To maintain the stability of continuous stochastic systems driven by the Lévy process, a discrete control method is proposed. Firstly, a sliding mode controller is designed, which ensures that the system can maintain mean square exponential stability under Lévy noise interference by designing the control law reasonably. Secondly, the continuous controller is discretized to meet the requirements of actual digital control systems. Thirdly, through deduction, it is concluded that the second-order moment of the difference between
the system state under the action of a discrete controller and the system state under the action of a continuous controller is bounded, indicating that the discretization process does not significantly increase the instability of the system. Finally, the stability of the closed-loop system under the action of a discrete controller is demonstrated through theoretical analysis. The experimental results show that the discretized controller can still keep the system stable.

The basis for predictive control of air conditioning terminal systems in data centers is the multi-step prediction of cabinet inlet temperature. To improve the precision and the portability of the prediction model, a nonlinear Takagi-Sugeno (T-S) fuzzy method is proposed to construct the temperature prediction model of data center server rooms. Firstly, the computational fluid dynamics (CFD) model of a server room is developed by using CFD numerical simulation method, and a data acquisition strategy is designed to capture the complete dynamic characteristics of the system. Then, to solve the problem that the fuzzy C-mean clustering algorithm is prone to local optimum, the improved beetle antennae search algorithm is proposed to optimize the identification of the forepart structure of the T-S fuzzy model. Finally, the cubature Kalman filter is adopted for the posterior part parameter identification and online correction of the T-S fuzzy model. The experimental results show that the T-S fuzzy model constructed by this method has higher computational efficiency and prediction accuracy than that of the conventional T-S fuzzy model, and the requirement of model portability can be met through the update of the posterior part parameters.

For the filtering problem of nonlinear systems with Gaussian noise and non-Gaussian noise, if the mixed noise is treated as non-Gaussian noise, the filtering accuracy will be affected by ignoring the characteristics of Gaussian noise. Therefore, based on the idea of “system split + algorithm fusion”, a new non-Gaussian nonlinear filtering algorithm is designed. Firstly, the system split weights and introduced to divide the nonlinear system under the mixed interference of multiple noise into several subsystems affected by a kind of noise. Then, according to the noise characteristics of each subsystem, the corresponding sub-filtering algorithm is designed. Finally, the filtering results of each sub-filtering algorithm are fused. Additionally, two kinds of weights, namely average weights and dynamically updating weights, are introduced. The simulation results show that, compared with the existing nonlinear filtering algorithms which regards the mixed noise as a kind of Gaussian noise or non-Gaussian noise, the proposed algorithm has a significant advantage in filtering accuracy.

There are problems in the optimization process of pelican optimization algorithm, such as the reduction of population diversity, the decrease of convergence speed, and the tendency to fall into local optimum. To solve these problems, multiple strategies are integrated to improve pelican optimization algorithm, and the improved pelican optimization algorithm (IPOA) is proposed. Firstly, the pelican population is initialized by using the tent chaos map and refracted opposition-based learning strategy, which not only increases the population diversity, but also lays the foundation for improving the optimization ability of the algorithm. Then, a nonlinear inertia weight factor is introduced at the stage when pelicans approach their prey to improve the convergence speed of the algorithm. Finally, the leader strategy of salp swarm algorithm is introduced to coordinate the global search ability and local optimization ability of the algorithm. The improvement effect of the single improvement strategy is tested and IPOA is compared with 9 other optimization algorithms in the experiment. The experimental results prove the effectiveness of each improvement strategy and the superiority and robustness of IPOA.

Cascade control systems have stronger anti-disturbance ability and adaptive ability than single-loop systems, and therefore are widely used in industrial systems, most of which are non-Gaussian systems. Entropy is often used as the performance index to assess the performance of non-Gaussian systems, but the entropy index has the problem of translation invariance. Therefore, the minimum entropy index is weighted and mixed with the Jensen-Shannon (JS) divergence index, and a mixed assessment index is proposed. The performance assessment of non-Gaussian systems requires the effective identification of system parameters based on closed-loop data. In order to obtain the accurate reference, based on the ideas of particle swarm optimization algorithm and estimation of distribution algorithm, and combining the advantages of both, a hybrid optimization algorithm is proposed for system parameter identification. Finally, the cascade control system is simulated under different non-Gaussian noise disturbances. The simulation results verify the effectiveness of the proposed hybrid optimization algorithm and hybrid assessment index.

In the control of bronchoscopic robots, the accuracy of the conventional proportional integral differential (PID) control is insufficient, and the back propagation (BP) neural network is prone to fall into local optimum. To solve the problems, a BP neural network-PID control method optimized by the improved whale optimization algorithm (IWOA) is proposed. Firstly, based on the conventional whale optimization algorithm, IWOA introduces a nonlinear convergence factor to dynamically balance the global search ability and local search accuracy, optimizes the population distribution by using tent chaotic mapping, enhances global optimization by using the Lévy flight strategy, maintains population diversity by combining the greedy selection mechanism, and provides the optimal initial connection weights for the BP neural network. Then, the BP neural network fuses the reference input, system output and tracking error at the input layer, and dynamically adjusts the PID control parameters through backpropagation. The simulation results show that, compared with PID control, BP neural network-PID control and their improved methods, the proposed method can significantly reduce the overshooting of the system, shorten the regulation time, and make the steady-state error approach zero. This method has good control accuracy and anti-interference ability, so it can significantly reduce mechanical vibration and tissue friction during operation, and improve the safety of bronchoscopic surgery.

The aircraft anti-skid braking system determines the safety and comfort of the aircraft landing process. Firstly, a dynamic analysis of the aircraft anti-skid braking system is conduct, and a model of the aircraft anti-skid braking system is established. Then, to solve the problem that the integrator in the conventional integral sliding mode control has the phenomenon of integral saturation, by setting the slip rate as the control target, the integral sliding mode surface is improved, and the controller is designed and verified through simulation. The simulation results show that the improved integral sliding mode controller can enable the slip ratio to quickly track the target slip rate with strong robustness, effectively reduce the output buffering of the controller, and improve the performance of the aircraft anti-skid braking system.

To solve the problem that the conventional central pattern generator (CPG) introduce too many coupled dynamic parameters in their applications, which makes them difficult to adapt to the robots with insect-like leg structure, an improved CPG control method combined with inverse kinematics to realize foot-end control is proposed. Firstly, at the system level, a linear converter, a function generator and an inverse kinematics solution module are designed, and the conventional CPG application is parametrically improved in a foot-oriented control manner. Secondly, in terms of details, the limit cycle of CPG is improved to enhance its adaptability to terrain. Finally, functional gaits such as linear steering are planned based on the improved CPG model, and the closed-loop control for attitude stabilization based on foot end deviation compensation is achieved. The experimental results show that the proposed method simplifies the gait control, enabling the insect-like hexapod robot to achieve linear steering on undulating terrain and maintain stable attitude.

Conventional simultaneous localization and mapping (SLAM) has poor robustness in weak-texture scenes and is disturbed by dynamic objects in dynamic scenes. To solve these problems, dynamic visual SLAM is proposed. Firstly, in the visual odometry, the geometric correspondence network version 2 (GCNv2) is used to extract feature points and generate binary descriptors, which improves the robustness of SLAM in weak-texture scenes. Then, the object detection network is introduced to detect dynamic objects, obtain the semantic information of the current frame, and combine multiple view geometry to eliminate dynamic objects, which removes the interference of dynamic objects on SLAM. The experimental results show that the proposed method can continuously extract a sufficient number of high-quality feature points in weak-texture scenes. In scenes with the interference of dynamic objects, the absolute pose error and relative pose error of the proposed method are small. In static scenes, the performance of the proposed method is still superior.

The manipulator system has problems such as unmodeled part, friction, and external disturbance. Therefore, an integral fast terminal sliding mode control method considering input saturation is proposed for the trajectory tracking control of joint angles. Firstly, the manipulator model is established, and regard the unmodeled part, friction, external disturbance, and input saturation error as are regarded as concentrated disturbance. Secondly, a fixed-time extended high gain observer is designed to estimate the concentrated disturbance, which solves the problem that the estimation error of the conventional extended high gain observer has a peak at the initial moment and can only be asymptotically stable. Thirdly, the integral sliding mode control is combined with the fast terminal sliding mode control, and the backstepping method is used to design the integral fast terminal sliding mode controller. The buffeting of the control output is reduced by the concentrated disturbance estimated through the observer. Finally, simulation experiments are conducted on the proposed observer and control method. The simulation results show that the proposed observer can accurately estimate the concentrated disturbance, and the proposed control method can improve the trajectory tracking speed and accuracy of the manipulator system, ensuring the safety of the manipulator system while considering the input saturation.

In order to improve the response speed and robustness of the AC microgrid inverter output voltage, a performance improvement control strategy combining linear quadratic optimal control and residual generator is proposed. Firstly, the state space model of the three-phase voltage inverter is established, and a quadratic optimal controller with a fast response is designed based on the model. Secondly, according to the double coprime decomposition and Youla parameterization theory, a performance improvement control structure based on the residual generator is obtained. Thirdly, based on the model matching principle, the performance improvement controller is solved to improve the robustness of the system. Finally, several experiments of step disturbance, three-phase imbalance load and nonlinear load are designed, and the experimental results verify the effectiveness of the proposed control strategy.

To obtain high quality and low cost cement raw meal, the optimization of raw material ratio is studied. Firstly, due to oxide content fluctuation of raw material and working condition change of vertical mill, the concept of equivalent value of oxide content for raw material is proposed. It is used as the relationship parameter between oxide content of raw meal and raw material ratio, and the method of system identification is proposed to calculate it. Then, a multi-objective optimization model of raw material ratio is established, with the goal of minimizing production cost and ratio adjustment. To ensure the feasibility of the solution, the quality control indicators of raw meal are added to the constraints. An improved multi-objective particle swarm optimization algorithm is proposed to solve the model. The experimental results show that, compared with the non-dominated sorting genetic algorithm II (NSGA-II) and manual proportioning, the proposed algorithm, when applied in optimizing the raw material ratio, can effectively control the quality control indicators of raw meal within the target range, and reduce the raw material cost.

The heating furnace system in the hot continuous rolling process has the characteristics of large inertia and large lag. In order to improve the temperature control performance of the heating furnace, a comprehensive optimization control strategy based on cyber-physical system (CPS) is proposed. Firstly, a traditional proportional integral differential (PID) temperature control system model with time-delay is established. Then, a modified state equation is constructed by introducing a shift parameter, the parameters of the controller for the neutral time-delay system are calculated, and a control strategy combining time-delay optimization control based on CPS and PID control is proposed. Finally, the proposed control strategy is applied in the stepping heating furnace of a steel plant. The experimental results show that this control strategy can enhance the intelligence level of temperature control, achieve a precise heating process, optimize the production process, improve the heating efficiency and reduce energy waste. This control strategy has practical engineering value and popularization significance.

To solve the problem of uneven thickness existing in the current production process of ultra-thin amorphous alloy, a multi-mode fuzzy human-simulated intelligent correction strategy is proposed. Firstly, the production site data is analyzed, and the thickness variation trend is divided by using thickness deviation information. Then, based on the experience of on-site experts in adjusting the pressure and temperature settings of the transition packet under different thickness variations, the multi-mode fuzzy human-simulated intelligent correction strategy is designed, including bang-bang, fuzzy, compensation, and maintain modes. Finally, the generalized regression neural network optimized by the improved particle swarm optimization algorithm is used to establish the thickness regression model of ultra-thin amorphous alloy for the simulation study of the thickness correction strategy. The simulation results show that the multi-mode fuzzy human-simulated intelligent correction strategy can stabilize the thickness deviation within ±1 μm, avoid the problem of uneven thickness, and has strong anti-interference ability.

Industrial process data have characteristics such as high dimensionality and imbalance, which can affect the accuracy of industrial process performance evaluation. To solve this problem, a multi-data space integration model of least squares support vector machine (LSSVM) based on latent variable technology and particle swarm optimization (PSO) algorithm is proposed for industrial process performance evaluation. Firstly, the sampled process variable data are divided into different data space according to performance levels. Then, feature mapping is performed on the data spaces of different performance levels to extract latent variables, and latent variables are screened by mutual information to achieve the goal of reducing the dimension of data spaces. Finally, sub-models of LSSVM are established in different data spaces, and PSO algorithm is used for their integrated optimization to obtain the offline model. The offline model obtains the performance evaluation results by calculating the similarity between the online data and each performance level. The proposed method is applied in the simulation of the operation performance evaluation of the ethylene cracking furnace, and the simulation results prove its effectiveness.

In the conventional islanding detection methods of microgrids, the passive method has problems such as large detection blind zones and difficult threshold setting, while the active method has the problem of disturbance in power quality. Therefore, an islanding detection method based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN)-Teager-Kaiser energy operator (TKEO) and optimized deep belief network (DBN) is proposed. Firstly, the voltage and current signals at the common coupling points are decomposed by the CEEMDAN algorithm to obtain a series of intrinsic mode functions (IMF), and the correlation coefficients are calculated to determine the effective IMF. Secondly, the effective IMF is subjected to product fusion, and the energy sequence of the fused IMF is calculated by TKEO to obtain the reconstructed island features. Finally, the DBN is optimized by the particle swarm optimization algorithm, and the extracted features are input into the optimized DBN for training and testing. The experimental results show that the proposed method can effectively distinguish the islanding and non-islanding states under different working conditions, the detection accuracy can reach 99.52%, the detection time is 25.326 ms, and it has a strong anti-noise ability.

The existing image inpainting models are unable to inpaint images with large-scale defects at a high quality. To solve this problem, an image inpainting model based on Transformer and the generative adversarial network is proposed. Firstly, a mask adapt input module is designed, which is used to extract the image blocks that are not masked from the input image. Secondly, the Transformer is used to extract global context information from the valid image blocks, thereby enhancing the model’s ability to inpaint missing areas. Thirdly, the fast Fourier convolution (FFC) modules are used to enhance the ability to inpainting details and eliminate artifacts in the output image. Finally, the performance of the whole network is improved under the adversarial training of discriminator network. The proposed model is used to inpaint the images of Place2 dataset. The test results show that when the mask ratio is 50%~60%, the peak signal-to-noise ratio of the inpainting results reaches 19.748 2 dB, and the structural similarity (SSIM) reaches 0.714 7.

Point cloud, an unstructured data form in three-dimensional space, is widely used in fields such as machining, robotic assembly, and autonomous driving. Current feature point detection algorithms are based on artificially designed rules, which impair the accuracy of subsequent classification and registration. Therefore, a feature point detection algorithm based on deep learning is proposed, which learns the feature points for subsequent classification and registration tasks in a weakly supervised manner. Firstly, the feature points in the point cloud are detected by the feature point extractor. Then, the feature points are used for subsequent classification and registration tasks. By minimizing the loss of classification and registration, the feature point detector can extract excellent feature points to overcome the unavailability of feature point labels. The experimental results show that, compared with the traditional feature point detection algorithms, the feature points obtained by the proposed algorithm can represent the overall state of the point cloud of the mechanical part, and improve the accuracy of subsequent classification and registration.

In order to improve the efficiency of transmission line selection and reduce the construction cost of transmission lines, an improved ant colony optimization algorithm based on geographic information system is proposed. Firstly, a raster model of the planned area is established to expound the application principle of the conventional ant colony optimization algorithm in the transmission line selection. Then, to solve the problems that the conventional ant colony optimization algorithm is prone to fall into local optimum and the searched path has many inflection points, an adaptive update mechanism of pheromone concentration and a node optimization mechanism are proposed to improve it. The experiment takes a certain area in Anhui Province as an example to select the transmission line. The experimental results show that, compared with the conventional ant colony optimization algorithm, the improved ant colony optimization algorithm has higher search efficiency, and the searched path has fewer inflection points, which can effectively reduce the construction cost of the transmission line.

In actual industrial production, the insufficient number of fault samples for rolling bearings leads to inaccurate fault diagnosis. To solve this problem, a fault diagnosis method based on the Glow-ECNN model is proposed by combining the convolutional neural network model with the efficient channel attention (ECA) mechanism (ECNN model) and the generative flow (Glow) model. Firstly, one-dimensional fault vibration signal is transformed into a two-dimensional time-frequency image that contains time-frequency feature information by continuous wavelet transformation. Then, the time-frequency images are input into the Glow model for data augmentation, generating a sufficient number of time-frequency images with a similar distribution to the original ones. These generated time-frequency images are combined with the original time-frequency images as training samples. Finally, the ECNN model is used to classify the faults. The experimental results show that the proposed method can achieve an accuracy rate of 99% in fault diagnosis of rolling bearings under small sample conditions, demonstrating its feasibility and effectiveness.