591 results about "Model learning" patented technology

The invention discloses a rapid heat transfer simulation method and device based on a neural network, and relates to the technical field of physical simulation. The method comprises the steps that a hybrid neural network model is trained, the model learns operator mapping from an input function to a temperature or heat flow field, and meanwhile physical constraints such as a heat conduction partial differential equation are coded into a loss function; for a new simulation task, single forward inference is carried out by using the operator mapping, and an initial prediction result is rapidly generated; then, according to physical constraints of coding, calculating a physical residual error of initial prediction, and when the residual error exceeds a preset threshold value, executing a small amount of optimization iteration by taking the prediction as an initial value to carry out rapid local correction; the problems that a traditional numerical method is long in calculation time and an existing neural network method is insufficient in physical fidelity are solved, and high efficiency and high precision of heat transfer simulation are achieved.

The invention relates to the technical field of deep counterfeit image detection, and discloses a deep counterfeit detection method and system, a storage medium and computer equipment. The method comprises the following steps: firstly, constructing a reference data set containing a forged image and an original real image; secondly, through an integrated model, generating antagonistic samples for the reference data set, and integrating the successfully attacked antagonistic samples into an antagonistic sample set; and finally, merging the reference data set and the adversarial sample set, and constructing a robustness enhanced data set containing four types of samples. In the model training stage, multi-classification cross entropy loss and comparative learning loss are combined, and expression of the model in a feature space is optimized through comparative learning constraint, so that the model learns discriminative features with more compact intra-class features and more dispersed inter-class features. The model trained by the method not only can effectively defend against attack and improve robustness, but also surpasses original detection performance on clean samples, and has remarkable technical advantages and application value.

Methods for training a series of neural networks to output driving behavior control parameters is disclosed. The training dataset for the neural networks includes sensor-based vehicle driving recordings that may be categorized by geographical area, by qualitative driving behaviors, or by some combination, such that various training data subsets are used to train the series of neural networks. By learning either city-specific driving behavior control parameters, qualitative driving behavior specific driving behavior control parameters, or both, the resulting parameters may then be provided to a motion planning model for use in modeling predictive control for an autonomous vehicle. Rather than relying on XYZ trajectories of agent vehicles when planning future trajectories of the ego vehicle, the motion planning model is adaptive, due to the use of the learned driving behavior control parameters.

The invention discloses an engine state estimation and system modeling correction method based on double-layer variational inference, which relates to the field of engine state estimation and comprises a variational inference stage aiming at component performance states and kinetic model parameters; a system output prediction stage based on an observation equation; and a solving stage of performing objective function optimization through an evidence lower bound. The structure clearly presents information flow and key calculation links of the proposed algorithm in state estimation and model learning. According to the method, combined reasoning of state variables and model parameters is achieved by building a probability modeling structure, the modeling problem when system dynamics is partially or completely unknown is solved by combining a modeling method of a stochastic differential equation, and while the state variables and the model parameters are optimized, the modeling efficiency is improved. Precise inference of component states and reliable identification of fault features are achieved, the fault detection accuracy of sudden gas circuit abnormity reaches the standard, and meanwhile the performance is better in the aspect of tracking long-term performance degradation.

The invention relates to the technical field of intelligent wharfs, in particular to a wharf container truck dynamic optimization scheduling method and system combining machine learning and path planning. Comprising a behavior data acquisition and feature coupling unit; a learnable incentive and behavior guide unit; a scheduling demand prediction unit; and a path planning and scheduling unit. According to the method, on the basis of the coupling characteristics, the excitation coefficient is optimized through reinforcement learning, the excitation instruction is dynamically pushed, and targeted guidance of the non-operation staying behavior of the container truck is achieved; according to the method, based on standardized time series data, an association rule of a historical staying period and a working condition is learned through an LSTM model, a prediction result is optimized in combination with real-time data, a prospective constraint basis is provided for scheduling, and meanwhile, a time, space, resource and priority multi-dimensional path constraint system is constructed through a structural causal model; container truck-berth matching and dynamic path planning are completed by matching with an improved A * algorithm fused with dynamic weights, and scheduling conflicts are effectively avoided.

The invention relates to the technical field of artificial intelligence, discloses a multi-modal self-supervision abnormal mode detection method and system, and aims to solve the problems that in the prior art, a large amount of annotated data is relied on, modal fusion is insufficient, the anomaly discrimination ability is weak, and the dynamic environment is difficult to adapt. The method comprises the following steps: synchronously acquiring videos, audios, sensor time sequences and log text data, and carrying out time alignment; extracting spatial-temporal characteristics of each modal through a modal specific encoder; constructing a contrast learning task under a label-free condition, generating positive and negative sample pairs by utilizing data enhancement, and driving model learning discriminative representation; and cross-modal feature alignment and dynamic weighted fusion are realized by adopting an attention mechanism, and joint representation is generated. According to the scheme, efficient anomaly detection without annotation data is realized, the multi-modal fusion representation capability is remarkably improved, the false alarm rate is reduced, the environmental adaptability is enhanced, and the real-time monitoring requirement is met.

The invention discloses an enterprise appeal intelligent perception and closed-loop processing system based on multi-modal AI, and relates to the technical field of intelligent government affairs, the system realizes accurate semantic understanding and deep intention recognition of multi-modal fusion, and the intelligent perception and closed-loop processing of enterprise appeals are realized through a cross-modal attention mechanism and a unified semantic representation model. According to the method, deep fusion and complementary analysis of multi-source heterogeneous data such as voices, texts, images and the like are realized, the limitation of a traditional single-mode or simple splicing mode is broken through, the accuracy and robustness of intention recognition are remarkably improved, and misjudgment is fundamentally reduced; the method comprises the following steps: constructing a dynamic self-adaptive routing mechanism based on Actor-Critic reinforcement learning, constructing a work order assignment problem into a sequence decision problem, driving a model to learn a dynamic fusion and weight distribution strategy of multiple decision factors through a reward mechanism, and adopting an online strategy iteration optimization mechanism to realize an optimal assignment decision, so as to improve the work order assignment efficiency. And the shunting accuracy and efficiency are obviously improved.

The invention discloses a multi-variable measurement sensor state lightweight evaluation method based on knowledge distillation, and belongs to the technical field of electric digital data processing and multi-sensor data fusion. The method comprises the following steps: firstly, carrying out time synchronization and physical consistency constraint modeling on original data of multiple sensors, and extracting feature representation; a high-precision teacher model is trained at the cloud end, and the high-dimensional mapping relation of the sensor state is learned; intermediate features, soft output and uncertainty information of a teacher model are extracted to serve as distillation knowledge, lightweight student model training is guided, and effective migration of discrimination knowledge is achieved in combination with soft label constraint, feature alignment and an uncertainty guiding mechanism; and finally, compressing, quantifying and optimizing the student model, and deploying the student model to a vehicle-mounted end to realize real-time evaluation and dynamic updating of the states of multiple sensors. According to the method, the model complexity is greatly reduced while the evaluation precision is ensured through a knowledge distillation framework, and efficient and reliable state perception and fault-tolerant control support is provided for an intelligent driving system.

The invention discloses a fusion method for balancing medical text positive and negative sample training, belongs to the field of medical text classification crossing natural language processing and medical informatics, and solves the problems of long text information loss, labeling noise interference and positive and negative sample imbalance in the prior art. The method comprises the steps of preprocessing a medical text, defining five types of medical entities such as disease diagnosis and the like, and labeling and establishing entity association; segmenting a long text by using a sliding window, extracting text features based on BioBERT, adding a gradient inversion layer, and setting a classifier and a discriminator; in positive and negative training, a weighted cross entropy loss function is used for learning real label mapping of a text in positive training, and a noise suppression loss function is used for reducing the influence of noise on model learning in negative training. According to the method, the delirium-symptom-containing medical record can be accurately identified, and the accuracy and robustness of medical text classification are improved.

The invention discloses a task scheduling method and system for a heterogeneous multi-operating system, and the method comprises the steps: periodically collecting task states, resource states and operating system capability data, and constructing a task scheduling heterogeneous graph comprising task nodes, resource nodes and operating system nodes according to the task states, the resource states and the operating system capability data; and then, inputting the heterogeneous graph into a pre-trained graph convolutional neural network, learning a complex relationship between nodes by a model, and outputting an adaptation rate of a task and each candidate scheduling combination. And the scheduling module determines an optimal target scheduling combination through multi-stage screening and optimization in combination with the adaptation rate, the real-time system state and the task constraint. Finally, through an efficient cross-system migration mechanism, the task is seamlessly migrated from the current operating system to the target operating system, and the target operating system calls corresponding resources to execute the task, so that self-adaptive task scheduling in a heterogeneous multi-core multi-operating system environment is realized, and the task scheduling efficiency is improved. And the resource utilization rate, the task execution efficiency and the cross-operating system cooperation capability are remarkably improved.

The invention discloses a federated backdoor defense method based on decoupling contrast learning, and the method comprises the steps: training a backdoor model based on a backdoor sample, and immediately stopping training after the backdoor model converges on the backdoor sample; respectively extracting a penultimate layer vector of the backdoor model and the local model from a sample pair held by the malicious client as a backdoor feature and a clean feature; comparing and learning the separated back door features and the clean features, and learning the clean features for the local model by using a sample weighting strategy to train the local model to obtain a trained local model; and sending local model parameters of the trained local model to a global server, and generating model parameters of a new global model based on the local model parameters through an aggregation function. The method aims at reducing information dependence between backdoor features and clean features through comparative learning, so that local model learning is free of backdoor representation, and the robustness of a global model is improved.

The invention discloses a wind measurement radar data correction method based on high-precision topographic data. According to the technical scheme, the wind measurement radar data correction method is characterized by comprising the steps of 1, data preprocessing; 2, modeling a high-precision terrain generation model; step 3, performing CFD pre-simulation and terrain correction coefficient matrix; 4, modeling a data correction model; and establishing a data correction model by using a space-time convolutional neural network coupled LSTM mixed structure, learning terrain-wind field nonlinear mapping, and outputting a corrected three-dimensional wind field. 5, correcting the data in real time; and inputting real-time wind measurement radar data into the data correction model, optimizing the radar data through online matching of the terrain correction coefficient matrix under the current wind direction and wind speed, and outputting a correction result. The method is mainly used for solving the problems of high wind speed measurement error and low data analysis efficiency in a dynamic environment.

The invention discloses a reinforcement learning intrusion detection method based on space-time attention and dynamic courses, and the method comprises the steps: carrying out the dynamic weight distribution of input features in space and time dimensions through a multi-level space-time attention mechanism, and enhancing the expression capability of key features; a dynamic curriculum learning strategy is adopted to realize progressive difficulty adjustment of training samples, and the learning ability of the model is gradually improved through initial difficulty estimation, dynamic threshold adjustment and curriculum sample selection; designing a composite reward mechanism to optimize a reward signal, and combining a basic classification reward and a stability reward to promote stable convergence of the strategy; a self-evolution target network module is introduced, and dynamic update management of a target network is realized through performance monitoring, emergency update triggering and frequency self-adaption. A space-time cooperative detection closed loop is constructed, and an end-to-end intrusion detection process is realized through environment interaction, strategy optimization and online detection. The method effectively improves the accuracy and robustness of intrusion detection, and adapts to a complex network attack scene.

The invention relates to the technical field of automobile cabin environment control, and discloses an intelligent cabin environment adaptive adjustment system based on an AI large model, which comprises a sensing module, an AI large model decision module, an execution module and a feedback module. According to the intelligent cabin environment self-adaptive adjustment technology based on the AI large model, a mixed decision sub-model of an AI large model decision-making module adopts an MPC-TD3 fusion algorithm, and a user behavior learning sub-model of the AI large model decision-making module learns user physiological parameters and operation records collected by a sensing module; the multi-physics field prediction sub-model predicts a cabin environment and a power battery state, and corrects a decision result in combination with comfort parameters fed back by a PMV-PPD evaluation model and an SET evaluation model of the feedback module in real time, so that the problems that a single MPC model is insufficient in robustness under complex working conditions and a single reinforcement learning model is difficult to quantify the comfort feeling of a user are solved, and the user experience is improved. And the dynamic response speed and the adjustment precision are considered.

A machine learning device includes an image set acquiring unit to acquire an image set including images, and an image set selecting unit to select an image set similar to the acquired image set from a plurality of image sets different from the acquired image set. In addition, the machine learning device includes a performance comparison unit, and a preprocessing acquisition unit to select a learning model from a plurality of machine-learned learning models based on a performance comparison result by the performance comparison unit and to acquire preprocessing performed on an image set used for machine learning of the learning model selected. Furthermore, the machine learning device includes a model learning unit to perform the acquired preprocessing on the acquired image set and to cause a learning model that has not yet trained to perform machine learning using the preprocessed image set.

The invention discloses a semi-supervised medical image segmentation method based on momentum prototype alignment and adaptive uncertainty estimation, and aims to solve the problems of unstable feature alignment and false label noise accumulation caused by a batch effect in an existing semi-supervised learning method in a scene of scarcity of medical image annotation data. The method comprises the following steps: firstly, constructing a dual-network model based on a mean teacher architecture; secondly, designing a mixed strong and weak disturbance strategy, respectively applying weak data disturbance and strong data disturbance to the unlabeled image, respectively inputting the weak data disturbance and the strong data disturbance into a teacher and student segmentation network, and forcing the model to learn local and global robust features; secondly, a self-adaptive uncertainty estimation mechanism is introduced to dynamically screen false labels, and isolated noise is eliminated in combination with maximum connected domain analysis; meanwhile, the prototype of the current batch is calculated through uncertainty weighting, and a global category prototype library is updated and maintained through momentum. According to the method, the segmentation precision and the boundary integrity are remarkably improved under a small amount of annotated data.

The invention is suitable for the technical field of intelligent traffic systems, and provides a trajectory anomaly detection method based on structure enhancement and comparison pre-training. The method comprises the following steps: firstly, carrying out road network matching on an original GPS track, extracting attributes such as a road type and a road length, coding the attributes into a structure lexical element sequence, and constructing a structured Prompt text; then, through structure-semantic comparison pre-training, enabling the model to learn an alignment relationship between path structure features and text description thereof, and obtaining stable shared representation; in the fine tuning and reasoning stage, the aligned path structure vectors are used for carrying out structure guiding fusion on large language model coding representation, track behavior categories are output through a classifier, and meanwhile key evidence fragments are given according to attention weights so as to enhance interpretability. Compared with an existing method, the method has the advantages that road structure information is explicitly fused, cross-modal alignment is achieved, and the accuracy and interpretability of track anomaly detection and the generalization ability in a city-wide cross-regional scene are remarkably improved.

The invention discloses a data center machine room temperature field equalization method based on a multi-scale model learning strategy. Fusing the equipment tolerance, the refrigeration system performance and the environmental thermodynamic boundary to construct a dynamic safe operation boundary; a machine room-cabinet-equipment multi-scale reverse model is constructed based on a physical guidance neural operator, and backstepping from observation of a temperature field to an optimal control strategy is realized; the method comprises the following steps: dynamically dividing a machine room into a plurality of logic control areas, deploying a reinforcement learning agent for each area, deciding and adjusting the air supply speed under the constraint of a safety boundary, and meanwhile, learning the minimum feasible air speed on line to optimize energy conservation; and when temperature abnormity which cannot be processed by local control is detected, generating and executing a global coordinated regulation and control instruction according to the global temperature field data and an analysis result of the multi-scale model. According to the method, on the premise of guaranteeing the safety of equipment, the temperature field of the data center is subjected to refined and self-adaptive balanced regulation and control, and the energy efficiency and the operation reliability of a refrigeration system are remarkably improved.

The invention provides a medical large model knowledge distillation method and system for medical image segmentation, and belongs to the technical field of artificial intelligence and medical image analysis. Comprising the following steps: respectively constructing a teacher model and a student model based on acquired training data; performing feature extraction on the medical image by using a teacher model, generating a global semantic embedding vector and a local embedding vector, calculating a semantic distance between the two vectors through a semantic grounding mechanism, constructing a quality scoring function, and generating a quality score for guiding student model learning; constructing a joint objective function to perform distillation training based on a staged optimization strategy on the teacher model and the student model; and after training is completed, only the student model is reserved for image segmentation. According to the method, reliable migration of large-model semantic knowledge of teachers in a light-weight model confidence space can be realized, so that the segmentation performance, generalization ability and clinical interpretability of medical images are improved.

The invention discloses a chained reasoning training and generating method and system of a medical diagnosis large model, and relates to the crossing field of artificial intelligence and intelligent medical treatment, comprising a first stage and a second stage: in the first stage, a knowledge-enhanced chained medical diagnosis data set is constructed, and a chained diagnosis normal form is learned by supervising a fine-tuning training large model; in the second stage, the model serves as a reinforcement learning strategy model, parameters are updated through a strategy optimization algorithm, and finally a chain type reasoning model considering the structure and diagnosis accuracy is obtained. The system further comprises a reasoning training module, a local knowledge base module, a reasoning control module and an interaction generation module. The method has the beneficial effects that the model is forced to internalize a hypothesis-positioning-reflection-decision chain reasoning structure through supervised learning, so that the diagnosis process of the model is completely presented in a structured and stepwise form, and the algorithm is optimized through a track-level composite reward function and a grouping relative strategy; and the model reasoning content is driven to evolve towards the direction of higher diagnosis accuracy and more reasonable knowledge reference.

The invention discloses a cast iron cylinder sleeve structure and alloy component optimization method based on friction and wear performance prediction, and belongs to the technical field of material science and machine manufacturing. The problems that a traditional cylinder sleeve alloy design method is low in efficiency and the material performance of the designed cylinder sleeve alloy is limited are solved. The method comprises the following steps: collecting an existing large-cylinder-diameter engine sample, and carrying out metallographic characterization and microhardness test on an engine cylinder sleeve to obtain cylinder sleeve structure characteristics and alloy component data; the wear volumes of different parts are obtained through a friction wear experiment, and then the wear rate is calculated based on the wear volumes. After tissue characteristic data, alloy component data and frictional wear test data of the cylinder sleeve are collected, the data are imported into different machine learning models, the models learn the corresponding relation between the tissue characteristic data and the frictional wear performance of the existing cylinder sleeve, and optimization of material tissue and alloy components is guided according to the prediction result of the optimal model. The method can be applied to optimization of cylinder sleeve structures and alloy components.

The invention provides a low-resource multi-language large model training method and system for personalized course learning, and belongs to the technical field of large language models, and the method comprises the steps: S1, collecting training samples of multiple languages; performing de-duplication and de-noising processing on each sample, then unifying data formats, and adding language attributes as language labels; s2, initializing parameters of an adaptive sampling scheduler and a dynamic loss scheduler; and S3, taking the pre-trained large language model as a base model, and adding an adaptive sampling scheduler and a dynamic loss scheduler in the training process. According to the method, the dependence on low-resource language annotation data is reduced, and the training weights of different language samples can be adaptively balanced; and dynamically matching the multi-language task difficulty with the model learning progress.

The invention discloses a fence piece production line monitoring device and monitoring system, and relates to the technical field of production line monitoring. Comprising a data acquisition device, an edge computing device, a feedback control device, a whole-process quality tracing module and a cloud platform data analysis center. The data acquisition device acquires multi-station multi-mode real-time data, the edge computing device performs data fusion, preprocessing and preliminary abnormality diagnosis, the cloud platform develops deep analysis and model learning, the feedback control device generates an instruction to adjust production line parameters, and the whole-process quality tracing module realizes production whole-link data binding tracing. According to the method, accurate diagnosis of production abnormity, predictive maintenance of equipment, dynamic optimization of the process and closed-loop control of quality are realized, the production efficiency and the product quality are greatly improved, and the failure shutdown loss is reduced.

The invention provides a coal-fired power plant SCR denitration system dynamic prediction method based on a parameter identification mechanism-data hybrid model, and relates to the technical field of intelligent modeling prediction. The method comprises the following steps: establishing an SCR dynamic mechanism model, completing parameter identification based on historical operation data, and obtaining mechanism predicted values of NOx concentration and ammonia escape amount; constructing a series model learning mechanism model prediction deviation of the liquid neural network and the long-short-term memory network; and carrying out weighted fusion on the mechanism prediction value and the prediction deviation to form a hybrid model with physical consistency and dynamic adaptive capacity, and realizing high-precision stable prediction of NOx emission and ammonia escape amount of the SCR outlet under the variable load working condition.

The invention provides an industrial product parameter extraction method based on domain knowledge learning, and the method comprises the steps: 1, collecting a plurality of technical documents of a preset industrial product field as knowledge sources, and constructing a structured domain knowledge model through a man-machine cooperation mode; 2, based on the domain knowledge model, generating a training data set with entity type labels, and training one or more deep learning extraction models by using the data set to enable the extraction models to learn parameter knowledge of the industrial product field; 3, receiving a new to-be-processed digital technology document, and processing the document by adopting the trained deep learning extraction model so as to identify and extract structured data corresponding to parameter entities in the domain knowledge model; and 4, storing the extracted structured data into a database, and selectively utilizing the data to carry out iterative updating and enhancement on the domain knowledge model.

The invention relates to the technical field of path planning and tracking, in particular to a heavy-duty car path planning and tracking method and system giving consideration to efficiency and energy-saving indexes, and the method comprises the steps: constructing a multi-dimensional state space fusing a vehicle state, environment information and real-time energy consumption; a reinforcement learning decision model is constructed and trained, the output of the reinforcement learning decision model is a layered action space, and a reward weight dynamic adjustment mechanism based on scene recognition is introduced to dynamically adjust the weight of safe, efficient and energy-saving rewards and drive model learning. And after training is completed, the model is deployed to a vehicle-mounted unit for real-time planning and control. During real vehicle operation, data are continuously collected, a model is updated through incremental learning, and an energy consumption-road condition association database is synchronously constructed. The database is inquired during real-time decision making, and model output is guided or finely adjusted by using a historical energy-saving strategy, so that rapid scene self-adaption of the energy-saving strategy is realized. Safe, efficient and energy-saving running of the heavy-duty car in a complex environment is achieved.

A system and method for distribution anomaly prediction for new mineral exploration using mineralogical understanding and artificial intelligence analysis of geochemical data are provided, the system comprising: a database development unit configured to collect regional distribution data including pre-collected geochemical data and geologic spatial data of a region and to integrate map data and data set to construct integrated data into a database; a model learning unit configured to generate and learn an exploration mineral distribution anomaly model based on machine learning, wherein the exploration mineral distribution anomaly model enables prediction of distribution anomaly based on geologic and ore deposit geological understanding of exploration mineral using the integrated data; and a distribution anomaly prediction unit configured to predict the distribution anomaly of the exploration minerals in an exploration region using the learned exploration mineral distribution anomaly model.

The invention belongs to the technical field of artificial intelligence, and particularly relates to a computer network intrusion detection method and system based on artificial intelligence, and the method comprises the steps: constructing a finite state automaton to carry out the state jump compliance verification of a connection flow, intercepting the illegal flow, only sending the compliance flow into a feature engineering module, and extracting the multi-dimensional behavior features; inputting a deep neural network classification model to determine whether the behavior is an intrusion behavior; according to the technical scheme provided by the invention, the data input to the artificial intelligence model can be ensured to have complete compliance on the protocol level, so that the model is prevented from learning false feature association caused by protocol violation, and the antagonistic attack based on protocol state jump can be effectively resisted on the premise of not depending on adversarial training.

Methods, systems, and apparatus, including computer-readable storage media for handling component error of datacenter infrastructure by training machine learning models to handle errors when the models are deployed. Rather than predict or mitigate errors in the hardware of a datacenter or another site deploying the machine learning model, the model is trained to perform a task with comparable accuracy and efficiency even when some hardware on which the model is deployed fails. Component error can instead be simulated during training to cause the machine learning model deployed on the infrastructure to learn to correct errors caused by the component errors. The model can continue to be trained to compensate for periods in which not all parts of the model are available at inference. Updates to the model can be backpropagated to correct errors for handling instances of component error.

The application relates to a facial video-based automatic depression recognition method, which comprises the following steps: obtaining facial video data to be recognized and performing pretreatment to obtain facial video sub-samples V c ; a deep learning model LDSML for combining depression label distribution and label measurement is constructed; the sub-samples V c are input into the LDSML deep learning model for training; a Softmax-T activation function with a temperature coefficient is used at the output end of the model to obtain a predicted depression label distribution p, and the mathematical expectation E of the label distribution is calculated; in the prediction stage, the mathematical expectation E of the predicted depression label distribution P is used to represent the final depression prediction result. Without increasing the training data set, the feature capturing capability of the deep learning model for facial depression video data is improved. The more detailed label distribution can effectively enable the model to learn the hidden label semantic information, and effectively improve the current situation of the lack of facial depression video data.