To address the mismatch between emergency supply demands and logistics distribution, how to simultaneously optimize the location selection of emergency supply centers and vehicle route planning under scenarios of demand uncertainty is investigated. Firstly, triangular fuzzy numbers are employed to characterize demand quantities, establishing an emergency location-route model. Secondly, the strengths of the dueling DQN and DDQN algorithms are integrated into the high-level selection strategy of a hyper-heuristic algorithm, proposing a reinforcement learning-based hyper-heuristic approach. This algorithm leverages its learning capability to evaluate the performance of underlying heuristic operators, assigning corresponding reward and penalty values. By integrating these values with an improved simulated annealing acceptance mechanism, it guides the heuristic operators to search for high-quality solutions within the solution space. Concurrently, an efficient encoding scheme is designed to enhance the algorithm’s computational efficiency. Finally, experiments validate the effectiveness and robustness of the proposed algorithm, demonstrating overall superior solution quality compared to competing algorithms.
A distributed model predictive control method is proposed to address the multi-objective control problem of vehicle platooning with first-order inertial dynamics under a unidirectional topology. The problem is converted into a local optimal control problem. Firstly, the dynamic model of the vehicle platoon and the communication topology structure are constructed. Secondly, based on the differences in the information received by the following vehicles in the communication topology, cost functions are designed and uniformly described, and system constraints are established. This approach enables each following vehicle to solve a local optimal control problem through rolling optimization based on the limited information received from the preceding vehicle, and apply the optimal acceleration to the system, achieving multi-objective cooperative control of the vehicle platoon. A Lyapunov function is employed to prove the system asymptotic stability. Finally, numerical simulation experiments are conducted for validation. The results show that the proposed control strategy can reduce the spacing error of the platoon and improve fuel economy while ensuring stability.
For issues in discovering energy-saving process routes for machining workpieces, the ant colony algorithm suffers from poor convergence, slow convergence speed and susceptibility to local optima. An improved ant colony algorithm that integrates immune algorithms, optimized pheromone increment and update rules, and a double elite collaboration strategy is proposed, which used for process route planning under a typical machining workpiece energy consumption prediction model. Firstly, the algorithm employs the immune algorithm for preliminary screening of process routes and updates the pheromone matrix, enhancing the search efficiency in the early stages. Subsequently, the pheromone increment is optimized, and pheromone updating strategies are enhanced to improve the algorithm’s convergence speed and global search capability. Finally, through the double elite cooperation strategy and introducing natural population update, the algorithm’s optimization capability has been significantly enhanced, while effectively avoiding the pitfall of local optima. Simulation experiments demonstrate that the improved ant colony algorithm can more quickly and effectively discover process routes with lower energy consumption, and its effectiveness and superiority are validated through comparisons with genetic algorithms and simulated annealing algorithms.
Considering the problem of container drayage operation in semi-automatic truck platooning mode, constraints such as driver fatigue and limited number of drivers are important factors to safe driving and transportation efficiency. The optimization goal is to minimize a logistics company’s total transportation costs including truck deployment cost and driver working time cost. A mixed integer nonlinear mathematical programming model is established based on graph theory, and then the original model is linearized by introducing auxiliary decision variables. Then an improved simulated annealing algorithm is designed. Finally, a series of experiments are carried out based on a large number of randomly generated small and large-scale instances. Experimental results show that optimization tool CPLEX can obtain optimal solutions in a short time for small-scale instances, and the improved simulated annealing algorithm is better than CPLEX in improving solution quality and speed for medium and large-scale instances. Experimental results verify the effectiveness of the proposed model and algorithm.
To address the shortcomings of the northern goshawk optimization (NGO) algorithm, such as slow convergence speed, low optimization accuracy, and a tendency to fall into local optima, an improved northern goshawk optimization algorithm (INGO) is proposed. Firstly, Bernoulli chaotic mapping is used to initialize the population and enhance its diversity. Secondly, a sine-cosine strategy, a nonlinear decreasing search factor, and a weight factor are introduced during the prey selection and attack stages to improve the algorithm’s local search ability and convergence speed. Additionally, a third stage is incorporated, which utilizes adaptive T-distribution variation to update the optimal solution and enhance the algorithm’s global optimization capability. Finally, the performance of the INGO algorithm is comprehensively evaluated through nine standard test functions and path planning scenarios on raster maps of varying complexity. Results demonstrate that the INGO algorithm exhibits excellent optimization accuracy and stability across all test functions. In path planning experiments on raster maps, this algorithm achieves significant improvements over the NGO algorithm in both path length and stability.
In order to solve the problem of false detection and missing detection caused by target occlusion and distant small targets in automatic driving scenes, an improved automatic driving target detection model based on YOLOv8n is proposed. Firstly, a deformable convolution network (DCN) is embedded in the C2f module of Backbone network, which strengthens the feature extraction ability of the model under complex background conditions. Secondly, the global attention mechanism (GAM) is added to the Neck network to highlight the important feature information of vehicles and pedestrians and improve the feature fusion ability of the model. Finally, dynamic head is introduced to enhance the detection ability of small targets. Experimental results show that the mAP value of the improved YOLOv8n algorithm on KITTI dataset reaches 91.0%, which is 3.7% higher than that of the traditional YOLOv8n algorithm, achieving better detection accuracy and effect.
In semiconductor production, traditional algorithms used for solving the reentrant hybrid flow shop scheduling problem often suffer from insufficient population diversity, leading to the issue of local optima. To address this, a hierarchical optimization algorithm (HOA) with improved population diversity is proposed. Firstly, the algorithm ensures adequate diversity in the initial population through a diversity threshold selection process. Then, it combines two global search techniques—extended order crossover and extended position crossover—to further enhance the global search capability and avoid local optima. A dynamic de-similarity mechanism is employed to maintain a well-distributed population during iterations, effectively preventing performance degradation. Finally, a multi-level evolutionary mechanism is applied to enhance the algorithm’s stability and efficiency. Simulation results demonstrate that the proposed algorithm performs excellently in reducing makespan, significantly outperforming traditional methods. It effectively resolves the issue of insufficient population diversity and exhibits higher efficiency and stability, providing a new approach for solving complex scheduling problems.
For the prevalent parallel processing characteristics across multiple workshops in prefabricated box girder production enterprises, the problem as a distributed permutation flow shop scheduling problem with variable processing times is formulated. A mixed-integer programming model is established with makespan and energy consumption as the objective functions, and a memetic algorithm is proposed to solve it. Firstly, a hybrid strategy is employed to generate a high-quality initial population. Secondly, multiple genetic and local search operators are introduced based on problem characteristics to enhance the algorithm’s exploration and exploitation capabilities. Simultaneously, a machine learning-based pre-selection strategy is used to balance the resource allocation of the two types of operators. Finally, the proposed algorithm is validated through experiments on a series of extended Taillard benchmark instances and real-world engineering cases, and compared with four evolutionary algorithms. Experimental results confirm the superiority of the proposed algorithm, offering a more scientific and effective approach to scheduling concrete box girder production in practice.
To effectively plan maintenance and control schedules for expressways and ensure safe traffic flow, an intelligent method capable of scientifically predicting road occupancy is urgently needed. This study proposes a residual decomposition recurrent neural network (RD-RNN) model for expressway occupancy prediction. Firstly, the autoregressive integrated moving average (ARIMA) method is employed to stabilize the residual sequence and extract potential trend and fluctuation features. Then, a long short-term memory (LSTM) network is introduced to capture long-term dependencies within the time series, enabling joint modeling of historical observations and residuals. Finally, the two components are integrated to enhance the model’s ability to predict complex traffic state variations. Comparative experiments with traditional LSTM and Conv-LSTM models, along with ablation studies, demonstrate that the proposed RD-RNN achieves higher prediction accuracy and sta
A quadruped wall-climbing robot with a terminal flexible mechanism is proposed to address the need for regular safety inspections of metal pipelines used in oil and gas transportation and architectural structures. The manual labor is replaced and improved both safety and efficiency. Based on a conventional quadruped robot configuration, a terminal flexible actuation mechanism is introduced that can clasp and adhere to pipe surfaces, enabling stable locomotion along external pipe walls. The D-H parameter method is employed to establish the leg model, and both forward and inverse kinematics along with dynamic analysis are conducted. The correctness of the kinematic solutions and trajectory tracking performance are verified via a 3D simulation model of a single leg. Inspired by the crawling motion of caterpillars, a peristaltic gait suitable for pipe climbing is designed, and the leg movement trajectory is planned using spline curves. Simulation results demonstrate the smoothness of the proposed gait and the feasibility of the motion strategy.
Conventional linear active disturbance rejection control methods exhibit limited performance, when applied to the magnetic levitation ball system containing high-frequency measurement noise. A linear active disturbance rejection control strategy based on cascaded extended state observer (CESO) magnetic levitation ball system is proposed. First, the total disturbance is decomposed into disturbance components of different frequencies. Then, leveraging the low-frequency filtering characteristics of the linear extended state observer and the positive correlation between its state estimation speed/accuracy and bandwidth, a linear cascade extended state observer is constructed, to achieve hierarchical and collaborative processing of the total disturbance. Thereby, enhancing the performance in handling comprehensive uncertain disturbances. Based on the analysis of the convergence of CESO, the performance of linear active disturbance rejection control is simulated and compared with PID and traditional linear active disturbance rejection control, respectively. The results show that the linear active disturbance rejection control based on CESO, can achieve high-precision estimation/control of uncertain disturbances, in the magnetic levitation ball system, and has strong robustness against multi-source uncertain disturbances with high-frequency measurement noise, which can effectively improve the system control performance.
In order to solve the problem that communication and computing resources are consumed in the time-varying leader-following formation for multi-agent systems with Lipschitz nonlinear terms, a distributed event-triggered mechanism is introduced and designed a time-varying formation control protocol for nonlinear systems. Firstly, the proposed event-triggered mechanism operates by updating the control protocol only at triggering instants, thereby avoiding the continuous communication required by conventional methods. Furthermore, theoretical analysis rigorously demonstrates both the convergence of the multi-agent system and the absence of Zeno behavior under the proposed triggering function. Finally, simulation results confirm that the event-triggered strategy significantly reduces inter-agent communication and computational resource consumption while successfully accomplishing the time-varying formation task.
The consensus problem in heterogeneous multi-agent systems with quantized information is addressed, and a bipartite consensus criterion applicable under the circumstances of unknown nonlinear dynamics and external disturbances is proposed. Firstly, a feedback observer is proposed with the help of the adaptive control method, and the leader can be tracked by the observer in a fixed time. Secondly, by the neural network approximation theory involving the pseudo-ideal weight matrix, inequalities and regulator equations, both output consensus protocol and neural network adaptive law are proposed under pinning control strategy. Finally, bipartite output consensus of heterogeneous multi-agent systems can be achieved via the Lyapunov stability theory. Moreover, the validity of the theoretical results is verified by a numerical simulation.
In recent years, with the large-scale integration of new energy sources into the grid, higher requirements have been placed on the grid’s peak regulation and frequency regulation capabilities. This necessitates that medium- and large-sized thermal power units have better control performance to quickly respond to changes in grid load. To address the control challenges of 660 MW thermal power units, which include nonlinearity, strong coupling, and large time delays, a multi-model nonlinear decoupling controller based on a back propagation (BP) neural network was designed. In the controller design, the generalized minimum variance control law was first used to select appropriate controller parameters. Then, a linear decoupling controller was combined with a neural network-based nonlinear decoupling controller to ensure the stability of the thermal power unit system, while a switching mechanism is designed to select the optimal decoupling controller suitable for the current conditions. Simulation experiments demonstrate that the adopted control method can achieve rapid tracking between different loads, which is of practical significance for ensuring the efficient, economical, and stable operation of the power system.
In the process industry, the presence of critical production operation indicators that are challenging to monitor online, coupled with the multimodal nature of industrial data, necessitates heightened complexity in traditional regression-based predictive models, thereby increasing the risk of overfitting. A method for terminal prediction via a hierarchical truncated Gaussian mixture regression model is proposed. Initially, essential variables pertaining to critical production operation indicators in the process industry are identified, followed by the introduction of a hierarchical tree structure to represent the multimodal data structure. Historical data of essential variables and indicators are utilized to build a hierarchical truncated Gaussian mixture regression model using a top-down training procedure, adapting to data distribution. Finally, the training model is utilized to forecast important indicators. Two sets of industrial data were utilized as experimental objects, and the findings demonstrated that the terminal prediction approach had excellent accuracy in predicting process variables.
With the wide application of manipulators in industrial manufacturing, deep-sea exploration, and aerospace fields, the requirements for their tracking accuracy and safety performance are constantly increasing. Therefore, the trajectory tracking control problem of manipulator systems with actuator faults under parameter perturbations and disturbance torques is studied. Firstly, an adaptive sliding mode fault-tolerant controller based on an improved reaching law is designed according to the system state equation. The adaptive algorithm is used to estimate and compensate for unknown disturbances and fault signals to further improve the dynamic performance of the closed-loop system. Then, based on Lyapunov stability theory, the finite-time reachability of the sliding surface and the fast exponential convergence of the tracking error are strictly proved. Finally, taking a two-link manipulator as an object, the simulation verifies that the designed controller has good anti-interference and fault-tolerant performance.
In order to solve the problem of system chattering caused by external interference, inaccurate model and discontinuous switching characteristics of sliding mode control, a sliding mode control of variable speed reaching law based on stochastic configuration networks (SCN) is proposed. Firstly, the switching function (dsat) with adjustable zero slope is designed, and the value of the switching term is adaptively adjusted by the distance between the system state and the sliding mode surface to effectively reduce the system chattering. Due to the uncertainty of the system, the nonlinear term of the system is approximated by SCN, and the approximation ability of SCN is improved by designing the configuration range of input parameters of hidden layer nodes. Finally, the simulation results of a second-order nonlinear system show that the algorithm can effectively reduce the system chattering.
For the low detection rate of industrial steel strip surface defects, an improved YOLO v5 steel strip surface defect detection algorithm is proposed. Firstly, the Squeeze and Exception (SE) attention mechanism is introduced into YOLO v5 backbone network to strengthen the model's ability to extract important defect information and solve the problem of large loss of feature information. Secondly, the four scale detection mechanism is used to increase the detection area of the network model and strengthen the integration of deep and shallow semantic information. Finally, the K-means++clustering algorithm is used to improve and optimize the detection anchor frame to solve the difficult problem of small target detection. The verification results on NEU-DET dataset show that the average accuracy of the proposed model is 79.2%, 3.3 percentage points higher than YOLO v5 model. The algorithm maintains the lightweight advantage of YOLO series, and achieves a better detection effect for small targets of industrial steel strip.
In order to predict the fault trend of complex industrial processes, an improved CNN-BILSTM fault prediction method based on RFKPCA and attention mechanism is proposed. Firstly, random forest algorithm (RF) is used to filter fault related features. Then, kernel principal component analysis (KPCA) is used to extract nonlinear features and feature reconstruction, and Hotelling (2T) statistics are constructed according to the reconstruction error to describe the operation state of the industrial system. The 2Tstatistics obtained and the variables collected form supervised learning time series data. In view of the inability to extract spatial features from the prediction model of bidirectional long short memory neural network (BiLSTM), convolutional neural network (CNN) is introduced and the activation function of CNN is improved. Then, the hidden output layer of BiLSTM is connected with the attention mechanism, and the model accuracy is improved by assigning weights to key features through the attention mechanism. The experimental results on the TE simulation platform data show that the model accuracy has been significantly improved, and it has good predictive performance.
A fixed-time fuzzy adaptive control strategy is proposed for a class of switched uncertain nonlinear systems with time-varying full-state constraints. Firstly, the fuzzy logic system is used to approximate the unknown nonlinear function in the switched nonlinear system, and the linear state observer is used to estimate the unknown state of the system, which eliminates the limitation that the state must be measurable and the function must be known in the controller design process. Then the barrier Lyapunov function is introduced to solve the problem of time-varying state constraints. In controller design, only one adaptive parameter is needed, which reduces the computational burden and solves the problem of over-parameterization. Finally, the simulation results show that the control strategy can ensure that the tracking error of the system converges to a small neighborhood of the origin in a fixed time, and the convergence time is independent of the initial state of the system, and at the same time, it can ensure that all the states of the system are within the constraint bound set in advance, and improve the transient performance and steady-state performance of the system.
