53 results about "Probabilistic modelling" patented technology

The invention discloses an engine state estimation and system modeling correction method based on double-layer variational inference, which relates to the field of engine state estimation and comprises a variational inference stage aiming at component performance states and kinetic model parameters; a system output prediction stage based on an observation equation; and a solving stage of performing objective function optimization through an evidence lower bound. The structure clearly presents information flow and key calculation links of the proposed algorithm in state estimation and model learning. According to the method, combined reasoning of state variables and model parameters is achieved by building a probability modeling structure, the modeling problem when system dynamics is partially or completely unknown is solved by combining a modeling method of a stochastic differential equation, and while the state variables and the model parameters are optimized, the modeling efficiency is improved. Precise inference of component states and reliable identification of fault features are achieved, the fault detection accuracy of sudden gas circuit abnormity reaches the standard, and meanwhile the performance is better in the aspect of tracking long-term performance degradation.

The invention discloses an articulated coordinate measuring machine error prediction compensation method and system based on a digital twin model. The method comprises the following steps: constructing a real-time synchronous articulated arm digital twinborn model; probability modeling is carried out on a plurality of error sources of the articulated coordinate measuring machine through a Monte Carlo method, and then error statistical characteristics are obtained. And inputting the error statistical characteristics, the joint angle during measurement of the real joint type coordinate measuring machine and the environmental parameters into an error compensation network. And the error compensation network outputs the end position compensation amount of the current coordinate measurement result. According to the method, the digital twinborn model of the articulated coordinate measuring machine is constructed, measurement point distribution is simulated, high-frequency online real-time compensation is realized in cooperation with an error compensation network, the whole process of'measurement-simulation-prediction-compensation 'can be completed through a unified system, and the real-time performance of error compensation is remarkably improved.

The invention discloses a Bayesian optimization-based hydrate reaction condition calculation method. The method comprises the following steps of S1, obtaining an initial data set through experimental measurement and numerical simulation; s2, establishing mathematical modeling of a hydrate reaction process based on chemical reaction engineering knowledge, and generating a simulation model; s3, constructing a proxy model by using a Bayesian optimization algorithm, and performing probability modeling and prediction on the target performance index to obtain performance distribution under a given experimental condition; s4, calculating and recommending optimal experiment parameters according to an expected improvement criterion; s5, performing an experiment or simulation under a recommendation condition, and merging new data into an existing data set; s6, repeating the steps of Bayesian optimization modeling, experiment recommendation and data updating until a predetermined termination condition is reached; and S7, outputting the optimized optimal reaction condition parameters and performance indexes. According to the method, the optimization efficiency and the data utilization rate are improved, multiple targets and complex constraints can be considered, and the scientificity and practicability of industrial process parameter optimization are remarkably enhanced.

This invention provides a probabilistic modeling method and system for characterizing cell capture performance in single-cell immunoblotting (scWB). First, considering the random cell settling process in a polyacrylamide gel microwell array under gravity, a probabilistic distribution model of the number of non-empty microwells is constructed. The evolution of the number of non-empty microwells under different cell loading amounts is analyzed, and the model's calculation results are validated using single-cell capture experimental results. Considering the constraint that trace protein bands in adjacent microwells may overlap, a probabilistic distribution model of the effective number of microwells is established, and a method for calculating the average number of effective microwells is given, revealing the variation of the average number of effective microwells with the spacing between microwell columns under different cell loading amounts. This invention provides a theoretical basis for further optimization of microwell array structures and significantly improves the cell sample utilization rate of scWB technology.

The invention relates to the technical field of machine vision and intelligent perception, in particular to a line laser center positioning method and system based on column-level probability modeling, and the method comprises the steps: obtaining an input image containing line laser, and carrying out the normalization processing of the input image; based on a preset line laser mask image, constructing a column-level supervision label used for representing laser existence and center position offset of each image column, wherein the column-level supervision label comprises a laser existence confidence label and a center offset label; a line laser prediction network is trained by using the column-level supervision label, and the line laser prediction network takes the normalized image as input and outputs the confidence coefficient and the central position offset prediction value of each column of line laser; and inputting an input image into the trained line laser prediction network, outputting the confidence coefficient and the center offset prediction value of each column of line laser, and reconstructing a line laser center track. According to the invention, efficient and controllable prediction of the center position of the line laser can be realized, and the technical requirements of real-time detection and continuous center extraction are met.

This invention discloses a video anomaly detection method based on scene-conditional normalized flow. The method includes the following steps: Step A, processing a continuous video frame sequence to obtain a foreground target sequence; Step B, dividing the foreground target sequence into blocks to obtain a block-level feature sequence; Step C, obtaining complete spatiotemporal structural information features; Step D, obtaining high-level semantic foreground features; Step E, extracting overall scene information corresponding to the foreground target from the original video frames; Step F, performing conditional probability mapping modeling on the scene features; Step G, mapping the foreground target sequence extracted from the video to be detected to the latent space and determining whether the corresponding foreground target or video is an anomalous event. This invention, through probabilistic modeling, effectively avoids the misjudgment problem caused by uniformly modeling the same behavior in different scenes in existing methods. The overall method is simple and can effectively reduce interference from background areas and irrelevant pixel information.

The invention relates to the field of medical data processing, and discloses an imaginary medical data generation method, which comprises the following steps: S1, constructing a dynamic medical knowledge graph containing rare disease knowledge; s2, constructing a virtual patient model, and carrying out probability modeling on a disease development process; s3, using a deep learning algorithm to generate virtual patient data under the guidance of the virtual patient model, and synchronously recording associated knowledge path metadata; s4, performing post-processing on the generated data; s5, providing the data to a medical diagnosis platform; s6, establishing a performance feedback module, and collecting and evaluating performance indexes of the downstream medical diagnosis model; and S7, executing attribution analysis and adaptive optimization. According to the method, the medical logicality of the data is ensured through the strong constraint of the knowledge graph, and the continuous iterative improvement of the data quality according to the application effect is realized through a closed-loop feedback mechanism.

The invention belongs to the technical field of industrial Internet of Things, and discloses a multi-sensor-based intelligent inspection method and system for a yarn twisting machine bearing, and the method comprises the steps: collecting data according to a multi-source sensor installed on a yarn twisting machine inspection robot, and obtaining the state information of the robot; according to the state information of the robot, the robot is driven to move to reach the to-be-detected bearing set of the yarn twisting machine; acquiring temperature distribution and acoustic signals of a to-be-detected bearing pack of each yarn twisting machine through an infrared thermal imaging sensor and an acoustic sensor mounted on a yarn twisting machine inspection robot; and fault diagnosis is carried out on the bearing pack according to the temperature and the acoustic signal. According to the method, a positioning framework of deep fusion depth probability modeling and physical law constraint is adopted, and the problem of high-precision pose estimation in an industrial isomorphic scene is effectively solved through a data-driven feature learning and environment priori knowledge collaborative optimization mechanism; and the practical engineering problem of fault early warning of the yarn twisting machine bearing is solved.

The invention discloses a strategy generation method and system for participation of an iron and steel enterprise in an electric power-carbon coupling market, and belongs to the technical field of industrial energy system optimization, and the method comprises the steps: obtaining multi-source heterogeneous data of the iron and steel enterprise; dividing a plurality of energy consumption units in the iron and steel enterprise into at least one production agent according to the energy consumption and carbon emission characteristics; probabilistic modeling is carried out on the energy consumption and carbon emission of the intelligent agent based on a production plan, and the adjustable potential of the intelligent agent is determined; fusing the market price information and the intelligent agent adjustable potential, and constructing an optimization model with the goal of comprehensive cost minimization and carbon quota income maximization; and solving by adopting a multi-agent reinforcement learning algorithm, and generating an optimal transaction and production adjustment strategy of the iron and steel enterprise participating in the electricity-carbon coupling market. The method solves the problems that a traditional method is difficult to dynamically respond to market fluctuation and cannot collaboratively optimize electric power cost and carbon assets, and provides scientific and efficient collaborative decision support for iron and steel enterprises.

The invention provides a soft measurement modeling method based on a VMD-BiLSTMA-GPR model, and relates to the field of artificial intelligence, and the method comprises the steps: S1, obtaining the direct measurement data of a process variable and the offline detection data of a target variable; s2, performing variational mode decomposition on the data of the target variable to obtain a series of intrinsic mode function subsequences and a residual sequence; s3, adopting a bidirectional long short-term memory network and self-attention mechanism fusion model to train and predict each intrinsic mode function sub-sequence and the residual sequence obtained in the step S2; s4, superposing the prediction results of the sequences obtained in the step S3, and reconstructing to obtain a point prediction value of the target variable; and S5, based on the point prediction value obtained in the step S4, adopting a Gaussian process regression algorithm to carry out probability modeling on the prediction residual error, and outputting an interval prediction result with a specific confidence interval. According to the method, high-precision point prediction and reliable interval prediction of variables difficult to measure in a complex industrial process are realized.

The invention provides a power system low-carbon economic dispatching method based on probabilistic carbon emission flow. The method comprises the following steps: determining input data of a probabilistic carbon emission flow model; constructing a carbon flow database according to the input data; performing random injection and linear regression modeling on the carbon flow database to obtain an updated database; probability carbon emission flow calculation is carried out on the updated database to obtain a probability density function; calculating a dynamic node electricity price based on a probability density function; performing demand response adjustment on the dynamic node electricity price to obtain a response node load; and performing low-carbon economic dispatching optimization on the power system according to the probability density function and the response node load. According to the method, iterative feedback optimization scheduling is carried out on the basis of combination of probability modeling and a price mechanism, and low-carbon economic operation of a full-power system is realized.

The invention discloses a micro-grid operation random optimization method based on source load probability prediction. The method comprises the following steps: firstly, performing energy-sensitive self-organizing segmentation on source-load historical time sequence data, and extracting morphological fingerprint features; and adopting an improved affinity propagation clustering algorithm fused with a power system operation constraint penalty mechanism to identify a typical operation mode. Secondly, a quantile regression model based on a gated pulse neural P system is established in each mode for probability prediction, and a probability scene library with weights is generated; and finally, constructing a two-stage stochastic optimization model with the goal of minimizing the expected operation cost, and solving by adopting a Benders decomposition algorithm to obtain a fixed equipment plan and a flexible operation strategy. According to the method, through refined mode recognition and probability modeling, on the basis of fully considering the uncertainty of the source load, robust optimization of economic operation of the micro-grid is realized, and the expected operation cost of the system is effectively reduced.

The invention provides a civil aviation operation-oriented airspace probability modeling and risk quantification method, which uses ADS-B data to construct layered space-time probability density, adopts an anisotropic kernel which is widened along the course and tightened in the transverse direction, and realizes short-time smoothing by using a time kernel. Inverse detection probability correction is introduced into the sample weight to reduce the coverage deviation. The reference density is subjected to multi-day exponential weighting generation according to hour data so as to describe day and night laws and workday characteristics. Under an information geometry framework, a relative entropy item, a Fisher information item, a capacity penalty item and a conflict kernel item are combined into a unified risk functional, and calculation of density evolution and a congestion dissipation direction is realized by a Wasserstein gradient flow. The system is composed of a data preprocessing module, a detection probability correction module, a layered anisotropic space-time kernel density estimation module, a reference density generation module, a risk functional and gradient flow module and an increment updating and index output module. The technology has unified modeling and clear interpretable and engineering landing characteristics, and can be used for situation analysis and collaborative decision support of air routes, sectors and terminal areas.

A method for probabilistic modeling for intention estimation in an operator-robot teleoperated assembly task is provided. The method may estimate the assembly task to be completed, wherein the assembly task comprises a sequence of actions. The method may predict a next action of the sequence of actions to be performed for the assembly task to be completed.

This invention discloses a stochastic optimization scheduling method for power systems based on LSTM-t distribution, belonging to the field of power system scheduling technology. The method includes: 1) constructing a two-layer LSTM wind power prediction model, obtaining predicted wind power values ​​and residual sequences after data preprocessing and model training; 2) fitting the residual sequences using a t-distribution, determining the distribution parameters through maximum likelihood estimation, and verifying the fitting effect using Q-Q plots and K-S tests; 3) generating initial wind power scenarios based on the t-distribution parameters through Monte Carlo simulation, reducing them to typical scenarios through K-means clustering, constructing a two-stage chance-constrained stochastic programming model with constraints, and solving the scheduling scheme using the Gurobi optimizer. This method accurately characterizes the heavy-tailed nature of wind power prediction errors, achieves deep integration of LSTM prediction and probabilistic modeling, and synergistic optimization of energy storage and thermal power, significantly improving the economy and robustness of power system scheduling.

The invention relates to the technical field of intelligent traffic, and discloses a multi-target path recommendation method considering congestion probability and application thereof, and the method comprises the steps: firstly calculating the congestion probability of a road section based on the probability distribution of a travel time index TT I, constructing a congestion amplification coefficient to correct a classical BPR function, and obtaining more accurate expected travel time; secondly, constructing a generalized road section impedance unified model fusing time value, energy cost, road toll and road type penalty; a toll station delay model and a bypass threshold mechanism are introduced in a path layer to screen effective candidate paths; and finally, taking minimization of total time cost and economic cost as double targets, obtaining a Pareto optimal solution set by using a multi-target genetic algorithm, and calculating comprehensive effectiveness in combination with user preference to perform final recommendation. According to the method, the congestion risk is quantified through probabilistic modeling, dynamic and personalized optimal path recommendation under the multi-dimensional cost constraint is realized, and the accuracy, practicability and user satisfaction of path planning are remarkably improved.

This invention discloses a DDoS attack detection method and system based on VAE-SAC collaboration, belonging to the field of network security detection technology. The invention includes probabilistic modeling of normal network traffic based on VAE, modeling DDoS attack detection as a Markov decision process, and performing feature extraction and VAE state representation calculation on real-time network traffic. This invention uses a variational autoencoder to probabilistically model the high-dimensional statistical features of normal network traffic, mapping the input features to a Gaussian distribution in the latent space, and calculating the normal deviation index based on the reconstruction error and the KL divergence of the latent space. The hybrid reward function, which integrates sparse label rewards and VAE deviation rewards, helps solve the policy optimization problem of reinforcement learning in sparse label environments, improves the stability and sample efficiency of model training, facilitates the synergy between feature extraction and decision optimization, and enhances the detection capability of DDoS attacks.

The invention discloses a volume rendering technology-based interactive three-dimensional visualization reconstruction system for a liver and gall system, which belongs to the technical field of medical image three-dimensional visualization and comprises an organ perception optimization transfer function adaptive optimization module, a multi-level adaptive ray casting rendering module, a virtual anatomy interaction control module and a quantitative measurement labeling module. An organ perception optimization transmission function adaptive optimization module performs probability modeling on CT value distribution of different tissues of the liver and gall system based on a Gaussian mixture model, and automatically generates an optimization transmission function; the multi-level self-adaptive ray casting rendering module adopts GPU acceleration and a self-adaptive sampling strategy to realize efficient volume rendering; the virtual anatomy interaction control module supports plane cutting and transparency adjusting operation; and the quantitative measurement labeling module provides distance, angle and volume measurement functions, and triggers adaptive adjustment of an optimized transmission function through a closed-loop feedback mechanism, so that the display quality and interaction performance of three-dimensional visualization of the liver and gall system are remarkably improved.

The invention discloses an infrared image compression method based on Mama linear recursive scanning, and belongs to the crossing field of computer vision image processing and infrared thermal imaging technologies. According to the method, an end-to-end deep learning framework is adopted, and the method comprises the steps that firstly, global physical parameter locking and normalization preprocessing are carried out; and 2, feature extraction and coding based on a thermodynamic state space module. And 3, carrying out super-prior feature analysis and probability modeling. And 4, entropy parameter prediction based on the bionic pulse decoder. And step 5, performing sub-pixel resolution reconstruction and physical field inversion. According to the method, through a full-process penetration and linear inversion mechanism of global gain and offset factors, based on the linear complexity advantage of the Mama architecture, and in combination with a four-way scanning causal deviation elimination and sub-pixel convolution artifact-free up-sampling strategy, reasoning delay and video memory occupation are remarkably reduced on the basis of ensuring global thermal field topological continuity, and the method has the advantages of high efficiency and high reliability. And a real-time and high-fidelity infrared image compression solution is provided for resource-limited edge end equipment.

The invention provides a dynamic characteristic modeling method and system for a feeding system based on data physical combined drive, and the method comprises the steps: building a parameterized dynamic model of the feeding system based on the change of the linkage characteristic of a dramatic milling force load and a feeding shaft in a machining track curvature sudden change process; and actual operation state data of the machine tool in the machining track curvature sudden change process are collected, probability modeling and correction are carried out on the prediction error of the parameterized kinetic model based on the actual operation state data, and a corrected kinetic model is obtained. According to the method, through data driving correction, the prediction deviation of a pure physical model under multi-source and random errors is effectively compensated, and the accuracy of predicting the dynamic characteristics of the feeding system under the complex curvature sudden change working condition is remarkably improved.

The invention belongs to the field of mechanical arm control safety, and discloses a mechanical arm planning confrontation attack method based on singular point induction, and the method comprises the steps: S1, obtaining a joint angle vector and a tail end pose of a mechanical arm, and generating a thermodynamic diagram based on the joint angle vector; s2, acquiring a singular region based on the thermodynamic diagram, performing multi-target optimization based on the singular region, and acquiring an obstacle position; s3, judging whether the position of the center of the obstacle is located in a reachable area or not, and if yes, moving the entity obstacle to the position of the center of the obstacle; and S4, the mechanical arm is controlled to conduct servo loop execution according to the original production track, and the curve of the minimum singular value and the tail end deviation are recorded. According to the method, singular probability modeling is superposed in an inevitable area of a grabbing path, small obstacles are arranged through a genetic algorithm, and the kinematics margin of taking and placing circulation is weakened specially; degradation can be triggered only by modifying the external environment, clamping jaw control logic or a visual recognition model does not need to be modified, and the compatibility risk of production line software is avoided.

This invention discloses a method for predicting the remaining service life of bearings based on inverse Gaussian process constraints, belonging to the field of mechanical equipment fault prediction and health management technology. This invention collects bearing vibration signals and constructs a temporal convolutional network to extract degradation features. Simultaneously, it introduces an inverse Gaussian process to probabilistically model the degradation increment, constructing a joint physical information loss function that integrates data fitting loss, inverse Gaussian negative log-likelihood loss, and monotonicity constraint loss. Through joint optimization training, the model extracts complex nonlinear degradation features while strictly adhering to the inverse Gaussian distribution characteristics and monotonic degradation physical laws. This invention maintains high accuracy and high stability in remaining service life prediction even with few samples and complex operating conditions, enhancing the physical consistency and reliability of the model. It is applicable to health monitoring and predictive maintenance of bearings in rotating machinery such as hydropower, wind power, and rail transportation.

The application provides a method, device, equipment and storage medium for improving retention through interactive scenarios, which comprises the following steps: obtaining current multi-modal interaction data flow between users in an interactive scenario, performing preprocessing such as noise filtering, frame processing, size normalization and time aggregation, and extracting emotional features to generate a real-time multi-dimensional observation vector sequence. The sequence captures the complementarity and time sequence dependence of multi-modal information and overcomes the limitations of single-modal static analysis. The observation vector sequence is input into a pre-trained hidden Markov model, which is trained based on historical sequences and defines a set of hidden states, initial probabilities, transition matrices and emission probabilities. Through Viterbi algorithm reasoning, the optimal emotional state path at the current time is obtained, realizing the probabilistic modeling and dynamic prediction of emotional transition uncertainty. The current emotional state is extracted from the path to drive the mimic appearance of the virtual pet in the interactive interface, forming intuitive feedback, enhancing emotional connection and improving user retention rate.

The invention discloses an active power distribution network multi-time scale coordinated optimization scheduling method based on MPC, and belongs to the technical field of power distribution network scheduling, and the method comprises the steps: carrying out the source load error prediction through deep reinforcement learning and a Bayesian confidence interval, carrying out the real-time quantification of the source load error into uncertainty intensity, and carrying out the fuzzy logic control according to the uncertainty intensity, according to the method, the complex relevance of historical source load data, meteorological data and a power grid operation state is dynamically learned through a deep reinforcement learning model, probability modeling is carried out on a prediction error in combination with a Bayesian confidence interval, uncertainty is quantized into a quantifiable confidence interval, and the reliability of the prediction error is improved. According to the method, the prediction precision and the uncertainty characterization capability are remarkably improved, the prediction error is mapped into normal distribution, the confidence interval width is calculated, and the uncertainty is further converted into an operable quantitative index.

A system for detecting unauthorized activity in blockchain transactions, consisting of: a data acquisition interface operationally connected to one or more blockchain nodes and configured to receive transaction data including transaction identifiers, address information, transfer values, and timestamps; a preprocessing unit consisting of a normalization circuit configured to transform heterogeneous blockchain data into a standardized internal representation stored in a storage unit; a graph construction processor connected to the storage unit and configured to generate and maintain a directed transaction graph, with nodes representing blockchain entities and edges representing value transfers with associated attributes;a classification processor with parallel computing circuitry configured to process the directed transaction graph using a variety of graph-based learning architectures, and a probabilistic modeling unit configured to compute higher-order dependencies between node features, structural relationships, and temporal attributes; an aggregation unit configured to combine the outputs of the classification processor to generate a risk assessment indicating illegal activity for at least one node or edge of the transaction graph; an optimization processor coupled to the classification processor and configured to iteratively select feature sets and determine weighting parameters to be assigned to the aggregation unit based on a predefined performance target;a perturbation generation processor configured to create modified instances of the directed transaction graph by modifying at least one of the following properties: graph topology, temporal attributes, or transaction values; a robustness assessment unit configured to compute stability metrics of the risk score among the modified instances of the directed transaction graph; and a cryptographic processor consisting of post-quantum cryptography circuits configured to secure the system's data transmission, storage, and audit logs.

This invention relates to the field of electronic digital data processing technology, specifically to a method and system for scoring and predicting risks of cross-chain bridges based on Bayesian networks. The method first addresses the common problems of long-tail distribution and scale inconsistency in cross-chain data through unified discretization, enabling comparison of data from different bridges and chains within the same state space. Then, a static Bayesian network is constructed to probabilistically model multi-dimensional performance indicators and their conditional dependencies, outputting a fine-grained, interpretable posterior probability distribution. Next, entropy-aware scoring is introduced, explicitly incorporating prediction uncertainty into the evaluation process, automatically shrinking the score towards a conservative prior in high-uncertainty scenarios to avoid decision-making risks caused by unreliable predictions. Finally, a dynamic Bayesian network is used to model the logical stage evolution of cross-chain transactions, combined with rolling inference and discounted risk aggregation, to achieve a forward-looking quantification of the temporal propagation and cumulative effects of risk.

The application discloses a kind of brain-like multimodal semantic probability alignment and integration measurement method and system, belong to artificial intelligence and computer technology field, to simulate the multimodal semantic co-occurrence mechanism of brain, improve the explanation and generalization ability of semantic modeling. Including: constructing ecological effective multimodal corpus, based on semantic naming task acquisition semantic response under single mode stimulation, construct standardized semantic probability distribution;In the unified semantic representation space, the semantic labels of different modes are aligned, and the multimodal semantic probability distribution is generated based on probability weighted fusion;Through information theory modeling, the entropy difference between single mode and multimodal semantic probability distribution is calculated, and the quantitative semantic integration degree index is obtained. The system is composed of four modules of semantic response collection, probability modeling, semantic alignment and integration measurement, and supports neural data verification. The application is suitable for multimodal semantic understanding, intelligent decision-making, human-computer interaction and other scenes.

This invention discloses a method and system for geographic mapping information acquisition based on big data, belonging to the field of geographic mapping information acquisition technology. By establishing an environmental disturbance data acquisition module, it actively collects non-topographic meteorological interference factors such as wind speed (Wv), relative humidity (Hr), precipitation particle size (Dp), and air pressure (Pa). Furthermore, it introduces a quantification mechanism for disturbance intensity values, breaking the limitation of traditional mapping systems that "only rely on laser body signals to identify anomalies." This enables the system to perceive and understand natural disturbance backgrounds, maintaining the stability of mapping signal judgment even under strong disturbance weather conditions. Through clustering and probabilistic modeling to identify anomalous samples, a noise identification function that can be used for real-time matching is constructed. This allows the system to make accurate judgments based on pattern similarity when facing nonlinear interference such as signal drift and echo abrupt changes, improving the reliability and intelligence level of noise identification.

The present application relates to the technical field of computer-aided processing, in particular to a kind of probability characterization method of ceramic matrix composite component mechanical property parameter, comprising: establishing target finite element model;Data set is constructed;Neural network model is constructed, and training is carried out;Error function is constructed;Likelihood function is constructed, and probability distribution model is obtained.The present application can provide accurate parameter basis for the uncertainty quantification, performance dispersion analysis and performance prediction of ceramic matrix composite through the analysis and probability modeling of posterior distribution.

This invention discloses an RPS grouping prediction method and apparatus based on deep representation and probabilistic modeling. The method includes: preprocessing an input point set to obtain a first embedding representation for grouping prediction; using a Gaussian mixture model (Gaussian Mixture Model), matching each embedding point in the first embedding representation with the Gaussian components corresponding to each unit in the Gaussian Mixture Model, and determining the predicted unit to which each embedding point belongs based on the matching results; mapping each embedding point to its corresponding supergroup based on its predicted unit to form a preliminary grouping result; constructing a graph structure based on the preliminary grouping result, and using an attribute prediction model to perform neighborhood aggregation on each node in the graph structure to obtain the attribute prediction result for the embedding point corresponding to each node. This invention solves the technical problem of inconsistencies in grouping results caused by the high degree of manual reliance in the RPS point grouping process in existing technologies.