60 results about "Reference modeling" patented technology

The present disclosure provides a training method and device of a query rewriting large model, equipment and medium. The method comprises: performing preliminary optimization on the query rewriting large model based on a fine-tuning dataset to obtain a reference model; and performing re-optimization on the reference model based on a reinforcement learning method to obtain a target model, wherein a reward signal in the reinforcement learning process is generated based on a deviation between a question and answer result of the reference model and a standard answer. Thus, the model can be quickly learned to the basic query rewriting mode by using the fine-tuning dataset for preliminary optimization. Then, the reinforcement learning is introduced, and the deviation between the question and answer result of the reference model and the standard answer is used as the reward signal for fine adjustment, which can effectively correct the generation deviation that may exist in the preliminary model, so that the final target model can generate a rewriting result closer to the real intention of the user in a complex query scenario, thereby improving the overall performance of the downstream question and answer system.

The application relates to the technical field of underwater detection, and discloses a 3D modeling method and device for an underwater target and an underwater robot. The method comprises the following steps: performing fixed-distance observation on a to-be-detected object according to a preset path, and acquiring a two-dimensional observation image in real time; performing real-time identification on the image, and confirming a target area; for the identified target area, based on two-dimensional visual data of the target area, a multi-view three-dimensional reconstruction algorithm is used to construct a local three-dimensional model, and corresponding navigation pose data of the underwater robot is recorded synchronously; after the observation is completed, based on each local three-dimensional model and the corresponding navigation pose data, an overall three-dimensional reference model of the to-be-detected object is constructed; finally, according to the navigation pose data, each local three-dimensional model is filled into the three-dimensional reference model, and a complete three-dimensional model is generated. The application combines selective local fine modeling and overall reference model construction based on the pose, greatly reduces the amount of image data required to be processed, and significantly improves the modeling efficiency.

The application specifically discloses a kind of wind turbine variable pitch device monitoring method and device, it discards the monitoring mode based on static historical baseline of tradition, instead constructs a kind of adaptive dynamic cross reference model.In any variable pitch event, the real-time dynamic response of any one blade to be measured is compared with the average dynamic response of the remaining blades of the same unit at the same time, thereby generating a deviation signal that can effectively filter out common operating condition interference and purely reflect the relative performance difference.Further, the deviation signal is decomposed into dynamic action error and dynamic consumption error two dimensions, by the information fusion analysis to these two dimensions, not only can the inconsistency degree of variable pitch action be accurately quantified, but also can the physical root cause behind inconsistency be preliminarily diagnosed according to the combination characteristics of different error components, thereby solving the technical problems of low monitoring accuracy, susceptible to operating condition interference and lack of diagnostic capability.

The application discloses a kind of distributed drive automobile inertial suspension semi-active control optimization method in the technical field of vehicle suspension vibration isolation, and the method comprises the following working steps: step one, establishing suspension controlled model and ideal reference model;Step two, construct three-order generalized ground semi-active control rule;Step three, phase-frequency collaborative optimization and topological structure screening;Step four, distributed drive automobile inertial suspension semi-active control optimization.The application has the advantages of reducing the performance deviation between actual response and ideal model, improving road friendliness while ensuring ride comfort, and adapting to time-varying road excitation.

The present application relates to the technical field of detection methods, in particular to a turbocharger impeller casting defect intelligent detection and control method, comprising the following steps: step one: based on the three-dimensional model of the impeller, the suction side and the pressure side of the blade are analyzed respectively, the reference picture and the respective single-side shooting axis set are determined; step two: the single-side shooting axis sets in the two recognition directions are fused into the same coordinate system, and the final coordinate system reference, the model limited coordinate set and the complete shooting angle set are obtained; step three: the impeller posture and the image acquisition equipment are adjusted according to the reference picture, and the impeller is rotated in batches according to the shooting angle set, and the surface image set is collected; step four: the sub-regions are divided, the apparent defects are detected by using the defect recognition model, and the defects repeatedly appearing in the same sensitive region are converted into processing abnormal signals and recorded.

This application relates to an adaptive learning-based robotic arm control method, apparatus, device, and storage medium. The method includes: jointly training a first initial model and a second initial model based on an offline dataset until a first preset condition is met, obtaining a first reference model and a second reference model; obtaining an online dataset corresponding to the robotic arm based on the first and second reference models; determining a hybrid dataset based on the offline and online datasets; jointly training the first and second reference models based on the hybrid dataset until a second preset condition is met, obtaining a first target model and a second target model; determining an adaptive control model based on the first and second target models; the adaptive control model is used to generate and filter optimal parameter pairs for the robotic arm in real time to achieve adaptive control of the robotic arm. Using this application can improve the adaptive capability of robotic arm control.

The application belongs to the field of gravity anomaly detection, and in particular to a gravity anomaly target body boundary estimation method and system. Based on the comparison and calculation of boundary functions and forward data sets, the characteristic line segment length vector is obtained from the extreme points of the boundary functions, a reference model is set in the observation range, the characteristic line segment length of each reference model is calculated, the boundary length and the characteristic line segment vector forward data set are established, the target to be measured is matched and compared with the data set, and the boundary estimation of the target body to be measured in the medium field gravity environment is realized. The problems of data attenuation, divergence and low resolution in gravity detection in the medium field environment are solved.

This invention relates to the field of wind power equipment monitoring and parameter identification technology, and particularly to a method and apparatus for screening and identifying operating conditions and parameters of wind turbine generators. During grid-connected operation, voltage and current at the grid connection point are collected, and a sliding data window is constructed according to the sampling period. A nonlinear state-space reference model containing the control parameters to be identified is established, and the current response within the window is calculated. Based on the model, the dynamic sensitivity of each parameter to be identified is calculated, and a Fisher information matrix is ​​constructed and accumulated. Completeness indicators such as the condition number are calculated on the Fisher information matrix and compared with a threshold, retaining valid data segments and discarding invalid ones. A nonlinear least-squares identification problem is constructed using only the valid data segments, and the proportional and integral parameters of the phase-locked loop and each control link of the converter are output using a damped Gauss-Newton iterative method. This method eliminates the need for shutdown frequency sweeping or manual fault injection, and utilizes daily random disturbances to achieve automatic screening of operating conditions, improving identification accuracy and reducing testing and data processing costs.

This invention discloses a method and system for verifying the functionality of a vector processor. The method unifies matrix, vector, and scalar data formats into a matrix type using a reference model. Multi-input interface data is stored in corresponding data queues in the form of address plus data. The required computational data is extracted according to the current instruction, and unused data is stored back to the original queue. Expected output data at the clock cycle level is generated and compared with the actual output data of the vector processor to determine functional correctness. The system includes an instruction generator module, a controller module, a data generation module, a reference model module, a data collection module, and a data comparator module. This invention can improve the accuracy of vector processor fault location, simplify verification logic, and improve overall verification efficiency.

The application discloses a chip design verification system and method, an electronic device and a storage medium, and relates to the technical field of integrated circuit design. Since the excitation descriptor chain table generated by the dynamic scheduler according to the target test scene can directly present whether it contains a transparent attribute, if the transparent attribute exists, the descriptor chain table is directly transmitted to the output comparator through a bypass channel, bypassing the complex calculation link of the reference model; if not, the traditional path is taken to process the reference model. The application provides a differentiated path selection mechanism, changes the data flow direction of the verification platform, and under the condition that the excitation descriptor chain table contains the transparent attribute, the comparison between the excitation descriptor chain table and the to-be-verified execution result can be performed in advance without waiting for the reference model operation, thereby improving the verification efficiency of the to-be-tested design.

The application relates to the technical field of fault prediction, and discloses a water level monitoring equipment fault prediction method based on multi-source data fusion, which comprises the following steps: performing stationary processing on multi-source time series data to obtain weakly stationary multivariate sequences; constructing a dynamic coupling relationship network through a dynamic conditional covariance model and a Granger causality test; extracting a topological feature vector therefrom, establishing a deviation degree reference model based on the topological feature vector of a historical normal period, and calculating the deviation degree of the topological feature vector to obtain an equipment coupling health degree index; and generating a fault prediction warning based on the evolution trend of the index and the matching degree of the index with a pre-stored fault mode; the application can improve the accuracy of a water level monitoring equipment fault prediction based on multi-source data fusion.

The present disclosure relates to a matching method of point cloud data, a matching device, and a computer readable medium. The matching device according to the present embodiment includes a generation unit that generates input models in which the amount of data of point cloud data is reduced, and generates an input model group including a plurality of the input models; an extraction unit that extracts a part of the input models from the input model group based on shape information of the input models; a calculation unit that compares the extracted input models with reference models based on reference point cloud data, and calculates a cost; a determination unit that determines whether the cost converges; and a change unit that changes the position and posture in a direction in which the cost becomes smaller in a case where the cost does not converge.

Embodiments of this disclosure relate to methods, apparatus, devices, storage media, and program products for constructing models. The method includes: determining a first feature map generated by a reference model based on sample data and a second feature map generated by a target model based on sample data; applying an attention mechanism to the first and second feature maps to determine a first weighted feature map and a second weighted feature map, the attention mechanism being associated with a channel dimension and / or a spatial dimension; determining a first loss based on a first difference between the first and second weighted feature maps; and adjusting parameters of the target model based on the first loss. In this manner, embodiments of this disclosure enable higher-quality model distillation.

The application relates to the technical field of data collection, and relates to a sports movement data collection system based on a wearable device, which comprises a signal synchronization fusion module, a reference modeling and calibration module, a signal enhancement and segmentation module, a feature matching and extraction module, a deviation quantization module and a comprehensive scoring module. In the progress process of a body measurement project, original inertia signals of a tested person are acquired; waveform features of standard actions are analyzed, baseline drift compensation is carried out on the analyzed reference waveforms, and measurement reference data are obtained; principal component signal reconstruction is carried out on the original inertia signals, and action stage signal segments of the tested person are segmented; the action stage signal segments are mapped to the measurement reference data, and key kinematic parameters are obtained; the key kinematic parameters are quantized, and action deviation degrees are obtained; comprehensive movement evaluation is carried out on the action deviation degrees, and movement performance data are obtained; and the application can improve the efficiency of sports movement data collection based on a wearable device.

The application relates to the chip verification field and provides a chip verification method, a device, equipment, a medium and a chip. The method comprises the following steps: a dynamic simulation verification environment of a design to be verified is established, a plurality of test scenes and corresponding test excitations are generated according to a design specification definition; the test excitations are sent to a driver through a sequencer, the test excitations are converted into signals conforming to an interface protocol of the design to be verified by the driver, and the signals are transmitted to the design to be verified; input signals and output signals of the design to be verified are collected in real time by a monitor, and the input signals and the output signals are converted into transactions and transmitted to a scoreboard; a reference model is compared with the output signals of the design to be verified, and a dynamic simulation verification result is generated; the state space is reduced according to the dynamic simulation result, the state space of a design part not covered by dynamic simulation in the design to be verified is traversed, and a formal verification analysis result is generated; and the design to be verified is evaluated according to the formal verification analysis result.

Apparatus and associated methods relate to a claim adjustment process management system (CAPMS) including a coverage comparison engine configured to generate a coverage analysis object (CAO) and a reference model interpretation engine configured to generate an initial coverage determination (ICD). For example, the CAPMS may receive claim object, a reference model, and a claim adjustment input (CAI). The reference model interpretation engine may, for example, apply a natural language processing (NLP) model to process claim object and the reference model to generate the ICD. The coverage comparison engine may, for example, analyze the ICD and the CAI to identify variances between the two and generate the CAO. The CAO may, for example, include an error and omission signal. The CAO may, for example, include a resolution object. Various embodiments may advantageously automate claim adjustment processes and reduce human error and ensure consistent, fair coverage determinations.

This invention provides an energy management method for hybrid vehicles. First, traffic condition information and vehicle status information are simultaneously acquired and a database is generated. Then, the acquired information is used as the state space, and the reference SOC trajectory is used as the action space. Combined with a lower-level energy management algorithm, the upper-level deep reinforcement learning agent is trained until convergence. Finally, the neural network of the trained deep reinforcement learning agent is extracted as the optimal reference model for battery SOC. Based on the neural network model and combined with the lower-level energy management algorithm, the energy consumption of the entire vehicle is allocated. This invention not only solves the problem of the difficulty in determining the SOC trajectory in energy management strategies, but also improves the effectiveness, generalization, and stability of energy management strategies for new energy vehicles under actual road conditions, thus possessing many beneficial effects not found in existing technologies.

This invention discloses an energy-saving water pump parameter adaptive scheduling method and system, relating to the field of water pump control technology. The method includes: collecting inverter active power data, pipeline pressure data, pipeline flow data, and motor speed data; performing time alignment processing; extracting response features from the time-aligned water pump operating condition data to obtain dynamic response features; performing dynamic behavior distortion analysis based on a dynamic response reference model and the dynamic response features to obtain dynamic behavior distortion analysis results; if the dynamic behavior distortion analysis results indicate the existence of behavior distortion, then canceling the modification of water pump parameters by the multiplicative correction coefficient. This invention can combine water pump operating condition data and dynamic response features to analyze behavior distortion and cancel water pump parameter modifications, thereby achieving adaptive scheduling of water pump parameters, improving accuracy and stability, and reducing energy consumption per unit of water volume.

The application relates to a kind of multi-mobile robot obstacle avoidance control methods based on dynamic control barrier function, comprising the following steps: S1, obtaining multi-mobile robot state information and dynamic obstacle point cloud data;S2, pre-processing is obtained with several obstacle point cloud clusters;S3, ellipse parameterization fitting is obtained with ellipse obstacle parameters;S4, based on ellipse obstacle parameters, the ellipse obstacle sequence in prediction time domain is obtained;S5, extended ellipse obstacle sequence is obtained;S6, establishment formation reference model generates reference trajectory;S7, construct discrete-time dynamic control barrier function constraint as safety constraint;S8, calculate obstacle avoidance risk index and obstacle avoidance responsibility coefficient and adjust formation keeping weight and obstacle avoidance constraint weight in model predictive control;S9, based on the weight after adjustment, establish model, obtain prediction control input sequence, and carry out rolling optimization.The beneficial effects of the application are: improve modeling accuracy, motion trend utilization rate and prediction error safety margin accuracy.

A controller that performs PI control on a control input to a controlled object on the basis of a difference between a target value and an output, the controller including a model converting part that converts an ultra-local model obtained by modeling the controlled object into a state space model by defining a differential of an output of the controlled object including measurement noise by the control input and a deviation amount with respect to a reference model, an estimation part that estimates the deviation amount of the ultra-local model on the basis of a Kalman filter constructed from the state space model, and a steering angle calculation part that obtains the control input on the basis of the estimated deviation amount, a proportional gain, and an integral gain.

This invention relates to the field of wind power equipment monitoring and parameter identification technology, and particularly to a method and apparatus for screening and identifying operating conditions and parameters of wind turbine generators. During grid-connected operation, voltage and current at the grid connection point are collected, and a sliding data window is constructed according to the sampling period. A nonlinear state-space reference model containing the control parameters to be identified is established, and the current response within the window is calculated. Based on the model, the dynamic sensitivity of each parameter to be identified is calculated, and a Fisher information matrix is ​​constructed and accumulated. Completeness indicators such as the condition number are calculated on the Fisher information matrix and compared with a threshold, retaining valid data segments and discarding invalid ones. A nonlinear least-squares identification problem is constructed using only the valid data segments, and the proportional and integral parameters of the phase-locked loop and each control link of the converter are output using a damped Gauss-Newton iterative method. This method eliminates the need for shutdown frequency sweeping or manual fault injection, and utilizes daily random disturbances to achieve automatic screening of operating conditions, improving identification accuracy and reducing testing and data processing costs.

This invention proposes a game theory-based distributed drive electric vehicle chassis coordinated control method, including: establishing a dynamic and kinematic model and a two-degree-of-freedom reference model for the distributed drive electric vehicle; in the design of the upper-level coordinated controller, a game-based distributed model predictive control is adopted, with the control objective of active front wheel steering being tracking accuracy, active rear wheel steering being lateral stability, direct yaw moment being yaw stability, and active suspension being ride comfort and rollover stability, while also designing a switching control for ride comfort and rollover stability; furthermore, an additional yaw moment is distributed to each wheel through a lower-level torque distribution controller. This invention solves for the optimal control quantity for each player through a game theory-based distributed model predictive architecture, achieving multi-objective optimization and effectively improving the vehicle's trajectory tracking accuracy, handling stability, rollover stability, and ride comfort.

This invention relates to the interdisciplinary field of artificial intelligence and media asset management, and particularly to a training method and system for a graded cataloging model of broadcast television media assets. The method includes: constructing graded preference data composed of preferred and unpreferred samples based on broadcast television media asset cataloging specifications; extracting and fusing multimodal features from and fusing video, audio, and text information of broadcast television media assets; constructing a strategy model and its corresponding reference model and freezing the reference model parameters; introducing a broadcast television sensitive word library and configuring dynamic regularization constraint parameters; training the strategy model using a direct preference optimization method based on the graded preference data; constructing a loss function and updating the model parameters based on the output differences between preferred and unpreferred samples; and fixing the model parameters when a preset training termination condition is met, thus obtaining a graded cataloging model for broadcast television media assets. This method effectively utilizes the preference information between cataloging results, improving the model's adaptability to the graded cataloging requirements of broadcast television media assets.

The application discloses a skin state evaluation method and system based on torque trigger data, and belongs to the technical field of skin state detection. The method comprises the following steps: acquiring skin part information and searching a database to obtain scene adaptation parameters; establishing a body surface coordinate system and collecting contact pressure data and probe attitude data; controlling the probe to move and performing dynamic torque balance regulation and control to obtain dynamic torque balance regulation and control parameters; performing feature matching based on the parameters and a reference model to obtain a feature trigger signal; recording a sliding distance and collecting synchronous data in response to the signal to obtain a complete torque displacement correlation data set; extracting skin biomechanics feature parameters; and obtaining skin state classification labels through fusion calculation. The application adopts a scheme of loading a reference model according to a skin part, performing dynamic torque balance regulation and control in sliding, and collecting a termination point based on torque curve feature trigger data, and can realize adaptive, accurate and objective evaluation of the skin biomechanics state.

The application provides a facial expression model generation method and device, an electronic device and a storage medium. The method comprises the following steps: determining an expressionless character model of a target character; determining a target expressionless reference model of a target reference character and a plurality of target expression reference models of different expressions of the target reference character from a model library according to facial features of the expressionless character model; and generating a plurality of expression character models of different expressions corresponding to the expressionless character model according to morphing data between each target expression reference model and the target expressionless reference model. The method can automatically generate a plurality of expression character models of different expressions of the target character, improve the generation efficiency of the expression character model, and determine the corresponding target expressionless reference model through the facial features of the expressionless character model, thereby improving the accuracy of the expression character model in expressing the expression and improving the generation effect of the expression character model.

This invention discloses an algorithm for identifying sudden changes in electricity load based on a multi-scale dynamic model, relating to the field of power load monitoring and anomaly identification technology. It performs time alignment, missing data correction, and anomaly smoothing on electricity consumption and power data within a continuous monitoring period, constructing a load feature vector containing load values, relative change rates, and relative change rate accelerations. Combining the temporal context of the current moment to be identified, it extracts recent trend samples, date-type samples, periodic similar samples, and seasonal same-period samples, establishing local feature models and forming a multi-scale dynamic reference model. Based on this, a target reference model matching the current scenario is generated. Finally, based on load value deviation, change rate deviation, and trend reversal deviation, it completes the identification of sudden changes in electricity load, quantifies the severity of anomalies, and outputs results. This algorithm is suitable for electricity consumption scenarios with complex load patterns and diverse fluctuation characteristics.

The present disclosure relates to a method and device for generating a three-dimensional character model of a virtual character and a medium, the method comprising: obtaining a two-dimensional character image of the virtual character; generating, according to the two-dimensional character image, a three-dimensional posture reference model representing a spatial posture of the virtual character and a plurality of independent three-dimensional components constituting a body of the virtual character; wherein the plurality of three-dimensional components correspond to different body parts of the body of the virtual character respectively; and performing spatial alignment and assembly on the plurality of three-dimensional components according to the three-dimensional posture reference model to generate the three-dimensional character model of the virtual character.

The present application is suitable for the technical field of building energy consumption analysis, and provides a building energy consumption dynamic response analysis method and system under meteorological evolution, which comprises the following steps: integrating a multi-source meteorological data set and constructing a building energy consumption reference model; generating sample combinations of uncertain parameters by using a sampling method, running the building energy consumption reference model, obtaining energy consumption output results under different climate scenarios, and statistically analyzing the energy consumption output results; based on the energy consumption output results, using a global sensitivity analysis method to analyze the contribution of each uncertain parameter to the energy consumption, and identifying key driving parameters; based on the sample set generated in the uncertainty analysis step, using a machine learning algorithm to train a building energy consumption prediction model, and outputting building energy consumption prediction results for target meteorological conditions. The present application solves the problem that the traditional method is inaccurate in evaluating under non-stationary climate, and can provide quantitative and robust decision support for climate-adaptive building design and energy-saving strategy formulation.

The application discloses a brushless direct current motor position sensor signal reconstruction method based on a reference model and relates to the technical field of motor control. The method comprises the following steps: collecting three-phase voltage and current signals of a motor and performing pretreatment; inputting the signals into a sliding mode observer and a preset reference model in parallel to obtain first and second rotor position estimation values respectively; calculating the difference between the two values to generate a position error signal, taking the signal as feedback, and dynamically optimizing the internal parameters of the sliding mode observer through an adaptive law; and reconstructing a high-precision rotor position signal in real time based on the optimized sliding mode observer, so that the system can be seamlessly switched to the reconstructed signal when a physical sensor fault is detected, and smooth fault-tolerant operation is realized. The application realizes accurate reconstruction of the position signal and smooth and undisturbed fault-tolerant switching by combining the sliding mode observer with the reference model.