138 results about "Si model" patented technology

The invention discloses a transverse federated learning system optimization method and device, equipment and a readable storage medium. The method comprises the steps that a target type of local modelparameter updating needing to be sent by each participating device in each round of model updating is determined from parameter updating types according to a preset strategy, and the parameter updating types comprise model parameter information and gradient information; sending indication information for indicating the target type to each participation device, so that each participation device performs local training according to the indication information and returns local model parameter update of the target type; and fusing the local model parameter updates of the target type received fromeach participating device, and sending the global model parameter updates obtained by fusion to each participating device, so that each participating device performs model updating according to the global model parameter updates. According to the method, the advantages of a gradient averaging algorithm and a model averaging algorithm are combined, and a hybrid federated averaging mechanism is realized.

The invention relates to an experiment method for jointly and synchronously measuring model motion track, object plane pressure and space velocity field during pitching-rolling two-degree-of-freedom dynamic simulation motion control process. A pitching motion mechanism and a model support rod are simultaneously driven through an industrial personal computer to respectively drive the model to make pitching and rolling motion along preset tracks. When the model moves to a position requiring data acquisition, the industrial personal computer sends an instruction to pressure measuring equipment and PIV (Particle Image Velocimetry) equipment at multiple points, and the actual motion position, the object plane pressure and the space velocity field of the model are collected synchronously. Experiment results show that two-degree-of-freedom coupled simulation motion has high precision under the control of PID (Proportion Integration Differentiation) closed loop feedback, and the error of synchronous measurement can be reduced by transmitting an external trigger signal of synchronous measurement according to a partitioning method. The joint and synchronous experiment technology provides an effective research means for the analysis of the large angle-of-attack unsteady flow mechanism.

The invention discloses a method and a system for efficiently estimating the near-surface PM2.5 (particular matter 2.5) concentration. The method includes a model building step and a model estimating step. The model building step further includes a data compressing sub-step, extracting main spectral signal structure characteristics of remotely sensed data; a data matching sub-step, extracting corresponding remotely sensed information according to spatial coordinates of PM2.5 ground monitoring data; a model building sub-step, building estimation models according matched data sets. The model estimating step further includes an estimation requesting sub-step, preprocessing estimation input data; an estimating sub-step, estimating the near-surface PM2.5 concentration according to estimation requests and outputting estimation results. The method and the system have the advantages that the near-ground PM2.5 concentration can be estimated according to the MODIS (moderate resolution imaging spectroradiometer) observation data on the basis of the artificial neural network models, and accordingly remote sensing operational monitoring requirements can be met; the method and the system can support importing of meteorological factors, and accordingly the near-surface PM2.5 concentration can be quickly and efficiently dynamically monitored on a large scale.

The invention discloses an image vehicle attitude estimation method based on a fine three-dimensional model. The method comprises steps: first, the position of the vehicle being detected from the image, and initializing the vehicle three-dimensional posture parameters, rendering the vehicle three-dimensional model to the image plane using the current posture parameters, extracting the contours ofthe image vehicle and the model vehicle respectively, using the matching error between the contours to construct the energy function, and adopting the Gauss-Newton algorithm to optimize the posture parameters, solving the energy function minimization problem, and obtaining the final result. In contour matching, the improved random sampling consistency (RANSAC) algorithm is used to fit the contourof the curve model with piecewise straight lines, and then the contour is matched with the real vehicle contour. The invention can accurately and robustly recover the three-dimensional posture parameters of the vehicle relative to the camera in the surveillance video in the complex surveillance scene, which is of great significance to the understanding and automatic processing of the surveillancevideo.

The invention discloses a multi-model control method based on self learning. The multi-model control method comprises the steps that (1) a model base is built, and the model base consists of a group of local models of a non-linear model; (2) a group of controllers are built, and a group of local controllers are designed according to the local model in the model base; (3) the performance evaluation is executed: output errors and differences between system output y and model output yi are observed, and a performance feedback or value function is calculated or sent to an API (application program interface) module on the basis of signals; and (4) a similar policy iteration algorithm is executed: performance feedback signals are observed, error signals between reference output and system output are received, the signals are used as the Markov decision process states, and meanwhile, the states are fed back to become return signals for enhancing the leaning. The multi-model control method has the advantages that the principle is simple, the application range is wide, the reliability is high, the general performance and the convergence of the control can be ensured, and the like.

The automatic modeling method of the business service model comprises the steps of constructing a model, training the model, evaluating the model and applying the model, and performing continuous dataexploration and model optimization in the process of training the model and the application model; Constructing a model: carrying out data cleaning, data mining, data conversion and data writing by uploading a training set, automatically and quickly matching an algorithm model, and providing multi-dimension assessment model quality; Training a model: training the model through the test set by using a machine learning technology; And model application: applying the model to a real business scene to verify the model, and outputting a visual chart report according to business requirements afterthe model is evaluated and optimized through a test set. The invention provides a big data technology automatic modeling method covering a data mining program module, a machine learning module and anartificial intelligence program module for the whole industry.

The embodiment of the invention provides a three-dimensional simulation method and a system based on virtual reality. The method comprises the following steps: creating an equipment initial model based on equipment basic information; optimizing the equipment initial model to obtain a model optimization result; rendering the equipment initial model to obtain a model rendering result; segmenting theequipment initial model to obtain a model segmentation result; importing the model optimization result, the model rendering instance and the model segmentation result into preset software to obtain preset model import information; performing simulation scene building on the preset model import information based on the environment elements to obtain a model simulation scene building result; and performing scene motion simulation interaction design on the model simulation scene building result by adopting a hybrid collision detection algorithm to obtain an equipment three-dimensional simulationresult. According to the embodiment of the invention, three-dimensional display is carried out by combining equipment with VR interactive operation, so that the authenticity of the equipment is improved, and the user experience is improved by using vivid simulation animation.

The invention discloses a multi-view three-dimensional model retrieval method and a multi-view three-dimensional model retrieval system based on pairing depth feature learning. The multi-view three-dimensional model retrieval method comprises the steps of: acquiring two-dimensional views of a to-be-retrieved three-dimensional model at different angles, and extracting an initial view descriptor ofeach two-dimensional view; aggregating the plurality of initial view descriptors to obtain a final view descriptor; extracting potential features and category features of the final view descriptor respectively; performing weighted combination on the potential features and the category features to form a shape descriptor; and performing similarity calculation on the obtained shape descriptor and ashape descriptor of the three-dimensional model in a database to realize retrieval of the multi-view three-dimensional model. According to the multi-view three-dimensional model retrieval method, a multi-view three-dimensional model retrieval framework GPDFL is provided, potential features and category features of the model are fused, and the feature recognition capability and the model retrievalperformance can be improved.

The invention provides a method, a device and equipment for executing an automatic machine learning process. The method comprises the following steps: providing a model training operator and a model prediction operator which are independent of each other; training a machine learning model by using the model training operator based on training data; and providing a prediction service for predictiondata by using the model prediction operator and the trained machine learning model.

The invention discloses an opportunity spectrum access method based on a sideband observation information multi-arm tiger machine model. The method comprises the following steps: firstly, aiming at asecondary user access problem under the conditions that cognitive network channel information is unknown and channel perception is imperfect, carrying out detection perception on N channels, and selecting a channel problem suitable for access to be modeled as MABP-SI model; sensing all authorized channels by a secondary user, recording a sensing observation result, updating a probability vector, and recording a channel set which is sensed to be idle; creating a candidate set of spatial probability vector estimated values, and selecting an estimated value of an idle probability vector; judgingwhether the idle channel set is empty or not at the time slot t: if the set is empty, not accessing the secondary user to any channel, and if the set is not empty, selecting k authorized channels withthe maximum coefficient to access; and finally, updating t to t + 1, and circulating. According to the method, the income loss of channel access under the statistical asymptotic condition is reduced,and the method has the advantage of statistical asymptotic effectiveness.

A method includes obtaining (402) operating data associated with operation of a cross-directional industrial process controlled by at least one model-based process controller (106, 204). The method also includes, during a training period (502a, 502b), performing (406) closed-loop model identification with a first portion of the operating data to identify multiple sets of first spatial and temporalmodels. The method further includes identifying (408) clusters (604) associated with parameter values of the first spatial and temporal models. The method also includes, during a testing period (504a, 504b), performing (410) closed-loop model identification with a second portion of the operating data to identify second spatial and temporal models. The method further includes determining (412) whether at least one parameter value of at least one of the second spatial and temporal models falls outside at least one of the clusters. In addition, the method includes, in response to such a determination (414), detecting that a mismatch exists between actual and modeled behaviors of the industrial process.

A method for compressing a machine learning model by an electronic device. The method may comprise determining a compression parameter of a set hidden layer in a model based on a pruning number of respective channels included in the set hidden layer and a pruning loss of each hidden layer of the model; and compressing the model based on the compression parameter of the set hidden layer. The compression parameter may be related to a pruning of the model.

The invention discloses a blast furnace molten iron silicon content prediction method and a device based on LSTM & DNN. The method comprises the following steps: dividing for obtaining a time lag attribute, a related attribute and a redundancy attribute based on a Pearson correlation coefficient; respectively constructing an LSTM blast furnace silicon content model and a DNN blast furnace siliconcontent model by utilizing the divided attribute data; performing weighted fusion on the LSTM model and the DNN model through a BP neural network to obtain a blast furnace molten iron silicon contentprediction model; according to the method, attribute division is carried out on the basis of the Pearson correlation coefficient, redundant attributes are removed, and the correlation attributes are stripped, so that the pressure of the LSTM model can be effectively relieved, the calculation speed is increased, and the model prediction effect is improved; the long-term and short-term memory capability of the LSTM network is utilized to effectively solve the large time lag characteristic of the blast furnace data; the DNN model is used for mining high-dimensional features of related attributes,so that the LSTM & DNN-based blast furnace molten iron silicon content prediction model has memory ability and generalization ability.

A technical system is described by several abstract or concrete models which have a mutual relationship. Information/attributes can be associated with model elements and “inherited” through these relationships. While the approach can basically be applied to different types and models of machines, it is described more particularly for hierarchical component models. Model elements in these models are referred to as components, assemblies, etc.

The embodiment of the invention discloses a searching method of a machine learning model and a related device, and equipment. Specifically, the invention relates to the technical field of artificial intelligence, the method comprises the following steps of: searching and quantifying a model; generating a plurality of pure bit models according to the to-be-quantized model, further obtaining an evaluation parameter of each layer structure in the plurality of pure bit models; further, selecting one candidate model from the candidate set for training and testing; obtaining a target model, the quantitative weight of each layer of structure in a target model can be determined based on the network structure of the target model and the evaluation parameter of each layer of structure in the targetmodel; therefore, the layer structure with the maximum quantization weight in the target model is quantified, the quantified model is added into the candidate set, frequent information interaction with the terminal can be reduced, and the efficiency of model search and model quantization is improved.

The invention relates to an optimized traffic flow prediction model based on a space-time diagram convolutional network. Traffic flow prediction is defined as follows: for a specific road network structure, traffic flow data of several time steps in the future are predicted according to traffic flow data of several time steps recorded historically, wherein the model establishment comprises spatial correlation modeling; the structure of the graph is represented through a self-adaptive adjacency matrix obtained through model training; time correlation modeling is carried out, the calculation process of the gate and the hidden state of the GRU is full-connection operation, and GCN is used for replacing the gate and the hidden state of the GRU; a TPA mechanism is introduced; loss function is adopted, and the purpose of designing and training the model is to minimize an error between a model prediction value and a real value of a road node. According to the method, the accurate prediction precision of the short-time traffic flow is improved, the capability of the model for analyzing the data of the graph structure is enhanced, and the time-space dependence of the traffic flow can be fully mined, so that the prediction precision of the short-time traffic flow and the convergence speed of the model are improved.

The invention provides a shipbuilding-oriented one-stop construction system and method for multiple industrial mechanism models. The system comprises a modeling preparation layer, a basic model construction layer, a model reconstruction layer and a model storage layer, wherein the modeling preparation layer is used for collecting and summarizing model demand characteristic information; the basic model construction layer is used for modeling a basic model; the model reconstruction layer is used for further editing and reconstructing the basic model; and the model storage layer is used for analyzing the types of the models generated by the basic model construction layer and the model reconstruction layer, and storing the models in corresponding model libraries in a classified manner according to the types of the models. According to the one-stop construction system for the multiple industrial mechanism models for shipbuilding, the multiple industrial mechanism models involved in the industry are rapidly constructed and regenerated through a one-stop basic model development and model regeneration technology, so the construction speed and quality of the models are improved, and high-quality development of shipbuilding is assisted.

The invention belongs to the field of computer graphics and statistics, and particularly discloses a casting three-dimensional feature extraction and similarity measurement method in combination withprocess parameters. The casting three-dimensional feature extraction and similarity measurement method comprises the following steps: S1, performing preliminary matching on a to-be-matched casting model and models in a model library so as to select a plurality of models from the model library as preliminary matching models; S2, obtaining the overall modulus, the average wall thickness and the maximum wall thickness of each preliminary matching model and the to-be-matched casting model, and calculating the overall modulus distance value, the average wall thickness distance value and the maximumwall thickness distance value of each preliminary matching model and the to-be-matched casting model respectively; and S3, calculating a weighted distance value between each preliminary matching model and the to-be-matched casting model, and sorting the preliminary matching models again, thereby determining a final matching model. The casting three-dimensional feature extraction and similarity measurement method completes similarity measurement of the casting model in combination with casting process parameters, overcomes the defects that an existing matching method is insensitive to model scale changes and individual models cannot be distinguished, and is good at the matching effect.

The invention discloses a wind tunnel test device which at least comprises a model, a wind tunnel test section and a model position adjusting part, the wind tunnel test section is located over the model position adjusting part, and the model position adjusting part is connected with the model and can send the model into or take the model out of the wind tunnel test section. The function of bringing the model into or out of the wind field after the flow field is established in the wind tunnel test section is realized, so that the impact load caused by the opening and closing of the supersonic wind tunnel is eliminated, and the test safety of the large-wingspan aircraft model is ensured.

The invention discloses a test device for measuring thrust of a shock tunnel. The test device comprises a supporting seat, a balance, a model balance adapter, a fairing and an accelerometer. The supporting seat, the balance and the model balance adapter are sequentially arranged from bottom to top, the fairing is fixedly mounted on the supporting seat and surrounds the outer sides of the balance and the model balance adapter, and the inner wall of the fairing is not in contact with the balance and the model balance adapter. The accelerometer is installed on the balance. According to the test device provided by the invention, the two balances are combined to measure the thrust acting on the model, the thrust measuring requirements of models with different lengths can be met by adjusting the distance between the two balances, the adaptability of the balances is good, the manufacturing cost of the balances is saved, the measurement signal is compensated through the accelerometer, and the measurement requirement of the shock tunnel in extremely short effective test time is met.

The invention relates to the related field of machine learning of privacy protection, in particular to a vertical federated modeling method based on a LightGBM algorithm, which comprises vertical federated modeling preparation work, initiator P1 data preparation work, partner P0 data preparation work, model training work and model evaluation work. The invention provides a novel longitudinal federated learning system structure based on a tree model, the two parties are allowed to construct a joint training model on the premise of protecting data privacy, and the two parties are enabled to train the joint model without the help of a trusted coordinator, so that a gradient value is protected, and the security of a protocol is improved; the architecture of the method is easy to expand, and besides two-party model training, the architecture of the method supports multi-party joint modeling; and according to the longitudinal federation learning based on the lightGBM algorithm, safety, speed, accuracy, support category features and continuous features are comprehensively considered, a large amount of training data can be processed under the same data set and features, and the method is suitable for engineering.

The invention relates to a method for applying a cut triangular patch to ridge repair of a three-dimensional model, which adopts a computer for processing and comprises the following steps: acquiring a to-be-processed model file; wherein the to-be-processed model file is a model file intersected with the target ridge; wherein the to-be-processed model file comprises corresponding model file data; wherein the model file data comprises geometric information of the model file and texture information of the model file; wherein the geometric information of the model file comprises coordinates of vertexes and texture coordinates of triangles in a spatial triangulation network; carrying out thread task allocation on the to-be-processed model file to obtain a plurality of thread tasks; executing a plurality of thread tasks to obtain a second ridge line; wherein the second ridge line is a ridge line after the target ridge line is repaired; wherein each thread task is to obtain a triangle corresponding to the model file according to a target ridge drawn in advance by one model file, and fuse the model files according to the triangle.

The invention discloses an engineering scene establishing method, and relates to the field of scene establishment; the method comprises the steps: S1, establishing BIM models of engineering scenes; S2, analyzing the BIM models, obtaining related information of the BIM models, storing the related information in a model database, and outputting a GIS readable model file; S3, acquiring geographic data, and establishing a geographic information scene through a GIS platform; S4, unifying a BIM model local coordinate system and a GIS coordinate system, and determining offset information; S5, importing all the BIM models and the model database according to the space coordinates and the offset information of the BIM models in the three-dimensional scene, and dynamically displaying the BIM models;and S6, completing seamless fitting of the BIM models and a DEM and sleeving of a terrain. The three-dimensional building information model is used as a core to be applied to the processes of design,construction and the like of an engineering project, so that the BIM design model is well applied to the field of civil engineering and is combined with geographic information; the problems that information borne by a BIM design model integrated into a three-dimensional geographic information system is lost, the loading speed is low, and the spatial position of a component in a structure deviatesare solved.

Techniques performed by a data processing system for analyzing the impact of training data changes on a machine learning model herein include training a first instance of a machine learning model with a first set of training data; modifying the first set of training data to produce a second set of training data; training a second instance of the model with the second set of training data; comparing the first instance of the model to the second instance of the model to determine features that differ between the first instance and the second instance of the model; identifying a subset of historical data associated with the features that differ between the first instance and the second instance of the model; and scoring the subset of the historical data to produce a report identifying differences in the output of the first instance and the second instance of the machine learning model.