The paper aims to study the problem of obstacle avoidance in air?ground cooperative tracking control for the unmanned aerial helicopter (UAH),in which a new approach of designing the path obstacle avoidance plan and controller design is proposed.Initially, as for the uncertain linear UAH,by processing and judging two?dimensional environmental information within the warning range for the UAH,an obstacle avoidance strategy is proposed with the help of wall?following algorithm,and the flight angle of obstacle avoidance path and the tracking speed that can make up for bypass distance are calculated.Secondly,the proposed obstacle avoidance method is extended to the three?dimensional case,and the flight angle of the UAH is determined based on the obstacle information in the horizontal and vertical directions,which can reduce the bypass distance caused by the obstacle avoidance link as possible.Thirdly,based on two derived obstacle avoidance algorithms above,the artificial neural network (ANN) is introduced to estimate model uncertainty,and then the tracking control design schemes are established by using feedforward compensation and optimal control technologies.some simulations demonstrate the effectiveness of the proposed obstacle avoidance strategy and control algorithm.

The COVID?19 epidemic is facing the influence of a variety of complex practical factors, which makes the development of the epidemic uncertain. In order to overcome the problem of large error in epidemic forecasting results due to the limitations of many ideal assumptions based on the infectious disease compartment model, a time series forecasting model based on deep learning is adopted to predict the epidemic development, and an informer model based on transformer model is established. Attention mechanism and distillation mechanism are applied to the time series forecasting of epidemic data. The threshold autoregressive (TAR) model and a variety of mainstream recurrent neural time series prediction models are used as comparison models. Through simulation experiments, the current number of remaining infections in the epidemic data of China, America and Britain is predicted in the short term, and RMSE and MAE are used as evaluation indicators, and then the best model is selected for medium ? and long?term prediction. The experimental results show that the indicator value of the informer model is optimal in both RMSE and MAE, further indicating that the prediction accuracy of the informer model is higher than that of other comparative models in China, America and Britain. Finally, the Informer model is used for the development of the epidemic in China，America and Britain medium and long?term prediction.

With the improvement of environmental uncertainty, the demand for supply chain stability of petrochemical enterprises in China is rising day by day. The evaluation of supply chain resilience of petrochemical enterprises has become an important means to judge the risk coping ability of petrochemical enterprises. This paper constructs a supply chain resilience evaluation index system in the petrochemical enterprises. And fuzzy analytic hierarchy process and BP neural network are used to evaluate the toughness strength of petrochemical enterprise supply chain, so as to determine the toughness level of petrochemical enterprise supply chain.It is found that the strength of supply chain toughness of petrochemical enterprises is uneven, and the overall level of supply chain resilience is low. On the basis of the research results, some practical suggestions are put forward for the forging of resilient supply chain in petrochemical enterprises.

Magnetic tomography method has been widely used for nondestructive external inspection of buried and submarine pipelines, which is based on the principle of metal magnetic memory to discern the danger level and location of the stress concentration zone by measuring the anomalies in the spatial magnetic field distribution outside the pipeline. The distribution characteristics and spatial propagation law of pipeline inspection signal detected by magnetic tomography method, the energy distribution and change law of spatial magnetic memory signal in the stress concentration zone of magnetized pipelines are studied in this paper. The magnetic dipole field is used to establish the magnetic field model in the stress concentration zone of the inner wall of the pipeline, and the magnetic energy and energy density of spatial magnetic memory signals under different lift?off values outside the pipeline are finite element calculated based on the magnetic energy theory to derive the distribution law of spatial magnetic field and the correlation of magnetic energy density of magnetic signals under different lift?off value is analyzed. The results show that the spatial magnetic field energy outside the pipe decays with the increase of lift?off value, and the decay is the fastest within the distance of 50 mm from the outer wall of the pipe to the physical force; the correlation of magnetic energy density of different lift?off values shows that the magnetic signal detected by magnetic tomography method outside the pipe is homologous with the signal in the stress concentration zone of the inner wall of the pipe. Theoretically, it explains the effectiveness of magnetic tomography method and also provides evaluation indexes for extracting effective signals from the detection data.

The data of chemical processes often contains dynamic timing characteristics, and traditional fault detection has low usage of dynamic information, which limits the fault diagnosis performance. To address this problem,a new method of chemical process fault diagnosis based on an attention?enhanced encoder?decoder network model (AEN) was proposed. The coding part uses the LSTM to extract the feature information of the process data and combine it with the attention mechanism to utilize the dynamic information among the process data more effectively； the decoding part uses the LSTM and combines the context vector provided by the attention mechanism to provide more accurate state information for the softmax regression, and finally, the softmax regression is used to obtain the probability value of the fault category for each sample data. The introduction of the attention mechanism improves the efficiency of the model in using process dynamic information in the time domain. The proposed method is experimented with using Tennessee Eastman process data and compared with the results of standard PCA?SVM, DBN and ResNet, and the results show that the proposed method is more effective in diagnosing faults.

Recognition of weather phenomena based on images is essential for the analysis of weather conditions. To address the problems that traditional machine learning methods are difficult to accurately extract various weather features and poor in classifying weather phenomena and the accuracy of deep learning for weather phenomena recognition is not high, a weather recognition model based on image block and multi?headed attention mechanism is proposed. The model introduces Swin Transformer into the field of weather recognition for the first time, and adopts a multi?headed attention mechanism combining window multi?head self?attention layer and shifted?window multi?head self?attention layer, whose regionally relevant features extraction capability makes up for the shortcomings of traditional methods and can extract complex weather features from images. The model is trained using transfer learning, and the fully connected parameters of the fine?tuned model are input to the Softmax classifier to achieve recognition of multi?category weather images with 99.20% recognition accuracy, which is better than several mainstream methods in comparison, and it can be applied to ground weather recognition systems as a weather recognition module.

In view of the irregularity of the bottom floor of working face and the diversity of the shape of the flying gangue in steeply dipping coal seam, based on the geographic information system data such as contour line of bottom floor of working face, the 3d grid model of bottom floor is established, combined with the energy tracking method(ETM) C + + programs, four typical shapes of flying gangue with the same mass and different shapes are simulated to obtain the motion trajectories of the migration of flying gangue in the actual working face, as well as the velocity, angular velocity and energy change curves at any time. The influence of the shapes on the motion of flying gangue is analyzed. In order to verify the accuracy and feasibility of the method in this paper, the trajectory simulated by Rockyfor3D software is compared. The results show that the transport capacity of ellipsoidal flying gangue is much higher than that of polyhedral flying gangue. Compared with common polyhedral flying gangue, the regular polyhedral flying gangue has farther migration distance and less energy loss due to collision. The number of edges of flying gangue of regular polyhedron is inversely proportional to the energy loss of flying gangue in collision, which indicates that flying gangue of regular polyhedron with multiple edges is most likely to cause danger．

The vibration signal during the operation of high voltage circuit breaker can reflect the mechanical state of circuit breaker. Aiming at the shortcomings of feature extraction and fault diagnosis accuracy of shallow vibration signal analysis model, a fault diagnosis method of high voltage circuit breaker based on convolutional neural network optimized by genetic algorithm was proposed. Using the global optimization ability of genetic algorithm, the optimal initial network structure parameters and the number of neurons in the whole connection layer were obtained through the selection, crossover and mutation of genetic algorithm to optimize the convolutional neural network, and the optimized convolutional neural network is applied to the fault diagnosis of high voltage circuit breaker. The results show that the diagnosis performance of the proposed network model is better than that of convolution neural network, dynamic support vector machine and multilayer perceptron.

Aiming at the problem of insufficient representation of interactive semantic information in the double human interaction behavior recognition method based on graph convolutional neural networks，a new double human interactive spatial?temporal graph convolution network (DHI?STGCN) was proposed for behavior recognition. The network contains spatial sub?network modules and temporal sub?network modules. Based on the 3D skeleton data obtained from the interactive action video, a spatial action graph of double human interactive action was generated for the representation of spatial information. In the graph, the connecting edges between double human were given different weights according to the joint point position information. The connection of context time information was added in the constructed adjacency matrix, and the joint points in the graph were connected with their nodes within a certain time range in time information processing. The generated spatial?temporal graph data was sent to the spatial graph convolution network module, and the temporal graph convolution network module was combined to enhance the continuity of inter frame motion features for modeling in time. The model fully considers the close relationship of double human interaction. The comparative experimental results on NTU?RGB+D dataset show that the algorithm has strong robustness and obtains better interaction recognition effect than the existing models.

In the research on the thermal efficiency of atmospheric tube heating furnaces, it is necessary to conduct effective online measurement of thermal efficiency and select relatively reliable advanced control methods. On the basis of the research on the combustion mechanism of the heating furnace, the online measurement method based on the principle and data processing is used to process the thermal efficiency, and the "dynamic matrix control" is introduced in the process of optimizing the control of the heating furnace efficiency, which is compared with the traditional control method. The introduction of dynamic matrix control makes the system have a better control effect. At the same time, the "particle swarm algorithm" was selected to optimize the parameters of dynamic matrix control. In the optimization process of dynamic matrix parameters, the particle swarm algorithm relatively shortens the optimization time and improves the control quality, so as to achieve a more satisfactory control effect. Finally, compare with internal model control, it shows that dynamic matrix control can achieve relatively better control effect.

Due to the huge amount of the customers'data the rise of various financial products and the short?term impact of the epidemic, banks are facing with increasing pressure resulting in the business volume declined sharply. The traditional classification tree model can not make more accurate prediction of long?term deposits and carry out accurate marketing to customers according to customer information. Therefore, this paper proposes a three?layer neural network model. Through the experiment, the customer data of grape Island banking institutions are predicted, and compared with the prediction results of a traditional decision tree, random forest model, AdaBoost model and XGBoost model. The experiment shows that compared with the other four models, the neural network model has a better effect of prediction, the model evaluation AUC reaches 0.977 7 and the accuracy reaches 99.06%.

Aiming at the problems of traditional neural network in mining transformer time series relationship, poor generalization of classification and low accuracy of classification for heterogeneous data, this paper proposes an improved transformer fault diagnosis model based on bidirectional recurrent neural network. The model extracts features through bidirectional recurrent neural network, fuses the features at the front and back time, classifies the feature data using multi?kernel learning support vector machine method, and fuses the features in multi?kernel learning support vector machine, so as to improve the accuracy of feature data classification. The accuracy and validity of the bidirectional recurrent neural network based on multi?kernel learning support vector machine model are tested through numerical simulation analysis of the temporal channel length for sequential network diagnostic performance, the influence of and multi?kernel learning on the generalization ability of support vector machines, and the influence on heterogeneous data processing capabilities. The experimental results show that the diagnosis performance of the bidirectional recurrent neural network based on multi?kernel learning support vector machine is better, and the prediction accuracy of the model is higher than that of several commonly used neural networks.

In order to solve the problems of short life cycle and low data throughput caused by too fast energy consumption in cluster head node selection of LEACH protocol in wireless sensor networks, a leach optimization algorithm based on cluster head node energy balanced selection is proposed. The algorithm selects ordinary nodes with high residual energy in WSNs as cluster head nodes, and considers the distance between ordinary nodes and cluster head nodes and between cluster head nodes and base station, as well as the residual energy and average energy of all nodes to select communication mode and transmission path. The proposed algorithm is simulated by MATLAB. The simulation results show that in the 100 m×100 m small and monitoring area and 200 m×200 m large and monitoring area, compared with LEACH, DEEC and IMP?LEACH algorithms, the algorithm reduces the energy consumption of nodes in WSNs, prolongs the life cycle and improves the data throughput.

In the field of point cloud processing, deep learning is a mainstream method, but the existing methods do not fully utilize the local structure information of 3D point clouds, and have less local shape perception. We proposes a 3D point cloud processing model based on improved PoinetNet. Network model introduces position adaptive convolution into PointNet. The position?adaptive convolution constructs the kernel function by combining the weight matrices in the weight bank in a dynamic way, in which the coefficients of the weight matrix are adaptively learned from the relative positions of the points through the position?relative coefficient network. The kernel function constructed in this way can better solve the problem of irregularity and disorder of point cloud data. The classification accuracy of the position?adaptive network in the 3D object classification experiment is 3.60% higher than that of PointNet, and the average intersection ratio in the 3D object part segmentation experiment is 2.20% higher than that of PointNet. In the 3D scene semantics In the segmentation experiment, the average intersection and union ratio is improved by 9.14% compared with PointNet.

The fast braking of resistant load and the falling speed control of the potential energy load are of great significance to improve the operation efficiency and the safety of motor drive equipments. In this paper, a method of constant current braking and dynamic adjustment of falling speed of lifiting objects for DC motor is proposed, in which the conventional energy consumption braking resistor is replaced by a power MOSFET and the equivalent resistance between its leakage and source can be adjusted by voltage control. Under the condition of no need to change the hardware, the double closed?loop control technology of current and speed is adopted to realize the requirements of the whole process constant current fast parking brake, the fast dropping of suspended objects and the falling speed continous adjustment control in the motor drive system. Theoretical analysis and experimental results show that the time for the parking brake and the time for the falling speed of the liftting object to reach stability are reduced respectively to 66.7% and 33.3% of the conventional braking method without chaning changing the hardware system, which effectively improves the dynamic working efficiency of the motor drive system and the flexibility of the control system.

The robustness of the particle swarm system is great, which is very helpful for solving ill?conditioned problems such as image reconstruction. However, the large number of pixels in the reconstructed image leads to a large dimension of particle and it is difficult for the particle to achieve the optimal solution in the optimization process. In order to solve this problem, a constraint is added to the particle position, imaging by Tikhonov regularization algorithm is used as the reference of particle position. The search for particles is constrained to the range of Tikhonov regularization algorithm reconstructs the image. Using the penalty function to solve the constraint problem to improve the particle search speed. Linearly decreasing weights as inertial weights for particle swarms optimization to realize the adaptive dynamic adjustment of the inertia weight and improve the flexibility of the algorithm; the chaotic operator is added to the position search process of the particle swarm optimization, when the particle falls into the local optimum, the chaotic variable will fluctuate within a certain range, reducing the missed rate of the optimal solution. The simulation results show that The improved particle swarm algorithm is more accurate and efficient than the traditional LBP algorithm and Tikhonov regularization algorithm.