9results about How to "Reduce computational efficiency" patented technology

The application discloses a hyperspectral anomaly detection method based on background clustering constraint and potential feature separation, and relates to the technical field of hyperspectral remote sensing detection applications. First, band selection and partition generation of a hyperspectral original image are carried out to generate fixed-size image blocks. Then, a basic image reconstruction framework is constructed based on a lightweight spectral-spatial feature extraction encoder and a convolution decoder. Then, a pixel-level deep clustering network is introduced to complete latent feature clustering modeling, calibrate the background clustering label of the image block, and obtain the intra-class RX detection score in the feature space. Then, the intra-class RX score and the reconstructed image are reversely mapped to the original image space to construct a global RX response map and a reconstructed full image. Finally, a weight matrix is generated by using the reconstruction error between the original image and the reconstructed full image, the global RX response map is modulated, and the final anomaly detection result is output. The application fuses the background clustering constraint and the potential feature separation mechanism, effectively improves the detection accuracy of small anomaly targets, and significantly suppresses false alarm interference.

This invention is a method for inverting azimuth difference seismic data in inclined fractured rocks. Previous AVAZ inversion methods for fracture parameters were mostly based on the HTI medium assumption, with insufficient consideration for fracture dip angle. The fixed form of the inversion constraint terms made it difficult to adapt to changes in parameter statistical characteristics, and the regularization parameters were all manually adjusted, affecting inversion accuracy and efficiency. This invention establishes an approximate formula for the reflection coefficient of the TTI medium, characterized by the P-wave and S-wave velocities, density, and fracture elastic parameters of the background medium. It utilizes pre-stack azimuth difference data to perform AVAZ inversion of fracture parameters. Dynamic sparsity constraints are introduced, dynamically adjusting the sparsity of the constraints based on data characteristics, and a covariance matrix is ​​introduced to consider the statistical regularity between parameters. Combining iterative reweighted least squares and adaptive regularization parameter acquisition methods, the dynamic constraint inverse problem is transformed into a weighted... L A 2-norm objective function is used to achieve a stable solution. Data testing shows that this invention can effectively improve the inversion accuracy of crack elastic parameters.

The technical scheme of the invention discloses a system for constructing a high-precision user behavior portrait through an improved federal learning algorithm, and the system is characterized in that the system comprises a multi-source privacy protection data obtaining module; a federated learning modeling module based on dynamic weighting and contribution proving; a multi-dimensional user behavior portrait module; a personalized vehicle network interaction strategy generation module; a dynamic execution and fair excitation module; and a feedback and model self-optimization module. According to the method, deep analysis and portrait construction can be carried out on historical data of user charging and discharging behavior habits, a high-precision and high-coverage user behavior portrait model is generated through an improved federated machine learning algorithm, the real-time income of users is improved, the user viscosity is enhanced, and the power grid stability and the renewable energy consumption rate are further improved.

This invention discloses a method for processing matrix multiplication weight data in model inference, adapted to large-scale matrix multiplication in BFP format for large models in the Transformer architecture. Addressing the pain points of existing architectures regarding BFP exponent and mantissa splitting, bandwidth, and resource consumption, this method employs (64,64) weight blocks, dual address generator 1:4 scheduling, waterline flow control, and double buffering techniques to achieve full-bandwidth read / write between DDR, BRAM, and computational units. Through multi-core parallelism and multi-round multiplication and accumulation, performance is improved by more than 3 times compared to FP16, reducing memory usage and design complexity while ensuring inference accuracy, providing an efficient hardware acceleration solution for LLM inference in resource-constrained environments.

This invention discloses a multi-robot path planning method and system based on hierarchical control. The method acquires model information and basic parameters of each robot in the system, and constructs a realistic map of the working environment using the SLAM algorithm. The grids of the constructed realistic map are merged, transforming a fine-grained grid map into a coarse-grained one. Considering the overall movement direction and target position of the robots, a global path input for the navigation system is obtained. This global path input is published to each robot, and robots can subscribe to other robots' paths to capture their global path information. Each robot perceives local environmental information in a local map and obtains information about obstacles around other robots. Based on the global path results from the MAPF algorithm, a local path planning method is used to generate local paths, enabling avoidance between robots. This provides a reliable solution for efficient collaboration in multi-robot systems.

The invention provides a mechanism deep learning dual-drive flow-slip disaster numerical simulation system, and relates to the technical field of geological disaster numerical simulation. The system integrates a physical mechanism model and a deep learning method. Based on a high-resolution DEM (Digital Elevation Model), by introducing a rock-soil type self-adaptive substrate entrainment mechanism, utilizing a shallow water wave equation to accurately calculate the terrain elevation, the flow direction, the flow velocity and the accumulation thickness of the first two time steps; and then taking the obtained multivariable field as four-channel input, and driving a CNN-GRU coupling network to perform rolling prediction. According to the method, the final accumulation form, the influence range and the dynamic evolution process can be quickly obtained only through a small amount of initial mechanism calculation, the simulation efficiency is remarkably improved on the premise of ensuring physical rationality, and the defects that a traditional pure mechanism model is long in calculation time consumption and a pure data driving model is physically distorted are overcome; the method is suitable for rapid risk assessment and emergency decision support of flow slip disasters such as landslide and debris flow.

The application provides a global recursive reconstruction method of a space robot recursive Newton-Euler dynamics algorithm, and relates to the technical field of space robot dynamics modeling. The method obtains the global closed matrix form of the mechanical arm movement screw and its differential by forward recursive reconstruction of the recursive Newton-Euler dynamics algorithm; the secondary diagonal matrix of the recursive operator is constructed to simplify the expression of the differential of the global closed matrix form of the mechanical arm movement screw; based on the global closed matrix form of the movement screw and its differential, the global closed matrix form of the required torque of each joint and the required force screw of the base is obtained by backward recursive reconstruction; based on the similarity of the backward recursive reconstruction results at the mathematical structure level, the global closed matrix form dynamics modeling of the space robot is completed. The method constructs an explicit matrix mapping form of the floating base space robot dynamics model, and realizes the direct mapping between the system state quantity and the input quantity.

The application discloses a mountain low-altitude wind field reconstruction system based on sparse observation data and a graph neural network, and relates to the technical field of meteorological prediction, artificial intelligence and low-altitude economic guarantee cross technology, and comprises a data preprocessing module, a multi-scale graph neural modeling module, a physical constraint generation reconstruction module, a terrain self-adaptive migration module and a closed-loop cooperative scheduling module which work in sequence. The data preprocessing module is used for sampling enhancement and feature fusion of sparse observation data and terrain data. The mountain low-altitude wind field reconstruction system based on sparse observation data and the graph neural network solves the problem of insufficient feature extraction of sparse observation data, realizes accurate deduction from discrete observation points to continuous three-dimensional wind field, reduces the prediction error of wind speed, wind direction and turbulence intensity, and significantly enhances the capturing ability of local wind field characteristics such as canyon pipe effect and mountain vortex.

This invention introduces a deep learning method for EDR estimation based on aircraft response characteristics. By combining staged calculation and neural network learning, this method provides an accurate and reliable EDR estimation approach. Unlike traditional EDR estimation methods, the method described in this invention features clearer stage division and dynamic adjustment capabilities. The method includes the following steps: systematically integrating flight data collected from Quick Access Recorders (QARs) of all flights nationwide, constructing time windows, and performing intelligent hierarchical sampling; reading flight data from CSV files and dividing flight stages using a state machine; calculating takeoff and landing stages and cruise stages separately, using raw QAR data to calculate high-level features; finally, training the calculated high-level features using a ConvLSTM-ELM neural learning network model to obtain the optimal parameters of the model. After neural network training and learning, this method can accurately estimate the EDR value of the entire flight route, improving the accuracy and real-time performance of EDR calculation using recorded QAR data in the aviation field.