57 results about "Temporal models" patented technology

The invention relates to the technical field of large model reasoning, in particular to a reasoning performance optimization method and system based on a large model, and the method comprises the following steps: initializing a reasoning performance optimization agent; collecting hardware environment indexes in real time, wherein the hardware environment indexes comprise a video memory utilization rate, a CPU (Central Processing Unit) exchange number, residual video card resources, storage IOPS (Input / Output Per Second) and network throughput; the method has the beneficial effects that related statistical indexes, including model types, model weight file total volume, model average sequence length, the number of tokens per second output by the model, first token time of the model, a display card list occupied by the model, the size of a KVcache block, the size of a KVcache sliding window and the like, of each model in a service system are comprehensively collected and analyzed; the system performance is comprehensively evaluated, and the defects that in the prior art, performance evaluation is not comprehensive, and a real-time monitoring mechanism for key indexes such as memory occupation and network bandwidth is lacked are overcome.

The invention provides an aerospace craft discrete trajectory precision reconstruction method based on a variable order, and the method comprises the following steps: calculating the order of a reconstruction model based on a given highest-order dynamic parameter of an aerospace craft discrete trajectory; calculating a coefficient of the reconstruction model; based on a given aerospace craft discrete trajectory, constructing a continuous time model of a flight trajectory; and substituting any moment into the continuous time model of the flight trajectory to obtain trajectory data of the aerospace craft at any moment. According to the method, reconstruction can be completed with higher precision based on given discrete trajectory data, and trajectory continuity is considered; order configuration can be flexibly carried out, and the adaptability to discrete trajectories with different dynamic change rules is improved.

In various examples, machine learning models that benefit from temporal context while being computationally efficient to train and use are described herein. For instance, the disclosed systems and methods may apply a temporal series of images to a model and use intermediate features output from one or more backbone layers of the model as training data. In some examples, one or more recursive layers and / or one or more head layers of the model—or another model—may be trained using the training data by applying the intermediate features to the recursive layer(s). The recursive layer(s) may output a state representative of a temporal combination of the intermediate features, and the state may be applied to the head layer(s) to make one or more predictions. During inference, the recursive layer(s) may, in some examples, continuously update the state based on previous states of the recursive layer(s).

The invention belongs to the technical field of control theories, and provides a ZNN model design method for solving a time-varying Sylvester equation based on a filter, and the method comprises the steps: building an error function based on the basic knowledge of the time-varying Sylvester equation; designing a predefined time function based on a high-gain method, introducing the predefined time function into a ZNN model, and considering unknown noise interference in a ZNN model solving process; for unknown noise interference, a filter containing a predefined time function is introduced, and a subsystem of a neural network model is constructed; a Lyapunov function is constructed, a ZNN is obtained through stability analysis, noise can be effectively suppressed, and a time-varying Sylvester equation is solved at predefined time; and solving a time-varying Sylvester equation by using the predefined time ZNN model based on the filter, and verifying the performance of the predefined time ZNN model based on the filter. The method is used for solving a time-varying Sylvester equation, and the influence of noise can be effectively suppressed.

The invention discloses a man-machine cooperation disassembly task dynamic planning method under uncertain operation time. The method comprises the following steps: 1) constructing a man-machine cooperation disassembly information model under the uncertain operation time; 1.1) constructing an uncertain operation time description model of the product parts; 1.2) constructing a disassembling tool, direction and station position switching time model by considering the operation characteristics of the robot and the disassembling personnel; 1.3) constructing a product part disassembly constraint relation model; 2) taking the man-machine cooperation disassembly information model under the uncertain operation time constructed in the step 1) as an environment of a reinforcement learning algorithm, and establishing a man-machine cooperation disassembly task dynamic planning model; and 3) using the trained man-machine cooperation disassembly task dynamic planning model to generate a man-machine cooperation disassembly task planning scheme under the uncertain operation time. The uncertainty of the structural adhesive softening time is quantified by constructing the uncertain operation time man-machine cooperation disassembly information model, and the characterization problem of the disassembly operation time is solved.

The invention relates to a two-dimensional multi-view brain tumor medical image segmentation method and a two-dimensional multi-view brain tumor medical image segmentation system based on a Vision Mama time sequence model, which utilize a novel visual representation model to complete focus segmentation of a brain tumor two-dimensional medical image. The method is used for solving the problems that when an existing two-dimensional segmentation method is used for processing two-dimensional brain medical images, space depth information cannot be fully utilized, and three-dimensional structure features are difficult to accurately capture. The method comprises the following steps: 1, acquiring and preprocessing data; 2, constructing a multi-view brain tumor segmentation network based on edge feature fusion and a spatial state model; 3, constructing a combined loss function of weighted cross entropy and weighted Dess loss, and meanwhile, storing an optimal model weight in training for prediction; and 4, predicting a brain tumor medical image by using the trained optimal model, calculating evaluation indexes and performing result comparison. Through the combination of the methods, the boundary feature extraction quality of the fuzzy edge of the complex focus and the two-dimensional segmentation precision of the model on the brain tumor are effectively improved.

According to the multi-sensor fusion evaluation algorithm and device based on the continuous time sequence, modeling is carried out based on continuous time, physical reality is better met, motion modeling with better precision is achieved, intra-frame motion can be accurately described, smoother and more accurate track estimation can be provided, and the method and the device are suitable for being applied to multi-sensor fusion evaluation. Especially, the advantages are obvious in high-speed and high-frequency vibration scenes; meanwhile, according to the technical scheme disclosed by the invention, when LiDAR / Camera matching fails, the constraints of the IMU and the GNSS are seamlessly and continuously applied to the whole track through a continuous time model instead of only acting on a discrete frame, so that error accumulation is greatly inhibited, system collapse is avoided, the robustness of a degraded scene is enhanced, application in specific fields such as surveying and mapping is greatly facilitated, and the method is suitable for popularization and application. And the method has the capability of flexibly adding constraints.

Systems and methods for evaluating risks for physical locations are disclosed. A request for risk evaluation of transactions associated with a physical location located within a predetermined region including a plurality of neighboring locations is received. A spatial model is implemented to generate a first risk score based on chargeback data associated with the physical location and chargeback data of the plurality of neighboring locations. A temporal model is implemented to generate a second risk score for the physical location. A final risk score for the physical location is generated based on the first risk score and the second risk score. The final risk score is transmitted to a computing device for detecting fraudulent transactions associated with the physical location in response to the request.

The invention discloses a simulation task load dynamic change-oriented end-cloud integrated cluster dynamic container capacity planning method, and relates to the technical field of computer services. Modeling a micro-service, a container, a user and a server; load sensing and triggering judgment, including calculation of micro-service total load and load variation, calculation of load intensity and triggering condition judgment; mDP differential capacity adjustment: constructing an MDP state space, formulating a container capacity expansion and shrinkage strategy, and calculating an adjustment correction coefficient; key index calculation including a response time model, a system total cost model and a fairness index model; and carrying out multi-objective solution and verification, namely constructing an objective function, setting constraint conditions and solving by adopting fusion MDP. Hierarchical modeling is carried out by adopting micro-services, containers, users and servers, MDP decision differentiation adjustment is triggered through load sensing, multi-target solving is carried out, a four-layer technical architecture of modeling-sensing-decision-solving is formed, and dynamic optimization of cluster capacity can be realized.

This invention relates to the fields of intelligent transportation and road construction management technology, and particularly to a multi-scale traffic operation state prediction method and system. The invention achieves high-precision traffic state prediction by constructing a three-layer spatial topology architecture, designing and building a dual-flow independent LSTM architecture, and establishing a recursive rolling and interactive input mechanism. Under complex conditions such as lane closures, temporary traffic diversions, and perception blind spots, this invention, based on a hierarchical spatial perception system and a dual-flow deep temporal model, can perform high-precision short-term predictions of micro, meso, and macro traffic states, providing decision support for lane-level refined management and control in construction areas.

This invention belongs to the interdisciplinary fields of marine engineering, marine geotechnical engineering, and artificial intelligence. Specifically, it relates to a method for predicting the seabed response around pile foundations based on a deep learning time-series model. The method includes generating multi-point time-series physical field data of multi-directional wave and seabed dynamic response through CFD-FEM coupled simulation, followed by random spatial sampling processing; enhancing wave nonlinear characteristics using multi-order difference; extracting spatiotemporal dynamic features using TimeDistributed-LSTM; constructing a four-dimensional spatiotemporal tensor based on relative coordinates and Euclidean distance; aggregating local and global features using a local multilayer perceptron and max pooling; and finally, using a multi-output decoding network to predict the seabed pore water pressure response in parallel. This invention eliminates dependence on fixed grids and measurement points, exhibits strong spatial adaptability and generalization capabilities, and can achieve real-time, high-precision prediction of the dynamic response of the seabed around pile foundations.

A predefined time ZNN model design method for solving pseudo-inverse of a time-varying matrix comprises the steps that a pseudo-inverse matrix is solved based on the time-varying matrix, an error function is established, the time-varying matrix is known, and the pseudo-inverse matrix to be solved is unknown; designing a predefined time function based on a high-gain method, and introducing the predefined time function into the ZNN model; aiming at bounded noise interference in a ZNN model solving process, introducing a disturbance estimator comprising a predefined time function, and constructing a subsystem of a neural network model; a Lyapunov function is constructed for stability analysis, and a ZNN model obtained through analysis can effectively suppress noise and solve a pseudo-inverse matrix of a time-varying matrix at predefined time. And applying the predefined time ZNN model to control of the permanent magnet synchronous motor chaotic system. According to the designed predefined time ZNN model, a complex parameter adjustment process is avoided in parameter adjustment. The method has the advantages of high robustness, high calculation precision and high solving speed.

The application discloses a terrain self-adaptive rainstorm waterlogging spatio-temporal joint prediction method and system, and belongs to the field of alarm responding to disaster events. In order to overcome the defects of the existing method, such as insufficient terrain adaptability of fixed feature weight, and prediction fragmentation caused by independent operation of the spatio-temporal model, the technical scheme of the application is as follows: on the one hand, a terrain self-adaptive dynamic feature weight mechanism is constructed, an adjustment coefficient is generated by quantifying the terrain undulation, and the feature factor weight of the geographical spatio-temporal weighted regression model is dynamically adjusted to adapt to the complex underlying surface; on the other hand, a spatio-temporal joint weighted fusion framework is proposed, a random forest spatial model and a PredFormer time series model are combined, the spatial prediction result is taken as the initial condition of the time series, and the time series collaborative prediction of the flooded range and the water depth is realized. The application is suitable for rainstorm waterlogging prediction of different terrains in cities, and has the advantages of strong terrain adaptability, high prediction accuracy and stability, and unified spatio-temporal rapid response.

This invention relates to a method, device, and medium for multi-agent coverage search and task allocation path optimization. The specific method includes: scenario construction and problem description, which describes the characteristics of the task area, the attributes of agents and targets, and clarifies the area coverage search problem and the target task allocation problem; area coverage model design, which establishes a minimum task time model objective function and constraints, and then solves the area coverage model; and task allocation strategy and path planning algorithm, which, considering agent speed and target attributes, combines a greedy strategy and a genetic algorithm to allocate target exploration tasks to agents, optimize target exploration paths, and reduce total task time. Compared with existing technologies, this invention improves the efficiency of marine area search and target exploration through multi-agent cooperation, achieving full area coverage search and accurate target exploration.

This invention discloses a method and system for evaluating the quality of immersive videos based on multimodal perception, relating to the field of video evaluation. The method includes: acquiring a multi-viewpoint texture and depth format video and extracting keyframes; extracting texture and depth features using ConvNeXt, processing texture features through a frequency-space texture enhancement module, and combining depth features with a cross-modal collaborative representation module to obtain structural texture coupling features; extracting semantic distortion features using a semantic distortion perception module; concatenating the coupling features and semantic distortion features and inputting them into a temporal modeling module to capture temporal features, then generating global perception features through a viewpoint fusion module; and finally outputting the final video quality score through a quality regression module. This invention achieves a more accurate and robust evaluation of immersive video quality by fusing multi-viewpoint texture and depth information and sequentially performing frequency-space texture enhancement, cross-modal collaborative representation, semantic distortion perception, temporal modeling, and viewpoint fusion.

The invention discloses a modular multi-level converter sub-module working condition simulation test method and a modular multi-level converter sub-module working condition simulation test system. The method comprises the following steps: establishing a mathematical model for describing the relation between current and voltage based on a submodule circuit structure and a working principle so as to obtain a continuous time model; discretizing the continuous time model to obtain a discrete state space model for MPC prediction and control optimization; designing a cost function; and in each control period, the system behavior is predicted again based on the current sampling state in combination with the discrete state space model and the cost function, and the optimal control input is calculated to adjust the on-off state so as to realize the dynamic control of the sub-module. By implementing the method provided by the invention, not only can the real switching behavior under NLC modulation be accurately simulated, but also the control system is simplified, and the requirements of the controller are reduced, so that the efficiency is improved and the system complexity is reduced while the test accuracy is ensured.

This invention discloses a fault analysis system and method for ship equipment. The system includes: a multi-source sensor synchronization module configured to synchronize the operating data of multiple devices through a hardware phase-locked loop mechanism; a multi-physics modeling module coupled to the multi-source sensor synchronization module, configured to construct a virtual state field of the equipment based on the phase-synchronized operating data of the multiple devices, and generate dynamic parameters characterizing the temporal misalignment of the states of the multiple devices; an adaptive diagnostic analysis module, including a deep learning fault analysis model, wherein the implicit feature extraction parameters and fault evolution temporal modeling parameters of the deep learning fault analysis model are adaptively adjusted in real time according to the dynamic parameters to output system-level fault characterization parameters; and a predictive decision-making module configured to generate decision information including fault development trend prediction and preventive maintenance windows based on the system-level fault characterization parameters. This system enables high-precision correlation diagnosis and predictive maintenance decision-making for early-stage faults of multiple ship equipment.

Systems and techniques for generating an enhanced estimated time of arrival (ETA) with a dynamic time window for last-mile delivery. A system includes an enhanced ETA model that ingests data from disparate data signals including historical data signals, projected traffic signals, and / or real-time signals. The enhanced ETA model considers station dwell time and travel time when generating the enhanced ETA. The system standardizes the format of ingested data and converts non-formatted data into a standardized format. The enhanced ETA model includes a station dwell time model to predict dwell time of trains within a station and a train travel time model to predict expected travel time from a source to a destination. The system generates a control signal based on the enhanced ETA to actuate movement of equipment. The dynamic time window of the enhanced ETA becomes smaller as the train approaches the destination.

The invention discloses a dynamic graph neural network method for simultaneously capturing local and global dependencies, and provides a novel multi-scale collaborative modeling architecture aiming at the problem of excessive localization of the existing dynamic graph neural network. The method comprises the following steps: a local time sequence modeling module based on node memory captures fine interaction in a neighborhood; the method comprises the following steps of: modeling long-range time sequence-frequency dependence of all events in a whole graph by adopting a time-frequency double-perception causal Transform global module; through an adaptive fusion module of the dynamic gating network, fusion weights of the two representations are adaptively learned according to sample characteristics. According to the method, full verification is carried out on 12 real dynamic graph data sets, the most advanced level is achieved on 9 data sets in a link prediction task, the RMSE is reduced by 39.7% at most in an edge regression task, and the learning representation ability of the dynamic graph is remarkably improved.

This invention relates to a satellite-ground collaborative edge computing method for 6G space-ground integrated networks, comprising: constructing a system model for a mobile edge computing scenario; based on the system model, constructing a computation offloading model, a low Earth orbit satellite communication time model, and a channel model; based on the computation offloading model, the low Earth orbit satellite communication time model, and the channel model, constructing an offloading optimization problem with the goal of minimizing the weighted sum of total system latency and energy consumption; and iteratively solving the offloading optimization problem using a block coordinate descent algorithm, wherein the offloading optimization problem is decomposed into two sub-problems: task offloading optimization and resource allocation, which are solved separately. This invention solves the task allocation problem between satellites and base stations for multiple users by establishing a refined satellite-ground geometric model and resource scheduling model, minimizing the total system overhead while satisfying task deadlines and satellite service window constraints.

This invention provides an intelligent drilling rate prediction method based on physical feature guidance and multi-source information fusion, belonging to the field of intelligent drilling rate prediction technology. Specifically, it includes the following steps: collecting multi-source heterogeneous data from a real-time drilling database, logging system, well logging system, and geological database; constructing a dual-channel deep learning prediction model, including: a dual-channel convolutional feature extraction module, a feature fusion module, a temporal fusion module, a temporal modeling module, and a fully connected layer connected sequentially; obtaining the predicted drilling rate using the dual-channel deep learning prediction model; constructing a joint loss function considering data-driven errors and physical constraint errors, calculating the error based on the joint loss function, and updating the network parameters through backpropagation; iteratively training until convergence; and using the trained dual-channel deep learning prediction model for drilling rate prediction. The technical solution of this invention overcomes the problems of insufficient accuracy of mechanistic models and poor reliability of data-driven models in existing technologies.

The invention discloses a broadcast and television building air conditioning system control method based on spatio-temporal model predictive control. The method comprises the following steps that 1, environment parameter data such as temperature, humidity and air quality of multiple positions in a broadcast and television building at different time points are collected; 2, a spatio-temporal model is constructed through the data fusion technology; 3, according to the historical environment data and the air conditioner operation data, a prediction model capable of predicting the future environment state is trained through machine learning, and the optimal operation parameter combination of the air conditioner system for achieving the target is calculated; and a prediction control algorithm is adopted to automatically regulate and control the operation state of the building air-conditioning system unit. According to the method, the indoor environment parameter change trend and the energy consumption demand of the air conditioning system in a period of time in the future are accurately predicted, so that the air conditioning system is optimally controlled, and the problems of energy waste and unstable indoor environment quality in the prior art are solved.

The application discloses a kind of surface layer sea temperature spatio-temporal prediction methods of fusing physical perception, belong to the spatio-temporal prediction field for sea temperature, including: selecting surface layer sea temperature spatio-temporal dataset, define surface layer sea temperature spatio-temporal prediction task;Build the surface layer sea temperature spatio-temporal prediction model of fusing physical perception;The model includes five main modules: feature embedding module, double-path space encoder, time attention module, spatial cross-attention mechanism and adaptive residual prediction module;In double-path space encoder, the global spatial features of surface layer sea temperature are captured by Fourier neural operator with physical perception;At the same time, gradient loss with physical perception constraint is also integrated in the loss function.The spatio-temporal model constructed by combining time and space attention mechanism can improve the accuracy and reliability of surface layer sea temperature prediction by considering global and local spatial features on the basis of fusing physical perception of surface layer sea temperature.

Systems and techniques for generating an enhanced estimated time of arrival (ETA) with a dynamic time window for last-mile delivery. A system includes an enhanced ETA model that ingests data from disparate data signals including historical data signals, projected traffic signals, and / or real-time signals. The enhanced ETA model considers station dwell time and travel time when generating the enhanced ETA. The system standardizes the format of ingested data and converts non-formatted data into a standardized format. The enhanced ETA model includes a station dwell time model to predict dwell time of trains within a station and a train travel time model to predict expected travel time from a source to a destination. The system generates a control signal based on the enhanced ETA to actuate movement of equipment. The dynamic time window of the enhanced ETA becomes smaller as the train approaches the destination.

The invention discloses a robot team coiling motion track planning method and system based on an infinitesimal method dynamic time sequence model, and belongs to the technical field of robot activity analysis, and the method comprises the following steps: obtaining structure parameters and spiral coiling parameters of a robot team, and carrying out model hypothesis; calculating infinitesimal displacement of a front handle of the head robot in each infinitesimal section, performing infinitesimal iterative calculation to obtain polar coordinates of the head robot, iteratively solving the polar coordinates of each handle in the direction from the head robot to the tail robot in combination with the structural parameters, converting the polar coordinates into rectangular coordinates, and calculating to obtain instantaneous speed of each handle; on the basis of the rectangular coordinates and the instantaneous speeds of the handles, position time sequence data and instantaneous speed time sequence data of the handles are generated, collision detection is carried out in combination with the structural parameters, and after the critical moment when the head robot does not collide is determined, position data and instantaneous speeds of the handles at the critical moment are obtained; and simulation is carried out to obtain a robot team disk-in motion track. According to the invention, the motion process of the robot team can be dynamically analyzed and simulated in real time.

A system generates a temporal model for a phishing resource from the database. The system evaluates the phishing resource based on the temporal model. The system determines that the phishing resource is non-phishing based on evaluation results generated during the evaluating. The system updates information about the phishing resource in the database.

The application discloses a workflow execution optimization method based on a random time Petri net, which comprises the following steps: S1, a service time model is constructed, total average service time of activities under different structures is calculated, and current load is updated to obtain a target function of minimum workflow service time; S2, according to the target model of minimum load balancing service time, executor capacity and actual task quantity allocated to the executor are converted into executor load to perform a traversal cycle to find an optimal candidate executor; S3, a waiting time model is constructed, and a target function of minimum workflow execution time is calculated based on the target function of minimum workflow service time; and S4, according to the target function of minimum workflow execution time, optimization of the random time Petri net execution time is performed. The application constructs a workflow execution time model based on the random time Petri net, which is used for analyzing, calculating and reducing workflow process execution time.