81 results about "Learning rule" patented technology

The invention provides an intelligent scheduling and control method and device for an integrated energy system. According to the method, power, gas and heat resource operation data are acquired, multi-scale layered modeling is performed according to a time scale and a space scale, and a power resource state space model, a gas resource flow continuity model, a heat resource heat balance model and a multi-energy coupling characteristic constraint model are established; carrying out feature extraction and dimension reduction representation by adopting a deep auto-encoder network; cooperative training of multiple groups of cognitive models is carried out through a split hierarchical federal learning framework, and a global intelligent model is obtained; constructing a neural architecture search network with a hybrid bionic learning rule, setting a hierarchical scheduling target, and generating a hierarchical intelligent scheduling strategy; and a fault-tolerant control mechanism is constructed, and error detection and correction of operation deviation are realized. According to the invention, multi-time scale collaboration, collaborative learning under multi-device group privacy protection and high-reliability fault-tolerant control are realized, and the operation efficiency and reliability of the integrated energy system are remarkably improved.

The invention discloses an event prediction and early warning method based on Bayesian deep learning. The event prediction and early warning method mainly comprises the following two parts: constructing a Bayesian deep learning rule model based on event characteristics, and designing a prediction and early warning model on the basis of constructing the Bayesian deep learning rule model. According to the method, the Bayesian statistical method and the deep learning model are fused, the advantages of the Bayesian statistical method and the deep learning model can be fully utilized, the uncertainty of a prediction result can be expressed in a probability distribution form by combining priori knowledge and observation data, efficient modeling and prediction are performed on complex and nonlinear time sequence data, early warning of potential events is realized, and the prediction efficiency is improved. The accuracy and reliability of event prediction and early warning are improved.

The invention is applied to the field of mechanical fault diagnosis signal processing, and particularly provides a pulse neural network mechanical fault diagnosis method based on STFT dimension transformation, and the method comprises the steps: collecting a one-dimensional mechanical vibration signal, carrying out the wavelet decomposition, carrying out the wavelet reconstruction of a low-frequency component and a denoised high-frequency component, and carrying out the wavelet reconstruction of the low-frequency component and the denoised high-frequency component; obtaining a denoised one-dimensional vibration signal; performing short-time Fourier transform, and converting the time-frequency two-dimensional matrix into a time-frequency two-dimensional matrix; inputting the time-frequency two-dimensional matrix into an improved HH threshold neuron model, carrying out Poisson sparse coding on the time-frequency two-dimensional matrix, and only carrying out pulse response on signal significant features; constructing a suprathreshold coding convolutional network with residual connection, inputting a sparse coding matrix, training by adopting an unsupervised learning rule based on STDP, and adaptively adjusting a network synaptic weight; and inputting to a trained above-threshold coding convolutional network, and obtaining pulse emission activity of neurons of an output layer through network forward propagation to determine a fault diagnosis result.

The invention discloses an atmospheric pollution trend analysis method and system, and relates to the technical field of atmospheric pollution trend analysis, and the method comprises the steps: collecting multi-source data in a target region, calculating the concentration change rate of all nodes by using the law of conservation of mass, updating the concentration of the nodes by using an Euler method, and inputting the updated concentration of the nodes into a pre-trained GCN model, calculating the optimized concentration and production abnormity list; modulating a propagation weight by using a Hebbian learning rule, modulating the optimized concentration by using the modulated weight, calculating a joint concentration by using a conservation type path flux method, combining the joint concentration by using a weighted linear combination, and calculating the joint concentration; and constructing a long-short-term memory network model to predict the future trend of the pollutants. The space gridding and the graph neural network are introduced, the recognition and control capability of the pollution source is improved, and the prediction precision and reliability of the analysis model are enhanced by combining the diffusion flux model and the convection flux model.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for enabling a user to conduct a dialogue. Implementations of the system learn when to rely on supporting evidence, obtained from an external search system via a search system interface, and are also able to generate replies for the user that align with the preferences of a previously trained response selection neural network. Implementations of the system can also use a previously trained rule violation detection neural network to generate replies that take account of previously learnt rules.

The invention discloses a spiking neural network driven robot dynamic adaptive control system, which relates to the technical field of robot control, and comprises seven modules: an environment sensing module which integrates various sensors and collects and transmits environment, attitude and interaction information; the signal preprocessing module processes data through composite filtering and feature extraction; the spiking neural network modeling module constructs a three-layer structure and performs training based on a fusion learning rule; the dynamic decision output module converts the pulse signal into a control instruction and adjusts gain; the actuating mechanism driving module drives the actuator to act; the state feedback monitoring module monitors and feeds back motion parameters and system states; and the adaptive optimization module optimizes the network and module parameters based on feedback data, and dynamically matches the environment. The control precision and the response speed of the robot in a complex environment are improved, the adaptive capacity is enhanced, the operation reliability and safety are guaranteed through multi-module cooperation, and the application scene is expanded.

The invention discloses a medical aid decision-making system based on a multi-modal large model, and relates to the technical field of medical artificial intelligence, and the system comprises a multi-modal data fusion module, a treatment scheme coding and management module, a prognosis prediction model module, a scheme simulation and deduction module, and a visual comparison module. The comprehensive state vector of a patient and the action vector of a candidate treatment scheme are jointly calculated, multiple long-term prognosis indexes after the scheme is executed are directly simulated, the model converts the treatment scheme into a computable variable by learning rules in historical treatment data, probabilistic prediction is carried out on a future result, and the prediction accuracy is improved. This enables a doctor to clearly see the risk brought by different selections and the long-term influence of treatment changes before making a decision, thereby converting the decision mode from experience-based inference to future simulation-based anticipation.

The provided are a federated reinforcement learning (FRL) end-to-end autonomous driving control system and method, as well as vehicular equipment, based on complex network cognition. An FRL algorithm framework is provided, designated as FLDPPO, for dense urban traffic. This framework combines rule-based complex network cognition with end-to-end FRL through the design of a loss function. FLDPPO employs a dynamic driving guidance system to assist agents in learning rules, thereby enabling them to navigate complex urban driving environments and dense traffic scenarios. Moreover, the provided framework utilizes a multi-agent FRL architecture, whereby models are trained through parameter aggregation to safeguard vehicle-side privacy, accelerate network convergence, reduce communication consumption, and achieve a balance between sampling efficiency and high robustness of the model.

The invention relates to an automatic driving decision planning model training method and device, and belongs to the technical field of automatic driving, and the method comprises the steps: carrying out the training of a first deep multi-agent reinforcement learning model based on the data of a first driving scene and a preset reward rule; wherein the reward rule comprises the steps that when the first deep multi-agent reinforcement learning model predicts and outputs a first decision behavior of at least one vehicle according to the data of the first driving scene, an instant reward is generated for each first decision behavior, and a local reward is generated for each first vehicle set; the first vehicle set is a set of the own vehicle and the adjacent vehicle; and based on the data of the second driving scene and a preset transfer learning rule, training the first deep multi-agent reinforcement learning model again to obtain a target deep multi-agent reinforcement learning model. The target depth multi-agent reinforcement learning model obtained through training has better generalization for different driving conditions and environments.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for enabling a user to make a conversation. Implementations of the system learn when depends on support evidence obtained from an external search system via a search system interface, and can also generate replies for the user that conform to preferences of a previously trained response selection neural network. Implementations of the system may also use a previously trained violation detection neural network to generate replies that take into account previously learned rules.

The invention discloses a system security access control method based on deep learning and dynamic risk awareness, which comprises the following steps of: generating a behavior state vector by collecting user identity, operation, equipment state and environment information, constructing an access behavior map, and establishing an access behavior map through a Hebbian learning rule; the method comprises the following steps: dynamically updating connection strength between behaviors in combination with historical behavior data, after receiving an access request, selecting an association path from a map, calculating an activation potential value, selecting an optimal path as an access decision basis by applying a Winner-Take-All selection mechanism, executing an access control operation, generating a risk mark and a behavior evolution mark, and establishing an access control strategy; the method is used for adjusting a selection threshold and an atlas weight. And finally, compressing and coding the related information, and recording the information in an access log. According to the method, intelligent judgment on the access behavior is realized by using deep learning and a dynamic risk perception mechanism.

The invention relates to a battery life and safety collaborative prediction method and system based on an improved pulse neural network algorithm, and belongs to the technical field of electric tool battery management. The method is based on a four-level SNN architecture, an input layer receives voltage, current, temperature, stress and change rate parameters of a battery, and neural pulse conversion of the parameters is realized through rate coding and time coding; the synaptic layer simulates a physical hysteresis effect between parameters by using a double-index model, and the hidden layer passes through Adaptive Iamp; the F neurons integrate historical memory and nonlinear coupling, and the output layer generates health status, residual life and risk index. The synaptic weight is optimized through a supervised STDP learning rule, the contribution degree of each parameter is quantified, a membrane potential linear accumulation mechanism of the life dimension and an abnormal pulse triggering mechanism of the safety dimension are established, and a joint early warning decision is realized. The system is deployed in a battery management system, the battery state can be monitored in real time, and accurate decision support is provided for battery full life cycle management.

A method, system, and apparatus, including a computer program encoded on a computer storage medium, for enabling a user to conduct a dialogue. System implementations can learn when to rely on supporting evidence obtained from an external search system via a search system interface and generate answers for the user that are consistent with the preferences of a previously trained response selection neural network. System implementations can also use a previously trained rule violation detection neural network to generate answers that take into account previously learned rules.

The invention relates to the technical field of data processing, and particularly discloses a financial information management system and method based on big data analysis. The system comprises a multi-source heterogeneous data acquisition module, a data semantic understanding and mapping module, a dynamic rule engine module, a real-time stream processing module and a financial decision support module, realizes data semantic unification through deep learning and a knowledge graph, and dynamically adapts to business change by means of a self-learning rule engine. And real-time calculation and anomaly detection are guaranteed based on a distributed stream processing framework, and finally the intelligence level and the operation efficiency of financial decision making are improved through visualization and predictive analysis.

The invention discloses a low-altitude dynamic target unmanned aerial vehicle track traceability identification method based on machine learning. The method comprises the following steps: step 1, constructing a known track state sequence set; 2, constructing an identity track neurograph; 3, updating a synaptic connection weight by adopting an improved Hebbian learning rule, and generating an identity track memory map through a synaptic enhancement and attenuation mechanism; 4, constructing a target trajectory neural map, executing an improved Hebbian learning rule to update the synaptic connection weight, and obtaining a target trajectory memory map; 5, constructing a candidate identity set; 6, calculating a resonance identification score; and 7, identifying the candidate identity track memory map with the highest resonance identification score as a target traceability, and outputting a mapping result of a corresponding identity tag and a synaptic path. According to the invention, improved Hebbian learning rules and atlas resonance identification are fused, and low-altitude dynamic target unmanned aerial vehicle track traceability identification is realized.

The invention discloses an AI model training method, an AI model using method, computer equipment and a storage medium, and the method comprises the steps: obtaining a strategy library, and initializing a running environment according to a scene in the strategy library; based on a unified AI model, multiple agents are controlled to play chess in the operation environment to obtain a training sample, and the training sample comprises a scene analysis feature, an instant state feature, a task instruction feature and a role feature; and carrying out model training on the AI model according to the training sample until the model converges, and obtaining the trained AI model. According to the method, task allocation is optimized on the basis of a unified AI model in combination with environment information and role information, conflicts and resource waste among multiple agents are reduced, meanwhile, a strategy library is added into AI model training, the AI model can learn rules and experience and make better decisions under different conditions, and therefore the team cooperation effect among the multiple agents is improved.

This invention belongs to the field of unmanned aerial vehicle (UAV) control technology, and particularly relates to a feedforward control method for multi-rotor UAVs based on a dynamic cascaded spiking neural network (RCN). The method includes: S1: constructing a network model of the RCN based on pulse signals; S2: constructing a cost function of the RCN based on the network model and pulse errors; S3: solving for the weights of the RCN; S4: setting a preset similarity threshold and determining the learning rules for the dynamic cascaded structure based on the preset similarity threshold; S5: obtaining the output signal of the RCN according to the weights and the learning rules of the dynamic cascaded structure, and implementing feedforward control of the multi-rotor UAV based on the output signal. This invention improves the adaptability and robustness of multi-rotor UAVs in complex flight environments.

A spiral electric radiant tube comprises a spiral electric heating wire, a spiral magnetic column, a magnetic disk and a radiant tube sleeve. A cavity structure is arranged in the radiant tube sleeve, a plurality of spiral magnetic columns are arranged in the cavity structure, a plurality of magnetic discs are arranged on the spiral magnetic columns in a sleeving mode, the spiral magnetic columns are fixedly installed in the radiant tube sleeve through the magnetic discs, and the spiral electric heating wire is evenly wound in the axial direction of the spiral magnetic columns. According to the method, the spiral electric radiant tube is accurately controlled through a model combining a single neuron model and a PID control model, a Delta learning rule is adopted, and the function relationship between the weighting coefficient and the input, output and output deviation of the neuron is considered; when the Delta learning rule is adopted, the connection weight can be adjusted according to the difference between the actual output and the expected output of the neuron, and the weight coefficient can be adjusted online according to the learning capability and the self-adaptability of the neural network, so that the method has a more stable tracking characteristic and better robustness, and has higher self-adaptability and higher control precision.

The present invention discloses a fully parallel vector outer product calculation operation method based on electrochemical random access memory, including constructing a memory array with electrochemical random access memory and its gate-select transistor as basic units, and a corresponding random pulse update method. Among them, the conductivity change of the electrochemical random access memory is highly linearly related to the number of stimulation pulses. When no stimulation pulse is applied, the device channel conductance remains unchanged, showing non-volatile characteristics. Through random pulse programming, the device's conductance update and online training can be performed in situ, the vector outer product calculation is completed in parallel, and the calculation result is stored in the device's conductance, and subsequent network in-memory calculation acceleration is performed, avoiding additional data handling, and effectively reducing operation complexity and device programming delay. In addition, this solution can be extended to the network training paradigm related to the Hebbian learning rule to accelerate the training process of related algorithms.

The application discloses a kind of memristor neural network circuit based on competition learning mechanism, it includes forward calculation module and reverse adjustment module.Wherein, forward calculation module is composed of memristor cross array module and leakage integral and discharge module, and reverse adjustment module is composed of weight adjustment module.Forward calculation module realizes the transverse inhibition and competition activation between two neurons by winner-takes-all algorithm, and only one neuron competes successfully and then generates output signal for each input.Reverse adjustment module carries out neuron weight adjustment by Hebb learning rule, realizes self-learning, it receives the output signal of forward calculation module, then generates adjustment signal to forward calculation module, and adjusts the weight of winning neuron.Therefore, the memristor neural network circuit based on competition learning mechanism proposed in the application can learn input data, and classify input data after learning is completed.

The invention discloses an intelligent auditing method and system based on machine learning, and relates to the technical field of intelligent auditing methods.The intelligent auditing method comprises the steps that a multi-source heterogeneous auditing time-space diagram composed of a plurality of time slice diagrams is constructed, each time slice diagram comprises an enterprise entity node, an account node, a personnel node and a document node, the nodes are connected through heterogeneous relation edges with timestamps; converting a preset auditing rule into structured logic expressions, generating a corresponding learnable rule embedding vector for each structured logic expression, and fusing the learnable rule embedding vectors with the original features of the heterogeneous relationship edge to obtain a rule perception edge representation; inputting a multi-source heterogeneous auditing time-space diagram and the rule perception edge representation into a time sequence diagram neural network, and updating the representation of each node through cross-time slice message transmission to obtain an abnormal sensitive representation; and inputting the anomalous sensitive representation into a classifier in which a random discarding mechanism is introduced.

The embodiments of the present disclosure disclose a training method of a semantic segmentation model, a semantic segmentation method and device of an image, wherein the method comprises: updating a first semi-supervised semantic segmentation model based on a first processing result of the first semi-supervised semantic segmentation model on first labeled training image data and a second processing result of the first semi-supervised semantic segmentation model on first unlabeled training image data, to obtain a second semi-supervised semantic segmentation model; determining at least one image to be labeled by using a preset active learning rule based on the second processing result; updating the first labeled training image data and the first unlabeled training image data based on the obtained each image to be labeled and the corresponding label; and training the second semi-supervised semantic segmentation model based on the updated second labeled training image data and the second unlabeled training image data to obtain a target semantic segmentation model. The embodiments of the present disclosure can effectively improve the distribution of labeled data, thereby effectively improving the performance of the semantic segmentation model.

The invention provides an AI-based virtual digital human teaching method and system and a storage medium, and relates to the technical field of intelligent teaching, and the method comprises the following steps: obtaining a student basic capability factor database, and constructing a teaching model learning rule; obtaining student interference data, and making a teaching plan of each subject; performing ability evaluation every preset time interval, performing score test on the students, and establishing an ability improvement system; according to the ability improving system, current ability parameters of the students are obtained and dynamically compared with target ability, whether learning early warning is generated or not is judged, and if yes, active correction teaching is carried out; otherwise, the operation is not executed. According to the invention, a whole set of student ability evaluation and teaching system is constructed, continuous guidance teaching is carried out according to the score and learning tendency of each subject of the student, the manual teaching cost is reduced, targeted active correction teaching can be carried out according to the learning ability improvement system, the teaching task is more accurate and reliable, and the ability of the student is effectively improved.

The invention relates to the technical field of ocean platform pipeline design, in particular to an ocean platform pipeline laying method based on reinforcement learning, and the method comprises the following steps: S1, dispersing a pipeline laying region into a three-dimensional grid matrix according to the physical size of an actual cabin of an ocean platform, and marking a pipeline starting point, a pipeline ending point and an impassable region; s2, performing Q-Learning algorithm parameter initialization configuration, defining an action space adaptive to the linear movement characteristics of the ocean platform pipeline, constructing a Q value matrix adaptive to three-dimensional space coordinates and actions, and performing initialization; s3, entering a training round, and continuously updating the value evaluation matrix by dynamically adjusting a greedy criterion, a multi-dimensional reward mechanism and a time sequence difference learning rule; and S4, after the training is completed, starting from the starting point based on the converged Q value matrix, selecting an optimal action through a greedy to generate a final pipeline path, and improving the quality and search efficiency of pipeline laying on the ocean platform.

The application discloses an implementation method and system for self-evolution learning of an intelligent agent, and belongs to the technical field of intelligent agent reinforcement learning, large language models, memory and cognitive intelligence, and transfer learning; the method comprises the following steps: acquiring an environment state; inputting the environment state into a pre-trained strategy to output an optimal action and executing the action; after executing the action, a feedback signal is acquired; the feedback signal comprises an environment response and a task completion degree; and a new strategy is obtained by optimizing the strategy according to the feedback signal. The task performance of the application is continuously upgraded: the intelligent agent can be continuously optimized when facing long-range or repetitive tasks, and the intelligent agent will become more and more skilled in processing cross-platform complex tasks. The application reduces the research and application cost: the intelligent agent can reduce the dependence on manual work, can autonomously discover reinforcement learning rules, does not need to continuously manually annotate data, and relies on environment feedback for iterative optimization.

The application discloses a simulation system supporting fast large-scale brain simulation, and belongs to the technical field of brain simulation.The simulation system comprises a hardware device layer, which provides a plurality of hardware device resources; a data communication layer, which provides a plurality of types of communication modes between the computing nodes; an operation abstraction layer, which provides a plurality of types of neuron models, synapse models, connection rules and learning rules; an API layer, which provides an API interface and receives user requirements through the API interface; a hardware abstraction layer, which calls corresponding hardware kernels; and a network abstraction layer, which firstly records the topological structure of a brain simulation network, performs resource allocation and neuron cluster mapping, then creates the brain simulation network on the hardware device, and finally performs training or execution of the brain simulation network.The application fully utilizes cluster hardware resources to realize faster and larger-scale brain simulation, and solves the problems of lack of resource scheduling and resource allocation, insufficient utilization of device parallelism, too long communication time consumption, and limited support of hardware and interfaces.

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.

A system and method for organizing information in neural networks using taylor series decomposition to create predictable, accessible information storage. A system analyzes training data to identify structural relationships including temporal, semantic, hierarchical, and ontological connections between data elements. These relationships are converted into continuous mathematical functions and decomposed using taylor series expansion to generate positioning coefficients that determine optimal spatial coordinates for each data element within the neural network. A composite learning rule trains the network and maintains spatial organization constraints, balancing prediction accuracy with structural integrity. The system generates a position index mapping data element to specific network layers and node ranges, enabling direct information retrieval without full network activation.

The application discloses a deep learning-based active power distribution network real-time voltage control method, relates to the technical field of power grid intelligent control, and comprises the following steps: collecting multi-source power grid operation data in real time, wherein the power grid operation data refers to bus voltage amplitude and circuit breaker switch state; performing characteristic density wave analysis on the bus voltage amplitude to output a pulse time sequence; extracting voltage deviation characteristics based on actuator feedback data; constructing a pulse neural network, combining a dynamic Hebb learning rule to update network weights; generating a control increment matrix and performing a consonance region safety constraint check; processing the checked control increment matrix through a time window moving average algorithm; and outputting a smooth control instruction to an execution terminal. Through the dynamic calculation mechanism of fusing characteristic density wave analysis and circuit breaker state Shannon entropy, the application realizes accurate perception of power grid topology changes, significantly enhances the overall performance response of voltage control, shortens the response time, and improves the stability of control instructions.

The invention relates to a multi-satellite distributed cooperative task planning method and system based on a multi-agent game, and belongs to the technical field of autonomous task management and control of spacecrafts. The method comprises the following steps: constructing a multi-satellite peer-to-peer network without a primary satellite; task whole network broadcast synchronization is carried out; each satellite generates an initial planning scheme based on self constraints; an inter-satellite communication exchange scheme is adopted; each satellite calculates an optimal self-adjustment scheme and a corresponding regret value under the current other-satellite scheme; selecting a part of satellites to update the scheme based on a preset game learning rule; iteration is carried out until the'regret values' of all satellites are zero, that is, the system reaches Nash equilibrium, and a final collaborative planning scheme is output. According to the method, the game model is established, the global optimization target is decomposed into individual local benefits consistent with the global optimization target, so that the satellite can be self-organized to converge to a high-quality collaborative solution only by means of local information interaction, the dependence on a central node is reduced, and the autonomy, robustness, response speed and expansibility of the system are remarkably improved.