25 results about "Parameter learning" patented technology

The application discloses a multi-behavior recommendation method and device based on contrast clustering learning and a medium. The method first uses a graph convolution network to learn user and item embeddings of each behavior, and then designs three types of tasks to improve the embedding quality: a) behavior-level embedding, an adaptive parameter learning strategy is used to obtain embedding weights of each behavior of each user, and a weighted method is used to aggregate embedding expressions of all behaviors of each user; b) instance-level embedding, a contrast learning method is used to optimize user embeddings and item embeddings under different behaviors; c) clustering-level embedding, a contrast clustering learning method is used to explore potential clustering information between user embeddings or item embeddings to obtain more comprehensive user embedding expressions and item embedding expressions, and to alleviate the problem of data sparsity. Finally, the three tasks are jointly learned by weighting. The application is simple and effective, and through comparison with other methods and testing on known data sets, the application has good performance.

This application relates to the field of geological engineering technology. To address the problems of time-consuming parameter learning and errors caused by reliance on subjective experience in undirected probabilistic graphical geological model construction, a method for constructing undirected probabilistic graphical geological models based on spatial location encoding is disclosed. This method includes: determining the geological modeling range based on known borehole data; discretizing the geological modeling range into a grid structure; filling the corresponding grid with stratigraphic data from the known borehole data according to soil type; spatially encoding the unknown grids based on the spatial relationship between the grid structure and the known borehole data; assigning an initial state to the unknown grids based on the spatial relationship between the grid structure and the known borehole data to obtain the initial stratigraphy and constructing an undirected probabilistic graphical geological model; and constructing a data-driven parameter optimization method based on the spatial location encoding results to learn the parameters of the undirected probabilistic graphical geological model and obtain the optimal parameters. This method improves the accuracy of undirected probabilistic graphical geological model construction.

This invention discloses a backstepping-based variable iterative learning control method for robotic arms, applicable to motor-driven robotic arm systems. This method converts the robotic arm system model into a third-order strict feedback form. By defining coordinate transformation error and introducing a command filter, an auxiliary system is constructed to compensate for the deviation between the filter and the virtual control law. A radial basis function neural network is used to approximate the unknown dynamics and disturbances of the system online. A backstepping controller is designed based on the compensation error, and a parameter learning law with variable iteration length is constructed to ensure system stability and convergence of the compensation error. By dynamically adjusting the iteration length during the control phase, this method can reduce computational resource consumption while maintaining high-precision tracking performance. When used for robotic arm control of repetitive tasks, it significantly improves the tracking accuracy and robustness of the system, demonstrating good engineering application value.

The application provides an anomaly recognition method and device based on post-processing fusion and a readable storage medium. The method comprises the following steps: obtaining the first output result of the MinSP, Max logit and Entropy post-processing recognition method according to a semantic segmentation network and an original training set, obtaining the first mean value and the first variance in the statistical stage; obtaining the second output result of the post-processing recognition method according to the semantic segmentation network after parameter learning on the original training set and an image test set, obtaining the summation result of the normalized second output result; performing normalization processing on the summation result and correcting the recognition output result; determining the region where the abnormal object is located in the test image in the image test set according to a preset threshold and the corrected recognition output result, and obtaining an abnormal recognition result. The application can automatically and efficiently recognize the out-of-class target stably.

The application is a large model multilingual security alignment system and method based on joint rejection direction. The application relates to the technical field of large model security alignment. The application constructs multilingual security parallel samples, uses the value flow mask on the frozen large language model to construct positive security state and negative security state, further extracts the joint rejection direction shared across languages from the internal representation difference of the two, and uses the layer-by-layer projection track in the direction as the internal description of the security behavior. Through lightweight parameter learning, the track of the low-resource language approximates the high-resource anchor language, thereby improving the security rejection ability of the low-resource language under the condition of few samples, and keeping the original general ability of the model as much as possible.

This invention discloses an image recognition method, system, and medium based on knowledge distillation temperature parameter optimization in the field of image recognition technology. The method includes the following steps: obtaining a teacher model, a student model, and a training dataset; dividing the training dataset into several mini-batch datasets; and constructing a distillation loss function; the distillation loss function includes cross-entropy loss and KL divergence loss, where the KL divergence loss is based on temperature parameters; obtaining the upper and lower bounds of the KL divergence loss; obtaining an adaptive threshold based on the upper and lower bounds of the KL divergence loss through weighted interpolation; alternately executing the following temperature optimization and parameter learning stages in batches; and recognizing images based on the optimized student model. This invention sets an adaptive threshold based on the upper and lower bounds of the KL divergence loss, using this threshold as a boundary to control the KL loss during distillation within an ideal range, avoiding underfitting of the student model and improving its generalization ability and accuracy.

The application provides an insulin resistance dynamic intervention method and system based on a metabolic flow graph network and reinforcement learning, and belongs to the technical field of insulin resistance intervention. The method comprises the following steps: acquiring first time series data and performing preprocessing; based on a feature matrix obtained through the preprocessing, performing individualized parameter learning through a heterogeneous graph neural network, constructing an individualized metabolic flow graph network model to generate a metabolic state vector; inputting the metabolic state vector into a reinforcement learning intelligent agent to generate an individualized intervention strategy vector; outputting and executing the strategy, and simultaneously performing online co-evolution updating on the metabolic network model and the reinforcement learning intelligent agent based on new data and feedback after the execution. The application solves the technical problems of the static, universal and unable-to-adapt-to-individual-metabolic-dynamic-fluctuation and multi-objective-weighting of existing intervention schemes, and realizes precise, real-time, adaptive and interpretable dynamic individualized intervention.

The application discloses a kind of self-determination quasi-state control and anti-disturbance method of intelligent shuttle vehicle of cold storage.The method constructs environment space gradient field by distributed optical fiber sensing network, and inverts heat flow disturbance to update dynamic environment model;Synchronous analysis driving current harmonic to identify ground phase change state, and utilize digital twin model to compensate mechanism hysteresis effect;Adopt reflection-adaptation-learning three-layer decision architecture: reflection layer millisecond level responds to burst risk;Adaptation layer adjusts motion parameters for seconds;Learning layer minutes level planning quasi-state navigation path;Finally, through model predictive control, the instructions of each layer are fused, and the output of collaborative control signal is obtained.The application enables the shuttle vehicle to actively perceive, predict and adapt to the non-uniform dynamic environment of the cold storage, improving the operation safety, positioning accuracy and work efficiency in low temperature, slippery and multi-disturbance working conditions.

The application provides a bridge disease chain diagnosis method based on physical constraints and a Bayesian network, comprising the following steps: S1, obtaining historical detection data of a target bridge to construct a standardized disease transaction data set; S2, mining a strong correlation disease item set meeting a preset threshold; S3, introducing a bridge physical constraint rule library, and constructing a bridge hazard chain evolution Bayesian network structure according to the strong correlation disease item set; S4, performing parameter learning on the bridge hazard chain evolution Bayesian network structure; and S5, inputting current performance disease detection results of the target bridge into the bridge hazard chain evolution Bayesian network structure after parameter learning as evidence variables, and calculating posterior probabilities of unobserved hidden diseases occurring by a probability reasoning algorithm as diagnosis results. The method can realize the fusion of data correlation characteristics and physical space constraints, automatically construct a disease chain evolution model, and quantitatively infer hidden diseases.

This invention provides a method, system, and training method for predicting the tunneling speed and cutterhead torque of a tunnel boring machine (TBM). The method includes: collecting on-site monitoring data during TBM construction to construct a sample dataset; preprocessing the sample dataset to form machine feature inputs and geological feature inputs, and dividing it into training and testing sets; constructing a prediction framework and learning parameters based on the training set; constructing a decoupled convolutional neural network model with two inputs, extracting and fusing machine and geological features respectively; constructing an adversarial training architecture and training the decoupled convolutional neural network model; and comparing the actual results with the model prediction results to evaluate the prediction effect. This invention is more suitable for active parameter optimization and assisted driving scenarios in TBMs, and can capture complex coupling relationships, thereby improving the accuracy and stability of joint prediction. It avoids errors caused by relying on geological identification or generative enhancement while exhibiting better generalization and robustness under cross-regional and cross-stratum conditions.

The present application relates to the technical field of mechanical arm control, in particular to a self-adaptive modular robot grabbing system and method, the system comprising a vision-guided registration module, a workpiece feature perception module, a contact state perception module, a grabbing parameter learning module and a dynamic grabbing control module. The present application can convert the workpiece contour position, width and spindle orientation into posture basis that can directly constrain the grabbing action through the cooperative registration of calibration images, end reference points and pose data. The contact process can be jointly represented by the opening width ratio, displacement rollback, current change and contact deformation variable, which can more accurately distinguish the contact state and clamping trend. The lifting following result and the trial grabbing sample are regressed into the opening and closing stroke, the closing rhythm, the holding current and the release timing, so that the formal grabbing stage can continuously roll around the end deviation and the contact change to correct the position, the closing increment, the current output and the closing direction, thereby compressing the alignment error and reducing the risk of slipping and extrusion.

This invention specifically relates to a cluster error correction parameter learning method and system, encompassing the fields of distributed cluster communication and error correction technology. It includes: a cell state space definition module; a dynamic neighborhood topology construction module; a local transition rule engine module; an asynchronous random update engine module; and an entropy monitoring and stability detection module. In this invention, cell neighbors are defined along both physical and logical dimensions to accurately adapt to dynamic cluster changes such as node online / offline status, link failures, and replica migration. Furthermore, direct reading from hardware registers enables microsecond-level acquisition of physical parameters. Discretized encoding and lightweight custom messages significantly reduce communication and computational overhead, meeting the core requirement of lightweight cluster interaction.

An apparatus of a vehicle comprises a memory storing at least one instruction and a processor configured to execute the at least one instruction. The at least one instruction may be configured to cause, when executed by the processor, the apparatus to: via a tuning parameter learning model for image correction, generate, based on received video data, a tuning parameter for adjusting image signal processing (ISP) for correcting the received video data; correct, based on the tuning parameter, the received video data; identify, via a video recognition model, at least one object in at least one image corresponding to the corrected video data; and control, based on the identified at least one object, autonomous driving of the vehicle.

The application discloses a pet toilet behavior monitoring method and device based on a millimeter wave sensor. The method comprises the following steps: a millimeter wave radar module continuously outputs continuous frame data containing target detection state, target distance value and motion energy value; a master control chip maintains a bidirectional buffer counter, the bidirectional buffer counter has an entry confirmation threshold and a departure judgment threshold, when a frame has a target, the count value increases towards the entry confirmation threshold, when a frame has no target, the count value decreases towards the departure judgment threshold, if a target is detected again during the decreasing process, the count value is automatically switched back to the increasing direction; when the count value increases to the entry confirmation threshold, an entry time stamp is recorded, when the count value decreases to the departure judgment threshold, a departure time stamp is recorded, it is determined that the present toilet event is over and a toilet event record is generated. The application also comprises motion energy integration to distinguish different behavior modes, mapping of indicator light state and counter state to realize algorithm visualization, and a parameter learning closed loop based on user manual correction driving to realize continuous optimization of detection precision. The device comprises a millimeter wave radar module, a master control chip, a wireless communication module and an indicator light module.

This invention provides a laboratory safety self-control method and system for fault self-diagnosis. The method includes: S1 periodically and actively generating health labels through self-checks; S2 dividing the time slice of a single processor into execution time slots and self-test time slots, calculating the CRC signature of key data in the execution time slot, executing test vectors and comparing them with historical signatures in the self-test time slot, and locking the output to a safe state via independent hardwired connection in case of a fault; S3 based on incremental causal graph and online parameter learning, collecting samples using fault labels, updating the conditional probability table through online EM, automatically expanding the causal graph structure through normalized mutual information, and outputting the fault type and confidence level using confidence propagation; S4 actuator closed-loop feedback and hardwired bypass protection; S5 dynamic reconstruction degradation fault tolerance; S6 emergency shutdown and solidification. This invention achieves high-coverage fault detection with a single processor, eliminates blind spots of common faults, and simultaneously achieves small-sample adaptive and accurate fault reasoning, significantly improving system reliability and intelligence.

There is provided a video processing method performed by a computing device, the method including the steps of: collecting video from an external device; generating preprocessed data by extracting two-dimensional or three-dimensional skeleton information from the video; pre-training a first artificial intelligence model including N transformer blocks from the preprocessed data by applying an attention from a body of an object to a plurality of joints, an attention from each of the plurality of joints to the body, and an attention between persons; and learning, when parameters determined as a result of the pre-training of the first artificial intelligence model are transferred, a method of recognizing an action from the video received from the external device, using a second artificial intelligence model including the N transformer blocks on the basis of the parameters, wherein N is a natural number equal to or larger than 2.

The application relates to the technical field of automobile sunroof control, and particularly discloses a self-learning calibration system and method for an anti-pinch force curve of an automobile sunroof. The system comprises a data acquisition and processing module, a parameter learning module, a self-adaptive curve generation module, an anti-pinch control execution module and a safety verification module. The system acquires motion data of the sunroof closing process in real time, and continuously updates internal parameters of a sunroof dynamics model by using an online recursive algorithm. After each parameter update, a self-adaptive anti-pinch force curve matched with the current mechanical state is calculated and generated according to the new parameters. Before the curve is put into use, safety verification is performed by comparing virtual simulation with historical data. The method solves the problem that a fixed anti-pinch force curve is inaccurate due to time-varying of the mechanical characteristics of the sunroof, realizes self-adaptive calibration of the anti-pinch threshold in the whole life cycle, and guarantees the reliability of the learning process through a safety verification mechanism, thereby improving the safety, accuracy and adaptability of the sunroof anti-pinch function.

This invention discloses a method for recommending points of interest (POIs) in different locations based on user preferences and personalized preference transfer, belonging to the field of terminal location-based recommendation. The method includes: constructing a heterogeneous hypergraph for five different types of nodes, and obtaining user preference representations through training the hypergraph; constructing a POI-category graph, and learning POI representations through a continuous skipping word model; constructing an attention network with POI representations as input to obtain user-transferable features; constructing a parameter learning network using a multilayer perceptron and user-transferable features as input, and constructing a transfer network with user preference representations as input and the output of the parameter learning network as parameters to achieve personalized user preference transfer; constructing a geographic map between POIs based on latitude and longitude, and learning the embedding representations of different POIs through a convolutional network; calculating the score for each POI by combining the user's transferred preferences with the embedding representations of different POIs, thus completing the final recommendation.

This application provides a method, system, equipment, and storage medium for analyzing the jamming risk of tunnel boring machines (TBMs), relating to the field of tunnel boring technology. The method includes: obtaining the causal hierarchy of construction risk factors for TBMs; converting the causal hierarchy into Bayesian network topological constraints; constructing a three-layer temporal risk Bayesian network based on the Bayesian network topological constraints, comprising a bottom-level causal layer, an intermediate symptom layer, and a top-level event layer; performing hybrid parameter learning on the three-layer temporal risk Bayesian network based on historical observation data to obtain the prior probability of the bottom-level causal layer and the conditional probability of the entire network; and performing dynamic risk deduction based on the prior probability of the bottom-level causal layer and the conditional probability of the entire network to identify the evolution path of the target risk. This application effectively improves the accuracy of TBM jamming risk assessment by explicitly and accurately quantifying the coupling effect of multiple factors through conditional probability.

This invention proposes a path tracking control method for polar unmanned surface vessels (USVs) in a polar ice-water mixed environment, belonging to the field of polar USV control technology. It solves the problem that existing polar USV path tracking control methods often suffer from reduced control accuracy, insufficient robustness, and poor disturbance rejection capabilities, as they struggle to balance high path tracking accuracy with the stability of the dynamic system. The method includes the following steps: Step 1: Constructing a kinematic-dynamic hierarchical control architecture for the polar USV and designing a kinematic controller based on the Model Predictive Control (MPC) algorithm; Step 2: Constructing a radial basis function neural network based on the minimum parameter learning method to estimate unknown coupled disturbances online; Step 3: Designing a dynamic controller based on adaptive dynamic sliding mode to drive the polar USV in path tracking; Step 4: Verifying the stability and robustness of the polar USV system based on Lyapunov theory. It is primarily used for path tracking control of polar USVs.

This invention discloses a communication sensing computing resource scheduling method for multi-point cooperative wireless sensing services, comprising: constructing a multi-point cooperative sensing model; analyzing the execution latency of sensing tasks and target sensing performance under the established multi-point cooperative scenario; constructing an optimization problem that minimizes task execution latency based on sensing, cooperative computing node selection, and synsensory computing resource allocation; transforming the node selection and synsensory computing resource scheduling problem into a Markov decision process; and obtaining the optimal node selection and resource scheduling scheme by learning and updating parameters based on a two-stage reinforcement learning algorithm. This invention overcomes the bottlenecks of resource contention and cross-domain cooperation by collaboratively optimizing multi-base station resources, achieving high-precision, low-latency, and high-reliability target sensing performance.

This invention relates to a method for predicting and coordinating flood hammer in a multi-pump water supply system, comprising: pump system initialization and characteristic curve calibration; real-time data acquisition and acceleration calculation; threshold judgment and event identification; control execution; multi-pump collaborative control; and online parameter learning and optimization. This invention, through joint judgment of flow rate, pressure, and acceleration, can provide early warning and initiate response before the arrival of water hammer or flood peaks, reducing the response time by more than 50% compared to traditional PID control. This effectively suppresses peak water hammer pressure, reduces the risk of pipeline rupture, and extends the service life of the pipeline network. This invention avoids frequent adjustments through dynamic threshold control, and combined with an energy-optimized response strategy, can reduce system energy consumption by 10%–20%. The multi-pump collaborative control effectively avoids resonance problems during parallel operation, improves system stability, and has adaptive control parameter capabilities, enabling online learning and adjustment based on actual operating conditions, making it suitable for various complex water supply scenarios.

This invention provides a method for identifying cable-stayed bridge damage based on a continuous Bayesian network (CBN), comprising the following steps: Step 1: Establishing a cable damage sample library; using finite element simulation combined with random sampling to establish a damage sample library for cable-stayed bridge cables; Step 2: Network topology and parameter learning; Step 3: Network inference and damage identification; using the rate of change of cable force of a certain cable or part of the cable as observation evidence, inputting it into the CBN, and inferring the posterior probability density distribution of other cables under the damage condition at time t; This technical solution can be used in practice when cable-stayed bridge monitoring data is incomplete, and to infer the cable damage condition through the rate of change of cable force.