69 results about "Dynamic neural network" patented technology

Systems and methods for adaptive edge computing using artificial intelligence (AI) include monitoring real-time accuracy of a neural network by using a feedback loop configured to detect changes in inference accuracy and dynamically adjusting the structure of the neural network by adding or removing hidden layers based on monitored error rates and predetermined computational constraints. A Kalman gain computation determines neural network weight adjustments based on monitored error rates. Weight matrices undergo incremental updates derived from these adjustments. Incremental weight adjustments remain stored in memory to enable low-bandwidth model updates. The neural network stores inference results and refined weights in an inference result database. Pre-trained models periodically receive incremental updates based on stored adjustments. Predictive holistic inference logic (PHIL) applied to stored inference results improves the accuracy of the inference results.

The invention discloses a dynamic DEM spatial interpolation method which comprises the following steps: performing depression filling, flow direction analysis and confluence cumulant calculation on DEM data, and extracting a natural sub-basin unit by adopting a minimum catchment area threshold method; constructing a topographic feature matrix, performing refined second-level classification on the first-level drainage basin through an improved self-organizing mapping network, generating a hydrological response unit through boundary processing, and establishing a hydrological attribute library; fusing multi-source data, supplementing attribute interpolation such as underlying surface and rainfall, and constructing an interpolation auxiliary parameter system; a drainage basin is divided into regular grids as neurons, a dynamic neural network containing dynamic states and static attributes is constructed, and nonlinear mapping of DEM correction parameters is achieved through optimization of a dynamic activation function and a loss function. The method overcomes the defects that a traditional interpolation algorithm does not consider hydrological boundary constraints, a neural network model topological structure is fixed and the like, and the DEM interpolation precision and the hydrological simulation effect of the complex terrain area are improved.

The invention discloses a short temporary rainfall forecasting method and system based on a dynamic neural network architecture. According to the method, training sample distribution is optimized through a space-time gradient driven resampling strategy, a time-frequency double-branch dynamic network architecture is adopted, a time domain branch extracts multi-scale space features through an encoder-decoder structure, a frequency domain branch adaptively activates an FFT calculation module through a content awareness dynamic network to capture multi-scale time features, and the multi-scale time features are obtained. And two key problems of meteorological data distribution imbalance and deep learning model optimization imbalance are solved in combination with a balance regression loss function. According to the method, adaptive computing resource allocation for meteorological events with different complexities is realized, the forecasting precision and computing efficiency of heavy rainfall events are remarkably improved, and an efficient and reliable technical solution is provided for short temporary rainfall business forecasting.

The invention discloses a rapid water quality detector and a water quality detection method. The detector comprises a digestion tube, a quantitative liquid feeding unit, a heating unit, a cooling unit, a temperature measuring unit, a colorimetric detection unit and a control unit, the quantitative liquid feeding unit quantitatively inputs liquid into the digestion tube; the heating unit and the cooling unit are respectively used for heating or cooling the digestion pipe; the temperature measuring unit is used for detecting the liquid temperature and the environment temperature in the digestion tube in real time; the colorimetric detection unit is used for detecting the absorbance of liquid in the digestion tube; and the control unit integrates a dynamic neural network temperature compensation algorithm and a temperature-sensitive adaptive multi-target particle swarm optimization algorithm to respectively realize accurate temperature control and detection value optimization. According to the invention, high-precision temperature control is realized through a dynamic neural network temperature compensation algorithm, and the detection error rate is reduced in combination with a multi-target particle swarm optimization algorithm; the liquid feeding precision is guaranteed through the design of the double quantitative loop pipes, and multi-parameter detection is supported; the incremental learning mechanism ensures long-term stability of the system and adapts to wide-temperature-range environment requirements.

The invention relates to the technical field of industrial automation control, and discloses a mixed PID temperature control system and method based on GDNN and multi-modal prediction. The system comprises a data acquisition module, an enhanced GDNN core controller, a prediction optimization module and an actuator module. And real-time adaptive tuning of PID parameters (Kp, Ki and Kd) is realized by fusing a general dynamic neural network, a genetic algorithm optimization initial weight, a variable learning rate and an additional momentum method, and a prediction model module of a multi-layer perceptron (MLP) and a radial basis function (RBF) network. According to the system, nonlinear, time-varying and interference factors can be processed for temperature control scenes such as a high-temperature furnace and a drying system, the mean absolute error, overshoot and stabilization time are remarkably reduced, and the energy efficiency is improved. A multi-modal sensor fusion and edge computing framework is innovatively introduced, and distributed industrial application is supported. The system is suitable for the fields of high-temperature furnaces, drying systems, intelligent buildings and the like, and provides high robustness and interpretability.

Apparatuses, systems, and techniques to assign a processing resource to an inference request directed to a neural network based on an amount of information to be inferenced indicated by said request. In at least one embodiment, an AI application is deployed with a software wrapper that intercepts inference requests and dynamically distributes such requests among available processing resources such as host processor(s) and / or AI accelerator(s) to improve execution performance of said AI application.

The invention provides an electrocardiosignal anomaly detection method fused with electrocardiosignal propagation simulation. The method comprises the steps that a space diagram structure is established based on ECG data; performing spatial position mapping on each electrode and establishing a heart simulation physiological model; feature extraction is carried out on each electrode node, and a dynamic space diagram structure is constructed; based on a dynamic graph structure, through multi-layer graph convolution updating and time aggregation, global space-time feature vectors are output, electrocardiosignal detection categories are output through a classifier, a space dynamic neural network model is constructed, a total loss function is established for model optimization, the electrocardiosignal detection categories are predicted, and quantitative recognition of abnormal propagation paths is achieved. And iteratively optimizing the dynamic graph structure. According to the method, unified modeling is carried out through time dynamics and spatial relevance of physiological signals, the defect that a static graph cannot capture abnormal changes of a propagation path in a pathological state is overcome, and the accuracy of cardiovascular disease diagnosis and the effectiveness of clinical decision support are remarkably improved.

A method for online target counting on unmanned aerial vehicles (UAVs) based on a multimodal dynamic neural network includes the following steps: First, the UAV uses an onboard visible light camera and a thermal infrared camera to capture visible light and thermal infrared images, respectively. Then, the segmented images are used as input to a target counting model. The target counting model is a target counting model based on a multimodal dynamic neural network. When thermal infrared images are unavailable, the visible light images are converted into thermal infrared images using a mode conversion network. Multimodal gating analysis is performed on the scene illumination where the images are located. If the illumination is good, only the visible light mode branch of the target counting model is selected; if the illumination is insufficient, both the visible light mode branch and the thermal infrared mode branch are selected, with the thermal infrared images used as auxiliary data. A feature extraction network extracts the modal features of the selected modal branches and fuses the features from the multimodal inference flow. The features obtained in step S3 are decoded to obtain the output of the target counting model.

The invention discloses a fan nonlinear load self-adaptive control method and device and a storage medium. The fan nonlinear load self-adaptive control method and device are used for efficiently achieving dynamic tracking control over fan nonlinear loads. The method comprises the steps of collecting operation state data of a fan system; constructing a dynamic neural network model containing a long short-term memory network-attention mechanism mixed structure, inputting the operation state data into the dynamic neural network model, and outputting a dynamic predicted value of the fan nonlinear load; constructing an adaptive model prediction controller based on the dynamic prediction value, and generating a control sequence; calculating a prediction residual error of the dynamic neural network model, and when the prediction residual error exceeds a standard threshold value, triggering a model updating mechanism; performing parameter updating on the dynamic neural network model by adopting an incremental learning algorithm with gradient constraint to obtain updated parameters; and the safety control quantity is output to a fan execution mechanism, and dynamic tracking control over the nonlinear load of the fan is achieved.

The invention provides a heterogeneous perception adaptive low-bit neural network deployment method, and relates to the technical field of artificial intelligence and heterogeneous computing, and the method comprises the steps: firstly obtaining the hierarchical computing feature information of each layer of a neural network and the dynamic feature parameter information of heterogeneous hardware, and forming a multi-level basic information set; and then a hierarchical efficiency association model is constructed to describe the association relationship among the calculation precision, the hardware dynamic characteristics and the layer calculation efficiency. During online operation, a hardware real-time load state and an energy efficiency constraint condition are tracked to generate a dynamic state monitoring result, and the dynamic state monitoring result and the dynamic state monitoring result are combined to generate a hierarchical deployment configuration scheme by adopting a multi-objective optimization algorithm. And finally, a lightweight runtime scheduler is called to allocate calculation tasks according to the scheme, and interlayer dependency data is loaded and executed, so that dynamic neural network deployment across hardware equipment is realized, and the deployment effect and the operation efficiency are improved.

The invention relates to the technical field of traditional Chinese medicinal materials, in particular to an intelligent detection, regulation and control method based on a traditional Chinese medicinal material planting environment, which comprises the following steps: S1, multi-modal data acquisition: acquiring environmental parameters and plant physiological parameters of a planting area in real time; s2, dynamic neural network modeling: performing space-time correlation analysis on the environmental parameters and the plant physiological parameters to generate a dynamic weight matrix; s3, fuzzy control and rolling optimization: according to the dynamic weight matrix, combining a fuzzy control algorithm to generate a multi-target cooperative regulation and control parameter; and S4, closed-loop feedback and model iteration: driving an execution mechanism to perform environment regulation and control, and correcting a dynamic weight matrix and regulation and control parameters in real time through an online feedback mechanism to form closed-loop control. According to the method, multi-mode sensing data are fused through the dynamic graph neural network, intelligent regulation and control of the traditional Chinese medicinal material planting environment are achieved, the light-temperature-water-fertilizer cooperation strategy is accurately optimized, the secondary metabolite synthesis efficiency is remarkably improved, energy consumption is reduced, and the method adapts to different medicinal material characteristics and environment changes.

A method for counting drone image targets based on a parameter-adaptive dynamic neural network, the steps of which include: 1) An aerial photography device on the drone collects ground images; 2) The collected images are preprocessed; 3) A feature extraction network is used to detect targets in the processed images; 4) The detected targets are counted. In step 2), the scale is dynamically divided and the image is segmented according to the flight altitude and shooting depression angle of the drone; in step 3), a feature extraction network with enhanced spatial information is used for the input feature map to extract the position information of the targets. The dilation rate parameter of the convolution in the receptive field selection module after the feature extraction network is determined according to the segmented image. A dilation rate parameter adaptive strategy is designed to calculate the scale range and segment the image according to the flight state, and to adjust the dilation rate parameter in the receptive field selection module. This method significantly reduces the inference latency, improves the MAE and MSE, and also has excellent counting accuracy.

The application provides a robot driving control method, device, equipment and storage medium, comprising: calculating a tracking error according to a desired trajectory and an actual running trajectory corresponding to each joint of a current time of a linkage robot; calculating the tracking error by using a sliding mode control algorithm to obtain a first input torque of the current time; taking an actual input torque of a previous time and the actual running trajectory as inputs of a dynamic neural network to predict an unknown torque in a robot dynamics model at the current time, and taking the predicted unknown torque as a second input torque of the current time; determining an actual input torque of the current time according to the first input torque and the second input torque, and controlling the linkage robot to move according to the actual input torque of the current time; and solving the problem of stability and accuracy decline of the linkage robot caused by various uncertain factors in the prior art.

The present invention discloses a wireless communication network resource allocation algorithm for dynamic adjustment on demand, which specifically includes the following steps: Step 1: Quantitatively describe the characteristics of tasks in the wireless communication network and represent them as input vectors of fixed dimensions #imgabs0# Step 2: Construct a high-reliability and low-latency resource allocation model based on a dynamic neural network with variable widths; Step 3: For the said model, give an optimization objective and a knowledge-driven solution; Step 4: Establish a knowledge base regarding the deployment environment and the optimal inference model; Step 5: Obtain the optimal model from the knowledge base to get the resource allocation scheme. The technical effect of the invention is to divide the network resource allocation problem into two stages: First, determine the optimal width of the decision network according to the task characteristics and the user's computing power; then, input the task characteristics into the optimal decision network to obtain the optimal resource allocation scheme.

The invention discloses an RSSI (Received Signal Strength Indicator) indoor positioning and ranging method based on neural network learning, relates to the technical field of indoor positioning and ranging, and remarkably improves RSSI ranging robustness in complex scenes such as underground parking lots and the like through multi-modal fusion and a dynamic neural network architecture. A double-flow neural network is adopted to separate modeling signal long-term attenuation and instantaneous abrupt change characteristics, the RSSI distortion problem caused by dynamic shielding is effectively relieved, and dependence of a traditional model on a uniform signal environment is avoided; an expansion rate and moving speed dynamic binding mechanism is introduced, a convolution receptive field is adjusted in a self-adaptive mode, time sequence information loss caused by low-speed sampling is compensated, and accumulated errors under high-speed moving are restrained; meanwhile, on the basis of a pre-training strategy of the generative adversarial network, diversified shielding scene data is synthesized through physical constraints, and the generalization ability of the model to multipath reflection and metal interference is enhanced.

The invention discloses a petrochemical engineering reaction process intelligent regulation and control method based on a dynamic neural network, and relates to the technical field of petrochemical engineering process control, and the method comprises the following steps: obtaining temperature distribution data of a catalyst bed layer, and fluid flow velocity field data, component concentration gradient data and stirring frequency data of a reactor in real time; and inputting the temperature distribution data and the stirring frequency data into a first branch of a dynamic neural network, and outputting an activity attenuation index of the catalyst bed layer. According to the scheme, by fusing the temperature standard deviation, the stirring dominant frequency component and the by-product weighted value, the dynamic relevance of mass transfer and heat transfer on the surface of the catalyst can be captured, meanwhile, the nonlinear influence of a side reaction path on activity attenuation is considered, the abnormal activity of the catalyst is warned in advance, and a more accurate input basis is provided for regulation and control of the reaction process.

The invention discloses a flexible load resource aggregation method based on an improved dynamic neural network, and relates to the technical field of flexible load resource aggregation and regulation, and the method comprises the steps: carrying out the analysis of the composition and adjustment characteristics of a flexible load, and obtaining a flexible load resource adjustable capability quantitative index, a generalization mathematical model is constructed based on the flexible load resource adjustable capability quantitative index, the generalization mathematical model comprises an electric vehicle model and an energy storage model, clustering analysis is carried out on the generalization mathematical model, a flexible resource pool model is formed through coupling, and the flexible resource pool model is optimized to obtain a flexible load resource aggregation optimization model; according to the flexible load resource aggregation optimization method based on the improved dynamic neural network, the adjustable potential of flexible loads such as electric vehicles and energy storage is accurately evaluated, and the problem that multi-type load collaborative optimization is insufficient in a traditional method is effectively solved; and high-precision resource support is provided for scenes such as peak regulation and frequency modulation.

The application provides a heterogeneous-aware adaptive low-bit neural network deployment method, and relates to the fields of artificial intelligence and heterogeneous computing technology. First, hierarchical computing feature information of each layer of a neural network and dynamic characteristic parameter information of heterogeneous hardware are acquired to form a multi-level basic information set. Then, a hierarchical performance correlation model is constructed to describe the correlation between computing accuracy, hardware dynamic characteristics and layer computing performance. During online operation, real-time hardware load state and energy efficiency constraint conditions are tracked to generate dynamic state monitoring results. A multi-objective optimization algorithm is used in combination with the two to generate a hierarchical deployment configuration scheme. Finally, a lightweight runtime scheduler is called to allocate computing tasks, load inter-layer dependent data and execute according to the scheme, thereby realizing dynamic neural network deployment across hardware devices and improving deployment effect and operation efficiency.

The present invention discloses a method for edge collaborative task division and resource allocation based on model segmentation, which relates to the field of edge computing video analysis, and integrates CNN channel pruning, bilinear sampling and affine quantization to multi-dimensionally compress intermediate transmission feature data; a parameter-sharing dynamic neural network is trained based on the sandwich rule and knowledge distillation technology to realize real-time switching of the feature data compression amplitude; user reasoning performance indicators in edge collaborative task division are modeled, and an optimization problem for maximizing reasoning accuracy that meets user performance requirement constraints and edge resource constraints is constructed; a solution algorithm based on dynamic programming is used to select the optimal model segmentation point and intermediate feature compression parameter configuration for the user, and at the same time, edge communication resources and computing resources are efficiently allocated to improve its reasoning accuracy under multi-user conditions.

The present application relates to the field of intelligent recommendation. In particular, it relates to a product recommendation method and system based on state updating. The method comprises obtaining product state time series data and consumer preference state time series data, inputting them into a recommendation model based on a double-state dynamic neural network, outputting product feature vectors and consumer preference feature vectors, and obtaining a pre-sale product recommendation list using a collaborative filtering method. According to the pre-sale product recommendation list, a corresponding adoption product is selected. After the product growth process is completed, the product state time series data and the consumer preference state time series data of the product that has not been adopted are input into the recommendation model based on the double-state dynamic neural network, real-time product feature vectors and real-time consumer preference feature vectors are output, and a formal product recommendation list is obtained using a collaborative filtering method. The present application can take into account the whole process of agricultural products and dynamically adjust the recommendation content according to the state of agricultural products.

The invention relates to the technical field of unmanned aerial vehicle cluster modeling and system stability analysis, in particular to an unmanned aerial vehicle network system prediction model phase point stability domain calibration method and system and a program product. The method comprises the steps of collecting phase point trajectory data, judging whether a data scale meets a modeling requirement or not, and if not, performing expansion in a linear interpolation and Gaussian disturbance mode; performing time sequence prediction on trajectory evolution by using an NARX dynamic neural network; and constructing a calibration data set in combination with latest observation data and a prediction result, and finally outputting a center vector and radial vector approximate value of a stability domain. The invention also provides a corresponding system structure and a computer program product, which have the advantages of high calculation efficiency, high precision and strong adaptability, can adapt to various typical scenes such as initial steady state, limit offset and steady state recovery, and is widely applicable to stability evaluation and control decision support of the cluster unmanned aerial vehicle.

The invention provides a gas turbine state intelligent early warning method and system based on a dynamic neural network. The method comprises the steps of gas turbine historical operation data acquisition, characteristic parameter screening, neural network prediction model construction and training, gas turbine real-time data acquisition, gas turbine key operation state characteristic parameter prediction and analysis, dynamic updating of a gas turbine historical database and a neural network prediction model and the like. And real-time monitoring and early warning of the running state of the gas turbine are realized. According to the method, through boundary condition-operation state characteristic parameter classification and neural network prediction model dynamic modeling and training, the problems of limitation of single-parameter early warning in traditional gas turbine state monitoring, lack of prediction model updating mechanisms and the like are solved, and a technical means can be provided for intelligent operation and maintenance of the gas turbine.

This invention discloses a zero-shot speech cloning method based on a dynamic neural network and feature modulation. The method comprises extracting a speaker style vector from reference audio using a speaker style encoder; performing feature modulation on the speaker style vector using the SGF algorithm; and inputting the modulated speaker style vector into a generator; and synthesizing the target speaker audio using a dynamic neural network. This method can clone the audio of any speaker in a zero-shot scenario, synthesizing smooth, natural, and highly similar target audio.

Neural network accelerators with one or more neural network accelerator cores. Each neural network accelerator core has hardware accelerators configured to accelerate neural network operations, an embedded processor, a command decoder, and a hardware feedback path between the embedded processor and the command decoder. The command decoder is configured to control the hardware accelerators and the embedded processor of that core in accordance with commands of a command stream, and when the command stream comprises a set of one or more branch commands that indicate a conditional branch is to be performed, cause the embedded processor to determine a next command stream, and in response to receiving information from the embedded processor identifying the next command stream via the hardware feedback path, control the one or more hardware accelerators and the embedded processor in accordance with commands of the next command stream.

The invention relates to the technical field of multi-modal processing, and discloses a multi-modal brain-like computing system which comprises an eye-like computing module, a skin-like computing module, a heart-like computing module, a nerve-like computing module and a brain-like computing module. The eye-like calculation module is used for capturing visual information and carrying out preprocessing and feature extraction; the skin-like calculation module is used for sensing physical stimulation and encoding the physical stimulation into tactile information; the class heart calculation module is used for identifying an emotional state of the user and generating an emotional response; the neural-like calculation module is used for realizing signal transmission and dynamic neural network adjustment; and the brain-like calculation module is used for integrating the multi-modal information and outputting a decision instruction. The multi-modal brain-like computing system aims to realize a computing system capable of efficiently processing multi-modal information and simulating a human brain processing mode by integrating a plurality of eye-like, skin-like, heart-like, nerve-like, brain-like computing modules and the like.

The invention discloses a coagulant self-adaptive adding control method and system based on a physical constraint neural network, and relates to the technical field of water treatment automatic control, and the method comprises the steps: collecting water quality data; preprocessing the water quality data to obtain a standardized water quality sequence; inputting the standardized water quality sequence into a pre-trained time sequence dynamic neural network model, and predicting future dosage and future effluent turbidity; at the end of each control period, based on a model prediction control mechanism, performing rolling optimization on the pre-trained time sequence dynamic neural network model by taking the optimization objectives that the future effluent turbidity predicted by the time sequence dynamic neural network model does not exceed the standard and the future dosing amount is minimum, and solving an optimal medicament dosing sequence; and converting the optimal medicament feeding sequence into a control signal to drive a dosing pump to feed medicaments. The problems that a traditional control method is weak in self-adaptive capacity, poor in predictability and high in medicine consumption when facing complex and changeable raw water quality are solved, and stable standard reaching of the effluent quality and accurate optimization of the coagulant adding amount are achieved.

The application discloses a motor driving sudden change friction characteristic modeling and feedforward compensation method, which realizes decoupling and independent description of non-sudden change characteristics and sudden change characteristics of friction, and constitutes a dynamic neural network friction model with a width neural network structure which integrates the width neural network structure characteristics and the dynamic recurrent neural network structure characteristics; and the motor control system feedforward control compensation based on the friction model: the motor rotation angle is used as the input signal of the friction model, the friction model obtains corresponding friction torque prediction and estimation, the output of the motor controller is corrected through the feedforward control compensation, and the influence of internal friction on the motor execution precision is offset.

An automatic driving target detection method based on self-distillation assisted dynamic neural network, the collected image is used as the input of the target detection model to realize target classification. The target detection model is based on a self-distillation assisted dynamic neural network, which is composed of a strategy network, a detection network and a hybrid distillation network. The detection network is composed of a series of multi-branch residual blocks. The detection network takes the collected image as the input and outputs the target classification result. The strategy network generates corresponding routing vectors for different inputs to determine the opening and closing of the residual blocks in the detection network. The hybrid distillation network includes two subnetworks. The two subnetworks are differentiated from the detection network in the initial training stage and are randomly initialized. The subnetworks after hybrid distillation are fused into a fusion network. Through knowledge distillation, the knowledge of the fusion network is compressed and migrated back to the detection network, completing a closed-loop optimization.

The application discloses a kind of bionic dynamic neural network, the neural network can form behavior state relation net by learning, and can be received when setting target task, state transfer and behavior activity control are carried out according to current state and the behavior state relation net formed, directly complete the target task.The bionic dynamic neural network provided in the embodiment of the application imitates the neural circuit topological structure and dynamic characteristics of biology, realizes the control to state transfer, so that state transfer and action control form the behavior state relation net with causal relationship, more in line with the way of human brain processing problem, improve the decomposition ability of complex task and the efficiency of reinforcement learning, so that reinforcement learning can be more realistically used in actual robot task.The bionic dynamic neural network of the embodiment of the application can be applied in reinforcement learning model, for building the reinforcement learning algorithm model with higher learning efficiency capable of decomposing task.

Apparatuses, systems, and techniques to assign a processing resource to an inference request directed to a neural network based on an amount of information to be inferenced indicated by said request. In at least one embodiment, an AI application is deployed with a software wrapper that intercepts inference requests and dynamically distributes such requests among available processing resources such as host processor(s) and / or AI accelerator(s) to improve execution performance of said AI application.