79 results about "Physical neural network" patented technology

The invention provides a diesel engine misfire fault diagnosis method based on a GRU and a PINN. The diesel engine misfire fault diagnosis method comprises the following steps of diesel engine multi-source data acquisition, data preprocessing and model sample generation; designing a gating circulation unit, introducing a physical constraint embedding module, constructing a misfire fault diagnosis model, and fusing the misfire fault diagnosis model with a long-short term memory model and a physical neural network; performing joint training and optimization on the misfire fault diagnosis model in the step S2 by using the samples in the step S1; and deploying a misfire fault diagnosis model and performing real-time diagnosis. The method has the beneficial effects that the time sequence data-physical model dual-drive diagnosis of the misfire fault of the diesel engine is realized; the accuracy of misfire diagnosis is improved; and the false alarm rate of complex working conditions is effectively reduced.

The invention belongs to the technical field of building energy conservation, and provides a real-time inversion method and system for a heat transfer coefficient of a building envelope, and the method comprises the steps: obtaining surface parameters and environment parameters of the building envelope; based on the obtained parameters and the physical neural network model, taking the minimum joint loss function based on the physical mechanism as a target, and performing inversion to obtain the optimal heat conductivity coefficient of the building envelope; and calculating the heat transfer coefficient of the building envelope according to the obtained optimal heat conductivity coefficient of the building envelope, and completing real-time inversion of the heat transfer coefficient of the building envelope.

The invention relates to a basin flood simulation optimization method based on a physical information neural network, and aims to construct a basin flood acceleration solution mode based on PINNs, and ensure the continuity and the physical characteristics of mass conservation during model operation, and the method comprises the following steps: simulating basin flood by using an urban flood space-time simulation model; obtaining a data set required for constructing the physical neural network; physical rules are integrated and simplified into limiting conditions, so that integration of a neural network structure is facilitated; establishing a partial differential equation (PDE) used for describing river flood routing, and taking the partial differential equation as a loss function of the physical information neural network model; constructing and training a PINNs drainage basin flood model constructed on the basis of a CNN (Convolutional Neural Network); and establishing a model evaluation index and performing evaluation. The method has the beneficial effect that the speed and precision of basin flood simulation are improved.

The invention relates to a mechanical arm optimization control method and device based on a coupled physical neural network, and belongs to the field of mechanical arm optimization control. The method comprises the following steps that S1, a mechanical arm optimization problem is extracted; s2, constructing accurate solution data; s3, establishing a constraint condition network and training; s4, establishing a target function network; s5, taking the objective function as a loss function, and establishing a dual-network coupling architecture; s6, training a dual-network coupling architecture; s7, outputting a predicted optimal solution result; s8, performing deviation correction on the predicted optimal solution by using a deviation correction method to obtain an optimal solution; and S9, mechanical arm parameters are adjusted according to the optimal solution to achieve control. According to the method, a multi-task objective function can be quickly coped with, so that a large number of repeated solving data sets are avoided, and the target required by a product is responded more timely and accurately; meanwhile, the innovative training method not only ensures the global search capability, but also effectively balances the calculation cost and the solving precision, and provides efficient technical support for solving a complex optimization problem.

The invention discloses a performance prediction model determination method, a performance prediction method and a performance prediction device of a fuel cell, mainly aiming at modeling simulation and performance prediction of the fuel cell, the method comprises the following steps: constructing a fitting coefficient training database of the fuel cell based on a semi-empirical physical model and a computational fluid mechanics model of the fuel cell, and training a neural network model based on the data in the fitting coefficient training database to obtain a target neural network model of the fuel cell, coupling the neural network model and a semi-empirical physical model, and obtaining a performance prediction model of the hybrid physical neural network architecture. And the performance prediction model is used for outputting a fitting coefficient of the semi-empirical physical model based on the input operation parameters, taking the fitting coefficient as the input of the semi-empirical physical model, and finally outputting performance parameters. Therefore, by effectively combining a machine learning method with a physical model, a hybrid physical-data driven performance prediction model with an efficient calculation characteristic and a high degree of detail can be formed.

The invention relates to the technical field of lithium ion battery health monitoring, and discloses a lithium ion battery health state estimation method based on a physical neural network, which comprises the following steps: reconstructing an SEI membrane growth process of a lithium battery, fitting the growth process through the physical neural network, and analyzing voltage time sequence data in a lithium battery charging process to obtain a lithium ion battery health state estimation result. Calculating a health index in a 71%-90% state of charge range of a lithium battery charging interval; according to the estimation method, the health indexes are extracted by analyzing short-period data of the charging state of the battery within the range of 71%-90%, the method is successfully applied to four data sets of different chemical systems including LFP, NCM, NCA and the like and multiple charging and discharging protocols, and the good applicability of the method to different battery types and working conditions is proved.

The invention relates to the technical field of environment monitoring, in particular to a building decoration environment harmful gas dynamic monitoring method based on deep learning, which comprises the following steps: acquiring temperature, humidity, volatile organic compound concentration, carbon dioxide concentration, energy consumption and equipment attitude in a building model coordinate system, and eliminating redundant channels according to spatial topology to form a simplified data stream; dense features are extracted from the simplified data flow through a space-time auto-encoder, query key values are generated through quantification, and adjacent historical features are retrieved and spliced into a fusion feature vector; inputting the fused feature vector and a sensing node coordinate into a physical neural network of an embedded gas diffusion and heat conduction equation, and predicting a temperature gradient and regional carbon equivalent emission; calculating a hazard index field according to prediction, heat flow simulation, construction progress and a material release rate, iteratively optimizing a sensing node set and writing the sensing node set into an environment mirror image body; and reconstructing a coupling field in the environment mirror image body, comparing a measured value to obtain a residual field, and updating a hazard index field and a sensing node set when the environment mirror image body is abnormal.

The invention proposes a steel slag asphalt interface adhesion work prediction method based on a physical neural network, and relates to the technical field of deep learning, and the method comprises the steps: obtaining chemical component parameters and environmental impact factors of a steel slag asphalt mixture interface sample; creating an original neural network prediction model based on a diffusion item, a reaction source item, a multi-mechanism dissipation item and a state variable corresponding to spatial-temporal evolution of the adhesion work; inputting the chemical component parameters and the environmental impact factors into an original neural network prediction model, and generating interface adhesion work prediction parameters in combination with physical constraint terms; training the original neural network prediction model by using the interface adhesion work prediction parameters and a gradient descent algorithm to obtain a target neural network prediction model; the physical constraint items comprise data loss, PDE residual error loss, state equation loss and boundary condition loss; and inputting the chemical component parameters and the environmental impact factors of a to-be-tested material into the target neural network prediction model to obtain target prediction parameters.

The invention discloses a real-time sea clutter simulation method based on a physical neural network, belongs to the field of sea clutter simulation, and particularly relates to a real-time sea clutter simulation method. The invention aims to solve the problems of poor real-time performance and long time consumption of the existing sea clutter simulation method. The method comprises the following steps: S1, initializing parameters and simulation; s2, numerical simulation of sea surface dynamics is realized according to a shallow water equation, and the obtained sea surface height field changing along with time, the flow velocity of the sea surface height field in the axial direction and the flow velocity of the sea surface height field in the axial direction are periodically stored; s3, constructing a physical neural network; s4, obtaining a trained physical neural network; s5, inputting the space coordinate and the time coordinate of the real-time sea clutter to be measured into the trained physical neural network, and outputting the sum; generating a real-time file in an STL format; and S6, obtaining sea clutter data by using the open source library.

The invention discloses a fiber composite material damage diagnosis method based on a physical neural network, and the method comprises the steps: firstly constructing a multi-dimensional physical parameter fused physical neural network model, comprehensively considering material, mechanical, thermal and electrical parameters, and fusing a physical mechanism to construct a network structure; input features are determined and preprocessed, acoustic emission, strain, temperature and other signals are collected, and input layer features are constructed after filtering and standardization processing. In the aspect of model architecture design, a neural network comprising an input layer, a hidden layer and an output layer is constructed, and a physical mechanism is integrated in the neural network. A loss function under physical constraints is constructed to ensure that the model accords with a physical rule, a final model is obtained through model initialization and training optimization, finally, real-time dynamic monitoring of damage is achieved, a diagnosis result is output, and the reliability of material use is improved.

The invention relates to a hip joint motion track prediction method based on a Lagrange physical neural network. The method comprises the following steps: inputting kinematics data of a hip joint at a current moment into a Lagrange physical neural network, and training a kinetic parameter estimation module based on a pre-constructed loss function according to a trajectory prediction value and a real trajectory to obtain a trained Lagrange physical neural network; and predicting a hip joint movement track at the next moment by adopting a Lagrange physical neural network. By adopting the method, the unification of the motion trail prediction interpretability and the algorithm robustness is realized.

According to the method, the system, the medium and the equipment for monitoring the abnormal state of the wind generating set, a self-adaptive noise reduction mechanism, feature extraction and model classification are combined, so that physical information features are effectively extracted, and accurate state detection is achieved. The method comprises the following steps: firstly, building a signal acquisition system, acquiring an acoustic signal in real time, and executing direct current offset removal preprocessing to eliminate low-frequency noise; secondly, noise reduction of the original sound signal is completed by using a self-adaptive noise reduction method; then, extracting the MFCC characteristics of the noise reduction signal; then, calculating an MFCC inter-frame correlation feature and an aerodynamic-harmonic imbalance factor, and forming a fusion feature vector; and finally, constructing a physical neural network model based on physical information to carry out abnormal state detection.

The invention discloses a bearing fault diagnosis method and system based on a simulated physical neural network. The method comprises the following steps: S1, preprocessing a collected vibration signal to obtain a to-be-diagnosed signal; s2, executing a hybrid optimization algorithm to obtain an optimal transmission band parameter and an optimal weight parameter; s3, configuring a feature extraction module according to the optimal transmission band parameter, and writing the optimal weight parameter into a classification module; s4, inputting the to-be-diagnosed signal into a feature extraction module to obtain a simulation feature vector; s5, inputting the simulation feature vector into a classification module, and outputting a classification voltage signal; and S6, determining a fault type, and outputting a result. According to the method, a hybrid optimization algorithm is adopted, the ratio of the inter-class dispersion degree to the intra-class polymerization degree of the feature vectors is calculated, optimal adaptation of feature extraction and classification tasks is achieved, and the diagnosis precision and generalization ability are improved; by constructing a full analog domain signal processing architecture, analog-to-digital conversion and digital calculation are not needed, and microwatt-level power consumption and microsecond-level real-time response are realized.

The invention discloses a photovoltaic power short-term prediction method based on an adaptive physical neural network. The method comprises the following steps: acquiring historical meteorological data and photovoltaic power data of a photovoltaic power station; selecting meteorological characteristic data from the historical meteorological data, and clustering weather types; taking the meteorological characteristic data and the photovoltaic power data corresponding to each weather category label as a group of training samples, and generating an original training sample set; performing oversampling on the minority class sample set to generate a new sample, and combining the new sample with the original training sample set to form an equalized training sample set; determining a meteorological characteristic combination of each weather type according to the physical characteristics of photovoltaic power generation under different weather types; based on the combination, extracting corresponding meteorological characteristic data from the equalization training sample set as input, taking corresponding photovoltaic power data as a training target, and training a photovoltaic power short-term prediction model; and optimizing network parameters and physical parameters based on the composite loss weight to realize photovoltaic power adaptive prediction under different weather types.

The present invention discloses a fluid topology optimization method and system based on a physical neural network. The method includes: based on the fluid topology optimization problem, determining the basic parameters of the optimized fluid pipeline through a two-dimensional steady-state control equation; based on the basic parameters of the fluid pipeline, constructing a fluid topology optimization mathematical model according to the target optimization function; based on the fluid topology optimization mathematical model, sampling the calculation domain sample points, and constructing a neural network model in combination with the basic parameters of the fluid pipeline; constructing a loss function and iteratively training the neural network model according to the loss function to obtain a topology optimization result. The present invention solves the problem of uneven density distribution of fluid pipeline materials, shortens the calculation time of fluid pipeline material density distribution, and realizes a method for finding the optimal material density distribution in the design space of uniformly distributed density materials. As a fluid topology optimization method and system based on a physical neural network, the present invention can be widely used in the field of fluid topology optimization.

The invention relates to the technical field of SiC MOSFETs, particularly discloses a SiC MOSFET I-V characteristic modeling method and system based on a physical information neural network, and aims at solving the problem that an existing SiC MOSFET simulation model cannot give consideration to high precision, high speed and practicability. Loss between a calculated value of a physical model and a predicted value of the neural network is added into a loss function of the neural network, and the model can be efficiently and accurately trained only through a small amount of data. According to the method and the system, the internal advantages of the physical neural network are fully utilized, and the actual physical behavior information of the device is embedded into the model, so that a modeling result is highly matched with experimental data in numerical value, strict physical interpretation is shown in theory, and key technical support is provided for industrial application of the silicon carbide power device.

The invention discloses a gas foil bearing wear high-fidelity modeling method based on a physical neural network, and relates to the technical field of rotor dynamics and fault diagnosis. The method comprises the following steps: firstly, constructing physical models of gas foil bearing wear, wherein the physical models comprise a gas foil bearing structure and lubrication mechanism model and a gas bearing rotor system nonlinear dynamic model; and then building a hybrid modeling framework fusing a physical mechanism and data driving, training a model mismatch item by adopting a neural network, representing an unknown parameter by adopting a Sigmoid function, introducing a displacement compensation factor, and finishing model parameter updating by combining forward calculation and reverse calculation. According to the method, high-fidelity modeling of wear evolution and key parameter collaborative identification can be realized, the model adaptability and the data matching precision under complex working conditions are improved, and technical support is provided for a gas bearing digital twin system and predictive maintenance.

The application provides a method and device for optimizing an improved network of a physical neural network. The method comprises adding an input scaling layer, an output scaling layer and a feature layer to the basic physical neural network to construct an improved physical neural network; using a plurality of observation points and the improved physical neural network to make a difference as a first loss function to perform first training on the improved physical neural network; and adding a residual term of the differential equation set to the first loss function to obtain a second loss function to perform second training. The method realizes the feasibility of using a physical neural network to solve a strongly rigid rate theory equation set, adds observation points as a supervised training term and performs pre-training to determine the optimization direction of the model, so that the model training has higher training efficiency. An optimization method for balancing each residual weight is adopted, so that the improved network has better regularization effect.

The invention discloses a biological image data processing prediction method based on a physical neural network, and belongs to the field of biological image processing, and the method comprises the steps: carrying out the denoising of a biological image: constructing a multi-layer memristor neural network structure which comprises an input layer, a plurality of hidden layers and an output layer, the memristors are distributed among all layers of neurons so as to realize weighting and processing of signals; preparing a large number of noisy biological images and corresponding clear image pairs as a training data set; according to the denoising method based on the memristor neural network, noise can be intelligently recognized, noise interference in a biological image can be effectively removed, image details can be reserved to the greatest extent, and high-quality image data can be provided for subsequent image analysis; by using the strong parallel processing capability and feature learning capability of the simulated VLSI neural network, a complex structure in a biological image can be accurately segmented, the segmentation precision and robustness are improved, and accurate analysis of a specific structure in biomedical research is facilitated.

The invention discloses a task scheduling method, device and equipment and a storage medium, relates to the technical field of resource management, is applied to a Kubernetes cluster, and comprises the following steps: determining a current cloud computing task to be scheduled; determining to-be-selected physical neural network processors according to a matching relationship between a virtual resource request corresponding to the current cloud computing task and physical resource information corresponding to the physical neural network processors in the current cluster; determining a target physical neural network processor and a target scheduling node from the to-be-selected physical neural network processors based on a preset processor scoring rule in combination with the virtual resource request; and creating virtual resource information through the target scheduling node to execute the current cloud computing task. Therefore, by means of the virtualization technology, the cloud computing task is scheduled through the virtual resources corresponding to the physical neural network processor, the virtual resources can be dynamically allocated according to the running condition of the cluster, and the resource utilization rate and the task processing efficiency are improved.

The application discloses a physical neural network-based electrical property tomography method, which spatially obtains permittivity and conductivity of all points through four steps.The application does not need to simplify the core equation of MR-EPT to solve, and does not need to use a numerical method to solve by discretizing the derivative of the dielectric property in space, thereby avoiding the discretization error, so that the application has higher precision in obtaining the permittivity and the conductivity.The physical neural network-based electrical property tomography method uses a neural network to construct a mapping relationship between a transmit field and its spatial derivative and electrical properties, so that the precision of the obtained result is higher.

The invention relates to a process industry system modeling method based on an anti-noise hard constraint physical neural network. The method comprises the following steps: firstly, constructing an anti-noise hard constraint physical neural network comprising a framework front end, an anti-noise hard constraint correction layer and an anti-noise loss function; based on the physical relationship between the input and the output of the required modeling system, constructing a linear equality constraint; then quantifying the uncertainty of each sensor in a system needing modeling, and configuring parameters of an anti-noise hard constraint correction layer based on the uncertainty and linear equality constraints; and finally, adopting an anti-noise loss function, and training the anti-noise hard constraint neural network after parameter configuration by using a noise data set to obtain a process industry system model. Compared with the prior art, the method has the advantages that the prediction accuracy of the built substitution model in the noisy input environment is improved while the physical consistency is ensured.

This invention discloses a method, apparatus, device, and storage medium for correcting measurement data of a thermal system, relating to the field of thermal system measurement and processing technology. The method includes the following steps: obtaining a simulation calculation model of the thermal system; constructing a known parameter correction model and an unknown parameter estimation model through a physical neural network; connecting the outputs of the known parameter correction model and the unknown parameter estimation model with the input of the simulation calculation model to obtain a fast correction model; inputting real-time measurement data and real-time operating data into the trained fast correction model to obtain the corresponding correction values ​​and unknown parameter estimates. Compared to classical data coordination methods, the method proposed in this invention effectively avoids the problems of slow calculation speed and poor iterative convergence, making it more suitable for the measurement parameter correction calculation of complex thermal systems in engineering application scenarios.

The present invention relates to an intelligent optimization method and system for the irradiation cross-linking process of flexible armor wires, including: S1: Establish a perception network for the wire irradiation cross-linking process through the industrial Internet to collect real-time data of wire irradiation cross-linking; S2: Construct a mathematical relationship model between process parameters and material properties through a cyber-physical neural network; S3: Combine the finite element simulation method to simulate the effects of the thermal field and dose field during the irradiation process and correct unreasonable assumptions in the mathematical relationship model; S4: Based on the corrected mathematical relationship model, under multi-objective process requirements, based on the data collected by the data real-time acquisition system, balance different objectives through an optimization algorithm to determine the optimal process parameters; S5: After applying the optimal process parameters to the production line, obtain the optimization results through the data real-time acquisition system and judge whether they meet the production objectives, thereby completing the intelligent closed loop. The present invention effectively improves production efficiency and reduces the costs and risks of quality management.

The invention provides a pipeline leakage detection method and system based on a domain adaptive physical neural network, and the method comprises the steps: inputting the space coordinates of a pipeline into a deep learning network model, and obtaining a physical quantity prediction value of the pipeline; inputting the features output by the previous layer of the deep learning network model into a domain classifier to obtain a feature source category; determining experimental measurement values corresponding to the space coordinates, experimental data loss and simulation data loss between CFD simulation values and physical quantity prediction values, determining physical equation residual loss and leakage boundary loss of the physical quantity prediction values, and determining domain confrontation loss according to feature source categories output by a domain classifier; training deep learning network model parameters, leakage position parameters and leakage aperture parameters; and determining the leakage information of the pipeline according to the physical quantity predicted value, the leakage position parameter and the leakage aperture parameter output by the trained deep learning network model. According to the invention, the precision, credibility and robustness of leakage detection are improved.

The present invention relates to the technical field of railway tracks. Specifically, it relates to a multi-objective fine-tuning method for high-speed railway tracks driven by an embedded physical neural network, which includes the following steps: 1) determining the dimension and length of the initial input data according to the track irregularity index; 2) establishing a latent feature expression layer to perform multi-layer non-linear low-rank transformation and feature activation on the input vector; 3) embedding a multi-track parameter constraint layer to perform hard constraints according to the management values; 4) designing a weighted loss function that embeds track features and chord measurement formulas; 5) adaptively iteratively optimizing and selecting the output of the scheme. The present invention can preferably perform multi-objective fine-tuning of high-speed railway tracks.

This invention discloses a photovoltaic power prediction method and system that considers physical mechanism correction. It acquires historical photovoltaic power plant data and future physical data, then selects parameters highly correlated with historical photovoltaic power as prediction inputs. These input parameters and future physical data are fed into a pre-trained physical-neural network joint model to obtain the final photovoltaic power prediction. The training process of this joint model includes inputting the prediction input parameters into the neural network model to obtain neural network prediction values, simultaneously obtaining physical prediction values ​​based on the physical data, and combining the two through weight correction to obtain the final prediction value. Finally, a joint loss function is determined based on the actual photovoltaic power and the final prediction value, and a GA genetic algorithm is used to optimize the weights and neural network hyperparameters, thereby obtaining the trained model. This method combines physical mechanisms and neural network technology, improving the accuracy and reliability of photovoltaic power prediction.

The invention discloses a method for solving a partial differential equation by a physical neural network of a time evolution residual, and belongs to the field of solving the partial differential equation by the neural network. The method comprises the following steps that on the basis of a physical information neural network, a weight factor is introduced into an equation constraint term in a loss function, and the factor dynamically changes along with the training process; a problem domain is divided into a plurality of areas through a time axis in the training process, and sub-domain training is sequentially carried out backwards in the direction of the time axis; the unsteady problem is better solved by using the mode of sequentially activating and solving the physical domain. According to the method, the network is designed through dynamic evolution of the time domain, and the problem that a traditional physical information neural network cannot effectively solve unsteady partial differential is solved.

The invention provides a flatness prediction method and system based on a wavelet and a physical neural network, and relates to the technical field of road engineering, and the method comprises the steps: carrying out the calculation through a scale function of a dbN wavelet, and obtaining a first weight coefficient of a pavement longitudinal contour elevation function; solving the quarter vehicle dynamic equation, determining unsprung mass displacement and sprung mass displacement, and obtaining a second weight coefficient and a third weight coefficient corresponding to the unsprung mass displacement and the sprung mass displacement respectively through wavelet transform; training a preset deep learning network model by using the obtained weight coefficient, and determining a loss function of the preset deep learning network model; performing iterative optimization on the preset deep learning network model by using a Bayesian optimization algorithm and the loss function to obtain a target deep learning network model; and inputting the fourth weight coefficient of the to-be-measured pavement into the target deep learning network model, and determining the to-be-measured unsprung mass displacement, the to-be-measured sprung mass displacement and the international flatness index.

The application provides a spacecraft lithium battery health state estimation method based on a physical information neural network, and has the characteristics that the following steps are included: a second-order hybrid equivalent circuit model of a battery monomer is established, and the relationship between the open circuit voltage of the equivalent circuit model and the state of charge of the battery is represented by a multilayer perceptron; the multilayer perceptron is trained by using the open circuit voltage and the state of charge of the battery in the collected sample data; the multilayer perceptron is used as a data driving layer of a physical neural network model, and the physical information neural network model is designed; the training of the physical neural network model is carried out by using the discharge curve in the collected sample data, and the estimation of the battery health state is realized by using the trained neural network model. The application does not need additional tests, and on the basis of using limited data to establish a physical information neural network model, the health state online estimation is accurately realized, and the stable operation of the spacecraft is ensured.