27 results about "Accuracy improvement" patented technology

The application discloses a self-distillation action quality evaluation method based on position awareness and reward feedback, which firstly acquires an action quality evaluation data set, obtains space-time features through a multi-scale space-time feature extraction backbone network based on a deep three-dimensional convolution; secondly, the space-time features are respectively input into corresponding time-channel attention modules to obtain enhanced space-time features; the enhanced space-time features are reconstructed to a standard feature dimension space through a feature reconstruction module, and adaptive weights are obtained by an adaptive weight generator; finally, the reconstructed space-time features are respectively input into an action category classifier and a shared regression head, and a loss is calculated by combining the adaptive weights for training; in the inference stage, the action category classifier and the adaptive weight generator are removed to obtain an action quality evaluation result. The application fuses bottom-level detail features and high-level abstract representations, guarantees the precision, improves the operation efficiency and enhances the overall knowledge distillation effect.

The application discloses a kind of based on wind power equipment manufacturing process quality prediction method and system.Method includes: obtaining process parameter data, equipment state data and parts detection data and so on Multi-source heterogeneous data and pre-processing, construct multivariate quality characteristic data set;Using the method based on CCO-JMD-SE to feature decomposition, screening and reconstruction, extract key quality feature sequence;The hybrid quality prediction model combining bidirectional long short-term memory network and support vector machine is constructed, and the model hyperparameter is optimized and trained using particle swarm optimization algorithm;Real-time data are input into model to obtain quality prediction value, and the prediction value is mapped into assembly quality index by quality measurement method based on QLF-SNR, and the quality grade is determined according to grading threshold and the disposal strategy is triggered.The application realizes quality evaluation quantification, multi-source data deep fusion, prediction accuracy improvement and key process accurate control.

The invention provides a training method of a fraud recognition model, a fraud recognition method and related equipment, and relates to the technical field of artificial intelligence, and the training method of the fraud recognition model comprises the steps: obtaining training data; performing feature extraction on the plurality of training samples to obtain a tag feature vector and a plurality of first feature vectors; performing feature screening on the plurality of first feature vectors to obtain N first feature vectors; performing feature dimension reduction on the N first feature vectors according to the association weight value corresponding to each first feature vector to obtain M first feature vectors; and inputting the M first feature vectors into a preset fraud identification model, and performing iterative training on the fraud identification model based on the tag feature vector to obtain a trained target fraud identification model. In the embodiment of the invention, the fraud recognition model is trained by applying the first feature vector strongly related to the fraud behavior, so that the recognition precision of the model is improved, and the accuracy of a fraud recognition result is improved.

The invention relates to the technical field of power equipment monitoring, and discloses a transformer iron core evaluation method, which comprises the steps of obtaining multi-source data and historical fault current grounding data of a to-be-detected transformer, and forming a multi-dimensional comprehensive score evaluation system by combining trend abnormal value analysis, two-factor correlation calculation and historical fault similarity evaluation, finally, the iron core evaluation level is determined based on the comprehensive score, the interference tolerance of a monitoring system is effectively enhanced, the false alarm phenomenon caused by false fluctuation is reduced, early warning of iron core faults is achieved through time sequence trend mining and early evolution law recognition, an optimal governance window is provided for operation and maintenance personnel, and the operation and maintenance efficiency is improved. Through a standardized data processing flow and a working condition differentiation evaluation mechanism, comprehensive evaluation is carried out in combination with key parameters such as operation load and oil temperature, the accuracy and practicability of evaluation results are remarkably improved, the monitoring precision of the operation state of the transformer iron core is enhanced, and the power supply reliability of a power grid is improved. And the large-area power failure risk caused by iron core faults is reduced.

The invention discloses a finite element simulation strain correction method suitable for a display backboard stamping process, and relates to the technical field of stamping strain correction, and the finite element simulation strain correction method comprises the following steps: obtaining geometric characteristics and stamping process parameters of a display backboard to be analyzed, and constructing a finite element model; historical stamping process processing data of display backboards of the same model are collected, and corresponding process parameters and environment data are recorded; the method comprises the following steps: mapping historical process parameters and processing data to form a training data set, performing finite element analysis based on the training data to obtain simulated strain data, calculating the relative difference between the simulated strain data and historical data, analyzing the correlation between the simulated strain data and the process parameters, environmental data and residual stress, and screening the parameters meeting the correlation requirements to obtain the simulated strain data. And a regression equation is established by utilizing a regression algorithm, the simulated strain data of the to-be-analyzed display is corrected, and the method improves the simulation precision and improves the adaptability to the change of the stamping process.

The present application belongs to the field of pile foundation static load test, and relates to multi-sensor fusion technology, which is used to solve the problem that the prior art cannot realize real-time sensing of the load distribution uniformity of a heaped load body, and specifically relates to a pile foundation static load test heaped load precision analysis method based on multi-sensor fusion, which comprises the following steps: sensor arrangement for the pile foundation static load test; data acquisition and preprocessing for the pile foundation static load test; after the pile foundation static load test starts, a collection time point T i is set i , and heaped load parameter collection is performed; the present application realizes multi-dimensional real-time monitoring and dynamic analysis of the heaped load precision of the pile foundation static load test, so that heaped load abnormalities can be found and located in time, load transmission deviation from the design axis is avoided to prevent bearing capacity data distortion, meanwhile, the hidden risks after test interruption are quantitatively evaluated to provide an objective basis for subsequent scheme adjustment, the test precision can be more comprehensively and accurately controlled, and the reliability of the pile foundation bearing capacity test result is improved.

The application discloses a text2sql accuracy improvement method and system based on a multi-path retrieval mode link, belongs to the natural language processing and database query technical field, and comprises the following steps: collecting and structuring metadata to construct a data knowledge graph, constructing an NLP dictionary and a domain knowledge base and vectorizing storage; receiving a user question and utilizing a large model to combine domain knowledge to perform semantic expansion and ambiguity elimination; recalling metadata nodes from a vector library through three parallel channels of original input, rewritten questions and NLP word segmentation; grouping the recalled nodes, searching and constructing multiple "path subgraphs" representing potential correlation in the knowledge graph; introducing a dynamic weight adjustment mechanism to calculate the correlation score of each path subgraph with the user query and select the path subgraph with the highest score, extract the correlation contained in the path subgraph, and provide the accurate mode link result to a subsequent SQL generation module, so that the accuracy and efficiency of text to SQL conversion are effectively improved.

This invention relates to the field of static testing technology, and discloses a support and boundary simulation device and method for static testing of a long straight cylindrical structure. The invention employs six-degree-of-freedom statically determinate constraints to constrain the rigid body displacement of the cylindrical structure without restricting the elastic deformation of the main beam. Active loads are applied to the outer docking fixture, while passive balancing loads are applied to the constraint points of the inner docking fixture. This ensures the accuracy of the boundary load simulation at the end face of the cylindrical structure while reducing the restriction on the elastic deformation of the cylindrical structure by the constraints. Furthermore, the change in constraint reaction force can be monitored in real time during the test loading process, improving the safety and reliability of the test. This invention optimizes the cantilevered fixed-support constraint of the current conventional testing method to a hinged constraint at both ends of the long straight cylindrical structure, shortening the deformation at the maximum deformation point of the long straight cylindrical structure. This improves the accuracy of test loading and structural assessment, enhances the safety of the test, and reduces the complexity of the test loading system.

This invention provides a finishing mill specification change control method and system. The method includes: within a single finishing mill work roll change cycle, if the historical number of rolled strips of the corresponding specification after the specification change is zero, then obtaining learning values ​​matching the target final rolling temperature, width, and thickness of the strip after the specification change from a short-cycle data table; wherein the short-cycle data table is used to record the learning values ​​within a single finishing mill work roll change cycle; and performing a finishing mill pre-setting calculation for the first strip after the specification change based on the matching learning values, thereby improving the accuracy of the pre-setting parameters for the head of the first strip after the specification change. This invention can effectively improve the control accuracy of the final rolling temperature and thickness of the head of the first strip after the specification change, and improve the quality control accuracy of the first strip after the specification change.

The present application relates to the field of steel rolling, and proposes a rail bottom width adjustment method, system, device and medium, the method comprising: obtaining historical data of rail production through VR simulation, and constructing an expert experience model based on the historical data; integrating the expert experience model in a simulation platform, and simulating the working condition of rail adjustment; in response to the deviation of the bottom width exceeding a preset value, and the upper leg tip thickness and the lower leg tip thickness being greater than the head thickness, adjusting the horizontal roller and the bottom vertical roller of the rolling mill; in response to the deviation of the bottom width exceeding a preset value, and the upper leg tip thickness and the lower leg tip thickness being less than the head thickness, adjusting the horizontal roller or the bottom vertical roller of the rolling mill. The present application aims at the deviation problem of the rail bottom width, realizes high-precision control of the rail bottom width, ensures the stability of the rolling process, and significantly improves the control precision of the rail bottom width; improves the timeliness of adjustment, and ensures the standardization of steel rolling adjustment; not only solves the problem of low calculation efficiency in manual adjustment, but also avoids errors caused by human factors.

Described herein are techniques for improving language model performance through collaborative interaction between specialized and general-purpose models. A specialized model with fewer than ten billion parameters undergoes supervised fine-tuning on domain-specific data and generates initial outputs. These outputs are refined by a general-purpose model having over one hundred billion parameters and advanced reasoning capabilities. The framework implements preference tuning where outputs from both models are evaluated to generate preference triplets that optimize the performance of the specialized model. This approach achieves significant accuracy improvements while maintaining data privacy and computational efficiency.

The invention discloses an equipment fault detection method and a related device, relates to the technical field of artificial intelligence, and aims to separate the detection opportunities of various equipment parts and consumables through different detection periods, combine daily period detection with detection opportunity evaluation, improve the detection efficiency and reduce the resource consumption, and improve the detection efficiency by configuring a visual weight sequence. The influence degrees of various indexes are flexibly adjusted, the adaptability and interpretability of fault detection are enhanced, the unified health degree evaluation model is used for processing multi-source state sequence data, and automation and accuracy improvement of fault prediction are achieved.

Arrangements for providing improved anomaly detection in time series data are provided. A computing platform may receive data from one or more servers. The data may be analyzed using natural language processing to identify features in the data. A machine learning model currently deployed in a production environment may be copied and updated based on the features. The production model may be executed to generate production model outputs and the updated model may be executed to generate updated model outputs. The outputs may be compared and, if no or insufficient differences exist, the production model may be maintained in the production environment. If differences exist and are sufficient, an accuracy improvement associated with the updated machine learning model may be determined. If the accuracy improvement meets a threshold, the updated machine learning model may be deployed to the production environment. If not, the production machine learning model may be maintained.

The invention provides a multi-source data-driven intelligent early warning method and device for a safety state of foundation pit construction, and the method comprises the steps: inputting a monitoring data sequence into a pre-created prediction model, and obtaining a deformation prediction value of each time point in a future preset time period; according to the construction plan of the foundation pit, key construction stages and planned completion time of the key construction stages are obtained, and control intervals with fixed duration are set before and after time nodes of all the key construction stages; performing mechanical analysis on the key construction stage, and generating a numerical simulation prediction result corresponding to the time point in the control interval; determining the difference between the numerical simulation prediction result and the deformation prediction value corresponding to the same time point, and correcting the deformation prediction value based on the difference to obtain a corrected prediction result; and the corrected prediction result is compared with a preset grading early warning threshold value, the safety level of foundation pit construction and a corresponding early warning signal are determined, the prediction precision of the sensitive area in the construction stage is improved, and the prediction lag phenomenon caused by process conversion is eliminated.

The present application relates to a space laser radar forest structure data correction method and system, and belongs to the technical field of space laser remote sensing and forest resource monitoring. In view of the four types of errors of GEDI data, such as geometric positioning deviation, slow variation distortion, terrain widening effect and crown structure influence, the present application adopts hierarchical layout strategy to layout UAV control points, establishes the same point relationship through waveform registration and calculates the weight; based on the regular grid control points, the overall geometric correction is carried out; based on the encryption control points, the local distortion correction is carried out; the physical compensation model is established to compensate the terrain widening effect; the statistical regression model is established based on the crown structure parameters and GEDI signal quality index to correct the crown height residual error; finally, the high-precision crown height convergence result is obtained through spatial smoothing processing, and the corrected data is output. The present application realizes the gradual accuracy improvement from the whole to the local and from the geometry to the height, has low cost, high precision and wide application range.

The application particularly relates to a precision improvement method for aero-engine performance digital twinning, which comprises the following steps: establishing a compensation model; setting two output channels to respectively output two output values; comparing and calculating the two output values with corresponding performance prediction target values respectively to obtain two loss parameters; iteratively inputting the loss parameters into the basic digital twinning model and the compensation model respectively until the error converges; and predicting the aero-engine performance parameters, and correspondingly inputting the monitoring data of the engine into the converged basic digital twinning model to obtain target performance parameter prediction values. The purpose is to gradually eliminate errors through the coupling training of the compensation model and the basic model, directly couple the compensation effect of the compensation model into the basic digital twinning model without increasing the complexity of the basic model, improve the prediction precision without increasing the prediction time, and ensure the real-time performance.

This invention discloses a multi-source data linkage AI analysis and accuracy improvement system, relating to the field of computer modeling technology. It includes a data acquisition module, a static offset generation module, an environmental feature generation module, a data standardization module, a training sample generation module, and a prediction model generation module. This application acquires historical time-series data from multiple MEMS sensor devices, generates a three-axis static offset label sequence and an environmental feature sequence. After data standardization, a local fusion feature vector is generated and concatenated with the original time-series feature sequence to form a fused time-series feature vector. Based on a multi-device training sample set, a bidirectional long short-term memory network is used to construct a multi-device static offset prediction model, outputting the predicted values ​​of the standardized three-axis static offset labels. The design employs the Adam optimization algorithm to update trainable parameters until the model converges, thus realizing the prediction of the three-axis static offset of multiple MEMS sensors under different environmental conditions.

This invention provides a cutting chatter identification method based on optimized VMD and CNN-ICBAM-MHSA-BiLSTM, belonging to the field of machine tool cutting monitoring technology. It solves the technical problem in existing cutting chatter identification methods that rely on manual adjustment of VMD parameters and have low recognition rates for minor chatter. The method includes: multi-sensor signal acquisition, AOA-VMD adaptive decomposition and IMFs screening and reconstruction, multi-attention fusion network training, and chatter condition identification. Specifically, it combines the arithmetic optimization algorithm AOA with variational mode decomposition VMD to achieve adaptive parameter optimization; designs an improved convolutional block attention mechanism ICBAM, enhancing spatial feature extraction through multi-scale convolutional kernels; and constructs a CNN-ICBAM-MHSA-BiLSTM fusion network to integrate local features, temporal dependencies, and key time step attention. Experimental verification shows that this invention can accurately identify three conditions: stable machining, minor chatter, and severe chatter, with an overall recognition accuracy ≥98.40%, a minor chatter recognition accuracy improvement of ≥3.9%, and strong anti-noise interference capability, providing reliable technical support for chatter identification in machine tool cutting.

This invention discloses a method, device, and storage device for RSOC system operating condition identification, comprising the following steps: an offline training phase and an online operating condition identification phase. The offline training phase involves retrieving historical data on the stack impedance, hydrogen production flow rate, and temperature in the system, along with the corresponding operating condition types; training and obtaining a denoised sparse autoencoder that maximizes the fitness value; and establishing an operating condition mapping model. The online operating condition identification phase involves acquiring the latest data from the system; processing the new data using the denoised sparse autoencoder; and outputting the operating condition type from the processed data through the mapping model. This invention utilizes the characteristics of stack impedance, hydrogen production flow rate, and temperature, and employs a denoised sparse autoencoder optimized by a heuristic cross-search algorithm for dimensionality reduction and noise reduction. This reduces interference from external signals, provides more reliable data for subsequent operating condition identification, and improves the accuracy of RSOC system operating condition identification.

This application discloses a method and system for evaluating the contributions of multi-dimensional participants, as well as a model aggregation and optimization method. The evaluation method includes: distributing an initialized global model to the multi-dimensional participants; each participant performing iterative training based on local data; determining the comprehensive accuracy improvement factor for each client based on the local model information after each iteration; then adjusting the weights of the magnitude, direction, and uniqueness dimensions based on the obtained model accuracy change rate; obtaining a dynamically adjusted comprehensive accuracy improvement factor based on the adjusted weights; dynamically aggregating the locally trained model parameters to obtain updated global model parameters, which are then distributed to the multi-dimensional participants for local training; stopping training when the stopping condition is met and obtaining the factor for the corresponding round; and simultaneously determining the model contribution information for each participant by combining historical decay cumulative contribution. This method can efficiently and accurately obtain the contribution information of each participant.

The invention provides a visual model accuracy improving method based on data enhancement and a related device, and relates to the technical field of data processing. The pre-trained visual model is used as an initial basis, and the training time and the resource cost are reduced while the model performance is remarkably improved through multiple times of fine adjustment of enhanced data. In the fine tuning process, a part of convolutional layers are frozen, and a regularization loss function is introduced, so that the model can quickly adapt to data distribution of a specific task, and meanwhile, overfitting is avoided. And an attention mechanism is adopted in the data enhancement process, so that the reliability of the generated pseudo-annotation data on semantic consistency and feature distribution is ensured, and the cost and difficulty of manual annotation are reduced. In the whole process, through forming closed-loop optimization of data enhancement and model fine tuning, iterative improvement of model performance is promoted, and an efficient and reliable solution is provided for complex tasks.

The invention provides an IR drop lookup table model for RRAM in-memory calculation and a rapid compensation method, relates to the field of in-memory calculation, and aims to solve the problems that the calculation precision is reduced due to an IR drop effect when the size of an RRAM cross array is enlarged, and an existing scheme is inaccurate in modeling, large in time overhead or needs to be retrained. According to the method, neural network input and weight parameter sparseness are analyzed, an IR drop noise matrix lookup table model matched with array size and sparseness is established, and a rapid IR drop inference simulation framework is established; on the basis, a batch normalization layer IR drop correction method without retraining is provided, and the precision can be recovered by correcting BN parameters layer by layer once after the network parameters are written in. Compared with a current advanced precise model, the time cost is saved by 54 times, the precision can be improved by 35.55% at most, and the error between the best reasoning precision and an ideal situation is only 0.16%.

To provide a label accuracy improving device, a label accuracy improving method, and a program that can improve the probability such that a label indicates a correct answer.SOLUTION: A label accuracy improving device includes a control unit. The control unit executes a unit process. The unit process contains a learning process, a determining process, and a label updating process. The learning process estimates, using data with a label, the label based on data without a label that is data to which the label is added among learning data, and updates a mathematical model which obtains a likelihood. The determining process determines whether or not there are satisfied conditions relating to a difference between the label estimated by the already-learnt mathematical model based on the data without a label and the label contained in the learning data, and to an inference score that is a value indicating that the likelihood is large. The label updating process updates the label when the conditions are satisfied.SELECTED DRAWING: Figure 1

The invention discloses a multi-modal semantic federation learning complex working condition equipment fault diagnosis method, which comprises the following steps that: a client extracts initial characteristics by using independent encoders of different modals, and realizes intra-modal redundancy removal through abnormal region mapping and spatial pyramid compression; in a cross-modal stage, utilizing a region-guided semantic interaction mechanism to reinforce local anomaly characterization and obtain fusion features; an anomaly scoring network is constructed, anomaly scores are distributed to the marks, and local representation with the maximum information gain is optimized by combining terminal bandwidth constraint and task emergency degree; and the client performs local decoding and prediction, and the server performs federal average aggregation and broadcasts update parameters. According to the method, under the constraint that original data of each client cannot be shared, through federal cooperation of semantic perception, the accuracy rate in a recognition task of multiple public data sets can be improved by 12.69%, the single-round communication overhead can be reduced by 71.04%, the overall efficiency of a complex working condition diagnosis system is remarkably improved, and the method has a wide application prospect.

The application discloses a method for tracking and discriminating unstable modes of Thevenin equivalence considering load characteristics, which comprises the following steps: defining an equivalent load node, calculating equivalent impedance of the equivalent load node, calculating parameters of the equivalent load node of Thevenin equivalence considering load characteristics, and discriminating unstable modes, and the method is based on a traditional Thevenin equivalence tracking algorithm and a load model of the equivalent node to improve the Thevenin equivalence tracking algorithm, improve the precision of Thevenin equivalence under dynamic load or heavy load, improve the accuracy of discriminating unstable modes of a system, expand the application range of the Thevenin equivalence tracking algorithm for discriminating unstable modes of a system, and play a more efficient role in subsequent discrimination of unstable modes or other purposes.

The invention discloses an accuracy improvement method for intelligent traffic target detection based on YOLO-World in the technical field of intelligent traffic, and the method comprises the steps: obtaining a closed vocabulary data set through employing a data classification method based on a first data set and a second data set, and obtaining a closed vocabulary data set based on an open set target detection model; the closed vocabulary data set is adopted to carry out model training with the training frequency not less than two times to obtain a fusion feature detection model, the first index and the second index are adopted to screen the fusion feature detection model, a preliminary detection model is obtained, a third data set is obtained, the preliminary detection model is adopted to train the third data set, and an updated detection model is obtained. According to the method, the recognition accuracy of the model in the specific field is enhanced, the zero sample reasoning capability of the model can be kept, the model can absorb semantic information in an open vocabulary data set while learning the intelligent traffic data set through cross training, and the recognition accuracy is improved.

The goal is to clearly demonstrate to users the concrete benefits of improved demand forecasting accuracy. [Solution] The information processing device includes: a current information acquisition unit that acquires current information related to the current cost of goods distribution; an assumed information acquisition unit that acquires assumed information including accuracy improvement information indicating the expected degree of improvement in the accuracy of demand forecasting for goods, and current improvement information indicating the expected degree of improvement in the current information; and an estimation unit that estimates the effect amount, which indicates the effect of improving the accuracy of demand forecasting, based on the current information and assumed information.