In the face of complex network attacks, in order to ensure network security, an automatic identification method of computer network illegal intrusion based on big data feature mining is proposed. In this method, K-means clustering algorithm is used to cluster computer network data to obtain abnormal data of computer network and reconstruct the phase space of abnormal data. It is used as the input of the least squares support vector machine to establish an automatic identification classifier for illegal intrusion. By constructing a classification plane to automatically identify the types of illegal intrusion in computer networks, the purpose of accurately and automatically identifying illegal intrusion in computer networks is realized. The experimental results show that this method has a good clustering effect on abnormal data of computer network, and can accurately and automatically identify illegal intrusion of computer network. By fully considering the complexity and computational efficiency of the algorithm, redundant calculation and unnecessary memory access are reduced, and this method is feasible to automatically identify illegal intrusion of computer network.
Due to the differences between communication protocols of different video devices,the communication data transmission of device interaction is incomplete,so a communication protocol conversion system for mass video device interaction is proposed.DS80C320 CPU is used to realize communication specification conversion and serial communication,and SDRAM memory is used to convert virtual address into physical address to ensure data integrity and security.The reset circuit is designed with RCM5700 as the core to complete power-on reset,manual reset and periodic automatic reset.The classification model of deep convolution neural network is constructed,and the characteristics of video data are extracted and classified.Read that classified label,setting corresponding function codes,jud whether key information is randomly inserted or not,and realizing a data conversion protocol according to the function codes of the key information.The results of the example show that the maximum error between the accumulated data of the designed system and the actual data is 5 bits,and the communication data can be completely transmitted with high conversion efficiency.
Facial paralysis refers to the abnormal behavior of facial muscles caused by a disorder of the facial nerve, mainly manifested as facial asymmetry. In recent years, deep learning has found extensive applications in facial paralysis detection research. However, most existing methods are constrained to assessing the severity of facial paralysis, thereby concealing crucial symptoms within black-box models. Compared to the severity of facial paralysis, the symptoms of facial paralysis are of greater significance to both physicians and patients. To address this issue, this paper proposes a facial paralysis symptom detection model based on facial action units (AUs). To enhance the accuracy of AU intensity prediction, a novel Difference Ensemble Method (DEM) is introduced. This method leverages differential information between frames within the same video to improve the accuracy of predictions for the current frame. Building upon the predicted AU intensity sequences for keyframes in a video, an interpretable model for detecting facial paralysis symptoms is designed. This model employs an active means to describe the asymmetry in facial muscle strength and utilizes co-occurrence matrices to detect synkinesis. It is noteworthy that DEM is exclusively trained on a dataset of normal faces but exhibits excellent performance when transferred to a facial paralysis dataset. Additionally, DEM exhibits higher accuracy in predicting AU intensity compared to existing methods. The F1 scores for detecting facial muscle function in the eyebrow, eye, and mouth regions with our proposed model are 80.0%, 79.23%, and 90.91%, respectively. To demonstrate the model’s performance, a synkinesis detection experiment is conducted, further validating its applicability in facial paralysis detection.
Aiming at the problem of intelligent supervision and control of power engineering projects,this paper proposes a power engineering data processing and risk identification model based on FCM-IFA-SVM algorithm.The algorithm uses FCM algorithm to cluster and analyze the index data of environmental risk,technical risk,economic risk and management risk,and then uses IF A algorithm to optimize the penalty parameters and kernel function parameters of SVM model.The clustered data set is input into the SVM algorithm after parameter optimization,and the risk grade evaluation result of power engineering project is obtained.The simulation test results of a power engineering data set show that the proposed algorithm has higher accuracy in risk identification.Compared with SVM and FCM-SVM algorithm,the average accuracy is improved by 7.1％and 2.9％,it shows that the proposed algorithm can more accurately identify and evaluate the risk level of power engineering projects.
A set of accurate timing constraints for both top and blocks is an important guarantee of static timing analysis and timing closure in the Integrated Circuit(IC)design procedure.To achieve high efficiency,this paper proposed an automated timing constraints propagation approach,which includes timing constraints demotion and promotion.Comparing to traditional method,the automatic approach avoids the manual process for propagating timing constraints that caused by design and timing iteration.Therefore,it reduces the human effort for the timing constraint generation and speeds up the timing closure period.The experiment result shows that this approach performs high accuracy for both test cases and industrial projects,and significantly improves both efficiency and automaticity for timing constraint generation.
In view of the shortcomings of intelligent triage technology,which is difficult to handle massive data and has strong subjectivity,this paper proposes a virtual triage model for medical institutions based on digital twin technology theory and intelligent perception algorithm.The model is composed of text classification algorithm and image classification algorithm.The text classification algorithm combines Word2Vec and VSM algorithm,which greatly improves the efficiency and quality of the generated text word vector,and also has disambiguation effect.The image classification algorithm uses CNN for feature extraction and RBM for feature classification.In order to reduce the dependence of the algorithm,the diagnosis results are output through the attention mechanism,and the data is fused using the Softmax layer.The experimental results show that the proposed algorithm can effectively and accurately output the diagnosis results,the text classification accuracy of the algorithm is about 3.5％higher than that of the comparison algorithm,and the text classification F1 value is about 3％higher,which shows that the algorithm has good performance and engineering application value.
The operation vulnerability of substation host will reduce the operation speed,cause program collapse and low amount of received data.Therefore,a detection method of substation host operation vulnerability based on state tracking is proposed.The state tracking is used to determine the data security state,and the five tuples of the security state are obtained through data attribute and mapping analysis.The security attributes and executed operations are extracted to determine the credibility of the data to be collected.At the same time,the data states of different time periods are recorded,combined with the conversion function to realize data identification,the state information is copied in memory,and vulnerability characteristics are detected for different types of characteristics.The experimental results show that the designed method can receive 39 GB of data,the vulnerability detection rate fluctuates between 84.52％and 97.33％,and the vulnerability detection error rate is less than 1.41％.
Based on the purpose of optimizing the design of LLC half-bridge resonant circuit and improving the control effect of LLC half-bridge resonant circuit.By replacing two capacitors and one reactor in the LLC half-bridge resonant circuit with analog signals to drive adjustable fixed value components,the experiment of intelligently optimized LLC half-bridge resonant circuit is formed by using one embedded 32bit development chip as the controller,five digital-to-analog conversion chips,and simultaneous access to current sensor and voltage sensor chip.It is concluded that the power consumption of the power supply decreases by 54.17％,the maximum temperature of the thyristor decreases by 29.01％,the voltage peak floating range of the output power supply is compressed by 60.22％,the valley floating range is compressed by 36.03％,and the average value of the adjustment time is compressed by 50.37％.The intelligent LLC half-bridge resonant circuit can provide higher quality power switch control effect.
The accuracy of lightning risk identification is low due to the influence of the identification range and area of large areas of buildings and structures.Design a lightning risk identification system based on Doppler weather radar telemetry for this purpose.The hardware part uses the dual channel receiver Doppler weather radar to receive the scattered wave component,designs the feeder structure composed of Cassegrain antenna,and collects the risk data through the orthogonal mode coupler.The software part obtains the longitude and latitude value of the echo by calculating the relative position of the echo and the radar station in the Cartesian coordinate system system.Establish a grid set of areas containing or partially containing the identified factors,and construct a Doppler weather radar telemetry grid model.Based on this,establish a lightning risk identification model to achieve lightning risk identification.The test results show that when the duration of the system is 11 minutes,the recognition distances of lightning current sequences 1,2 and 3 are 125 m,200 m,and 380 m,respectively,which are consistent with the actual distance and have accurate recognition effects.
There are many problems in the storage of live working tools and instruments,such as high storage requirements,many types of instruments,and difficult access records.An intelligent warehouse for live work tools and appliances based on a distributed control system is designed and built.The temperature and humidity are dynamically adjusted by the temperature and humidity control system to reduce the impact of environmental factors on the preservation of tools.The warehouse dispatching system based on RFID technology is introduced to control AGV to complete the business of entering and leaving the warehouse of tools,avoid the interference caused by non-standard operation.Through the B/S architecture warehouse background management system overall control,clarified the real-time state of the warehouse.Previous tests and trial run show that the design of distributed control system reduced the difficulty of warehouse management,lowered the risk of work industry,prolonged the service life of tools,and finally improved the level of warehouse intelligence.The distributed control system is used to provide a new construction idea for the development of smart warehousing.
The power grid data system is relatively large,and mining the correlation between the high dimensional spatiotemporal data can help to improve the quality of power grid operation.Therefore,a high dimensional spatiotemporal data association mining method based on improved DTW algorithm is proposed.The improved DTW algorithm is used to calculate the distance matrix between the high dimensional spatiotemporal data,to clarify the cumulative result of the distance matrix,and to complete the data preprocessing.According to the frequent items in the recursive item header table in descending order,we calculated the support of spatial data and temporal data and compared them with the set threshold.On this basis,it mines the correlation of high dimensional spatiotemporal data.The experimental results show that the mining results of this method are relatively complete,and the average error is 1.6％,which can provide a reliable basis for the operation of the power grid.
In order to solve the problem that the controllability of multi-agent systems is affected by the destruction of edges,a new method is proposed to identify different types of edges in multi-agent systems based on multi-agent control system theory,combined with state space model and graph theory.According to the failure of edges in the system,the edges are divided into three types and four different combination types,and an algorithm for identifying edge classification is given.In addition,based on this classification,the controllability of edge failure to the system and the change rule of Laplace matrix rank are given.When different types of edge failures occur,the leader selection method is given to ensure that the multi-agent system is controllable after edge failures.The transformation rule given in this paper is verified by an example.
For the gesture control system,a gesture control robot based on Arduino is designed to meet the requirements of reducing the manual labor of traffic police.The system is based on Arduino development board,mainly composed of PAJ7620 gesture control module,ESP8266 wireless transceiver module,SG90 steering gear,power module and so on.Analyze and simplify the gestures performed by the traffic police,and to obtain the angle change data of each joint of the human,and the initial data of each joint of the robot is obtained to realize the gesture planning.The PAJ7620 gesture recognition module detects control gestures and transmits data through ESP8266.After receiving the gesture control data,Arduino controls nine steering gears to make robot has eight kinds of traffic police gestures.The experiment and application shows that the system has the characteristics of correct gesture,beautiful appearance and practicability.
In order to improve the efficiency of line loss control,data fusion and intelligent computing technology are deeply studied in this paper.Based on the circuit superposition principle and Newton-Raphson calculation method,a fusion model based on DBN is constructed.The model takes the active power,reactive power and node voltage of the power grid model node as the model input and the line loss value of the line as the output.It is trained by the gradient rise method.In the training process,the training parameters are adjusted by online learning according to the structure of different networks,the convergence speed of loss function and the updating efficiency of network parameters are guaranteed.The simulation results of line loss calculation in the actual power operation environment show that compared with BP network,the MAPE of this model is increased by 4.019％,the RMSE is reduced by 74.40％,and the calculation efficiency of line loss is significantly improved.
In view of the complexity of human motion posture,and the difficulty and low efficiency of motion posture recognition algorithm based on single feature,a correlation intelligent recognition algorithm based on multi feature fusion is proposed in this paper.The algorithm uses image denoising and other methods to preprocess the collected original image data to highlight the key information in the image.At the same time,Hu moment invariants and other features in the data are extracted by OpenCV,and the extracted features are fused by neural network algorithm.Then SVM is used to establish the classification model,and realize the intelligent recognition of human motion posture.The results of two experiments show that the proposed algorithm can accurately recognize human posture on the premise of ensuring the processing efficiency,and the accuracy rate can reach more than 93％.
In order to solve the problem of poor quality of large-scale data parallel mining caused by excessive data dispersion,a large-scale data parallel mining method based on multi-dimensional assoc-iation rules is proposed.Follow the multi-dimensional association idea to establish the association tree structure,improve the multi-dimensional algorithm according to the RFM value calculation formula,and use the multi-dimensional association rules to build the data set.Calculate the values of nearest neighbor value index and inverse nearest neighbor value index,so as to determine the discrete mining coefficient,and mine large-scale data in parallel with this coefficient.The experimental results show that under the action of multi-dimensional association rules,the value of data dispersion is less than 35％,and the data distribution is no longer sparse,which can effectively improve the quality of large-scale data parallel mining.
Under the background of continuous improvement of big data analysis technology,aiming at the problem of low accuracy of power engineering data analysis and prediction,this paper proposes a power engineering data information analysis and prediction method based on GBDT algorithm.In this method,the loss function is used to represent the negative gradient value,and the regular term is added to the function.By continuously reducing the residual value to increase the authenticity of the data,the regression function is constructed to complete the iterative cycle operation,and then the data is mapped to a new feature space.The two-way comprehensive filling method is adopted to deal with the data with missing features,and the principal component analysis method is used to extract the features of power engineering data,so as to improve the prediction accuracy of the algorithm.The results of example analysis show that when the proposed algorithm is used to analyze and predict power engineering data,its prediction accuracy is high,and the error is less than 5％,which has a certain engineering application value.
Aiming at the problem that BP learning algorithm of least square method requires high acuracy of training data.It is proposed that apply the fair estimation function to the neural network for optimizing the calculation of the neural network.When the fitting data contains gross errors and random errors,the fitting effect of the learning algorithm based on the fair estimation function is significantly better than the traditional BP learning algorithm based on the least square method,and has high robustness.When the algorithm is applied to fit the output characteristics of the thic-kness sensor the experimental results show that the thickness sensor can accurately measure paper thickness with the relative error less than 1.5％.
In federated learning,exchanging model parameters or gradient information is generally considered safe.However,recent studies have shown that model parameters or gradient information can also lead to the leakage of training data.To protect client data security,this paper proposes a federated learning algorithm based on generative model.In order to verify the effectiveness of the proposed algorithm,the simulation experiment was carried out on the DermaMNIST dataset,and the gradient leakage attack was used to verify the algorithm.Experimental results show that the proposed federated learning algorithm based on generative model is only 0.02％different from the classical federated learning algorithm,and can be judged by MSE,PSNR,SSIM and other evaluation indicators so that the algorithm can effectively protect data privacy.
Aiming at the problem of inaccurate voice calibration results of intelligent robot,an automatic voice calibration system of intelligent robot based on deep learning is studied.Design the A/D circuit of voice automatic calibration engine,collect and control the original audio information through the transmission range of analog signal,and receive the audio information by using a compact embedded audio receiver.Sort out and identify the audio information content,and obtain the statement text sample set.The input part of the correction model is constructed by using the sine and cosine function coding processing method of deep learning.The input samples are trained by the feedforward neural network of deep learning to complete the construction of the output part of the correction model.Input the trained samples into the correction model to get the corrected text,and realize the automatic voice calibration of intelligent robot.The experimental results show that the amplitude fluctuation range of the system under the two commands is 9～22 dB and 7～21 dB respectively,which is consistent with the actual amplitude fluctuation and has accurate calibration results.
