16results about How to "Ensure diversity" patented technology

The application relates to the technical field of aluminum alloy structure design, and discloses an aluminum alloy net shell structure design method and system based on a genetic algorithm. The method receives design requirement parameter input by a user, including span, height and load conditions of the net shell structure; divides the net shell structure into multiple design units corresponding to nodes or rod pieces; obtains historical performance data of aluminum alloy material, such as strength, rigidity and durability, from a database; determines the importance level and constraint level of each design unit based on the above data; generates an initial population containing multiple design schemes by using a genetic algorithm; iteratively executes selection, crossover and mutation operations to optimize the design individual; collects structure response data such as stress, strain and displacement in real time in the iteration; confirms whether a preset constraint condition is violated based on the data, and adjusts the genetic algorithm parameter if the constraint condition is violated; and outputs an optimal design result and generates a structure analysis image for the user to view.

The invention relates to the technical field of power system intelligent maintenance, and discloses a high-voltage circuit breaker fault intelligent diagnosis method and system based on deep learning, and the method comprises the following steps: 1) data collection: collecting multi-dimensional operation parameters of a high-voltage circuit breaker through a sensor system; 2) data preprocessing: carrying out noise filtering, abnormal point elimination, data normalization and other operations on the collected data; 3) deep learning model training: inputting historical fault data and normal operation data into a deep learning network for training, and optimizing model weight through back propagation; and 4) real-time diagnosis: inputting the data acquired in real time into the trained deep learning model, and analyzing the data by the model to obtain a fault diagnosis result. According to the method, technologies such as noise filtering (such as wavelet analysis) and abnormal point elimination (such as an isolated forest algorithm) are adopted, so that the data quality can be effectively improved, interference in model training is reduced, and the diagnosis accuracy is enhanced.

An aircraft equipment cabin visual field blind area augmented reality auxiliary assembly method comprises a depth camera self-positioning based on point cloud registration, an assembly object pose estimation based on deep learning, an assembly object pose estimation based on an edge region, a pose estimation method adaptive switching and an auxiliary assembly guide visualization; the adaptive switching method is used to fuse the pose estimation method based on deep learning and the pose estimation method of traditional image processing, so that the robustness, real-time performance and accuracy of the pose estimation are balanced; meanwhile, based on the pose estimation result, the visual field blind area of the assembly process can be visualized by using the augmented reality technology, and the assembly result is detected, which is beneficial to the user to quickly locate the assembly object and complete the assembly task, so as to improve the consistency, real-time performance and immersion of the aircraft assembly operation guide, and further improve the artificial assembly efficiency and quality.

The invention relates to the technical field of autonomous planning navigation, in particular to an anti-interference multi-route autonomous planning navigation method in a polar region environment, which comprises the following steps: environment perception and data acquisition: collecting polar region environment data by using a millimeter wave radar, a laser radar, a meteorological sensor, an infrared night vision device and a high-definition camera. According to the method, a dynamically updated environment situation map is constructed through a multi-source data fusion technology, an accurate and reliable basis is provided for planning, a multi-route parallel generation and multi-target optimization mechanism is adopted, the diversity and robustness of a scheme set are ensured, the anti-interference capability of a decision-making system is enhanced, and the decision-making efficiency is improved. And a whole-course linkage mechanism from real-time interference evaluation to dynamic rolling optimization to closed-loop verification is also established, and self-adjustment and continuous optimization can be carried out according to environmental feedback, so that the autonomy, safety and adaptability of polar navigation are remarkably improved.

The application belongs to the technical field of neural network model training, and discloses a neural network model training method and system based on an improved random configuration algorithm, which comprises the following steps: under the condition of reducing the root mean square error of the neural network model output, screening suitable neurons as candidate hidden layer nodes through the inequality constraint condition of the random configuration algorithm; screening K neurons with the fastest training error reduction from the candidate hidden layer nodes, and selecting the neuron least related to the previous L-1 hidden layer nodes from the selected K neurons as the optimal hidden layer node; calculating the output weight through the least square method, and updating the structure of the random configuration network; and judging whether the network structure is completed by using the maximum allowable number of hidden layer nodes and the maximum allowable output error. The application can reduce the overall calculation amount of the algorithm while ensuring the prediction accuracy, and is suitable for application scenarios with high real-time requirements.

The embodiments of the present invention aim to provide an aperture module and a camera module including the aperture module that can reduce magnetic interference between the AF magnet and the ring magnet. By appropriately adjusting the magnetic pole positions of the AF magnet and the ring magnet, magnetic interference is reduced, making the drive of the aperture module smoother. The correlation between the number of magnetic poles of the AF magnet and the ring magnet can be utilized to ensure the diversity of the number or position of multiple coils, fixed shafts, and rolling components inside the aperture module by increasing the number of magnetic poles of the second magnet. This allows for more free design of the aperture module. Furthermore, two or more position sensors can be asymmetrically arranged according to the coil arrangement to more accurately measure the rotational displacement of the ring magnet. The thrust of the aperture module can also be adjusted by the size, number, and number of turns of the coils.

The invention provides a language model training method and device, and the method comprises the steps: obtaining original dialogue type text data, and converting dialogue type data in the original dialogue type text data, so as to obtain a plurality of pieces of text type data; wherein the original dialogue type text data comprises at least two types of text data; performing text data mixing on the multiple pieces of text type data in a preset mixing mode to obtain target training data; wherein the preset mixing mode comprises a title-free mixing mode and / or a multi-title mixing mode; and inputting the target training data into a to-be-trained language model, and training the to-be-trained language model by using the target training data to obtain a target language model. Through the method and the device, the calculation overhead of the model is controlled while the text expression diversity is ensured, so that the training efficiency and the final performance of the language model are remarkably improved.

The invention discloses a rape phenotype inversion model generation method and device, and relates to the technical field of model training, and the method comprises the steps: collecting rape data of a preset region in a preset time period through an unmanned plane and manual sampling in the preset region of rape growth, vegetation index data collected by the unmanned aerial vehicle and ground actual measurement phenotype data sampled manually are obtained; analyzing the ground actual measurement phenotype data to obtain a phenotype main influence factor; matching and associating the vegetation index data with corresponding ground actual measurement phenotype data to obtain an initial sample set; performing sample layering and sampling on the initial sample set based on the phenotype main influence factor to obtain a training sample set and a test sample set; training the initial machine learning model according to the training sample set to obtain an initial rape phenotype inversion model; verifying the initial rape phenotype inversion model through the test sample set to obtain a verification result; and optimizing the initial rape phenotype inversion model according to a verification result to obtain a target rape phenotype inversion model.

The full-localization video decoding board comprises a processor adopting a domestic chip, an FPGA (Field Programmable Gate Array), a storage unit, an interface unit and a power supply system, the interface unit comprises a first connector and a second connector; a serial port of the processor is connected to the first connector through the serial port conversion chip; the output end of the processor is connected with the receiving end of the FPGA; the FPGA is connected to the second connector through the first conversion chip and the second conversion chip in sequence; the first conversion chip is used for converting an RGB signal output by the FPGA into a VGA single-ended signal and accessing the VGA single-ended signal to the second conversion chip; the second conversion chip is used for converting the VGA single-ended signal into an XGA signal and outputting the XGA signal to the second connector; the FPGA is further connected to the second connector through a third conversion chip, and the third conversion chip is used for converting RGB signals output by the FPGA into VGA single-end signals and outputting the VGA single-end signals to the second connector. The utility model has the advantages of simple circuit structure, low cost, strong compatibility, high transmission stability and the like.

The application discloses a kind of network type converter and electric heat cold energy storage system collaborative optimization control method, contain steps as follows: construct "network type converter-electric heat cold energy storage-flexible load" multi-objective optimization model;Establish the collaborative constraint of network type converter and energy storage system;Establish demand response and network state linkage mechanism;Solve the multi-objective optimization problem by NNC algorithm and combine TOPSIS method to filter optimal operation scheme;According to optimal operation scheme, dynamically adjust the virtual inertia and droop coefficient of network type converter, synchronously dispatch electric energy storage charge-discharge power and cold energy storage and / or heat energy storage electric drive equipment operating state.The method can quickly and accurately suppress new energy output fluctuation and load mutation, strictly constrain system voltage deviation and frequency deviation within a more optimal operating range, fundamentally enhancing the transient stability and dynamic power quality of the grid with high proportion of new energy access.

The pipe network robot path and data station deployment joint optimization method disclosed by the application belongs to the technical field of smart city infrastructure operation and maintenance, and specifically comprises the following steps: firstly, a pipe network-robot-data station system model is constructed, and the pipe network topology is mapped into an undirected graph; secondly, combined with EPANET hydraulic simulation data, the pipe monitoring time is calculated under the conditions of downstream flow and upstream flow, and a hybrid graph with time weight is constructed; then, a weighted optimization mathematical model with the double objectives of minimizing the maximum monitoring time and minimizing the maximum data delay is established, and constraint conditions such as pipe full coverage and robot collision-free are defined; finally, the NSGA-II algorithm is used to generate a Pareto frontier, and the best compromise scheme is selected through weight distribution. The application solves the technical problems that the robot path planning and data station deployment lack cooperation in the existing pipe network monitoring technology, and the pipe network hydraulic characteristics are not coupled, so that the monitoring time and data delay are difficult to consider.

This invention discloses a method for generating Text2SQL datasets based on dynamic difficulty adjustment. The method proposes a complexity metric to evaluate the complexity of SQL statements and text. A large language model is used to generate a Text2SQL dataset based on a prompt word generation template, incorporating dynamic constraints into the template. The complexity of the data generated by the large language model is evaluated using the complexity metric, and compared with a preset difficulty target distribution. The dynamic constraints are then modified to guide the model to generate data that conforms to the expected target distribution. Simultaneously, validation and correction templates are set to ensure the execution and semantic correctness of the generated data, improving the efficiency and quality of dataset construction, ensuring dataset diversity and relevance, and providing the model with more comprehensive and practical training data. This helps improve the generalization ability and performance of the Text2SQL model in different scenarios.

The invention provides a few-sample fracture recognition model training method and system for orthopedic images, and the method comprises the steps: obtaining an orthopedic image original data set with a label, generating a sample global feature embedding vector, and calculating a complexity score of a sample local region; a support set and a query set are constructed in each round of training: the support set firstly extracts a predetermined category, and initial samples are selected from various sample feature spaces by using farthest point sampling; for a fracture positive sample, generating a homologous hard negative example by using a generative image restoration algorithm, transplanting a normalized fracture region to a position corresponding to a healthy sample with a complexity score difference value smaller than a threshold value, generating a heterogenous positive example, and supplementing to form a support set; the query set carries out non-uniform sampling according to the complexity score of a local region from the remaining samples of the extracted category, and the selected probability of the high-complexity sample is in negative correlation with the average prediction loss of the query set; and calculating a prediction result and a loss function, updating network parameters, and calculating the average prediction loss of the query set in the current round for the next round of training.

The application discloses a random collection-based e-commerce platform sales scheme generation method and system, relates to the technical field of e-commerce sales, and generates a first data identifier and a first strategy parameter by using a random parameter generation algorithm through a merchant client, and generates a data collection instruction based on the same and sends the same to an authentication server; the authentication server sends the instruction to a data collection server after verification; the data collection server randomly collects product sales data from multiple e-commerce platforms according to the instruction, determines a second data identifier and a second strategy parameter, and sends the second data identifier and the second strategy parameter to the merchant client after encryption; the merchant client generates multiple sales schemes in combination with the first and second data identifiers and the strategy parameters, and supports individualized adjustment. The application also includes a system, a computer device and a computer readable storage medium for implementing the method. The technical scheme randomly collects and analyzes e-commerce platform data, combines individualized needs of merchants, generates optimized and compliant sales schemes, and improves e-commerce sales efficiency and effect.

An embodiment of the present invention aims to provide an aperture module capable of reducing magnetic interference between an AF magnet and a ring magnet by appropriately adjusting magnetic pole positions of the AF magnet and the ring magnet, and a camera module including the aperture module, so that driving of the aperture module is smoother, and the aperture module is more compact. The aperture module can be designed more freely by increasing the number of magnetic poles of the second magnet by using the correlation between the number of magnetic poles of the AF magnet and the number of magnetic poles of the annular magnet to ensure the diversity of the number or positions of the plurality of coils, fixed shafts, and rolling members inside the aperture module. And more than two position sensors can be asymmetrically arranged according to the arrangement of the coils so as to more accurately measure the rotation displacement of the annular magnet, and the thrust of the aperture module can be adjusted through the size, the number and the number of turns of the coils.

The invention provides a three-dimensional point cloud semantic segmentation method and device and a storage medium, and belongs to the technical field of computer vision and image processing, and the method comprises the steps: obtaining three-dimensional point cloud data in a target scene, constructing a semi-supervised average teacher network, and training the semi-supervised average teacher network through the point cloud data; performing dual-view uncertainty evaluation on each sample in the unlabeled point cloud data in the target scene by using the trained average teacher network to obtain a fusion uncertainty score of each sample, and screening out a to-be-labeled sample from the unlabeled samples by using a dual-drive selection strategy based on the score and a sample confusion degree score; and labeling the to-be-labeled sample, adding the labeled data set, and carrying out iterative training on the average teacher network until a preset condition is met to obtain a final point cloud semantic segmentation model suitable for the target scene. Based on the method, the invention further provides corresponding equipment and a storage medium. According to the invention, efficient point cloud sample selection is realized.