A novel decomposition method that combines the improved LHBA (Levy Honey Badger Algorithm) and VMD ( Variational Mode Decomposition) algorithm is proposed. This method is designed to solve the unsatisfactory noise reduction effect caused by inaccurate parameter selection of the VMD algorithm during signal decomposition. Firstly, the LHBA algorithm is utilized to optimize the decomposition mode number K and penalty factor α of VMD. Secondly, the optimized parameters are applied to decompose the VMD signal. Finally, the effective modal component for signal denoising is chosen after calculating the HD ( Hausdorff Distance) between each modal component and the original signal. The experimental results indicate that the proposed method can entirely distinguish the signal component from the noise for simulation signals. Thus, the four evaluation indices of the method are superior when compared to HBA( Honey Badger Algorithm) -VMD, GA(Genetic Algorithm) -VMD, and PSO( Particle Swarm Optimization) -VMD, demonstrating the algorithm's efficiency and superiority. 

For the course “ error theory and data processing ”, which is highly theoretical and has many calculation formulas, while traditional teaching focuses on theory and ignores practical problems, based on Matlab APP(Application) Designer, a comprehensive experimental system APP is designed. The basic concepts such as random error, systematic error and gross error and typical algorithms such as least square fitting are realized and visualized respectively. Based on the actual engineering data of ground nuclear magnetic resonance, the methods for removing systematic errors and gross errors are presented. This comprehensive experimental platform cultivates students ' application ability and completes the organic combination of scientific research and teaching. The experimental platform is convenient for students to understand and master abstract concepts, and improve students ' interest in learning.

In order to solve the chattering phenomenon in sliding mode control of nonlinear the ball and plate system and the multi-disturbance problem in trajectory tracking, we studied a sliding mode control scheme based on a three-step method. Firstly, establish the ball and plate system state-space model, and design the sliding mode control law to overcome the uncertain disturbance of the system. Secondly, combining the three-step control law with the sliding mode control algorithm, design a three-step sliding mode controller to avoid the chattering phenomenon in the sliding mode control. Finally, prove the closed-loop control system stability by Lyapunov function. Compared with the simulation results, this method's trajectory tracking control effectiveness is further verified. 

When the host has intrusion data with delayed response characteristics, the existing judgment mode is disconnected from the delayed data, resulting in distorted judgment of data association confidence between nodes and failure of intrusion detection. A method to judge the confidence of intrusion data association is proposed. Under the host security protection framework, the host firewall packet filtering technology is used to eliminate abnormal data. The security node is placed in the host by distributed deployment, and intrusion detection is carried out by using mathematical model technology. By analyzing the association between normal data, the association confidence between data is determined, and then the intrusion judgment is completed. The experimental results show that the security and effectiveness of the host security protection system are verified by testing the successful times of virus and Trojan attacks with delay characteristics, the time used for packet monitoring, and the functional coverage. 

 The output power of a photovoltaic array exhibits a multi-peak state under partial shading, and traditional MPPT(Maximum Power Point Tracking) control can not solve the multi-peak problem, resulting in the system being trapped in a local optimum affecting the photovoltaic power generation efficiency. To address this issue, a hybrid algorithm is proposed for photovoltaic maximum power point tracking. This method optimizes the initial population of the sparrow algorithm and combines with reverse learning strategy to enhance the algorithm's global search ability. When the algorithm searches near the maximum power point of photovoltaic power generation, the perturbation observation method is used to quickly search the maximum power point by utilizing its fast convergence characteristics. Using Simulink simulation and hardware experimentation, the global search ability and fast convergence ability of the proposed hybrid algorithm are verified. Compared with the sparrow algorithm and perturbation observation method, the hybrid algorithm has significantly improved accuracy and speed

An improved artificial potential field method is proposed to solve the problems of local minimum and unable to reach the target in the path planning of mobile robots. Firstly, in order for the robot to reach the target point when there are obstacles near the target point due to the large repulsive force, a safe distance factor is introduced into the potential field, and this parameter is optimized, so that the robot can maintain a proper distance from the obstacles and reach the target point smoothly. Secondly, in order to solve the local minimum problem, the local minimum discriminant condition is introduced, and the local minimum region is circum- navigated when the condition is triggered, so that the robot can reach the target point smoothly. The simulation results show that the improved algorithm has strong robustness when operating in the map environment with different number of obstacles. The proposed algorithm can make the robot bypass the local minimum area in the U-shaped obstacle environment, and successfully solve the local minimum problem in the mobile robot path planning.

In order to improve the accuracy of short-term power load prediction, a CNN-BiLSTM-Att (Convolutional Neural Network-Bidirectional Long Short-Term Memory-Attention) short-term load prediction model based on variational mode decomposition VMD(Variational Mode Decomposition) is proposed. In this model, the historical load data is decomposed into multiple sub-sequence loads using VMD and combined with weather, date, type of working day and other factors as input characteristics. The predicted value of each sub- sequence load is predicted by this model, and then added and reconstructed to form the actual load prediction curve. By comparison with other models, the VMD-CNN-BiLSTM-Att model has a decrease in the test set. In the continuous weekly load prediction, the average absolute percentage error of daily load prediction is basically maintained between 1% ~ 2% . In the non-working days with complex load changes, the mean absolute percentage error is reduced by 0. 13% compared with the CNN-LSTM model. It is proved that VMD-CNN- BiLSTM-Att short-term load forecasting model can improve the accuracy of power load forecasting. 

In traditional VSG (Virtual Synchronous Generator) voltage and current double closed-loop control, the anti-disturbance performance is poor, and the influence of system parameter changes and uncertainties is great. In order to solve the problems, a nonlinear PID(Proportion Integration Differentiation) based on tracking differentiator is used to control the outer voltage loop and the inner current loop, and the dynamic response of the system is adjusted in real time according to the feedback value of the output signal, so as to achieve the effect of stable output. The correctness and effectiveness of the double closed-loop control of microgrid virtual synchronous generator based on nonlinear PID are verified by simulation. 

In order to solve the problem that experiments of communication engineering professional course focus mainly on simulation, and practice are insufficient, a set of optical camera communication experimental system is designed. Using the host computer to edit information, the system employs a STM32 single-chip microcomputer to drive the high-frequency flicker of the strip-typed LED(Light Emitting Diode) illumination via an amplifying circuit, thus realizing the modulated transmission of information. And the mobile phone with a smart camera acts as a receiver to collect, demodulate and decode the strip information to get real-time information to display, which is implemented by a mobile application using Android Studio. The system integrates a variety of professional and practical technologies, such as embedded software, hardware development and Android development, to stimulate students' interest in learning. The experimental platform realizes coding, modulation, transmission, demodulation and decoding step by step, so that students can have a deep understanding of the principle of optical camera communication system through experiments, which is conducive to the study of professional education theory courses. 

In order to better protect the ecological environment and increase the economic value of recyclable waste, to solve the problems faced by the existing garbage identification methods, such as the complex classification background and the variety of garbage target forms, a residual semantic enhancement network for garbage image classification is proposed, which can strip foreground semantic objects from complex backgrounds. Based on the backbone residual network, the network uses visual concept sampling, inference and modulation modules to achieve visual semantic extraction, and eliminates the gap between semantic level and spatial resolution and visual concept features through the attention module, so as to be more robust to the morphological changes of garbage targets. Through experiments on the Kaggle open source 12 classified garbage dataset and TrashNet dataset, the results show that compared with the backbone network ResNeXt-50 and some other deep networks, the proposed algorithms have improved performance and have good performance in garbage image classification. 

Aiming at the problems that dynamic objects (such as pedestrins, vehicles, animals) appear in visual SLAM(Simultaneous Localization and Mapping) in real scenes, affect the accuracy of algorithm positioning and mapping, the YOLOv3-ORB-SLAM3(Oriented FAST and Rotated BRIEF-Simultaneous Localization and Mapping 3) algorithm is proposed based on ORB-SLAM3. The algorithm adds a semantic thread on the basis of ORB- SLAM3, and the thread uses YOLOv3 to perform semantic recognition target detection on dynamic objects in the scene. The outliers are removed from the extracted feature points on the tracking thread, and the static environment area extracted by the ORB feature, thereby the positioning accuracy of the visual SLAM algorithm is improved. The TUM(Technical University of Munich) data set is used to verify the positioning accuracy of the algorithm in monocular and RGB-D(Red, Green and Blue-Depth) modes. The verification results show that the dynamic sequence of the YOLOv3-ORB-SLAM3 algorithm in monocular mode is about 30% lower than that of the ORB-SLAM3 algorithm in RGB-D mode, the dynamic sequence decreases by 10% , and the static sequence does not decrease significantly.

To address the problem that traditional language models can not solve the problem of deep bidirectional representation and the problem that classification models can not adequately capture salient features of text, a text classification model based on BERT-BiGRU-CNN ( Bidirectional Encoder Representation from Transformers-Bidirectional Gating Recurrent Unit-Convolutional Neural Networks) is proposed. Firstly, the BERT pre-training model is used for text representation; secondly, the output data of BERT is input into BiGRU to capture the global semantic information of text. The results of BiGRU layer again are input into CNN to capture the local semantic features of text. Finally, the feature vectors are input into Softmax layer to obtain the classification results. The Chinese news text headlines dataset is used, and the experimental results show that the BERT-BiGRU-CNN based text classification model achieves an F1 value of 0. 948 5 on the dataset, which is better than other baseline models, proving that the BERT-BiGRU-CNN model can improve theshort text classification performance. 

 Passive acoustic monitoring is carried out during a continuous day and night for cage culture fish in a large semi-submersible platform, named ‘Penghu爷. Basing on the results of time-frequency analysis, the noise frequency band and sound pressure levels of four moments, e. g, manual feeding, automatic feeding by machine, ship disturbance, and quiet state in the middle of the night, are obtained. The results show that the activity of fish measured under large-scale automatic feeding is obviously higher than that of other states, and the sound intensity is about 30 dB higher than the background noise of the marine environment. When the ship passing by, the fish noise intensity is about 9 dB higher than that of the artificial feeding moments. Fish are in a quiet state and not active during late night, and noise sound pressure level is about 70 ~ 75 dB. The time-domain signal is decomposed by VMD( Variational Mode Decomposition). The obtained Hilbert spectrum analysis shows that IMF1 is the high-frequency noise component caused by fish flapping and swimming noise. IMF2 is the vocalization of golden pomfret with a frequency band of 1 100-3 000 Hz. The frequency band of grouper sound is 300 ~ 1 100 Hz, which is the main component of the third order IMF (Intrinsic Mode Function) component. The peak of Hilbert marginal spectrum is ranged in the 600 ~ 700 Hz, which is the frequency band that the highest energy proportion of bio-noise produced by fish. It is expected to provide a basis for bait regulation and population classification of the fish cultured in the large cage by studying the relationship between the vocal characteristics, behavioral state and environmental background.

This work aims to deploy a CNN onto resource-constrained MCUs(Microcontroller Units) to achieve image classification and recognition for scenarios that require simple image recognition tasks, low image recognition accuracy, and low cost. Firstly, a lightweight deployment strategy on resource-constrained MCUs is proposed. To reduce the number of model parameters, a lightweight neural network algorithm is proposed. To ensure that the model size can fit into limited RAM(Random Access Memory), a storage replacement algorithm is presented based on FLASH ( Flash Memory ) sectors. Secondly, the strategy on embedded devices is implemented. The camera peripheral circuit is designed for image quality, but the acquisition speed does not match. The collected images are binarized by an adaptive threshold based on Gaussian distribution and the integrity of image samples is verified. Experimental results show that the system can achieve better image classification and recognition accuracy when applied in the above practical scenarios. 

 In order to solve the problem of unstable lighting conditions, dust and other interference factors when the camera monitors the distribution of mineral materials on the conveyor belt of the coal mine, the effect of directly applying binarization to the camera image to obtain the distribution of mineral materials is unstable and prone to missed inspections, a conveyor belt spill detection technology based on Cycle GAN (Cycle Generative Adversarial Networks) image enhancement is proposed. First, the image of the coal mine conveyor belt collected by the camera is used as input, and the image is enhanced through Cycle GAN; after that, the binary method is used to segment the image to accurately obtain the target area of the conveyor belt; finally, the threshold method and morphological processing are used to analyze the conveyor belt. The belt spraying area is judged and detected. The experimental results show that this technology can effectively monitor the spillage on the conveyor belt, and can improve the monitoring accuracy on the basis of traditional monitoring methods.

In response to the problem of poor classification performance faced by current imbalanced big data classification algorithms, a random forest model based imbalanced big data classification algorithm is proposed. Firstly, the SVM(Support Vector Machine) algorithm is used to filter information on imbalanced big data, and then the anti k-nearest neighbor method is used to detect and eliminate outliers. The singularity of the covariance matrix in imbalanced big data is removed through incremental principal component analysis. And based on the entropy method, weight analysis is carried out to extract imbalanced big data feature information. The CART (Classification and Regression Trees) decision tree is used as the base classifier for imbalanced big data, and a random forest decision tree classifier is constructed. The extracted imbalanced big data feature information is input into the classifier to achieve imbalanced big data classification. The experimental results show that the proposed algorithm has good sampling performance, high classification accuracy, high stability, and high performance for imbalanced big data. 

To solve the problem that brute-force attack poses a major risk to network security, this paper proposes an open source brute-force attack recognition algorithm for big data in cyberspace. The open source network space data information model is constructed, the set of parameter vectors is obtained, and the calculation results of model variables are optimized. Based on ant colony algorithm, the feature optimization is transformed into a path search problem. First, the brute-force attack feature is regarded as a location to be visited by ants, and then the state transition probability is selected to refine part of the search to obtain the global optimal feature. Using the information gain method to measure features, the gain of each feature information in the data set is obtained. By calculating the function of the values between single data sets, the sample difference is measured, the outlier value in the data set is reduced, and the attack behavior is identified by comparing with the threshold value. The experimental results show that the proposed algorithm can accurately identify the brute-force attack, the recognition rate is above 95% , the false positive rate is low, and the recognition effect is the best. 

Due to the excessive number of path hops and the large distance in the optical network, it is more difficult to find the available spectrum resources, which leads to lower dynamic spectrum utilization, less network benefits and higher blocking rate in the optical network. Therefore, a dynamic spectrum allocation method based on frog jumping game optimization algorithm is proposed for the optical network. The OHM(Optimized Link State Routing Protocol using the Highway Model ) routing algorithm is used to select the candidate path that corresponds to the service request and meets the minimum hops and the highest modulation level. The available spectrum resources are found. According to the obtained spectrum resources in the optical network, the minimum of the maximum frequency slot number in all links is used as the target to construct the objective function of the dynamic spectrum allocation of the optical network. Under the constraint conditions, the frog jump game optimization algorithm is used to solve the objective function. The obtained solution is the optimal result of dynamic spectrum allocation in optical networks. The experimental results show that the proposed method has low blocking rate, high spectrum utilization and high network revenue, and is practical. 

Due to the current rigidity of network data planes, which leads to imbalanced data flow in the network, a programmable load balancing algorithm based on data planes is proposed. Firstly, INT ( In band Network Telemetry) technology is used to obtain real-time network status information, and then the proposed BD- ECMP(Bandwidth and Delay Equal-Cost Multi-Path Routing) algorithm is used to select the optimal transmission path for the data stream. Using P4(Programming Protocol Independent Packet Processors) language to optimize the data flow of SDN network data plane, network load balancing is achieved. The simulation results show that compared with the traditional ECMP algorithm, the BD-ECMP algorithm has significant advantages in terms of average flow completion time, network throughput, and network packet loss rate.

To ensure stable network access and reduce resource waste, an open-source big data load balancing optimization algorithm based on real-time node load is proposed. An open-source big data node computing capability model is established, timely feedback and adjustments based on the size of node load are provided, the next action based on the number of requests received by servers in the region is predicted, exponential smoothing method is used to calculate the predicted number of server requests per second, the lag deviation problem of first- order exponential smoothing method is improved, and the comprehensive server load is calculated. Add a load agent and load monitor on the node to balance the number of blocks and the load of sharded nodes, and place undeleted shards and blocks into the minimum unit candidate list to achieve load balancing optimization. Through experiments, it has been proven that the proposed algorithm can improve network resource utilization and load balancing, ensuring a more stable and secure network during access.

A network intrusion detection algorithm that combines systematic data pre-processing and hybrid sampling is proposed for the problem of class imbalance in intrusion detection datasets. Based on the feature distribution of the intrusion detection dataset, the feature values are systematically processed as follows: for the three categorical features, “Proto’’,“Service’’ and “State’’, minor categories within each feature are combined to reduce the total dimension of one-hot encoding; the 18 extremely distributed numerical features are processed with logarithm and then standardized according to the numerical distribution. The class imbalance processing technology, which combines Nearmiss-1 under-sampling and SMOTE ( Synthetic Minority Over-sampling Technique) is designed. Each class of samples in the training dataset is divided into sub-classes based on the “Proto’’,“ Service’’ and “ State’’ categorical features, and each sub-class is under-sampled or oversampled in equal proportion. The intrusion detection model PSSNS-RF ( Nearmiss and SMOTE based on Proto, Service, State-Random Forest) is built, which achieves a 97. 02% multiclass detection rate in the UNSW-NB15 dataset, resolving the data imbalance problem and significantly improving the detection rate of minority classes.

 In response to the problem of many class hours and many students attending classes in public computer laboratories each semester, a set of public computer laboratory information system has been developed independently. The system consists of three parts: student side, teacher side and server. The student side is a desktop program based on Python. The teacher side and the server are implemented in a web project written by JSP(Java Server Pages). In terms of function, the platform can be divided into three basic modules: student sign in, lost and found, feedback. It integrates other common functions. The application results indicate that this system can utilize information technology to provide convenience for students to learn in the laboratory. It truly implements the teaching concept of “student-oriented”

To avoid the gradient vanishing or exploding issue in the RNN(Recurrent Neural Network)-based sequential recommenders, a gated recurrent unit based sequential recommender DeepGRU is proposed which introduces the residual connection, layer normalization and feed forward neural network. The proposed algorithm is verified on three public datasets, and the experimental results show that DeepGRU has superior recommendation performance over several state-of-the-art sequential recommenders ( averagely improved by 8. 68% ) over all compared metrics. The ablation study verifies the effectiveness of the introduced residual connection, layer normalization and feedforward layer. It is empirically demonstrated that DeepGRU effectively alleviates the unstable training issue when dealing with long sequences. 

The traditional methods of manually identifying electrocardiogram signals have problems such as high workload and recognition errors. The existing electrocardiogram monitoring equipment still faces drawbacks such as limited recognition types of electrocardiogram signals, low diagnostic accuracy, and excessive reliance on network services. In order to improve the performance of electrocardiogram monitoring systems, ECG (Electrocardiogram Signals) analysis and detection system is designed based on deep learning technology. SENet-LSTM ( Squeeze-and-Excitation Networks-Long Short Term Memory) network model is built to realize automatic diagnosis of seven categories of ECG signals. The model is deployed on an intelligent hardware platform which uses ADS1292R as the ECG acquisition module, STM32F103 as the data processing module, and Raspberry PI as the central processing module. The system uses the integrated high-performance microcomputer Raspberry PI for calculation and analysis, and provides users with offline AI(Artificial Intelligence) services. The preciseness of the model can reach 98. 44% , and the accuracy can reach 90. 00% , realizing the real-time monitoring and accurate classification of ECG, and providing accurate disease diagnosis for patients.