performance degrades under physical perturbations such as array misalignment and reverberation mismatch. To address these limitations, we propose Spectral Adaptive VAST (SA-VAST), which stabilizes the perturbed generalized-eigenvalue spectrum through linear shrinkage and selects the subspace dimension at each frequency using a conservative RMT-informed spectral separation criterion augmented by an empirical finite-size safety margin. Across simulations involving sensor noise, spatial misalignment, and reverberation variability, SA-VAST achieves higher acoustic contrast than an explicitly preconfigured Fixed-VAST baseline under matched perturbation conditions, with gains of 5.92 to 6.01 dB, while maintaining both normalized reproduction-error metrics below 0 dB. Additional sweeps over transfer-function signal-to-noise ratio and array configuration further characterize the adaptive behavior of SA-VAST. Together, these results provide evidence that SA-VAST supports autonomous, empirically conservative subspace selection for robust sound-zone control in Internet of Audio Things deployments.

Photovoltaic (PV) power generation exhibits significant stochasticity and volatility due to meteorological influences, challenging grid stability and energy management. This paper proposes a Spectral-Laplacian Enhanced Inverted Transformer (SLEIT) for accurate PV power forecasting. The architecture incorporates three key innovations: (1) a dual-domain fusion stack that adaptively integrates time-frequency features through parallel Fourier and temporal processing paths; (2) an inverted embedding mechanism that redirects attention to variable dimensions, enhancing multivariate dependency modeling; (3) a Laplace stack that approximates differential properties through cascaded nonlinear transformations, improving sensitivity to transient fluctuations. Experiments on six years of PV station data across multiple prediction horizons demonstrate that SLEIT achieves 3.63%-15.84% reduction in MSE and 0.43%-14.13% reduction in MAE compared to state-of-the-art baselines, with superior performance during extreme weather-induced power variations.

This paper proposes an intelligent vehicle local path planning algorithm for complex scenarios. Firstly, the Frenet coordinate system is used to decouple the motion of vehicle. According to the initial configuration and target configuration, the trajectory set are obtained through a polynomial method. Secondly, the feasibility of these trajectories is checked by collision, minimum distance between vehicles, curvature peak, and acceleration peak. Finally, a novel cost function based on the driving risk value and the rate of acceleration change is used to evaluate the feasible trajectories. The trajectory that minimizes the loss function is selected as the optimal solution. The results show that the proposed method can obtain a reasonable driving plan in a specific scenario.

Electroencephalography (EEG) signals can be used as a neuroimaging indicator to analyze brain-related diseases and mental states, such as schizophrenia, which is a common and serious mental disorder. However, the main limiting factor of using EEG data to support clinical schizophrenia diagnosis lies in the inadequacy of both objective characteristics and effective data analysis techniques. Random matrix theory (RMT) and its linear eigenvalue statistics (LES) can provide an effective mathematical modeling method for exploring the statistical properties of non-stationary nonlinear systems, such as EEG signals. To obtain an accurate classification and diagnosis of schizophrenia, this paper proposes a LES-based deep learning network scheme in which a series of random matrixes, consisting of EEG data sliding window sampling and their eigenvalues, are employed as features for deep learning. Due to the fact that the performance of the LES-based scheme is sensitive to the LES’s test function, the proposed LES-based deep learning network is embedded with two ways of combining LES’s test functions with learning techniques: the first is to have the LES’s test function assigned, while, using the second way, the optimal LES’s test function should be solved in a functional optimization problem. In this paper, various test functions and different optimal learning methods were coupled in experiments. Our results revealed a binary classification accuracy of nearly 90% in distinguishing between healthy controls (HC) and patients experiencing the first episode of schizophrenia (FES). Additionally, we achieved a ternary classification accuracy of approximately 70% by including clinical high risk for psychosis (CHR). The LES-embedded approach yielded notably higher classification accuracy compared to conventional machine learning methods and standard convolutional neural networks. As the performance of schizophrenia classification is strongly influenced by test functions, a functional optimization problem was proposed to identify an optimized test function, and an approximated parameter optimization problem was introduced to limit the search area of suitable basis functions. Furthermore, the parameterization test function optimization problem and the deep learning network were coupled to be synchronously optimized during the training process. The proposal approach achieved higher classification accuracy rates of 96.87% between HC and FES, with an additional 89.06% accuracy when CHR was included. The experimental studies demonstrated that the proposed LES-based method was significantly effective for schizophrenic EEG data classification.

In this paper, a new sub-pixel intensive particle image identification method used for particle tracking velocimetry (PTV), is discussed. With aiming to improve the performance of accuracy and robustness, a two-stage deep learning framework consisting of two independent convolutional neural networks (CNN) is proposed in this method. The first neural network is to segment the particle blobs from the image, and the second one is to locate the position of the segmented particles at the sub-pixel level. A synthetic dataset containing particle images and the ground-truth positions is generated for network training. The effect of different characteristic parameters (e.g., the particle conditions; the tuning of the loss function; the noise level) is evaluated on synthetic images. To further verify the generalization capabilities of the technique, we also apply the proposed network to real-world images. Both synthetic and real-world experiment results strongly demonstrate that the proposed method possesses better accuracy and robustness than other conventional methods.

The vertical cement mill is widely used in the cement industry due to low power consumption, high energy efficiency, and its compact size. The intelligent fault detection using reducer vibration data for vertical cement mills is of great importance for safe, stable, and efficient operation. In this paper, a deep learning based fault diagnosis method using 1D samples is proposed. For fast and low cost detection, the convolution neural network is improved for real-time implementation. The depthwise separable convolution and inverted residual block are applied for light-weighted network construction. The feature extraction is performed by the wavelet kernel based convolution layer. Three kinds of wavelet basis have been applied, namely Laplace, Mexhat, and Morlet basis. To improve the feature extraction performance of the first convolution layer, different wavelet basis functions are testd. The number of convolution layers is reduced to extract more features. Measured signals from a cement slurry factory in China are used to verify the effectiveness of proposed detection method. The light-weighted network structure provides 75% reduction of the memory space. Compared with conventional cement mill fault diagnosis methods, which acquire high-dimensional data and high frequency sampled signals, the proposed method adopts data measured in a simpler way with compatible detection accuracy.

The damage stability of ship is a key performance metric in ship design. The challenge of damage stability optimization is a complex calculation process with multi scene dynamic inflow simulation relying on the cooperation of ship design software such as NAPA. Our work is to establish a collaborative optimization framework to effectively coordinate reinforcement learning (RL) with particle swarm optimization (PSO) and NAPA. The collaboration of RL and PSO realizes the update direction selection of watertight bulkhead position scheme. The collaboration of PSO and Napa promotes the iterative process of watertight bulkhead position and damage stability value A. The experimental results show that compared with the traditional damage stability optimization method, the collaborative optimization method improves the damage stability by 2.36% and reduces the calculation time significantly.

Globally, wind power plays a leading role in the renewable energy industry. In order to ensure the normal operation of a wind farm, the staff will regularly check the equipment of the wind farm. However, manual inspection has some disadvantages, such as heavy workload, low efficiency and easy misjudgment. In order to realize automation, intelligence and high efficiency of inspection work, inspection robots are introduced into wind farms to replace manual inspections. Path planning is the prerequisite for an intelligent inspection robot to complete inspection tasks. In order to ensure that the robot can take the shortest path in the inspection process and avoid the detected obstacles at the same time, a new path-planning algorithm is proposed. The path-planning algorithm is based on the chaotic neural network and genetic algorithm. First, the chaotic neural network is used for the first step of path planning. The planning results are encoded into chromosomes to replace the individuals with the worst fitness in the genetic algorithm population. Then, according to the principle of survival of the fittest, the population is selected, hybridized, varied and guided to cyclic evolution to obtain the new path. The shortest path obtained by the algorithm can be used for the robot inspection of the wind farms in remote areas. The results show that the proposed new algorithm can generate a shorter inspection path than other algorithms.

Recently, run-to-run controller (R2R) has been widely used in process manufacturing, such as fine chemicals industry and semiconductor manufacturing. Numerous control schemes have been developed for various industrial processes to enhance product quality by reducing the output variation. However, previous work was mainly focused on optimizing the steady process while the transient process is seldom considered. Since the production mode is changing from mass production to more diversified and small lot-size production, the transient process will become more and more critical. In this paper, a weighting tuning control framework based on R2R control is proposed which uses the information among the successive batches. The superiority lies in selecting better weighting factors for EWMA controller. Compared to previous work, we derive a better analytical expression for evaluation and after that a recursive algorithm is proposed to estimate important parameters of the process on which the performance of the transient process is closely dependent. The numerical results manifest significant improvement compared with state of the art solutions.

Abstract Particle image velocimetry (PIV) is an essential method in experimental fluid dynamics. In recent years, the development of deep learning‐based methods has inspired new approaches to tackle the PIV problem, which considerably improves the accuracy of PIV. However, the supervised learning of PIV is driven by large volumes of data with ground truth information. Therefore, the authors consider unsupervised PIV methods. There has been some work on unsupervised PIV, but they are not nearly as effective as supervised learning PIV. The authors try to improve the effectiveness and accuracy of unsupervised PIV by adding classical PIV methods and physical constraints. In this paper, the authors propose an unsupervised PIV method combined with the cross‐correlation method and divergence‐free constraint, which obtains better performance than other unsupervised PIV methods. The authors compare some classical PIV methods and some deep learning methods, such as LiteFlowNet, LiteFlowNet‐en, and UnLiteFlowNet with the authors’ model on the synthetic dataset. Besides, the authors contrast the results of LiteFlowNet, UnLiteFlowNet and the authors’ model on experimental particle images. As a result, the authors’ model shows comparable performance with classical PIV methods as well as supervised PIV methods and outperforms the previous unsupervised PIV method in most flow cases.

Combining the advantages of both integer-order and fractional-order complex chaotic systems, we propose a hybrid-order complex Lorenz system. We demonstrate its abundant chaotic characteristics, including symmetry and dissipation, fixed points and their stability and Lyapunov exponents, with 0-1 test. Then we show that, as the initial value, parameters and the order are varying, the system exhibits diverse dynamical behaviors, with fixed points, limit cycles and chaotic attractors. We further show that the system has coexisting attractors and para-metric attractors. In addition, we find that the system generates different chaotic attractors as the system hybrid order varies, referred to as order attractors. Finally, we examine the dynamic transport of the hybrid-order complex Lorenz system and design a piecewise continuous controller to realize offset boosting control. By varying the initial value, parameters or orders, we realize the dynamic transport of the system. Our simulation results confirm the dynamic transport of the hybrid-order complex Lorenz system.

Due to the intuitiveness of image transmission information, the application of images can be ubiquitous, correspondingly, the security of image information has become more significant. As chaos is compatible with cryptography due to some of its characteristics such as sensitivity of initial values, randomness, etc., it is extensively used in image encryption. In this paper, based on artificial images and 2D lagging complex Logistics mapping (2D-LCLM), a novel image encryption algorithm was proposed. Firstly, the pseudorandom sequences generated by 2D-LCLM is used to construct an artificial image. Artificial images can replace the traditional iterative encryption mechanism. Secondly, adopt chaotic sequences to position the original and artificial images. Finally, DNA XOR manipulation is used to merge the two images. Experimental simulation of the standard test graph, through the analysis we can know the encryption algorithm based on artificial image has good encryption effect and security performance.

Industrial processes are often nonlinear and multivariate and suffer from non-Gaussian noise and outliers in the process data, which cause significant challenges in data-driven modelling. To address these issues, a robust soft-sensing algorithm that integrates Huber's M-estimation and adaptive regularisations with multilayer perceptron (MLP) is proposed in this paper. The proposed algorithm, called RAdLASSO-MLP, starts with an initially well-trained MLP for nonlinear data-driven modelling. Subsequently, the residuals of the proposed model are robustified with Huber's M-estimation to improve the resistance to non-Gaussian noise and outliers. Moreover, a double L1-regularisation mechanism is introduced to minimise redundancies in the input and hidden layers of MLP. In addition, the maximal information coefficient (MIC) index is investigated and used to design the adaptive operator for the L1-regularisation of the input neurons to improve biased estimations with L1-regularisation. Including shrinkage parameters and Huber's M-estimation parameter, the hyperparameters are determined via grid search and cross-validation. To evaluate the proposed algorithm, simulations were conducted with both an artificial dataset and an industrial dataset from a practical gasoline treatment process. The results indicate that the proposed algorithm is superior in terms of predictive accuracy and robustness to the classic MLP and the regularised soft-sensing approaches LASSO-MLP and dLASSO-MLP.

This paper studies the process of industrial App development for cement industry and the basic framework of the support platform for network collaborative manufacturing. The key processes of App development are realized with a concrete case of multi-base collaborative manufacturing in cement industry. In the framework, a complete closed-loop is formed starting from the definition of collaborative problem, abstraction of manufacturing resources, integrated model design, optimization algorithm and low-code development. Finally, in order to verify the feasibility of the supporting platform framework, we design a simulation software to stimulate the overall Apps development environment.

Using battery power is an effective way to improve the efficiency of IOT system construction. However, the existing battery State of Health (SOH) prediction methods can't achieve the timeliness of prediction under accuracy ensuring. It brings difficulties to the maintenance of IOT devices. This paper focuses the SOH prediction of batteries in IOT devices. This paper proposes an prediction approach based on the combination of one dimension convolution neural network (1DCNN) and Bi-directional Long Short-Term Memory (Bi-LSTM) - the 1DCNN_Bi-LSTM model. The parameters in 1DCNN_Bi-LSTM model are optimized with Whale Optimization Algorithm (WOA). The high-dimensional features are derived by 1DCNN. Then, these features are transferred to Bi-LSTM to explore the effective memory information. Finally, the prediction results of battery SOH are output from the fully connected layer. In the training process, parameters, such as the number of nodes of each layer and the number of epochs, are optimized with WOA. The WOA_1DCNN_Bi-LSTM model is tested with NASA battery circulation dataset, and the results show that the model has accurate and stable prediction effect on battery SOH. The mean absolute error (MAE) and mean square error (MSE) of the model are 2.1% and 0.624% respectively.

A large body of hydrogenation reactions is frequently running in fed-batch reactors now, which is very important in the industrial field, as we know. However, in the face of such a complex reaction system, relying only on traditional control algorithms will cause many problems in getting stable control. With the rise of intelligent control algorithms, a more effective control method is introduced based on the gaussian process(GP) to rebuild a model and combined with a predictive model method to achieve a stable control during operation and learning with more minor data set, compared to the neural network. The proposed technique’s effectiveness is verified on the simulink platform. The results show that this method can control the corresponding state to an acceptable range and has a more effective control result.

船舶不沉性是衡量船舶生命力的重要性能,也是优选水密分舱策略的关键指标,但时间成本高仍是制约不沉性寻优实用化的难点.随着机器学习技术应用不断深入,将提供更有效的途径.该研究基于强化学习的粒子群算法求解不沉性分舱优化问题,实现集成机器学习模块和不沉性设计模块的优化系统开发和界面设计,讨论算法中不同参数的设置对寻优能力的影响.通过寻优解的分析,表明该方法能够高效地找到较优的分舱方案,为制定科学的分舱策略等方面提供依据.

This paper aims at the multi-base cooperative production and scheduling problem in cement industry. Considering the supply of raw materials, manufacturing capacity of production lines, inventory capacity and customer demand, a MIP (Mixed Integer Programming) model is established. Then AMPL (A Mathematical Programming Language) is used to formally describe this MIP optimization model. The problem then is solved by Gurobi solver. Finally, the collaborative optimization problem is visually configured and solved based on the developed network collaborative manufacturing simulation software platform. The computational results verify the effectiveness of the optimization problem modeling and the feasibility of the simulation platform.

The question for understanding the correlation between genetic expression and diagnosis of schizophrenia patients has always been challenging. The DNA methylation can provide an effective approach to solve the problem. However, limited by the size of datasets and the high dimensionality, conventional feature selection and pattern recognition methods cause a severe overfitting phenomenon. In this paper, the DNA methylation data are analyzed to filter out the most important gene points for classification. To alleviate the negative overfitting impacts the optimization-based meta-learning i.e., the MAML algorithm, is developed to solve the multi-classification and feature selection problems. The proposed method combining MAML with attention-based mechanism is capable of conducting feature selection with fast generalization and quick adaptation for small samples. The major work lies in that the proposed model is pretrained with TCGA which can be tailored to our circumstance with priori information. To test the efficiency of the proposed algorithm, the DNA methylation data are used including homicidal schizophrenics and violent schizophrenics from some authoritative institute. Our method achieves a recognition accuracy of 91.89% (homicidal schizophrenics, schizophrenics without violent behaviors, and normal people) and 67.96% (schizophrenics with violent behaviors, schizophrenics without violent behaviors, and normal people}) respectively, which is a great improvement compared to the best-known results based on the traditional method with 82.52% and 63.76%. Furthermore, 1000 significant feature gene points are selected from more than 400 thousand feature points.