65 results about "Neural Network Simulation" patented technology

The invention discloses an ice lake feedback monitoring and early warning method based on artificial intelligence and multi-source data fusion, and belongs to the technical field of natural disaster monitoring, and the method specifically comprises the steps: extracting ice lake state feature vectors of multiple spatio-temporal data through cross-modal self-supervised learning, and constructing a three-dimensional virtual ice lake model; based on the feature vectors, constructing a glacier three-dimensional stress field model by using a finite element method, constructing a seepage channel graph model by using a graph neural network, and fusing the features of the glacier three-dimensional stress field model and the seepage channel graph model to form mechanical-seepage coupling state vectors; constructing a pulse neural network to simulate glacier fracture extension and seepage mutation pulse events; constructing a heterogeneous graph federated architecture and outputting a federated weight matrix in combination with a dynamic weighted aggregation strategy; the federal learning weight and the three-dimensional model output result are fused, the outburst posterior probability is calculated through a Bayesian neural network, and a fourth-level early warning signal is generated; according to the invention, the multi-dimensional characterization and coupling process simulation of the state of the ice lake is realized, and the early warning accuracy and timeliness are improved.

The invention discloses an adaptive optical system non-common-path aberration correction method based on a neural network. The adaptive optical system non-common-path aberration correction method comprises the following steps: constructing a lightweight neural network CP-Net; simulating and generating a perfect point spread function image; generating a deformable mirror control voltage matrix based on the image; taking the voltage matrix as the input of CP-Net for training, and ending the training when the loss function value is smaller than a preset minimum value; the trained CP-Net is connected to an actual control loop, real-time images acquired by a scientific camera are received, and perfect reference voltage is generated and directly used for controlling and changing the mirror surface type of the deformable mirror. According to the method, the deformable mirror control command can be directly generated, aberration does not need to be measured, real-time automatic adjustment can be achieved, compared with a traditional phase diversity method and a stochastic gradient descent algorithm, the operation difficulty can be greatly lowered, the correction time can be greatly shortened, and an operator does not need to manually adjust related parameters in the correction process.

The invention relates to the technical field of large model algorithms, in particular to a large model dynamic compression optimization method and system based on sparse pruning, and the method comprises the steps: capturing original weight fluctuation data generated by resource fluctuation in reasoning, and obtaining sparse weight reference data through sparse processing; analyzing calculation complexity through model reasoning delay data, and separating reasoning delay amount caused by a model scale; dynamically controlling the model compression ratio within a preset performance range based on the delay amount and the sparse reference data, and collecting reasoning precision distribution data under different compression parameters; evaluating a model performance state under each parameter by means of a neural network simulation method, and generating a performance state simulation result; determining a model quality optimization compensation parameter based on a simulation result by combining resource fluctuation data acquired in real time in a compression process; and finally, the compression strategy is adaptively regulated and controlled through the compensation parameters, and collaborative optimization of model calculation complexity, reasoning precision and delay during dynamic change of hardware resources is realized.

The invention discloses a three-dimensional magnetotelluric deep learning inversion method, and relates to the technical field of three-dimensional magnetotelluric inversion in electromagnetic exploration, and the method comprises the steps: constructing a three-dimensional layered underground resistivity theoretical model, and forming a sample pair through the structure data of the underground resistivity theoretical model and the corresponding visual parameter data; the method comprises the following steps: constructing a three-dimensional neural network based on a Swin Transform module and jump connection; a forward modeling sub-network is trained for the multiple visual parameters, and the weight of the forward modeling sub-network is frozen to serve as a fixed forward modeling operator, so that rapid forward modeling is achieved; after each fixed forward operator is migrated and spliced to the inversion sub-network, each fixed forward operator is used as an additional loss constraint term, and physical driving of neural network simulation is realized; and end-to-end mapping from each apparent parameter to the underground resistivity is established by fitting the inversion sub-network, and quasi-physics and data dual-drive three-dimensional magnetotelluric deep learning inversion is realized. The three-dimensional magnetotelluric inversion method has high practical value and popularization value in the technical field of three-dimensional magnetotelluric inversion with crossing of deep learning and electromagnetic exploration.

The invention relates to the technical field of artificial intelligence, in particular to a method, device and equipment for simulating a finite-state machine based on a spiking neural network and a storage medium. The method comprises the following steps: acquiring an input signal; according to the input signal, a corresponding output signal is output through a DTSRNN model which is trained in advance, the DTSRNN model is a model which simulates the state transition process of an FSM by combining the characteristics of the DTSRNN and an SNN, and the output signal is a response signal generated by the DTSRNN model according to the input signal and internal state transition logic. According to the embodiment of the invention, the DTSRNN model combining the characteristics of the DTRNN and the SNN is provided, and the information can be processed in a discrete and sparse mode, which is highly matched with the discrete state conversion process of the FSM, thereby achieving the efficient simulation of the FSM behavior, and greatly improving the performance of a neural network model for simulating the FSM.

An automated insulin delivery system provides an insulin injection that is computed by adding (i) the recommendation of an artificial neural network trained to mimic a constrained model predictive controller dosing rule from a neural network implementing an artificial pancreas (ii) a hypoglycemia mitigation system includeds a correction (one dose computed as the correction down to 110mg / dl based on prevailing continuous glucose monitoring and once an hour at most, unless there is a triggering BPS and G>180 mg / dl), and (iii) the current basal rate (output of the Performance Assessment System, (PAS)); that amount is then saturated by the Safety Supervision System, SSM. Finally, a priming bolus from the Bolus Priming System, BPS, is added to the total if the conditions for large glycemic excursions are detected, also saturated by SSM.

The invention discloses a population antibody concentration dynamic evaluation method based on neural network simulation calculation. The method comprises the following steps: S1, acquiring a multi-modal immunology related data set; s2, generating a complete multi-modal feature data set; s3, constructing a deep learning model based on the multi-modal feature data set; s4, training the segmented neural network model; s5, inputting the multi-modal immunology related data of the to-be-evaluated crowd into the trained sectional neural network model, and generating a prediction result of the individual antibody concentration changing along with time; s6, summarizing prediction data of a plurality of individuals according to the prediction result of the individual antibody concentration changing along with time, and establishing a population overall antibody concentration dynamic distribution model; and S7, providing vaccination optimization strategy suggestions and crowd immune intervention schemes in combination with data output. The invention provides key technical support for immune monitoring and vaccination strategy optimization.

The invention relates to the technical field of intelligent site selection of new energy automobile charging facilities, and discloses an intelligent site selection base station management system and method, and the system comprises a city multi-dimensional data collection and preprocessing module, a knowledge base construction module, a city evolution dynamics simulation module, a future scene generation module, and a charging network optimization model construction module. A robust optimization solving module; the method comprises the following steps: collecting and constructing a knowledge base of multi-dimensional city data; simulating city dynamics by using multi-agent modeling and a space-time diagram neural network to generate a plurality of future scene sets; and finally, constructing a multi-target robust optimization model based on the scenes, and solving to generate a group of Pareto optimal robust charging network planning schemes. According to the method, the overall scientificity, foresight and robustness of charging base station site selection planning are improved, so that the investment decision risk of the charging base station site selection planning is effectively reduced.

Systems and methods are provided for analog hardware realization of neural networks. The method incudes obtaining a neural network topology and weights of a trained neural network. The method also includes transforming the neural network topology into an equivalent analog network of analog components. The method also includes computing a weight matrix for the equivalent analog network based on the weights of the trained neural network. Each element of the weight matrix represents a respective connection between analog components of the equivalent analog network. The method also includes generating a schematic model for implementing the equivalent analog network based on the weight matrix, including selecting component parameter values for the analog components.

The invention relates to the technical field of underground water monitoring, in particular to a high-precision underground water pollution plume migration prediction and source identification method, which comprises the steps of constructing a graph structure, simulating an underground water convection-dispersion process by using a space-time graph neural network, introducing a physical equation residual error constraint to train a prediction model, and converting source identification into an optimization problem. A multi-task strategy is adopted to synchronously identify a sparse source position and reconstruct a release history of the sparse source position, a model is finely adjusted and updated according to gradient descent cooperative solution, and uncertainty is quantified. According to the high-precision groundwater pollution plume migration prediction and source identification method, source position identification and release history reconstruction are synchronously realized through a multi-task strategy; organizing data by using a graph structure, constructing static and dynamic feature vectors, and comprehensively describing pollution migration space-time characteristics; a multi-stage curriculum learning, a splitting algorithm and a Bayesian optimization initialization strategy are adopted, gradient is calculated in combination with automatic differential, and a multi-source uncertainty decomposition framework is constructed to quantify uncertainty.

The invention discloses a deep learning-based precise sorghum fertilization amount dynamic adjustment method and system, and relates to the technical field of deep learning-based fertilization amount control. According to soil nutrient concentration and sorghum root physiological index data, a time sequence analysis method is adopted to generate a time sequence change sequence of root absorption capability; and inputting a pre-trained deep learning model to determine a nitrogen-phosphorus-potassium demand difference value. And when the demand difference value exceeds a preset demand threshold value, optimizing a fertilization amount ratio by using a convolutional neural network, simulating a response of a root system to fertilization adjustment in combination with a recurrent neural network, and outputting a target fertilization regulation and control scheme when it is detected that a predicted response accords with an expected yield index, so that equipment is controlled to perform fertilization subsequently according to the target fertilization scheme. According to the method, a nonlinear mapping relation between the fertilization amount and the physiological response of the sorghum root system is established through a deep learning technology, and the fertilization proportion of nitrogen, phosphorus and potassium is dynamically adjusted in combination with time sequence data, so that precise sorghum fertilization amount regulation is realized.

The invention relates to the technical field of network-on-chip and brain-like computing, in particular to a hierarchical routing network facing a large-scale brain-like computing system and a dynamic arbitration method.The hierarchical routing network comprises a bottom routing network layer, at least one middle routing network layer, a top routing network layer and a dynamic routing control module, and the dynamic routing control module comprises a pressure sensing unit, a routing path selection unit and a data scheduling unit. According to the scheme, the communication efficiency, the fault-tolerant capability and the energy consumption control level of the brain-like system are improved, and the method is suitable for high-performance low-power-consumption computing architecture such as a neuromorphic processor, a brain-like intelligent chip and a large-scale pulse neural network simulation platform; the network is divided into a bottom routing network layer, a middle routing network layer and a top routing network layer, and a dynamic congestion sensing routing control module of a three-level assembly line mechanism is combined, so that burst data streams in the network are effectively scheduled, and the data transmission efficiency is improved.

The invention discloses a nonlinear multi-agent system double-clock asynchronous hybrid neural adaptive control method, and relates to the technical field of nonlinear multi-agent system cooperative control. According to the method, a double-clock asynchronous framework is provided, time decoupling is carried out on updating of the leader observer and updating of the local observer, the limitation of synchronous updating of all assemblies is broken through, and the leader observer and the local assembly are made to operate independently; an event trigger pulse mechanism is designed, sampling is carried out only when a specific event occurs, the communication and calculation cost is remarkably reduced, the event trigger mechanism allows each agent to autonomously determine a trigger moment, and self-adaptive resource allocation between heterogeneous dynamic agents is achieved; and the numerical value of the nonlinear function is simulated and estimated by using the neural network, so that approximate state information can still be obtained under different conditions. And the weight of the neural network is only updated at the triggering moment, so that the resource consumption in the learning process is reduced.

The invention discloses a digital brain data assimilation system and method based on multi-modal data fusion. The system comprises an EEG forward model construction module, an MRI-EEG space mapping construction module, a network structure generation module, a multi-modal data assimilation module, a neural network simulation module and a digital brain similarity evaluation module. The multi-modal data assimilation module is improved on the basis of an existing hierarchical mesoscale data assimilation algorithm, and the model state is filtered on the basis of an augmented observation vector (an experiment / system simulation BOLD signal and an experiment EEG signal) in each updating step, so that estimated parameters can fit two observation signals at the same time. According to the method, fitting of multi-modal data with different temporal-spatial resolutions can be realized in the same digital twin brain system.

Disclosed herein are related to a system, a method, and a non-transitory computer readable medium for simulating, predicting, or estimating, based on machine learning neural networks, wall stress of a body part. In one approach, a first neural network automatically detects features in multiple images of a body part. For example, the first neural network may detect, for each image, a lumen and a wall of an aorta. According to the detected features, a second neural network may simulate, estimate, or predict wall stress of the body part in response to pressure applied to the body part. For example, a model generator can generate a three-dimensional model of the body part according to the detected features in the multiple images, and the second neural network can simulate, estimate, or predict wall stress of the body part according to the three-dimensional model.

The application relates to a high-definition image explainable classification method and device, computer equipment and a storage medium. The method simulates the process of image perception, feature extraction, inductive reasoning and learning of the human brain. The method is based on the latent space of a pre-trained StyleGAN, a conversion network is designed to simulate the feature extraction of the visual cortex of the brain, effectively converting high-dimensional latent encoding into low-dimensional and decoupled classification features, and the problems of rule explosion and calculation collapse are alleviated; a fuzzy neural network is used to simulate the reasoning function of the parietal lobe and the prefrontal cortex of the brain, and an improved method of the fuzzy neural network structure suitable for the high-definition image explainable classification model is designed, so that the explainable classification of the low-dimensional features is realized; the fuzzy rules used for classification are visually displayed and analyzed, and a feature visualization method is designed based on the pre-trained StyleGAN generator, so that the explainability is further improved.

The application discloses an automatic needle insertion method combining a blood flow optimization model with a neural network. The method first calculates the blood vessel position and blood flow velocity distribution of a needle insertion area according to blood flow information measured by automatic ultrasonic scanning by means of the Doppler ultrasonic principle, then establishes a blood flow relative velocity optimization model and a neural network with a selection function, and decides the optimal needle insertion point and needle insertion angle on the basis of physical measurement results, and finally simulates the needle insertion path by means of a neural network with a learning function according to the needle insertion method of a professional. The method can be used for needle insertion of subcutaneous blood vessels of limbs and other body parts, and overcomes the difficulty of manual needle insertion when the blood vessel is not visible to the naked eye, thereby providing technical assistance for self-rescue and first aid in an environment lacking professional personnel.

The present invention relates to the technical field of atmospheric pollution source tracing, and specifically to a pollution source tracing method based on high spatiotemporal resolution pollution concentration distribution, comprising: constructing a deep neural network simulation model for the spatial distribution of fine particulate matter concentration; determining the sites and time periods where the fine particulate matter monitoring concentration exceeds the standard according to the air quality monitoring standard, and generating a gridded fine particulate matter concentration distribution within the exceeding time period by the deep neural network model for the spatial distribution of fine particulate matter concentration; cyclically stepping to generate pollution transmission trajectories, and when the tracing termination condition is met, superimposing the pollution source coordinates in the emission inventory data to achieve the final pollution tracing. The present invention, from the perspective of geographic spatial association, based on the gridded fine particulate matter concentration in continuous time periods, iteratively traces back the pollution diffusion path time by time period, obtains the fine particulate matter transmission and diffusion trajectory, intuitively reflects the atmospheric pollution diffusion process, provides directional guidance for fine particulate matter pollution source tracing, and after superimposing the spatial distribution of pollution sources, can lock the atmospheric pollution emission source.

The embodiment of the invention provides a distributed energy storage regulation and control method, device and system and a medium, and belongs to the technical field of electric power. The method comprises the steps of performing data processing on various random factors associated with distributed energy storage regulation and control based on multi-algorithm collaboration to generate a random scene for distributed energy storage regulation and control; processing battery operation parameters associated with distributed energy storage regulation based on a neural network model to estimate battery aging cost, the neural network model taking a battery capacity fading rate as an output; and determining a distributed energy storage regulation and control strategy based on the generated random scene and the estimated battery aging cost. According to the embodiment of the invention, scientific regulation and control and resource optimization configuration of the energy storage system are effectively realized, random factors and the battery aging cost of neural network simulation are comprehensively considered, and efficient application and accurate regulation and control of distributed energy storage in a power system are promoted.

The invention relates to the technical field of production management, and discloses a collaborative scheduling method, device and equipment for a digital manufacturing platform, and the method comprises the steps: generating a real-time sensing matrix through simulating the operation condition of a factory in real time, predicting material shortage information according to a real-time production work order and a material demand, obtaining the positioning data of an automatic carrying unit, and carrying out the real-time monitoring of the real-time sensing matrix; according to the method, a transport target map and a material carrying characteristic matrix are constructed, a graph neural network is utilized to simulate a transport path and optimize and drive an automatic carrying unit to carry out transport, risk assessment is carried out through multi-modal sensing data, an optimal adjustment strategy is generated, and through real-time data driving and path optimization, the precision and efficiency of production scheduling are improved, and the production efficiency is improved. The risk in the transportation process is reduced, the utilization rate of production resources is increased, and the problem that in the prior art, material demand changes are difficult to respond rapidly is solved.

The application belongs to the technical field of neural network simulation and analog, and particularly relates to a large-scale cortex pulse neural network asynchronous parallel simulation algorithm based on GPU. The application utilizes the advantages of multi-thread and texture memory of a computing graphics card, combines the general form of a biological brain receiving external stimulation and the general connection mode between neurons in the cortex, designs an asynchronous parallel algorithm framework of GPU and CPU, GPU is responsible for parallel evolution of neuron dynamics equations and block parallel calculation of isotropic connection in a local network, CPU is responsible for processing anisotropic long-range connection, and different neuron dynamics equations and plasticity learning rules can be compatible. Compared with the prior art, the application can effectively improve simulation speed, provides a tool for simulating a biological brain cortex in a single computing node, and is suitable for a single node multi-graphics card and a multi-node multi-graphics card distributed operation model.

This application provides a method for preparing a dry binder, achieving stable quality control through dynamic sensing and multi-level intelligent adjustment of the moisture content of the raw material powder: First, moisture content and environmental humidity data are collected in real time, and time series analysis is used to predict the trend of moisture content changes; then, support vector machine is used to classify seasonal and batch differences to determine the main influencing categories; for high humidity interference scenarios, neural network is used to simulate the dynamic impact of material residence time in the mixing equipment on moisture content and determine the risk level; based on the risk level, historical data is retrieved, and a random forest algorithm is used to generate a preliminary adjustment plan for process parameters; then, a digital simulation model is used to simulate and verify the entire process and iteratively optimize it to finally determine the optimal process parameters; during production, the moisture content prediction model is dynamically updated using real-time feedback data to achieve closed-loop correction and continuous optimization. This invention effectively reduces the quality risks caused by environmental fluctuations and batch differences.

This application belongs to the field of delay differential equation modeling, specifically disclosing a delay differential equation modeling method based on Bayesian optimization and neural networks. The method includes: generating multiple trajectory data of the system; searching for the optimal delay term using a Bayesian optimization algorithm, where the delay term to be solved is the optimization variable, and the error based on neural network simulation is the objective function, obtaining the optimal delay term through iterative optimization using a surrogate model; constructing a state matrix and a delay matrix from the trajectory data based on the optimal delay term, using the concatenated matrix as input to a neural network to construct a neural network for approximating a nonlinear function; integrating a linear multi-step method into the loss function of the neural network, training the neural network using the trajectory data to obtain a nonlinear function approximation model; and separating the numerical discretization error and the neural network approximation error based on this model to construct a total error bound. This application can achieve high-precision, high-efficiency delay differential equation modeling with quantization error guarantee.

The invention relates to the technical field of public safety management, in particular to a panoramic video fused public safety risk intelligent early warning method and system, and the method comprises the steps: obtaining panoramic video data of a target monitoring area in real time under a unified space-time framework; extracting spatial features and time sequence features from the acquired panoramic video data; fusing the spatial features and the time sequence features, and constructing a space-time diagram of the target monitoring area based on the fused spatial features and time sequence features; and simulating propagation and evolution of the risk in the space-time diagram by using a graph neural network, and carrying out risk decision making and early warning based on propagation and evolution characteristics.

The invention relates to the technical field of artificial intelligence decision analysis and behavior modeling, in particular to a figure decision mode analysis method and device based on a situation-reasoning one graph, and the method comprises the steps: constructing a decision mode situation-reasoning one graph, extracting and quantifying decision influence factors, and extracting and quantifying the decision influence factors. The method comprises the following steps of: firstly, converting mass unstructured open source data into a computable and deducible situation-reasoning graph by utilizing the structure of the situation-reasoning graph; then, according to a decision theory, quantifying intrinsic psychological cognition characteristics of a person, and taking the intrinsic psychological cognition characteristics as key constraint conditions of a reasoning network; and finally, constructing a decision-making mode neural network, and simulating a'perception stimulation-cognitive screening-logical reasoning 'decision-making process of a person under a specific situation. According to the method, deep fusion and accurate analysis from an external objective situation to internal subjective cognition are realized.

This invention relates to the field of large model algorithm technology, and in particular to a dynamic compression optimization method and system for large models based on sparse pruning. The method captures raw weight fluctuation data caused by resource fluctuations during inference, and obtains sparse weight baseline data through sparsification. It analyzes the computational complexity of model inference latency data to separate the inference latency caused by model size. Based on the latency and sparse baseline data, the model compression ratio is dynamically controlled within a preset performance range, and inference accuracy distribution data under different compression parameters are collected. The model performance status under each parameter is evaluated using neural network simulation methods, generating performance status simulation results. Combined with real-time resource fluctuation data acquired during compression, model quality optimization compensation parameters are determined based on the simulation results. Finally, the compression strategy is adaptively adjusted through the compensation parameters to achieve coordinated optimization of model computational complexity, inference accuracy, and latency when hardware resources change dynamically.

The present invention discloses an adaptive sliding mode control method for a sway robot to resist false injection attacks, belonging to the technical field of sway robot control. The method comprises obtaining the initial trajectory tracking error equation of the sway robot based on the robot's dynamic model, a tracking trajectory reference signal, and the trajectory the sway robot needs to track; simulating the actuator attack signal using a neural network, obtaining the weight estimate of the neural network through a linear filter based on the Lyapunov stability theorem, and determining the actuator attack signal; determining the trajectory tracking error equation of the sway robot after being attacked by a false injection actuator based on the attack signal and the initial trajectory tracking error equation; and setting an adaptive integral sliding mode controller with the trajectory tracking error converging to 0 within a fixed time as the control goal to control the sway robot to achieve trajectory tracking. This method can ensure that trajectory tracking is not affected by attacks and interference, resolving the problem of insufficient protection against targeted attacks.

To provide methods, systems and recording mediums for generating suggested communications during a multi-agent interaction using a language model neural network.SOLUTION: A method comprises: identifying multiple first and second strategies for selecting between parameterizations for communications for first and second actors; parametrizing a set of context variables for the communications of the actors during an interaction; for each of multiple pairs including respective first and second strategies: generating simulations of the interaction using a language model neural network; determining, from the simulations, a score for the pair that indicates a degree to which the first and second actors satisfy objectives for the interaction; selecting, using the scores for the multiple pairs, a parametrization for the first actor; and generating, by processing an input that conditions the language model neural network on the selected parameterization, a suggested communication for the first actor.SELECTED DRAWING: Figure 2

An automated insulin delivery system provides an insulin injection that is computed by adding (i) the recommendation of an artificial neural network trained to mimic a constrained model predictive controller dosing rule from a neural network implementing an artificial pancreas (ii) a hypoglycemia mitigation system includeds a correction (one dose computed as the correction down to 110mg / dl based on prevailing continuous glucose monitoring and once an hour at most, unless there is a triggering BPS and G>180 mg / dl), and (iii) the current basal rate (output of the Performance Assessment System, (PAS)); that amount is then saturated by the Safety Supervision System, SSM. Finally, a priming bolus from the Bolus Priming System, BPS, is added to the total if the conditions for large glycemic excursions are detected, also saturated by SSM.

The invention discloses a carbonate rock acid fracturing effect main control parameter analysis method and system, relates to the technical field of acidification, and solves the problem that main control parameters are difficult to identify in existing carbonate rock acid fracturing effect assessment. Calculating the correlation of each parameter based on the parameter data, and filtering out the parameters with extremely high correlation; executing a random forest algorithm and a recursive elimination algorithm based on the parameters to obtain a first weight and a second weight of the parameters; performing weighted average on the first weight and the second weight to obtain a comprehensive weight of the parameter; selecting parameters according to the comprehensive weight to construct a plurality of schemes, inputting the schemes into a BP neural network to simulate the yield to obtain the correlation between the actual yield and the predicted yield, and selecting the parameters contained in the scheme with the highest correlation as main control parameters of the acid fracturing effect; and analyzing carbonate rock acid fracturing effect master control parameters based on a random forest, recursive elimination and a BP neural network algorithm.