22results about How to "Reduce calculation" patented technology

This invention discloses an InSAR interferometric network optimization method based on a multi-factor coherence surrogate model. The method includes: constructing a multi-factor feature dataset containing SAR imagery, baseline data, and soil moisture data of the study area; constructing a candidate interferometric pair set, and randomly selecting a subset of sample interferometric pairs from the candidate set; constructing a coherence surrogate prediction model, which uses the multi-factor feature data corresponding to the sample interferometric pair subset and the ground truth values ​​of sample coherence to train the model parameters; inputting the multi-factor feature data corresponding to each interferometric pair in the candidate set into the trained coherence surrogate prediction model and outputting the predicted coherence of each interferometric pair and the optimized interferometric pair network. This invention improves the stability and accuracy of unwrapping and time-series inversion, while also reducing the difficulty of interferometric processing and coherence calculation under large-scale data, and can support the needs of large-scale, long-term, and near-real-time monitoring.

The application provides a road surface state recognition method and device, and solves the problems of single road surface state recognition category, coarse result granularity and inability to distinguish specific road surface abnormal types caused by dependence on simple threshold judgment and shallow features in the prior art. The method comprises the following steps: acquiring a vertical vibration acceleration data segment of a vehicle in operation; extracting a characteristic value according to a preset optimal feature subset, wherein the subset comprises a frequency domain peak value and a spectrum energy; inputting the characteristic value into a classification model to obtain a recognition result, wherein the model is obtained by training based on data comprising at least three road surface state labels. Through fixed feature optimization, the application realizes fine recognition of multiple states such as well frame difference, expansion joint and deceleration strip, and significantly improves the calculation efficiency while ensuring the accuracy.

This invention discloses an adaptive self-triggered optimal control method for catalytic rod reactors, specifically comprising: a temperature field model of the catalytic rod reaction process described by a high-dissipation partial differential equation system based on the control input of the cooling material under time triggering, and obtaining a low-order slow subsystem characterizing the high-dissipation partial differential equation system after order reduction processing; constructing a corresponding Hamiltonian function based on the performance index of the low-order slow subsystem, obtaining the optimal control strategy under time triggering control based on the Hamiltonian function, and then obtaining the optimal control strategy under self-triggered control, converting the Hamiltonian function into HJB equations; constructing a self-triggered controller based on an evaluation-execution dual-network structure using data-driven adaptive dynamic programming, and outputting the adaptive self-triggered optimal strategy based on the optimal value function of the HJB equations.

The application discloses a multi-agent dynamic real-time path planning method and device based on signal timing logic, and relates to the technical field of control methods. The method comprises the following steps: acquiring multi-robot path planning task information and a preset space-time constraint task demand; constructing a multi-robot task specification based on signal timing logic, and defining a corresponding robustness evaluation function; constructing a comprehensive path cost function based on the robustness evaluation function; introducing the path cost function into a path expansion and reconnection process based on a real-time rapid expansion random tree algorithm, updating a search tree structure, and generating a candidate path meeting environment constraints and timing constraints; planning paths for robots in sequence according to a preset multi-robot cooperative planning strategy, taking the path information of the robots that have been planned as dynamic obstacles and performing progressive cooperative obstacle avoidance, and outputting a conflict-free trajectory adaptive to a dynamic environment. The application can solve the path planning problem in a multi-agent scene.

This invention discloses a 3D target detection method based on sparse dynamic attention and star-shaped interaction. The method obtains basic voxel features from the original LiDAR point cloud through voxelization and sparse convolution, then introduces a sparse dynamic parallel attention module. This module achieves efficient enhancement of global context and channel dimensions through dynamic attention branches and parallel channel interaction branches. A sparse star-shaped interaction module is then used to construct a star-shaped neighborhood interaction structure with a central voxel, completing local geometric modeling and nonlinear feature interaction only on non-empty voxels. Finally, keypoint sampling, RoI pooling, and a detection head output the 3D detection box, category, and confidence score. This invention, through the synergistic complementarity of SDPA and SSB, significantly improves the detection accuracy of long-distance, small-scale, and occluded targets while maintaining linear growth in computational complexity and meeting real-time requirements. It achieves balanced performance optimization across multiple categories, including vehicles, pedestrians, and cyclists, and is suitable for 3D perception scenarios with high precision and real-time requirements, such as autonomous driving.

The present disclosure provides a new energy station power prediction method and system based on longitudinal federal decision tree, relates to the field of federated learning and information security technology, and comprises the following steps: acquiring real-time environmental characteristic data; inputting the environmental characteristic data into a power load prediction model to output a predicted power value; wherein the power load prediction model is a longitudinal federal decision tree model, and the training process of the longitudinal federal decision tree model comprises the following steps: each participant generates a key and a hyperparameter for each column of characteristics according to a security requirement, generates a random topological mapping for each column of characteristics, constructs a nonlinear conversion function based on each mapping relationship, encrypts the data in a cascade manner using the generated nonlinear conversion function and a lightweight order-preserving encryption, concentrates the encrypted data in a coordination party for ciphertext state model training, and obtains the longitudinal federal decision tree model after the training is completed. The present disclosure can improve the accuracy of joint power prediction modeling and realize more accurate scheduling optimization.

The application belongs to the technical field of pattern matching and mining under the hypergraph data model, and particularly relates to an incremental dynamic hypergraph matching method centered on overlap, which comprises: on the basis of OHMiner, when the data of a pattern hypergraph changes, partial updating is performed on OIG; when a hyperedge is inserted, new overlap relationships brought by the new hyperedge are calculated, and the new overlap relationships are added to OIG as new nodes in an incremental manner, and the same nodes are merged; when the data of a target hypergraph changes dynamically, DDAL is updated incrementally, and pattern matching is performed in an incremental manner; when a new hyperedge is added, the existing partial matching affected by the new hyperedge is searched in a partial matching state storage library, each partial matching instance is incrementally expanded, and the size of a new overlap is calculated; the size of the new overlap is compared with a requirement in an execution plan to verify, and invalid candidates are discarded, which is incremental pruning. The application introduces a brand-new computing paradigm for dynamic hypergraph matching, and optimizes the performance of dynamic pattern matching.

The application discloses a multi-modal satire detection method based on a large language model, and comprises the following steps: constructing a large-scale high-quality text data set; constructing a pre-training language model; adopting a supervised fine-tuning strategy to optimize parameters of the pre-training language model, training the pre-training language model through a cross-entropy loss function of self-recurrence language modeling, and obtaining a multi-modal large language model; and inputting the large-scale high-quality text data set into the multi-modal large language model for processing, and obtaining a detection result.

This invention discloses a parametric generation method for structured meshes of a three-dimensional pipeline system, comprising: acquiring the geometric structure of the three-dimensional pipeline system; dividing the geometric structure into multiple parts and acquiring the geometric parameters of each part; acquiring the coordinates of the starting section and the coordinates of the center point of the starting section in the coordinate system based on the diameter of the starting section of the geometric structure; calculating the coordinates and normal vector of the center point of the starting section of the geometric structure; performing a spatial matrix transformation on the coordinates of the starting section of the geometric structure based on the geometric parameters and the normal vector of the starting section of the geometric structure to obtain the interface coordinates and the center coordinates; acquiring the coordinates of the interface coordinates; the interface of the geometric structure includes the starting section of the geometric structure; generating a dictionary file of the interface coordinates and corresponding geometric parameters of each part of the geometric structure in the three-dimensional pipeline system according to the blockMeshDict rule using Python; and executing the dictionary file to generate the structured mesh of the three-dimensional pipeline system.

This invention belongs to the field of path planning, specifically relating to a method for obtaining the shortest path between nodes. The method first obtains the sets of cut vertices on the original graph through several traversals, and then maps the cut vertex set subgraphs corresponding to these sets of cut vertices into a high-level tree of nodes. Preprocessing and parallel shortest path queries are then performed using the high-level tree with the cut vertex set subgraphs as the smallest unit, thereby obtaining the overall shortest path. This method transforms the shortest path analysis for a large number of nodes in a topologically complex original graph into a shortest path analysis for cut vertex sets. It eliminates a large number of useless nodes when obtaining the shortest path, saving related calculations and reducing computational load, thus improving the speed of shortest path acquisition. Furthermore, preprocessing is performed on the subgraphs formed by cut vertex sets, avoiding the computationally cumbersome problem of directly obtaining the shortest path for complex topological structures of large subgraphs.

This invention discloses a distributed multi-view visual radar UAV positioning method and system integrating sensing and control, including radar, a visual camera, a data transmitter, a data processor, a distributed sensing terminal, and a computing platform. It designs a sensor control method and a multi-view collaborative fusion wide-area UAV positioning algorithm, utilizing multi-view modalities for UAV positioning to achieve collaborative fusion of distributed cross-modal sensing data and high-precision wide-area UAV positioning. The technical solution provided by this invention can avoid UAV sensing failures and meets the needs of wide-area applications, significantly improving the positioning range and accuracy of UAVs.

This invention provides a fault information conversion and extraction device and a vehicle. The fault information conversion and extraction device includes: a fault diagnosis module, an upload decision module, a fault information conversion module, an on-board bus, and an on-board terminal. The fault diagnosis module is used to diagnose vehicle faults and generate diagnostic messages; the upload decision module is used to send diagnostic messages; the fault information conversion module is used to convert the diagnostic messages sent by the upload decision module into application messages and send the application messages; the on-board bus is used to transmit the application messages sent by the fault information conversion module; and the on-board terminal is used to extract application messages from the on-board bus and send application messages to a cloud server. This fault information conversion and extraction device can process and convert pre-collected vehicle data, converting diagnostic messages into application messages, ensuring data validity, saving bandwidth, facilitating cloud data analysis and rapid response, and reducing additional computation by functional controllers or the cloud.

This application discloses a code processing method, apparatus, terminal, and computer-readable storage medium. The method includes: acquiring initial front-end code; performing code analysis on the initial front-end code using a target code processing model to obtain code risk information; then, determining target risk code segments in the initial front-end code based on the code risk information; performing intermediate representation optimization processing on the target risk code segments using the code risk information to obtain target intermediate representations corresponding to the target risk code segments; then, generating target bytecode corresponding to the target intermediate representations; and replacing the target risk code segments in the initial front-end code based on the target bytecode to obtain target front-end code. This solves the technical problem of relying on manual code optimization when optimizing completed front-end code.

Embodiments of the present application provide a decoding method based on an attention mechanism and a text generation method of a large language model. The decoding method based on the attention mechanism is applied to a first decoding layer of a model based on the attention mechanism. The model includes a plurality of decoding layers in cascade. The first decoding layer is any decoding layer except the first decoding layer among the plurality of decoding layers. The decoding method based on the attention mechanism includes: performing linear transformation on an input vector of the first decoding layer to obtain a first value vector; obtaining an attention matrix of a second decoding layer in the large language model, the second decoding layer being before the first decoding layer; and performing weighted summation on the first value vector according to the attention matrix of the second decoding layer to obtain an attention output of the first decoding layer. The method decodes by obtaining the attention matrix of the second decoding layer, saves the calculation process of obtaining the attention matrix based on the input vector of the first decoding layer, and improves the decoding speed based on the attention mechanism.

The application relates to a kind of implementation methods of verifiable Internet of Things device data asset transaction, based on the data seller, blockchain and data buyer involved in Internet of Things data asset transaction system, the process for realizing verifiable data asset transaction is as follows: data seller publishes encrypted data on blockchain through smart contract, data buyer requests data according to needs and submits verification request.Smart contract is responsible for verifying the integrity and authenticity of data, and transmits decryption key to data buyer after verification.The data buyer decrypts the data using the key and completes the transaction.If the buyer finds that the decrypted data is abnormal, the smart contract is called to initiate arbitration.Through the confusion encryption mode, RSA accumulator technology, hash verification mechanism and on-chain arbitration mechanism, these attack behaviors are effectively resisted, and the privacy, integrity and verifiability of data transaction are ensured.

The invention discloses a catastrophe scene large-range modeling method and device based on unmanned aerial vehicle multi-source scanning data fusion. The method comprises the steps that S110, acquisition and unified organization of catastrophe scene multi-source scanning data are completed through an unmanned aerial vehicle multi-source data acquisition module; s120, sequentially completing data preprocessing, geometric prior generation and multi-source fusion alignment optimization through a multi-source data fusion and geometric optimization module; and S130, completing three-dimensional model achievement generation, model lightweight and block-level incremental updating through a scene reconstruction and incremental updating release module. According to the method, a catastrophe scene large-range three-dimensional model of a unified scale can be rapidly obtained under complex conditions of weak texture, smoke shielding and the like, the modeling calculation overhead is reduced, rapid on-site iteration updating is supported, and the research, judgment and command supporting capacity of an emergency rescue scene is improved.

This invention discloses a method for efficient fine-tuning of a visual Mamba model based on selective prompts, belonging to the field of image processing technology. This invention transforms the fine-tuning process of a pre-trained visual Mamba model into a prompt-based information propagation optimization method by introducing a selective prompt module. This method utilizes the selective prompt module to generate adaptive prompts based on the sample dataset, enhancing the model's responsiveness to input data and thus effectively improving the performance of classification tasks. Specifically, this invention employs a dual-path structure, including cross-layer prompts and intra-layer prompts, optimizing the propagation of shared information between model layers and specific information within layers, respectively. Cross-layer prompts achieve information sharing between different layers, ensuring feature consistency; secondly, intra-layer prompts focus on extracting and maintaining specific features of each layer, thereby improving the model's discriminative ability. This method not only avoids the problem of catastrophic forgetting but also improves the training efficiency and performance of the visual Mamba model in downstream classification tasks.

The application discloses a multi-modal three-dimensional target detection method based on structural feature and semantic feature fusion, and belongs to the technical field of target detection.The application solves the problems of low precision, poor real-time performance and insufficient robustness of the existing method.The application constructs an explicit matching fusion mechanism of low-layer structure guided initialization, high-layer semantic optimization and time sequence expansion, projects 3D boundary boxes of a laser radar and a camera to multi-view image planes, calculates geometric similarity and category consistency constraints of 2D projection regions, and constructs an explicit matching graph in combination with low-layer structure information and high-layer semantic features.The application guides weighted aggregation of cross-modal target features with high confidence through a sparse matching graph, aligns historical frame targets to a current frame coordinate system to construct a space-time matching graph, aggregates historical information through expansion to a time sequence dimension, and realizes efficient and accurate three-dimensional target detection of the laser radar and the visual multi-view camera.The application method can be applied to multi-modal three-dimensional target detection.

This invention proposes a continuous learning method and evaluation method to mitigate the double forgetting of visual language models. The method includes: adding two parallel paths to the backbone network of a pre-trained visual language model—one a task-agnostic expert path and the other a task-related expert group path—to obtain a model based on heterogeneous expert hybridization; training the model using a two-stage training strategy, including: a first stage, freezing the task-related expert group path and training only the task-agnostic expert path, using contrastive learning to solidify the pre-trained general knowledge of the visual language model; a second stage, freezing the task-agnostic expert path and training only the task-related expert group path, using a cross-entropy loss function to learn task-specific knowledge; and an inference stage, dynamically fusing the outputs of the two parallel paths through a hierarchical routing mechanism to achieve collaborative learning of pre-trained knowledge and task-specific knowledge.

The application discloses a traditional Chinese medicine formula effective component analysis system and method, and belongs to the technical field of traditional Chinese medicine analysis. The system comprises a data acquisition interface, a characteristic bin sampling module and an effective component content calculation module. The data acquisition interface acquires a mass spectrum response signal of a traditional Chinese medicine extract and converts the mass spectrum response signal into a digital chromatography-mass spectrum data matrix comprising a plurality of component detection points. The characteristic bin sampling module divides the plurality of component detection points into a plurality of characteristic bins based on chromatography retention time values and mass spectrum mass-to-charge ratio values in the data matrix, assigns corresponding characteristic bin identifiers, and generates a plurality of representative detection points. The effective component content calculation module determines the content or relative peak area ratio of a target effective component. The application can dynamically adjust the characteristic bin granularity, support multi-detection channel parallel sampling, and realize multi-component synchronous identification and quantification in combination with deep learning, and is suitable for efficient and accurate component analysis of a traditional Chinese medicine complex system.

The application discloses a high-voltage switch cabinet based on a modular structure, which has a cabinet body and an external pipe extending from the cabinet body, an instrument room, a handcart room, a busbar room and a cable room are arranged in the cabinet body, an instrument for measurement is arranged in the instrument room, a handcart is arranged in the handcart room, a contact box is arranged in the interior of the cabinet body, the upper contact box is connected with a busbar connecting piece through a busbar, the lower contact box is connected with a current-voltage transformer, a non-flat partition plate is arranged between the busbar room and the cable room, a handcart room cover plate, a busbar room cover plate and a cable room cover plate are arranged above the handcart room, the busbar room and the cable room respectively, a cover plate joint piece which is not integrated with the cover plate is arranged on the handcart room cover plate, the busbar room cover plate and the cable room cover plate, and the safety of the high-voltage switch cabinet is improved.

The application provides a kind of RAG knowledge base retrieval threshold self-adapting dynamic optimization method and system thereof, comprising reading system basic configuration parameter, and based on parameter dynamic calculation maximum shard containing number;Load test question set, execute threshold step scanning to generate candidate threshold;For each candidate threshold, retrieve test question, record the original data of retrieval result under each candidate threshold;Statistical analysis is carried out on the retrieval result, and effective coverage rate and average result number are calculated;Apply constraint-based nonlinear scoring algorithm, combine effective coverage rate and average result number, and introduce load adaptability as penalty boundary, calculate the comprehensive score of each candidate threshold;According to the comprehensive score result, output the threshold with the highest score as the recommended parameter, and generate an analysis report containing the recommended reason.The application effectively controls the context noise and Token cost input to the large model while ensuring the retrieval recall rate, and realizes the automatic performance optimization of RAG system.