70 results about "Dependent model" patented technology

The invention discloses a field specialized large model fine tuning and deployment method and system based on adaptive optimization, belongs to the technical field of artificial intelligence and machine learning, and aims to solve the technical problem of how to improve the rapid adaptation capability of a large model on small-scale field data. In order to meet the requirements of special fields of law, medical treatment, finance and the like on professional terms and complex contexts, the technical scheme adopted by the invention comprises the following steps of: data processing and sample generation: performing cleaning, feature extraction and small sample expansion on field data, and generating a high-quality training sample through a field feature guide mechanism; small-sample fine tuning and migration: realizing rapid field adaptation of a large model through a small-sample learning technology and dynamic Prompt optimization, and reducing dependence on large-scale annotation data in combination with cross-field migration learning; compressing and optimizing the model; performing automatic deployment and collaborative reasoning; and monitoring and adaptive optimization are carried out.

The invention discloses an IT asset fault propagation prediction method and system based on dynamic evolution of a knowledge graph, and relates to the technical field of cloud computing and large-scale IT operation and maintenance management. Through an asynchronous message bus and a logic clock, the knowledge graph is updated immediately when resources are abnormal and a scheduling event occurs; the knowledge graph uniformly integrates physical connection, logic dependence and multi-copy redundancy, so that the cross-machine-room asset relationship is clear at a glance. And then, based on a weighted logistic regression model, node features and relation weights in the knowledge graph are fused, the node fault probability is accurately calculated, the limitation of traditional single-dimensional analysis is solved, self-healing operation is supported, end-to-end intelligent operation and maintenance from fault detection to prediction and early warning to closed-loop self-healing are realized, and the fault detection efficiency is improved. The problems that in a cross-machine-room and multi-live-site environment, resource topology is split, real-time state and alarm information cannot be fused with an asset dependence model, and large-scale real-time deployment of a traditional single-dimensional fault analysis and high-complexity prediction algorithm is difficult are effectively solved.

The invention relates to the field of optical energy storage technology, and discloses a method for predicting the state of charge of an optical energy storage battery based on CNN-LSTM-SA. The key points of the technical solution include the following steps: collecting all state data of an optical energy storage battery; standardizing the state data to obtain a standardized state data matrix, performing feature extraction, and obtaining a time series data matrix; inputting the time series data matrix into a convolutional neural network to extract a spatial feature vector sequence; inputting the spatial feature vector sequence into a long short-term memory network to establish a model of the time dependence of the battery state of charge, and obtaining time-varying trend data; taking the spatial feature vector sequence and the time-varying trend data as inputs, citing them for processing by an attention mechanism, and obtaining a feature output after attention weighting; and calculating a state of charge prediction value by applying a weighted fusion algorithm based on the spatial feature vector sequence, the time-varying trend data, and the feature output after attention weighting.

The invention discloses a reinforcement learning training method and system for relieving illusion of a multi-modal large model, and belongs to the field of reinforcement learning training of a multi-modal large language model. Firstly, a planning and visual description generation step is introduced in an early stage to guide a model to perform structured reasoning, then a grouping relative strategy optimization algorithm is used, reward values are calculated for multiple candidate responses generated by the model after cold start, and particularly, a visual perception reward mechanism is set. The reward mechanism evaluates the consistency of the generated text description and the visual information by using an external large language model. Then, based on a vision description attention score advantage distribution method, learning of the model on key vision signals is dynamically enhanced, and the perception ability of the model on the vision signals is improved; and finally, the perception and reasoning performance of the model is further improved by adopting multiple rounds of rejection sampling and supervised fine tuning. The scheme does not depend on a model architecture, the extra overhead is small, the illusion problem caused by early image-text inconsistency is effectively solved, and the accuracy and the reliability are improved.

The invention discloses an automatic testing method oriented to a large industrial system RPC API (Remote Procedure Call Application Program Interface). According to the method, API documents and historical calling data of tenants are used as input, and automatic API testing is achieved. The method mainly comprises three stages of API dependency model construction, API test sequence generation and API sequence parameter rendering. The construction of the API dependency model is realized based on an API document and historical calling data of tenants, and comprises three steps: static analysis of the API document, parameter clustering based on a large model, and dependency relationship filtering based on the historical calling data of the tenants. The generation of the API test sequence is based on an API dependency graph generated in the previous stage, a to-be-tested API is taken as a starting point, a front dependency API is searched for continuously in an iteration mode to generate the sequence, and finally filtering is carried out through CRUD semantics. The rendering of the API sequence parameters is realized by constructing a dynamic dictionary and recommending large model parameters aiming at the API sequence generated in the previous stage.

A method can include receiving seismic data of a seismic survey of a subsurface region; performing a full waveform inversion using the seismic data and a model of the subsurface region to determine one or more characteristics of the subsurface region, where the model of the subsurface region includes one or more angle-dependent model parameters; and outputting the one or more characteristics of the subsurface region.

The invention discloses an electromechanical industry digital process evaluation method based on data analysis, and particularly relates to the technical field of digital evaluation, a behavior path map, a time sequence causal dependence model and an operation frequency density model are constructed based on information provided by the electromechanical industry, whether structural deviation or behavior abnormity exists is analyzed, and the evaluation result is obtained. If the initial deviation label is generated, calculating a behavior consistency score and a process integrity score, inputting the scores into a fuzzy logic device for risk reasoning, and outputting a credible level; according to the credibility level, judging whether the data is suitable for digital process evaluation, and if the data does not meet the preset requirement, feeding back a corresponding optimization strategy; according to the method, digital behavior structured analysis is realized by constructing a behavior path map, a time sequence causal dependence model and an operation frequency density model; a deviation triggering mechanism is set, and abnormal operation behaviors are accurately recognized; the credibility level is output based on the scoring result, the information applicability is judged according to the credibility level, an optimization strategy is fed back, and the evaluation intelligence level is improved.

The invention relates to the technical field of automatic control systems, in particular to an intelligent substation sequence control method and system, and the method comprises the following steps: based on an equipment operation instruction of an intelligent substation, analyzing the logic sequence of equipment operation, extracting the state change information of equipment, and carrying out the operation analysis of equipment state and time correlation; and judging the dependency relationship through the time difference of the equipment state, and generating an equipment operation dependency model. According to the method, the resource occupation condition of the equipment is monitored through a data stream compression algorithm, load balancing is carried out, the load level of the equipment is evaluated in real time, dynamic distribution and adjustment of loads are achieved, the pressure of high-load equipment is effectively relieved, potential conflicts of the equipment during parallel execution are recognized and solved in real time by introducing an operation conflict judgment mechanism, and the safety of the equipment is improved. Smooth execution of parallel tasks is guaranteed, delay and conflict possibly occurring in the equipment execution process are effectively reduced, and continuity and reliability of a power system are guaranteed.

The invention discloses a multi-site photovoltaic power generation power prediction method and system based on a graph attention network. The method comprises the following steps: obtaining a spatial dependency model containing a site spatial relationship according to a site dependency relationship based on the graph attention network; constructing a time dependence model containing a site time relationship based on an autoregression long-short-term memory network in combination with a probability framework; constructing a first missing data filling layer according to a spatial feature compensation strategy, constructing a second missing data filling layer according to a time feature compensation strategy, and constructing a third missing data filling layer according to the spatial-temporal correlation of the neighbor site set so as to construct a spatial-temporal data processing layer; and constructing a photovoltaic power generation power prediction model according to the spatial dependence model, the time dependence model and the spatio-temporal data processing layer by utilizing historical operation data of a photovoltaic station. The method has the beneficial effects that the requirement on training data is reduced, the pertinence of photovoltaic power generation power prediction is improved, and the accuracy of photovoltaic power generation power prediction and the applicability of the model are improved.

The invention provides a direct / gas composite aircraft model-free attitude control method considering model uncertainty. According to the method, in a control law design process, an LSTM neural network is utilized to fit system items, so that the system items do not depend on prior information of a model any more. On the basis of a Lyapunov method, an online updating rule of a neural network weight is deduced, and the system stability under the dual effects of a control law and a network weight updating law is proved. In the simulation process, ASMFC is applied to an aircraft attitude control scene with model uncertainty and time-varying parameters. A simulation result shows that the performance of the ASMFC is superior to that of other existing adaptive sliding mode control methods.

The invention relates to the technical field of variable subtitles, and particularly provides a multi-variable subtitle method for adaptive balanced perception. The method comprises the following steps: obtaining training set visual features of changed subtitles; constructing a vocabulary and a tag file; in the first stage of the encoder, image features are obtained; in the second stage of the encoder, the obtained common characteristics are as follows; in the third stage of the encoder, extracting difference features of the image; splicing the two groups of image features, and constructing a comparison expression of two differences; finally, modeling the overall difference and the original image respectively; in the first stage of a decoder, grammar correlation is established for words in text data one by one through a text dependence model; in the second stage of the decoder, a group of subtitle information describing image differences is obtained, according to grammar rules and context information, the method ensures that sentences are complete and smooth, visual information and language information are effectively combined, and the accuracy of semantic information is improved.

The invention discloses a multi-granularity data cascade updating method and device based on a dynamic topological graph and a medium, and relates to the technical field of data processing. The method maintains a multi-level data dependency model (DAG) in memory. The system recursively updates a target node by monitoring the change of a source node and utilizing an incremental propagation algorithm. For a high-concurrency scene needing isolation verification, the method introduces an overlay shadow topology mechanism: on the premise of not destroying a main graph structure and not copying a total graph, only instantiating a shadow copy of an affected node based on a Context ID, and constructing an overlay Mapping pointing to a father node; when the propagation path is calculated, the numerical value of the shadow node is preferentially read based on the context, and the propagation path of the original node is logically blocked in the current context. The invention further provides a path convergence mechanism and rendering frame synchronization technology based on topology in-degree, and the problems of resource competition, data consistency and front-end rendering flicker under the complex dependence network are effectively solved.

The invention relates to the technical field of software engineering automation of digital data processing, and discloses a game development automation method based on a multi-agent system. The method is used for solving the problem that a machine-readable cross-product dependence model and consistency verification are lacked in a traditional method. According to the method, based on a task list freezing project context, a constraint packet and a threshold aperture, and based on a demand account book and registration demands, constraints, codes, configuration, resources and other entry abstracts, version information and strong and weak reference relationships; the multiple agents only submit change packets in a temporary storage area by writing tokens and occupying fragmentation write-in gating, and an influence range report binding verification list is generated for each change packet; boundary compliance, reference integrity, compilation verification, determinacy operation verification and acceptance assertion verification are sequentially carried out on the temporary storage area snapshots, and when conflicts occur, rollback or confinement repair is triggered to output traceable and acceptance deliverables based on reproducible evidence judgment.

The invention discloses a GPU resource aware matrix multiplication parallel performance analysis model construction method, which relates to the technical field of high-performance calculation, depends on the Roofline principle, combines a bandwidth model, a resource load model and an instruction dependence model to construct a performance analysis model, and is successfully applied to a matrix multiplication scene. And the parallel performance quantification under the matrix multiplication application of resource awareness is realized. The method is not limited by a platform, resource awareness can be achieved, maximized resource use and parameter setting of optimized performance can be distinguished through a performance analysis model in a matrix multiplication scene, and a foundation is laid for follow-up parallel program optimization work; the hardware parameters used in the method are completely based on the GPU hardware parameters which can be obtained publicly, so that non-GPU professionals can also use the model to evaluate the performance of the program.

The invention discloses a battery off-line and on-line hybrid modeling method and device, a storage medium and a product. The method comprises the following steps: constructing a state dependent model of a battery; determining a state vector according to the state parameter of the battery; taking the state vector as input, and determining a function type coefficient of the state dependent model; converting the state-dependent model into a nonlinear least square model; according to the obtained battery measurement data, performing offline optimization on parameters of the nonlinear least square model by adopting an optimization algorithm to obtain all parameters, and further obtaining a battery offline model; and keeping the non-linear parameters of the battery offline model unchanged, and updating the linear parameters in the battery offline model online according to the real-time measurement data of the battery so as to obtain a real-time updated state dependent model. According to the method, high-precision modeling of the lithium ion battery is realized on the premise of ensuring real-time performance.

The invention discloses a learner cognitive diagnosis method oriented to an intelligent education scene, and belongs to the technical field of cognitive diagnosis and education data analysis. According to the method, concept-level multi-dimensional modeling is carried out on the ability and exercise difficulty of a learner by constructing multi-dimensional representation of concept perception so as to describe mastering characteristics of the learner on different cognitive levels; and meanwhile, distinguishing a pre-correction dependency relationship and a semantic approximation relationship, establishing a relationship-perceived concept dependency model, and deducing a potential exercise-concept association structure according to the relationship-perceived concept dependency model. Further, the annotated exercise-concept incidence matrix and the inference incidence matrix are fused in a unified diagnosis layer, and comprehensive evaluation of the knowledge mastering state of the learner is achieved. According to the method, an end-to-end mode is adopted for optimization, the fineness, knowledge coverage and stability of cognitive diagnosis can be effectively improved, better prediction performance and generalization ability are shown on multiple real education data sets, and the method is suitable for intelligent education application scenes such as learning analysis and personalized teaching.

The invention discloses a method and device for determining an operator shape of an AI model and related equipment, and relates to the technical field of artificial intelligence. Determining the degree of influence of each parallel strategy on the shape of a first operator included in the AI model in a plurality of parallel strategies provided by a computing system running the AI model; obtaining a target parallel configuration of the AI model, wherein the target parallel configuration comprises a configuration value of each parallel strategy in a plurality of parallel strategies; and determining the shape of the first operator under the target parallel configuration according to the configuration value of each parallel strategy in the plurality of parallel strategies, the basic shape of the first operator and the influence degree of each parallel strategy on the shape of the first operator. Therefore, according to the influence degree of each parallel strategy on the shape of the operator in the AI model, the shape of the operator under the target parallel configuration can be determined without depending on the specific structure of the AI model, so that the generalization and universality of determining the shape of the operator can be effectively improved, and meanwhile, the efficiency of determining the shape of the operator can also be improved.

The invention provides a method and a model for predicting the temperature continuity characteristic of a low-temperature MOSFET (Metal-Oxide-Semiconductor Field Effect Transistor), and the method comprises the following steps: designing and preparing a verification structure which covers a full box of the model based on the process size range of the MOSFET; performing direct current characteristic test on the verification structure at a plurality of test temperatures, acquiring actual measurement data, and completing low-temperature characteristic characterization of the verification structure; the method comprises the following steps: defining a sub-circuit based on a cryo-BSIM model architecture, and establishing a sub-circuit temperature dependent model driven by sub-box parameters; according to the low-temperature electrical characteristic parameters, through parameter fitting and discrete model integration, establishing a sub-circuit temperature continuous intensive model for low-temperature electrical characteristic simulation; when the consistency verification of the fitting result and the measured data fails, backtracking and adjusting the sub-box parameters, and completing the model iteration of the sub-circuit temperature dependent model; and when the consistency verification of the fitting result and the measured data is successful, obtaining the temperature continuous model of the low-temperature MOSFET. According to the invention, the electrical characteristics of the transistor in a full-temperature zone can be reflected more accurately.

Disclosed herein is an apparatus associated with a network controller in an Open-Radio Access Network (O-RAN). The apparatus is configured to receive a request from a RAN function application from among a plurality of RAN function applications associated with the network controller. The request is indicative of a change in one or more parameters associated with the RAN function applications. The apparatus is configured to determine, based on a parameter dependency model, whether a conflict will occur as a result of the change in the one or more parameters. The parameter dependency model defines dependencies among a plurality of parameters associated with the plurality of RAN function applications. The apparatus is further configured to trigger, based on a set of mitigation rules, one or more reconciliation actions upon a determination that the conflict will occur as a result of the change in the one or more parameters.

The application discloses a cross-media model error attribution and auxiliary correction method based on causal decoupling, which comprises the following steps: image segmentation is performed on an input image, and the input image is divided into a plurality of sub-regions; a target function is solved according to the sub-regions, the minimum key region which supports the generation of a target word element to the greatest extent and removes the generated ability is sorted according to the contribution to the generation decision, and an attribution saliency map is generated; an influence score is calculated according to the change of the generation probability of the target word element in the process of inserting the sub-regions in the ordered subset; the ordered subset, the attribution saliency map and the influence score are evaluated from the dimensions of fidelity, positioning ability and error correction guiding ability, and an evaluation result is obtained; the evaluation result is obtained without relying on the internal gradient, attention weight or activation map of the model, the dependence on the internal structure of the model is reduced, the regional attribution of an arbitrary word element set is realized, and the relative dependence of the generation process on visual evidence and language prior is quantified.

A computer-implemented method for ascertaining an item of environment information is provided. The item of environment information relates to material in surroundings of a third article and / or an interaction of the third article with the material. The method includes receiving an X-ray image and an item of sensor information, and determining model parameters or limiting possible parameter values of the model parameters of a three-dimensional model of the third article as a function of the X-ray image in order to specify an X-ray-dependent model. The three-dimensional model describes a three-dimensional shape and / or pose of the third article as a function of the model parameters. The method includes ascertaining the item of environment information as a function of the X-ray-dependent model, where the item of environment information and / or the X-ray-dependent model additionally depends on the item of sensor information. The item of environment information is provided.

The application relates to a periodic operation component multivariate perception abnormal mode time sequence dependence mining method, which comprises the following steps: S1, a component multivariate perception variable abnormal event capture platform is constructed; S2, abnormal mode time sequence data of the perception variable is captured, and an abnormal time sequence database Event_seqDB is constructed; S3, a multivariate perception abnormal mode time sequence database KDD_SeqDB is constructed; and S4, multivariate perception abnormal mode time sequence dependence mining is carried out based on an improved PrefixSpan. Through the construction of a key component multitransmission variable perception mode time sequence dependence model and algorithm, the time sequence cause-effect relationship formed in the component operation and maintenance process can be effectively mined, thereby providing support for preventive maintenance and design iteration optimization.

The present invention relates to the field of computer application technology, and discloses a software product component version control method, system, storage medium and device, including: responding to a project code compilation instruction, processing the code modified by annotations, obtaining a dependency tree of the project code, building a dependency model according to the dependency tree, assembling several components and their version numbers in the dependency model into version record information; obtaining software and hardware information under the compilation environment, obtaining compilation environment information; assembling the version record information and the compilation environment information into a reporting object, and requesting an encrypted public key, using the encrypted public key to encrypt the reporting object, obtaining a ciphertext, reporting the ciphertext and receiving the returned component information. The mental burden of developers is reduced, and it is convenient for developers to timely understand component updates, perform updates and vulnerability repairs, and improve the overall stability and security of the project.

The invention relates to the technical field of reliability management of electromechanical equipment, in particular to an electromechanical equipment reliability evaluation method based on a switching Markov process and fractional Brownian motion, which comprises the following steps: constructing a switching Markov process model, the description module is used for describing a multi-stage degradation process experienced by equipment in a life cycle and state transition thereof; based on the external working condition parameters and the switching Markov process, constructing a stage dependency model; establishing a nonlinear degradation model of the electromechanical equipment based on the fractional Brownian motion process, and introducing a random effect into the model to represent individual differences; and constructing a two-stage parameter estimation method, and sequentially estimating unknown parameters in the switching Markov process and the degradation model. The performance degradation of the electromechanical equipment is accurately modeled, and the nonlinearity, randomness, individual difference, long-term memorability and stage dependence of the multi-stage degradation of the electromechanical equipment are effectively captured, so that the reliability of the electromechanical equipment is accurately evaluated.

This invention relates to the field of data processing technology for predictive purposes, and discloses a method for adjusting the power generation performance of an energy system throughout its entire life cycle. The method determines the parameter boundaries of a wind-solar-storage power supply system, divides the entire life cycle planning period into multiple consecutive life stages, and determines the load demand input for the entire life cycle. It sets a baseline power generation performance at the initial commissioning stage, calculates the degradation maintenance level for each stage, establishes a monotonic mapping relationship between the degradation maintenance level and maintenance intensity, constructs a stage-dependent model for calculating available power generation capacity coupled with degradation and maintenance, and obtains available power supply capacity parameters for each stage of the entire life cycle. It generates a maintenance time series distribution through Monte Carlo simulation and constructs a set of life cycle power generation performance evolution scenarios. With the goal of minimizing the total cost throughout the entire life cycle, a mixed-integer linear programming model is constructed to obtain the optimal capacity configuration and adjust the power generation performance throughout the entire life cycle. This method solves the problems of ignoring degradation maintenance and inaccurate planning, achieving the goal of accurate and reliable life cycle data at a low cost.

A method can include receiving seismic data of a seismic survey of a subsurface region; performing a full waveform inversion using the seismic data and a model of the subsurface region to determine one or more characteristics of the subsurface region, where the model of the subsurface region includes one or more angle-dependent model parameters; and outputting the one or more characteristics of the subsurface region.

The invention provides a decoration project intelligent construction planning system based on artificial intelligence, and relates to the technical field of construction planning, and the system comprises the steps: obtaining a resource library and a construction task sequence in a decoration project; constructing a resource task constraint model by using the resource library and the construction task sequence, determining a regulation and control optimization feature matrix, and performing constraint optimization on the resource task constraint model according to the regulation and control optimization feature matrix to obtain a resource task optimization model; constructing a construction process dependence model in the decoration engineering process, and performing decoration core measurement on each process dependence node in the construction process dependence model to obtain a decoration core coefficient of each process dependence node; and generating a resource scheduling planning sequence in the decoration project on the basis of the decoration core coefficients of all the process dependent nodes and the resource task optimization model, and planning and scheduling resources in the decoration project according to the resource scheduling planning sequence. Therefore, resource collaborative scheduling in the decoration project is realized.

The present invention provides a pedestrian trajectory prediction method based on a sparse spatiotemporal graph Transformer network, which belongs to the field of computer vision technology. The technical problem of low efficiency in predicting pedestrian trajectories in crowded environments and over long distances is solved. The technical solution is as follows: it includes the following steps: S1: obtaining the position information of pedestrians from image frames; S2: using a dynamic spatial Transformer, and using a multi-head attention mechanism to jointly model multiple spatially dependent models; S3: using a self-attention mechanism to achieve bidirectional temporal dependency modeling across multiple time steps; S4: obtaining a spatiotemporal Transformer network with sparse transformations; S5: designing a model framework of a spatiotemporal block chain based on a spatiotemporal Transformer. The beneficial effects of the present invention are: the present invention improves the accuracy of pedestrian trajectory prediction in crowded environments and over long distances.

This document can use a "Perception Statistical Performance Model" (PSPM) that models perception slices of the runtime stack of an autonomous vehicle or other robotic system, e.g., for safety / performance testing. The PSPM is configured to: receive computed perception ground truth; determine a probabilistic perception uncertainty distribution from the perception ground truth based on a set of learned parameters, the parameters being learned from a set of actual perception outputs generated using the perception slice to be modeled. The PSPM includes a time-dependent model such that the perception output sampled at the current moment depends on at least one of: an earlier one of the perception outputs sampled at the previous moment, and an earlier one of the perception ground truth computed for the previous moment.

The invention discloses a crane track seam real-time detection method and system, and belongs to the technical field of cranes, and the method comprises the steps: obtaining seam images under various different track surface conditions, extracting a region of interest, combining a model activation consistency index and background disturbance sensitivity, calculating a background dependency index, and carrying out the dependency classification of a model; according to the method, the recognition accuracy of a high-dependency model in a new environment is predicted based on track background features, a prediction result and dependency are subjected to joint analysis, and directional optimization is performed on the model by adopting an attention-guided confrontation decorrelation training mechanism and the like, so that the problem that the recognition capability of a traditional CNN model is reduced in a background change environment is effectively solved, and the recognition accuracy of the model is improved. The generalization ability and deployment robustness of the model are significantly improved, and the environment migration cost and the misjudgment risk are reduced.