Patents
Literature
Patsnap Eureka AI that helps you search prior art, draft patents, and assess FTO risks, powered by patent and scientific literature data.

329results about How to "Robust" patented technology

Rainfall kinetic energy space-time prediction method based on dual-polarization weather radar observation parameters

PendingCN119247517AAchieve high-precision forecastingAccurate modeling of spatial and temporal changes characteristicsRainfall/precipitation gaugesWeather condition predictionWeather radarSpectrometer
The invention relates to a rainfall kinetic energy space-time prediction method based on dual-polarization weather radar observation parameters, and the method comprises the steps: firstly processing minute-resolution raindrop spectrometer data, calculating raindrop diameter distribution, and calibrating the data of a rain gauge; secondly, dual-polarization radar parameters (ZH, ZDR and KDP) are calculated based on a physical formula, rainfall kinetic energy KE is calculated in combination with micro-physical characteristics, and rainfall types are marked; and then, after atmospheric field variables are extracted and physical normalization is carried out, a bidirectional time series kinetic energy prediction network (BTKE-Net) is constructed, and rainfall kinetic energy is accurately predicted through BiLSTM and a deep regression network. And finally, after attenuation and offset correction, inputting the data into the trained BTKE-Net model, and carrying out real-time space-time prediction. The innovation of the invention lies in that the application potential of the dual-polarization weather radar in the water environment field is excavated, large-range high-precision rainfall kinetic energy space-time prediction is realized, and more accurate scientific basis and decision support are provided for water and soil conservation and meteorological disaster prevention and control.
Owner:SUZHOU UNIV OF SCI & TECH

Federal learning robust aggregation method based on backdoor attack defense

PendingCN118965415ASolve the problem of distance measurement failure in high-dimensional spaceRobustDigital data protectionBiological modelsRobustificationAlgorithm
The invention relates to a federated learning robust aggregation method based on backdoor attack defense, and belongs to the technical field of artificial intelligence security. The method comprises the following steps: 1, constructing an initialized federal learning global model; 2, training the federal learning global model; 3, aggregating the federal learning global model; and 4, repeating the step 2 and the step 3 until the federal learning global model aggregation reaches the expected precision or the pre-negotiated training round number, and completing the federal learning model training. According to the method, the similarity of key parameters of a federated learning global model is analyzed, dimension reduction and unsupervised clustering are performed on the model parameters, a local agent model is obtained through calculation, a malicious model and a benign model are divided through the cosine distance between the model parameters after dimension reduction, and the performance of outlier detection and unsupervised clustering is guaranteed. And the local agent model is cut through the Euclidean distance, so that the high-amplitude malicious backdoor attack can be effectively resisted, and the robustness of aggregation is improved.
Owner:BEIJING INST OF TECH

Power data enhancement analysis method for enabling Text2SQL (Structured Query Language) based on large language model

The invention relates to an electric power data enhancement analysis method based on large language model energizing Text2SQL. The method comprises the following steps: S1, collecting natural language texts related to electric power and corresponding SQL query pairs; s2, preprocessing the data collected in the step S1; s3, using a pre-trained large language model, performing model fine tuning based on the training data set, and reinforcing the generalization ability of the model in combination with natural language understanding and information extraction to obtain an electric power large language model; s4, expanding the training data set and the test data set; s5, evaluating the performance of the electric power large language model; s6, using adversarial training to enhance the robustness of the model, and using an automatic parameter adjustment technology to optimize model parameters to obtain an optimized electric power large language model; and S7, packaging the optimized electric power large language model into a container for deployment, setting a real-time monitoring mechanism after deployment, and tracking a conversion effect from natural language query to SQL in the electric power system. According to the invention, the natural language query in the power system can be processed more effectively.
Owner:STATE GRID INFO TELECOM GREAT POWER SCI & TECH

Carrier frequency offset estimation method based on Newton orthogonal matching pursuit

ActiveCN119583282AHigh precisionMeet the complexity requirementsMulti-frequency code systemsOff-the-gridEstimation methods
The invention belongs to the technical field of carrier frequency offset estimation, and particularly relates to a carrier frequency offset estimation method based on Newton orthogonal matching pursuit. The method comprises the following specific steps of: splitting CFO into two parts of frequencies, namely off-grid deviation and grid point frequency, and further modeling an observation signal as a parameter weighted compressed sensing problem; using a Newton orthogonal matching pursuit algorithm to iteratively optimize the off-grid deviation of the CFO and the estimation of the grid point frequency, and obtaining a CFO candidate frequency set; and obtaining a required CFO estimation result from the candidate set by using the lead code information of the target user. According to the method, the complexity requirement of an actual system is met, meanwhile, the precision of CFO estimation is remarkably improved, and good robustness is achieved.
Owner:FUDAN UNIVERSITY

Marine weather prediction method and system based on neural network

ActiveCN118962852AReal-time weather forecast resultsHigh precisionWeather condition predictionBiological modelsData setSea waves
The invention provides an ocean weather prediction method and system based on a neural network, and the method comprises the steps: obtaining a cloud layer image data set and a sea wave time series data set of a specific ocean region in a plurality of historical time periods, storing the cloud layer image data set and the sea wave time series data set as a historical data set, and carrying out the preprocessing; constructing a marine weather prediction model; training a marine weather prediction model by using the preprocessed historical data set; and finally, real-time cloud layer image data and sea wave time series data of a specific ocean area are obtained and input into the trained ocean weather prediction model for prediction, and a real-time weather prediction result is obtained. According to the method, data sources are comprehensive, data sources in two dimensions of time and space are combined, weather prediction is carried out based on cloud layer image data and sea wave time series data, and the weather condition of a specific region of the sea can be comprehensively described; meanwhile, the neural network model is trained based on more comprehensive data, and the trained model is higher in weather recognition and prediction precision and has higher robustness.
Owner:SHENZHEN HENGYUAN ZHIDA TECHNOLOGY TRANSFER CENTER CO LTD

Solid-state laser radar and IMU robust tight coupling system under assistance of point cloud intensity

The invention provides a solid-state laser radar and IMU (Inertial Measurement Unit) robust tight coupling system under the assistance of point cloud intensity, which comprises the following steps of: firstly, pre-processing a solid-state laser radar point cloud by the system, eliminating point cloud defect points, correcting a point cloud intensity value and performing down-sampling; and then geometric feature extraction and intensity feature extraction are carried out on the preprocessed point cloud, and intensity calibration of the laser radar point cloud is carried out. And then compensating the motion of the point cloud by using the high-frequency prior information predicted by the IMU. For geometric and intensity feature point clouds, a classical registration method and a fast plane fitting method are respectively adopted to calculate residual errors, and an intensity feature assisted point cloud residual error measurement model is constructed. And a robust strategy based on an IGGIII robust weight function is designed by using a residual error estimation value in a filtering algorithm. According to the invention, on the premise of keeping relatively low calculation cost, the positioning and mapping capabilities and robustness of the small-view-angle laser radar tight coupling system in a complex environment are improved.
Owner:SOUTHEAST UNIV

Multi-target detection method based on multi-source high-frequency track and high-precision map

The invention discloses a multi-target detection method based on a multi-source high-frequency track and a high-precision map, which realizes conversion of a video coordinate system, a millimeter wave radar coordinate system and a world coordinate system and fuses a radar target and a video target. The method comprises the following steps: automatically calibrating video and thunder vision; performing automatic deviation correction on a north corner and an equipment position based on multi-straight-line regression; transforming the video sensor into a world coordinate system; and carrying out decision association of the Leiyu tracking target. According to the method, the target trajectory and the high-precision map data are tracked by utilizing the thunder vision high frequency, the complexity of calibration between the sensors and between the sensors and a world coordinate system is reduced, meanwhile, the precision errors of coordinate transformation between the thunder vision sensors and thunder vision data world coordinate transformation are reduced, and compared with the prior art, the method has the advantages that in the application of complex traffic roadside vehicle detection scenes, the detection efficiency is improved. The precision is higher; and the data association has robustness.
Owner:LIANYUNGANG JARI ELECTRONICS CO LTD +1

Vehicle camera pose determination method and device, equipment, medium and program product

The invention relates to a vehicle camera pose determination method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps of obtaining a calibration image, performing image adjustment on the calibration image according to image adjustment parameters, and performing image detection on the adjusted image to obtain image coordinates of calibration feature points; clustering the calibration feature points according to the image coordinates of the calibration feature points to obtain a plurality of clustering centers and the calibration feature points belonging to each clustering center, removing the calibration feature points whose distances to the corresponding clustering centers are greater than a preset threshold value, and performing calibration on the remaining calibration feature points according to the image coordinates of the calibration feature points; obtaining a position mapping matrix according to the corresponding image coordinates and world coordinates; the calibration feature points with the similar position mapping matrixes are screened out from the remaining calibration feature points, the camera pose of the camera is calculated according to the image coordinates and the world coordinates of the screened calibration feature points, and the calibration precision and the calibration efficiency of the camera pose can be improved under the condition that the image quality is poor.
Owner:深圳市欧冶半导体有限公司

A-ECMS energy management method based on predictive road traffic information

The invention belongs to the technical field of vehicle energy management, and discloses an A-ECMS energy management method based on predictive road traffic information, and the method comprises the steps: obtaining the historical driving data of a vehicle, and calculating the feature parameters of the historical driving data; inputting the characteristic parameters into an off-line driving data recognition library, and obtaining current optimized power control parameters; acquiring predictive road data in front of the vehicle, and inputting the predictive road data into an off-line driving data recognition library for scene recognition to obtain a predicted scene and a corresponding predictive optimization power control parameter; calculating a long-term energy track based on an online dynamic planning algorithm; calculating a final power control parameter based on a DRL algorithm in combination with the current optimized power control parameter, the predicted optimized power control parameter, the long-term energy track and the current energy track; and calculating an optimal control sequence of the vehicle power system in real time according to the final power control parameters. According to the technical scheme, global optimal energy management can be realized.
Owner:DALIAN UNIV OF TECH

Distribution perception test self-adaption method, distribution perception test self-adaption system and distribution perception test self-adaption terminal

PendingCN119106617Areduce updateReduce the number of trainable parametersBiological modelsDesign optimisation/simulationData classData stream
The invention discloses a distributed sensing test adaptive method, system and terminal, and the method comprises the steps: inputting batch data into a deep learning model, determining a first data feature, statistical information and data type of the batch data in a first batch normalization layer, adjusting batch normalization parameters by using a preset weight, and determining a second data feature, statistical information and data type of the batch data in a second batch normalization layer; carrying out batch normalization processing on the batch data by utilizing the adjusted batch normalization parameters; and if the batch data is homogeneous data, updating the batch normalization layer. According to the method, the adaptive capacity of the model during testing is realized through a plurality of batch normalization technologies, so that the model can effectively deal with dynamic data streams under the condition that source data and real labels are not needed, and the model is ensured to keep good performance and robustness in various testing environments.
Owner:SHENZHEN TECH UNIV

Feature selection and multi-scale residual network-based turn-to-turn short circuit fault diagnosis method

ActiveCN119149984AImprove short circuit fault identification accuracyautomatic extractionNeural learning methodsPermanent magnet synchronous motorTest set
The invention discloses a turn-to-turn short circuit fault diagnosis method based on feature selection and a multi-scale residual network, which belongs to the field of motor fault diagnosis, and comprises the following steps: collecting a current signal of a permanent magnet synchronous motor under a turn-to-turn short circuit fault condition in an internal insulation fault of the permanent magnet synchronous motor, and preprocessing the current signal to obtain a pre-processed current signal; obtaining a training set and a test set; constructing a permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis model, training the permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis model by adopting the training set, and keeping the optimal structure of the permanent magnet synchronous motor turn-to-turn short-circuit fault diagnosis model; and putting the test set into the trained permanent magnet synchronous motor turn-to-turn short circuit fault diagnosis model for feature learning, and then inputting the test set into a Softmax classifier for fault diagnosis. The method can automatically and efficiently extract weak fault features in current signals, gets rid of dependence on traditional signal processing technologies and diagnosis experience, and has good robustness for noise-containing signals.
Owner:HUNAN UNIV OF SCI & TECH

Multi-model fusion power daily load online prediction method and system

The invention relates to a multi-model fusion power daily load online prediction method, which comprises the following steps of: perfecting a data set by adopting an auto-encoder and moving average, extracting high correlation characteristics by adopting a unified information coefficient, realizing parameter selection of a KELM model by adopting a hiking optimization algorithm, determining single-day load prediction by utilizing the KELM model, and obtaining a next-day prediction result based on a rolling strategy. And finally, applying a multi-index evaluation model to predict a result.
Owner:CHINA THREE GORGES UNIV

Illumination adaptive road surface detection method and system considering small target area recognition

The invention discloses an illumination adaptive road surface detection method and system considering small target area identification, and relates to the technical field of intelligent perception. The method comprises the following steps: receiving real-time image data, building a horizon line detection network, and inputting the image into the network to obtain an upper half region image containing environmental elements and a lower half region image containing a road surface and obstacles; the day and night state of the driving scene is judged by using the upper half area image, so that the system automatically adjusts model parameters according to illumination conditions; performing semantic segmentation by using the lower half region image, and intercepting a small-resolution image block with road surface information amount according to a semantic mask; a multi-block fusion decision network is built, blocks are input into the network, and the network can adaptively distribute weights of features of the blocks and perform fusion decision on prediction probabilities of the blocks. According to the invention, the road surface detection is expanded to an all-weather scene, the method has higher practical application value, the identification difficulty of a small target road surface area is overcome, and the accuracy of road surface detection is improved.
Owner:SOUTHEAST UNIV

Satellite attitude adaptive control algorithm and system based on DSAC-T

The invention relates to the related field of satellite communication, discloses a DSAC-T-based satellite attitude adaptive control algorithm and system, and solves the problem that the angular velocity of a satellite continuously swings between a positive value and a negative value in the existing algorithm, so that the stable operation of the satellite can be better controlled, and meanwhile, compared with the prior art, the satellite attitude adaptive control algorithm and system have the advantage that the satellite attitude adaptive control efficiency is improved. The DSAC-T algorithm has a faster control speed and faster algorithm convergence, improves the real-time performance of satellite attitude control, improves the anti-interference degree, has good control stability in different interference experiment environments, has good robustness for interference which is not seen in the training process, and can be applied to the field of satellite attitude control. The trained agency also has good adaptability to satellites with different rotational inertia, so that the trained agency can be conveniently migrated to a new satellite, and has better universality.
Owner:MINJIANG UNIVERSITY

Heavy duty gas turbine intelligent control method integrating error active disturbance rejection control and improved hyperbolic sine and cosine optimizer

The invention discloses a heavy duty gas turbine intelligent control method integrating error active disturbance rejection control and an improved hyperbolic sine and cosine optimizer. The heavy duty gas turbine has the characteristics of numerous monitoring parameters, strong nonlinearity and high coupling degree. The method is used for solving the problem that the control effect is not ideal when the traditional PID control strategy faces the system. The method comprises the steps of collecting and preprocessing data of the heavy-duty gas turbine, establishing a Wiener model of the heavy-duty gas turbine based on variational Bayesian identification, designing an extended state observer and an error active-disturbance-rejection control strategy, and designing a multi-target operation optimization target function of the heavy-duty gas turbine; and then solving the optimal solution of the decision variable of the heavy duty gas turbine controller by using an improved hyperbolic sine and cosine optimizer. The invention provides a scheme and thought for the intelligent control method of the heavy duty gas turbine.
Owner:NORTH CHINA ELECTRIC POWER UNIV

Attention area marking method and device based on machine vision, equipment and medium

The invention relates to the field of image processing, and discloses an attention area marking method and device based on machine vision, equipment and a medium, and the method comprises the steps: obtaining an original image of a detection object in a preset range; obtaining priori knowledge of the detection object, and extracting a plurality of key feature sets in the original image according to the priori knowledge; generating a plurality of corresponding feature maps based on the key feature sets, and fusing the feature maps corresponding to each key feature set to obtain a target saliency map; acquiring a detection specification of the detected object, selecting a fixation point from the target saliency map according to the detection specification, and acquiring a plurality of sub-images according to the fixation point; and judging whether the sub-images contain the detection object or not, and marking the sub-images containing the detection object as suspicious areas. The problems that an existing machine vision system is low in processing efficiency, poor in detection precision and prone to being interfered by redundant data when small targets are detected in a large background are solved.
Owner:GUANGDONG COMM POLYTECHNIC

Variable water depth damper and application thereof

PendingCN118959489AReduce the amplitude of motionrecover as soon as possibleNon-rotating vibration suppressionMachines/enginesClassical mechanicsEddy current
The invention discloses a variable water depth damper and application, and belongs to the technical field of ocean engineering. The damper comprises a damper shell with a void, a damper cover plate and a damper base. A combined magnet mass block is arranged in a void of the damper shell, is hung on the damper cover plate through a cable, and comprises an upper metal mass block and a lower magnet block; an arc-shaped copper plate is arranged on the top face of the damper base and located below the combined magnet mass block. During application, the damper is connected with a platform through a mooring rope, and the whole damper serves as a pendulum type tuned mass damper to restrain platform movement. The combined magnet mass block is arranged in the void of the damper shell, so that the whole structure forms the multi-mass tuned mass damper, and the built-in combined magnet mass block can enable the copper plate to generate eddy current and generate corresponding damping force when the whole platform tilts, so that the function of kinetic energy form conversion is achieved, and the structure is kept stable to the maximum extent.
Owner:HARBIN ENG UNIV

Backdoor attack method based on frequency domain, terminal equipment and storage medium

The invention discloses a backdoor attack method based on a frequency domain, terminal equipment and a storage medium. The method comprises the following steps: firstly, converting an original image from a spatial domain to a frequency domain through discrete cosine transform; then selecting an area covering high-frequency information and low-frequency information, and reducing the frequency domain base amplitude in the area; and finally, converting the image from the frequency domain to the spatial domain by using inverse discrete cosine transform to obtain a back door image. Backdoor triggers generated by existing methods are often presented as increments of frequency domain base amplitude, and defense measures based on preprocessing can exactly eliminate the increments, so that existing backdoor attacks are vulnerable to the defense measures. The backdoor trigger mode generated by the method is similar to the existing defense based on preprocessing, the trigger mode has high robustness for the defense and is not easily perceived visually, and the limitation of the existing backdoor attack is broken.
Owner:CENT SOUTH UNIV

Container packing photo detection method and system based on computer vision

The invention relates to a container encasement photo detection method and system based on computer vision, based on a YOLO network and a CRNN network, a container opening and closing state, a container number position, cargo packaging and other target detection model and a container number identification model are respectively constructed, an interface receives encasement photo address, container number and other data uploaded by a business system, and the data are sent to a server; after the corresponding picture is successfully loaded, the specific content of the box type and the box number can be comprehensively and accurately obtained through each optimized target detection model and each optimized box number identification model, finally, the information identified by each model is sorted, and the information corresponding to the box number in the business system is found according to the identified box number for cross verification. According to the technical scheme, the compliance of the uploaded photos is judged, intelligent detection of the boxed photos is completed, the workload of customer service staff is greatly reduced, illegal or problematic photos are effectively recognized, the risk and cost caused by wrong detection or missing detection of the illegal photos are reduced, and the boxed photo detection efficiency and accuracy can be greatly improved.
Owner:SHANGHAI COSCO INFORMATION & TECH

Goods label identification method based on intelligent logistics warehouse management

The invention relates to the technical field of image recognition, and particularly discloses a cargo label recognition method based on intelligent logistics warehouse management, which comprises the following steps: S1, collecting an image in a warehouse by taking a vertex of a three-dimensional grid as a standard point and any standard point as a collection point, and determining a target shape in the image; s2, collecting a to-be-detected image, and determining a to-be-determined shape in the to-be-detected image; determining a standard point corresponding to the to-be-detected image, determining a to-be-compared shape, and determining a comparison shape based on the Euclidean distance between the center of the to-be-compared shape and the center of the to-be-determined shape; and S3, determining the similarity between the comparison shape and the to-be-determined shape, and judging whether the cargo label is an abnormal label or not according to the similarity. According to the invention, the abnormal cargo label is recognized in advance and prompted, so that the inventory management is prevented from being delayed.
Owner:武畅物流科技(上海)有限公司

Partial discharge detection method and device, electronic equipment and storage medium

ActiveCN119044703ARobustSolve the problem of difficulty in detecting weak discharge signalsTesting dielectric strengthAcoustic emissionNoise
The invention discloses a partial discharge detection method and device, electronic equipment and a storage medium, and the method and device are applied to the electronic equipment, and specifically comprise the steps: collecting an acoustic signal generated by the partial discharge of to-be-detected high-voltage electrical equipment, and obtaining acoustic data; based on the acoustic data, extracting a high-frequency acoustic signal through high-pass filtering, and calculating the power spectral density of the energy envelope of the high-frequency acoustic signal; and identifying the discharge type of the high-voltage electrical equipment based on the energy distribution of the power spectral density of the energy envelope at a specific frequency. The scheme has robustness to external noise interference, the problem of noise sensitivity existing in a traditional partial discharge detection method (such as PRPD atlas analysis) based on the acoustic emission technology is effectively solved, and therefore the detection capacity of weak discharge signals is remarkably improved.
Owner:BEIJING DISHENG TECHNOLOGY CO LTD

Water-wind-light short-term complementary scheduling method under finite probability distribution information

The invention discloses a water-wind-light short-term complementary scheduling method under finite probability distribution information, and relates to the field of short-term scheduling of a novel power system. Collecting and sorting initial calculation data, and constructing a moment fuzzy set of new energy output prediction errors by using finite probability distribution information; according to the moment fuzzy set, constructing a hydroelectric flexibility reserve constraint condition based on distribution robust chance constraint planning, and quantifying the hydroelectric reserve capacity; taking the minimum average absolute value of the average distance of the residual load of the power grid as a target function, and providing flexible reserve to compensate a new energy output prediction error while peak regulation is carried out on cascade hydropower; introducing a Chebyshev inequality to convert the distribution robust opportunity constraint to obtain a deterministic linear programming model; and solving the model by using an optimization solver. According to the method, the peak regulation flexibility and the standby flexibility can be provided by reasonably utilizing the cascade hydropower, so that the problem of reverse peak regulation caused by large-scale new energy grid connection and the problem of power imbalance caused by new energy output prediction errors are effectively solved.
Owner:DALIAN UNIV OF TECH

Dry-method pole piece calender control method based on AI algorithm and calendering system

The invention discloses a dry-method pole piece calender control method and calendering system based on an AI algorithm, the calendering control system comprises a roll gap control subsystem, a belt speed control subsystem, a micro-tension control subsystem and a deviation rectification subsystem, a control center of each subsystem adopts a zebra optimization recurrent neural network algorithm (RNN-ZOA) to calculate signals, and the signals are sent to the control center of each subsystem; according to the control method, reasonable control quantity is predicted, the obtained control quantity serves as an input vector of a PID controller, output quantity obtained through calculation of the PID controller is output to a corresponding execution mechanism, and corresponding parts are adjusted. Each control subsystem of the calendering system introduces a zebra optimization recurrent neural network algorithm to be deployed in the PID controller, so that the control performance of the system is improved, the stability and the accuracy of the system are improved, and the production efficiency and the product quality of the calender are also improved.
Owner:WUXI RICH INTELLIGENT EQUIP CO LTD

Fault classification system and method based on distributed device DTW method

The invention discloses a fault classification system and method based on a distributed device DTW method. The fault classification system comprises a knowledge sample module, a data preparation module and a fault classification module. The knowledge sample module is connected with the fault classification module, the knowledge sample module can be updated according to faults provided by the fault classification module through interconnection of the knowledge sample module and the fault classification module, so that the intelligent degree of the knowledge sample module is automatically improved, and the data preparation module is connected with the fault classification module. The knowledge sample module is used for initializing a knowledge sample library and dynamically updating data of the knowledge sample library. The data preparation module is used for signal acquisition and fault detection and positioning. And the fault classification module is used for weighting coefficient calculation, sample similarity calculation, category similarity calculation and classification result determination. According to the method, DTW method similarity calculation is carried out by using fault data and traveling wave current data in a system sample library, the similarity of each fault category is selected by using distributed sensor data results at different positions, and finally, the category with the highest result similarity is determined as the fault type.
Owner:CHENGDU FOURIER ELECTRONICS TECH

Drainage tunnel environment intrusion event identification method based on distributed optical fiber monitoring

The invention discloses a drainage tunnel environment intrusion event identification method based on distributed optical fiber monitoring, and the method comprises the steps: firstly laying a distributed optical fiber monitoring system which adapts to a drainage tunnel environment, and calibrating the position of a distributed optical fiber sensor in a knocking manner; judging whether the signal at the drain pipe is a seepage water release event or not by using a template matching method, and qualitatively identifying the seepage water flow according to a threshold value; then constructing a distributed optical fiber sensor array signal library adapted to a drainage tunnel environment, realizing DAS array signal time-frequency conversion by using continuous wavelet transform, constructing an intrusion event identification model based on a LeNet convolutional neural network and a random forest, and finally outputting the type and position of an intrusion event. According to the method, seepage water release is recognized as an independent intrusion event, and the problem that the seepage water release event exists for a long time is solved; influences of intrusion events outside the drainage tunnel are eliminated by separation of the sensors inside and outside the drainage tunnel.
Owner:THREE GORGES JINSHAJIANG CHUANYUN HYDROPOWER DEV CO LTD

Visual repositioning method and system and storage medium

The invention provides a visual relocation method and system and a storage medium. The method comprises the steps of obtaining image information of an environment where a vehicle is located at the current moment; performing feature extraction on the image information to obtain pixel-level feature points, local features and global features; wherein the local features and the global features are extracted according to a deep learning model, and the pixel-level feature points are extracted according to a pixel-level feature detection algorithm; matching the pixel-level feature points, the local features and the global features according to a map model, and selecting candidate frame image features; and performing pose recovery on the vehicle in the map model according to the candidate frame image features, and outputting visual relocation pose information. According to the method, the problem of mismatching of visual relocation is solved, the problem of candidate frame selection abnormity in scenes such as visual scene degradation and illumination variation is solved, the pose recovery precision is improved, and the relocation success rate is improved.
Owner:SHENZHEN DESAY SV AUTOMOTIVE CO LTD

Electric power communication system robust routing method based on multi-agent deep reinforcement learning

The invention discloses an electric power communication system robust routing method based on multi-agent deep reinforcement learning, and the method comprises the steps: building an information flow model of an electric power communication system, taking a routing node as an agent, designing an observation space, an action space and a reward function of the agent according to a routing optimization target, and carrying out the routing optimization. A routing optimization problem is converted into a multi-agent deep reinforcement learning problem, a routing strategy network and an improved commentator network are designed for each agent, parameters of the routing strategy network and the improved commentator network of each agent are alternately updated by adopting a Soft-Actor-Critic algorithm, an optimal routing strategy network of all agents is obtained, and the optimal routing strategy network of all agents is obtained. And the routing module is used for outputting an optimal routing decision, and when the power communication system is subjected to network attack or has a fault, different intelligent agents maintain the transmission delay of a normal measurement data packet within an allowable range through the cooperation of the optimal routing decision. The method can be applied to millisecond-level control electric power communication scenes, and the communication delay and the packet loss rate of an electric power communication system are minimized.
Owner:SOUTH CHINA UNIV OF TECH

Image deblurring method based on multi-scale feature weighting

The invention discloses an image deblurring method based on multi-scale feature weighting, and the method comprises the steps: collecting fuzzy ship sign picture data of a real scene to obtain a preliminary data set, and estimating a basic fuzzy kernel and a mixing coefficient from the preliminary data set through employing a fuzzy estimation network; obtaining a group of clear ship license pictures, converting the clear ship license pictures into fuzzy ship license pictures by using a basic fuzzy kernel and a mixing coefficient, and gathering each clear-fuzzy ship license image pair to obtain a clear fuzzy data set; constructing a deblurring model which is based on the multi-scale feature extraction network and is fused with a feature weighted fusion module; training a deblurring model by using the clear fuzzy data set, and optimizing the deblurring model through a loss function based on a frequency domain; and inputting a blurred image to be processed into the trained deblurring model to obtain a clear image. According to the method, detail features can be better restored, and the deblurring effect and the overall definition and accuracy of the generated image are improved.
Owner:HANGZHOU YINGLIAN TECH CO LTD

Data expansion method based on hierarchical question generation and self-gaming

The invention discloses a data expansion method based on hierarchical question generation and self-gaming, which comprises the following steps: firstly, designing a hierarchical question generation model, and predicting a next question by utilizing a target query statement and a current query statement through four steps of clause decomposition, predicted clause set generation, sub-question conversion and question combination. On the basis of existing multi-round medical Text-to-SQL data, new data is circularly generated by means of a self-game thought. The layered question generation can better process complex SQL statements, more natural and accurate questions are generated, and the diversity of dialogue streams is increased. The expansion data set generated through self-gaming can effectively relieve the problem of scarcity of multiple rounds of Text-to-SQL data sets in a medical consultation scene, the generalization ability of the retrained Text-to-SQL model is improved, and then the model performance is improved.
Owner:HANGZHOU DIANZI UNIV

Offshore wind plant centralized active harmonic suppression method and device and storage medium

The invention discloses an offshore wind power plant centralized active harmonic suppression method and device and a storage medium, and the method comprises the following steps: building a grid-connected system overall impedance model based on an offshore wind power plant grid-connected system according to the transmission relation between an AC harmonic voltage vector and an AC harmonic current vector, adding a control model to the frequency domain impedance model of the single SVG, obtaining the optimal parameters of the control model based on the control model and the overall impedance model of the grid-connected system, and obtaining a control signal according to the control model and the optimal parameters; and adding a control signal in a control loop of a corresponding single SVG in the offshore wind plant grid-connected system so as to suppress harmonic waves in the offshore wind plant grid-connected system. According to the centralized active harmonic suppression method for the offshore wind plant, additional sampling equipment does not need to be added, the economical efficiency is good, the robustness is high, the operation stability of a grid-connected system of the offshore wind plant is improved, and the electric energy quality is improved.
Owner:NORTH CHINA ELECTRIC POWER UNIV +1