88 results about "Gradient method" patented technology

The invention discloses a source load storage dynamic strategy verification method based on double-layer reinforcement learning, and the method comprises the following steps: S1, collecting the operation data of a source load storage system, and constructing a standardized operation data set; s2, constructing a double-layer reinforcement learning model, generating a global scheduling strategy by an upper-layer strategy network, and outputting an action decision strategy by a lower-layer strategy network; s3, performing joint training on the double-layer reinforcement learning model by adopting a strategy gradient optimization method, and outputting a scheduling strategy; s4, introducing an integral gradient method to analyze a scheduling strategy, and constructing a key scheduling state node set; s5, optimizing the generation logic of the global scheduling strategy to obtain an optimized double-layer reinforcement learning model; s6, constructing a plurality of source-load-storage system operation scenes to form a typical operation scene set; and S7, deploying the optimized double-layer reinforcement learning model in the typical operation scene set, and outputting a strategy verification result. According to the invention, through combination of double-layer reinforcement learning and an integral gradient method, source-load-storage dynamic strategy verification is realized.

The invention discloses a high-precision static aeroelastic model optimization design method based on a model correction technology, and relates to the technical field of aircraft design, and the method comprises the following steps: S1, firstly constructing an initial model, and carrying out statics pre-analysis to verify integrity; s2, executing SOL 101 statics analysis based on the initial model and outputting a physical field result; s3, carrying out consistency analysis in combination with test data and generating a correction decision; s4, screening high-priority correction parameters through local or global sensitivity analysis; s5, correcting model parameters by adopting a mixed algorithm of a gradient method and an agent model, and verifying precision and generalization ability; s6, the corrected model is output as a Nastran file and a reduced-order model in a standardized mode, and a parameter change log is recorded; s7, executing static aeroelastic coupling and flutter analysis, and feeding back a result to drive optimization iteration; s8, constructing a multidisciplinary coupling optimization model in combination with aeroelastic and flutter results to realize collaborative optimization; and S9, finally performing engineering standardization packaging on the optimization model and outputting a verification report.

The invention relates to the technical field of fault diagnosis, and provides a graph network and knowledge driven interpretable fault diagnosis method and system, and the method comprises the following steps: extracting intra-class and inter-class features for obtained operation data, and constructing a data driving graph; constructing a knowledge graph based on prior knowledge, constructing a knowledge-driven graph, and fusing the data-driven graph and the knowledge-driven graph into a unified graph structure; constructing a plurality of graph attention network sub-models, and identifying and fusing the obtained unified graph structures to obtain a diagnosis result; and performing interpretability analysis on the diagnosis result based on an input feature gradient method and a graph neural network interpreter to obtain a comprehensive diagnosis result. According to the method, more accurate, more stable and more explainable fault diagnosis and decision support is provided through data driving, knowledge driving, intra-class and inter-class feature joint extraction, multi-model fusion and fusion of an explainable analysis method.

The invention relates to the technical field of artificial intelligence, in particular to a class activation mapping interpretability method fusing a spatial perturbation mechanism, which comprises the following steps: a first stage: capturing contribution of each channel to prediction, and obtaining a channel attention weight of a deep learning model through a gradient method; in the second stage, disturbance analysis is introduced, and space importance weights given to image samples by the model are obtained; and a third stage: fusing the interpretation information of the two view angles through weighted integration, and generating an interpretation saliency map of the region-level granularity. According to the invention, the dual positioning capability of an interpretable technology in a channel domain and a space domain can be improved.

The invention provides a coal ash composite material goaf closed filling parameter intelligent matching method, and belongs to the technical field of deep learning and mining engineering crossing. According to the method, dynamic optimization and accurate matching of the filling parameters are realized through combination of data driving and an intelligent algorithm. The method comprises four core links: multi-source data perception and fusion, a material performance prediction model, key parameter identification and boundary constraint, and intelligent matching and optimization decision. According to the method, the modeling capability of the model for the complex coupling relationship is improved through the attention mechanism and the feature cross network; the performance evolution trend of the material under different proportions and process conditions is accurately predicted, key regulation and control parameters are automatically identified, and efficient search and optimal matching of a parameter space are achieved through a strategy gradient method. According to the method, multi-source data can be fused, key parameters can be dynamically identified, and the intelligent matching and optimizing capability is achieved.

The invention provides a structured feedback risk constraint LQR solving method based on optimal control optimization, and belongs to the technical field of control optimization, and the method comprises the steps: obtaining an initial feedback gain matrix meeting a microgrid communication topology sparsity constraint; a zero-order strategy gradient method is adopted, through an outer layer iteration optimization step and an inner layer sampling estimation sub-step, a gradient direction is estimated by utilizing observation change of a target function value; according to the distributed frequency control method, strategy updating is carried out on the basis of an optimal control algorithm, a sparse feedback gain matrix obtained after N rounds of iteration is output, the sparse feedback gain matrix is directly deployed to a local controller of the micro-grid, and distributed frequency control meeting the risk constraint is achieved. Compared with a classical stochastic gradient descent method, the OCP algorithm obtains a satisfactory optimality gap, the convergence speed is higher, and the stability is also improved through dual redescription of maximum and minimum problems.

The invention discloses a load-balanced low-orbit satellite general computation resource collaborative allocation method and system, and belongs to the technical field of low-orbit satellite resource allocation, and the method comprises the steps: building a model with the maximization of the task completion rate as the target, obtaining a computation scheduling sub-model and a flow scheduling sub-model through Lagrange dual decomposition decoupling, and achieving the collaborative allocation of low-orbit satellite general computation resources. The calculation scheduling sub-model determines a calculation decision through linear solution and node preliminary screening, the flow scheduling sub-model is converted into a maximum flow model based on a storage time aggregation graph, a link rate and a calculation frequency distribution decision are obtained through solution, continuous iteration is carried out to convergence by utilizing a gradient method, and an optimal decision of calculation scheduling and flow scheduling is obtained; according to the method, the problem of load imbalance of the low-orbit satellite is solved, space-time joint scheduling of communication and computing resources is realized, network throughput and resource efficiency are improved, congestion and overload are avoided, the task completion rate is remarkably improved, and the performance is superior to that of an existing algorithm.

The invention belongs to the technical field of computers, and particularly relates to a personalized recommendation method and system based on a dynamic heterogeneous graph and reinforcement learning. The method comprises the following steps: firstly, constructing a global heterogeneous information graph of multiple types of nodes offline, and learning static embedding of the nodes by using a graph neural network; secondly, dynamically constructing a session history into a session graph in a real-time interaction process of the user, and aggregating by adopting a graph convolutional network to generate a dynamic state vector of the user; inputting the dynamic state vector into an actor and commentator reinforcement learning framework; and finally, using a dominant function calculated by the commentator network as a stable learning signal, and performing end-to-end joint training on the whole model to optimize long-term cumulative return. According to the method, by introducing the session graph volume accumulation device, the accuracy of dynamic state representation is remarkably improved; and an actor commentator framework is adopted, so that the problem of high variance of a traditional strategy gradient method is effectively solved, and the training stability and efficiency are improved.

The invention relates to the technical field of structure optimization, in particular to a topological optimization method and device for a GPU acceleration fiber reinforced composite material structure, and the method comprises the steps: obtaining parameters such as material attribute parameters and design domain size; 21 basic stiffness matrixes are calculated; respectively filtering the density variables and the angle variables; a geometric multi-grid preprocessing conjugate gradient method iteration solver is adopted to calculate the current displacement of each node of the densest grid level; and calculating an objective function according to the current design variable and the node displacement, performing sensitivity analysis on the objective function and the constraint condition to update the density and angle variables, judging whether a topological optimization iteration convergence condition is met or not, and outputting an optimization result if the topological optimization iteration convergence condition is met. Therefore, the problems that the stiffness matrix of each element in the topological optimization problem of the large-scale fiber reinforced composite material structure cannot be stored due to the limited memory of the global memory of the GPU, and the calculation efficiency of the stiffness matrix of the dynamic calculation element in the finite element solving process is low are solved.

The invention discloses an improved particle swarm optimization algorithm-based unequal-area facility layout method with interference. The method belongs to the field of intelligent optimization algorithms, and comprises the following main steps: S1, constructing a mathematical model with an interference unequal area facility layout problem; s2, initializing a population, and calculating an objective function value; s3, selecting a global optimal particle by using a target space division method; s4, executing heuristic mutation operation by using an adaptive gradient method to improve the search precision; and S5, using a tabu search rule and a neighborhood transformation rule to prevent the algorithm from being caught in a local optimal solution trap too early. Compared with the existing method, the method has the following main advantages: (1) the influence caused by a physical interference object is considered in the unequal-area facility layout problem, so that the applicability of the problem is enhanced, and the method can be more suitable for actual application scenes such as factory workshop transportation scenes; and (2) a tabu search rule is introduced into the multi-target particle swarm optimization algorithm, so that the convergence speed and the global search capability of the algorithm are improved.

The invention relates to the technical field of flood control optimal scheduling, in particular to a reservoir group flood control optimal scheduling method and system based on quadratic programming, and the method comprises the following steps: systematically combing the spatial distribution and hydraulic connection of key nodes according to hydraulic engineering facilities and hydrological elements in a drainage basin, constructing a flood control scheduling topology network, and constructing a flood control scheduling topology network; a reservoir group flood control optimization scheduling topological graph is obtained; establishing a reservoir group flood control optimal scheduling quadratic programming model according to the reservoir group flood control optimal scheduling topological graph; and on the basis of the MINOS library of the GAMS, performing optimization solution on the reservoir group flood control optimization scheduling quadratic programming model by using an approximate gradient method and a quasi-Newton method. According to the method, propagation of the flood among the reservoir groups is coupled into the mathematical equation set, complex simulation sequence coding is avoided, modeling is easy, universality is good, the model is easy to seek an optimal solution, the flood control capacity and the peak clipping effect of the reservoir and reasonable scheduling of the water storage capacity of the reservoir can be balanced, and the flood control precision and efficiency are improved.

The invention discloses a black box directional attack countermeasure method based on a general interaction mode, and belongs to the field of attack countermeasure. According to the method, the PGD-based gradient method is adopted to update the adversarial sample in the directional adversarial attack under the condition of the white box irrelevant to the prompt, so that the adversarial sample meets the l-infinity norm limitation, and the directional adversarial attack PATA under the condition of the white box irrelevant to the prompt is realized. According to the method, a PATA algorithm is constructed to aim at an attacked model in a general interaction mode, the attacked model generates a target mask to achieve interference, transferability of an adversarial sample among different models is achieved by adding a new regularization loss function, and defense and robustness of the model can be pointedly improved and enhanced. According to the method, a single competitive sample is used, but the competitive sample is changed in each iteration, so that the calculation overhead of the attack method is reduced, the black box directional attack countermeasure method with high calculation efficiency can be realized, and the dependence on the sample is lower.

The invention discloses a flux linkage interpolation model adaptive gradient scheme-based switched reluctance motor sensorless control method, which comprises the following steps of S1, designing a flux linkage interpolation model FLIM by representing flux linkage characteristics of rotor misalignment and alignment positions of a switched reluctance motor SRM; s2, designing an adaptive gradient optimization algorithm, and estimating the optimal position of the rotor by updating an interpolation function in real time; and S3, aiming at the phase fault problem, providing a fault-tolerant strategy, namely fault phase flux linkage reconstruction FPFR, and reconstructing the flux linkage at the fault phase. According to the method, FLIM magnetic flux link modeling, adaptive gradient optimization and FPFR fault-tolerant control are combined, and the FLIM model not only significantly reduces the modeling complexity, but also lays a foundation for high-precision observation. According to the adaptive gradient method, real-time dynamic optimization is realized through a magnetic flux link error, and the inherent problems of calculation complexity and response lag of a traditional observer are effectively solved. The FPFR strategy reconstructs the fault phase magnetic flux characteristics, and a feasible solution is provided for high-reliability application.

The invention discloses a power transmission line three-dimensional meteorological field reconstruction method and system considering terrain influence, and the method comprises the steps: constructing a physical constraint projection function according to a terrain normal vector and an actual measurement value of each station in a power transmission line; constructing a computational physical operator system oriented to sparse sites; constructing a variational optimization objective function of each station according to the physical constraint projection function and the computational physical operator system; constructing a POD basis matrix according to historical data, and performing iterative solution by adopting a preprocessing conjugate gradient method according to the POD basis matrix to obtain a modal coefficient vector enabling the variational optimization objective function to be optimal, namely an optimal modal coefficient vector; and reconstructing a three-dimensional field according to the optimal modal coefficient vector, and synchronously calculating a space gradient. Reconstruction accuracy, physical consistency and reliability can be improved, and the requirements for a high-precision three-dimensional meteorological field in multiple aspects such as severe convection weather monitoring and early warning of a power transmission line and weather forecast verification can be met.

This invention discloses an intelligent air combat decision-making method combining imitation learning and reinforcement learning, belonging to the field of air combat. The method includes: processing battlefield situation information through an intelligent air combat decision-making model to obtain decision results for guiding the aircraft. The pre-trained intelligent air combat decision-making model is obtained through the following steps: for coarse-grained sparse expert policy data, a behavior cloning algorithm is used to train a neural network architecture for imitation learning and reinforcement learning to obtain a policy network Q1; the policy network Q1 is used as the initial network in a generative adversarial imitation learning algorithm to perform imitation learning on fine-grained dense expert policies to obtain a policy network Q2; the policy network Q2 is used as the initial network in a reinforcement learning algorithm framework for decision network training, and the policy gradient method is used to train the network until convergence to obtain the intelligent air combat decision-making model. This invention is based on the ability to effectively improve sample utilization and reduce cumulative error.

The invention provides a generation method about human driving data trajectory generalization, comprising the following steps: step 1, acquiring a driving data set, driving trajectory data comprising a state vector; step 2, constructing a space-time decoupling attention mechanism ST-DAM, and decoupling the state vector based on ST-DAM to form a potential feature; step 3, constructing a deep enhanced generation network DRGN which comprises a generator G and a discriminator; the discriminator calculates a global utility score of the prediction trajectory sequence based on the utility function; 4, performing optimization training on the DRGN based on the potential features by adopting a strategy gradient method in reinforcement learning RL; and step 5, inputting the new state vector into the trained deep enhanced generative network DRGN through a space-time decoupling attention mechanism ST-TAM, and obtaining a prediction trajectory. According to the method, the high fidelity of the generated trajectory is ensured, and meanwhile, the method has high strategy explosiveness and generalization ability under global dynamics constraint.

This invention aims to optimize the input sequence of actuators with significant time delays in marine vehicles (such as variable buoyancy hydraulic pumps and valves, and slow-response actuators like rudders). Under complex sea conditions and model uncertainties, this method achieves rapid and stable attainment of the target operational state with minimal action cost, while simultaneously satisfying safety and lifespan constraints. The method disclosed in this invention uses a parameter identification model as the prediction kernel, incorporating terminal task deviation, the number of execution actions and non-zero duration, and the safety volume boundary into a multi-objective cost function. It employs an improved LM algorithm with adaptive damping for sequence-level optimization and rolling updates, thereby overcoming the shortcomings of conventional closed-loop compensation and pure gradient methods in terms of slow convergence, oscillation, and difficulty in balancing safety and wear on time-delayed objects. This significantly improves convergence speed, control accuracy, and reliability, while reducing energy consumption and mechanical wear.

The invention discloses a power system oscillation traceability analysis method and system based on multi-method probability fusion, which can effectively solve the problem of oscillation source positioning error caused by sparse measurement data and compressed sensing. Firstly, the oscillation source probability of each node is calculated through an oscillation energy method, a frequency spectrum characteristic method, a correlation analysis method, a voltage fluctuation method and a phase angle gradient method; secondly, calculating the adaptive weight of each method based on an information entropy theory, and completing multi-method probability fusion; the fusion probability is optimized in combination with a power grid topological structure, and an oscillation source candidate sequence is determined; and finally, a comprehensive probability distribution diagram and a detailed analysis report are output through a visual module, and reliable technical support is provided for safe and stable operation of a power system.

The application relates to the technical field of digital site selection planning, and particularly provides a reservoir resettlement area site selection method and system. The method comprises the following steps: constructing a reservoir resettlement area planning model under multi-target optimization, generating an initial site selection planning scheme; performing topological representation on the initial site selection planning scheme to obtain a current resident gathering point-resettlement area distribution matrix, and constructing a solution space; estimating an initial resettlement area quantity; realizing global search by using a simulated annealing algorithm, avoiding local optimization by probabilistically accepting a poor solution, combining the current resident gathering point-resettlement area distribution matrix and an adaptive multi-step neighborhood generation algorithm meeting the resettlement area quantity constraint, iteratively optimizing and adaptively adjusting a search strategy under the feasibility constraint, and converging to obtain an optimal resettlement area site selection scheme. The application organically combines a policy gradient method in reinforcement learning and a simulated annealing framework, and solves the problem of insufficient intelligence of reservoir resettlement area site selection.

The invention relates to the technical field of power system optimization scheduling, in particular to a multi-microgrid system operation optimization method and system considering distributed energy transaction, and the method comprises the steps: constructing a multi-microgrid system composed of M interconnected microgrids; on the premise that all micro-grids in the multi-micro-grid system interact through energy exchange, a target function with the total running cost minimization of the multi-micro-grid system is built; s2, establishing an internal energy balance constraint and a power constraint of the micro-grid, and establishing an operation optimization model of the multi-micro-grid system in combination with the objective function in the step S2; the method comprises the following steps of: decomposing an operation optimization model of a multi-microgrid system into a plurality of local sub-problems for reducing complexity, finding out a dual problem of the local sub-problems, and then solving the dual problem by adopting an iterative sub-gradient method to find out the minimum operation cost of the multi-microgrid system and an energy value exchanged among microgrids; and controlling the operation of each micro-grid according to the energy value exchanged between the micro-grids. According to the invention, the stability and economical efficiency of the multi-microgrid system are improved.

The invention discloses a power grid power flow analysis method and device based on a quotient gradient method and a neural network, and the method comprises the steps: converting a nonlinear algebraic equation set of flow calculation into an ordinary differential equation based on the quotient gradient method, and constructing a physical information neural network model according to the ordinary differential equation. A system state reference trajectory is generated through an ordinary differential equation solver to serve as a training data set, a composite loss function containing a predicted value and reference trajectory error and a predicted value time derivative and theoretical derivative error is defined, and a training model is iteratively trained. And inputting to-be-calculated working condition parameters and steady-state set time into the trained model, and outputting a steady-state power flow solution through forward propagation. According to the method, the convergence, the physical interpretability and the generalization ability under the ill-conditioned or extreme working condition are improved, and the real-time online analysis requirement is met.

The invention discloses a dynamic equipment layout method based on a quasi-object strategy and a multi-target ant colony optimization algorithm. The method comprises the steps of layout initialization, layout legalization operation, a movement strategy based on reference equipment, configuration optimization operation and Pareto optimal configuration selection based on a maximum and minimum target distance method. N configurations Xl (l = 1, 2,..., n) are randomly generated, each ant l represents one configuration Xl, and initial configurations of the n ants are obtained; performing legalization operation on the configuration X1 by adopting a gradient method based on dynamic step length, executing a movement strategy based on reference equipment to obtain a group of compact and legal configurations, and recording the configurations as a configuration library BL; selecting all non-dominated configurations in the set BL, and storing the non-dominated configurations in an external document CS; and for the current ant l, generating a random number o, and judging the size relationship between the random number o and the parameter p to select an optimization strategy, when o is greater than p, selecting to execute a local search strategy based on improved pseudo-random proportion, otherwise, using a global optimization strategy based on ecological niche.

The invention relates to the technical field of hard capsule polishing, and discloses an automatic polishing system based on hard capsules and a device thereof. A data acquisition module of the system acquires surface characteristic data such as hard capsule surface roughness, stain distribution, capsule size and the like; the data analysis module receives the data and then obtains a polishing demand analysis result through random forest regression analysis; the optimization module adopts a genetic algorithm to optimize parameters and rules of the polishing controller according to the result; the control module carries out fuzzy logic calculation on the polishing amount through particle swarm optimization according to the optimized parameters and rules and converts the polishing amount into polishing control signals; after the execution module receives the signal, the polishing amount constraint condition is adjusted through a near-end gradient method, and integer programming is conducted on the polishing process in combination with a branch and bound method so as to automatically adjust the polishing amount; and the detection module detects surface characteristic data after polishing through Bayesian filtering and feeds back the processed data to the control, data analysis and optimization module so as to adjust a polishing strategy.

The invention provides a PI controller parameter optimization method, a PI controller and a grid-connected inverter, and the method comprises the steps: firstly, determining an allowable range of a current loop bandwidth, taking the value range of controller parameters as a constraint, and taking the minimum overshoot, the highest response speed and the minimum reactive current change as targets; and parameters of the current loop dynamic performance model are optimized according to a simple gradient method. According to the method, high accuracy can be achieved through less test data through the current loop dynamic performance model, the relation between controller parameters and output variables can be accurately reflected, a reliable basis is provided for parameter optimization, the parameter optimization process is more efficient due to application of the simple gradient algorithm, and the method is suitable for large-scale popularization and application. The optimal solution can be found in a short time, the optimized controller can achieve the good decoupling performance of a current loop, and especially when the active current changes greatly, the optimized controller can remarkably reduce the change of the reactive current, so that the dynamic performance of the system is improved.

The invention relates to the technical field of label noise of table data, a distance measurement-based method and a neural network-based method are currently common label noise screening methods, and the methods are difficult to distinguish wrongly labeled samples and difficult samples with fuzzy categories near a decision boundary; according to the table data noise recognition and correction method based on Stein scoring, the difference of the attribution degree between sample and data feature distribution and label export distribution is analyzed, and the logarithmic probability density gradient of a sample is estimated through a diffusion model; moving the sample along a gradient represented by a scoring function by using a gradient method until the sample is converged to a centroid, forming a moving track moving from an original position of a sample space to a high-density centroid of data of a domain, and calculating the data feature distribution by comparing the differences of the directions and lengths of the moving track of the sample in the data feature distribution and the distribution exported by the label. Samples of potential tag errors in the tabular data are identified and tags thereof are corrected.

This application relates to the field of wireless communication technology, providing an antenna robustness design method and apparatus based on hybrid deep learning. This invention simulates manufacturing process errors by combining sampling methods within the range of process errors, obtaining design variables and... S Sensitivity analysis was performed on the Gaussian distribution curves of the statistical mapping relationship between parameter responses to identify variables more sensitive to manufacturing errors, thereby reducing the dimensionality of variables and decreasing the complexity of subsequent antenna robustness optimization. A hybrid deep learning model was used to construct an antenna response substitution model, replacing traditional electromagnetic simulation, significantly shortening the optimization cycle while ensuring the accuracy of response prediction and avoiding distortion of optimization results due to model errors. By constructing a robustness objective function and employing a genetic algorithm to optimize antenna parameter robustness, a highly robust optimal solution was found, avoiding premature convergence of the gradient method in multi-peaked environments caused by random errors.

A generator set control method and apparatus, and a device. Comprising: constructing a state transition model for sub-problems of a single unit, and adding as a state in the model a penalty price corresponding to a Lagrange multiplier for each time period Using a reinforcement learning algorithm to train a startup / shutdown strategy and a power increase / decrease strategy for each unit; using a surrogate sub-gradient method to relax constraints coupled to different units in a UC problem, using the surrogate sub-gradient method to perform iteration and Lagrange multiplier updating, solving sub-problems in the iteration process using a trained reinforcement learning agent to perform sequential decision-making, and iterating repeatedly until convergence, so as to obtain an optimal solution to a dual problem; and performing a feasibility operation on a resulting unit commitment state, and controlling generator set nodes.

The present invention relates to the field of personalized push technology, specifically a personalized music teaching content push system, the system including a user ability assessment module, a teaching content matching module, a learning path planning module, a push strategy optimization module, a behavior analysis and prediction module, a teaching effect feedback module, a dynamic adjustment and optimization module, and a personalized recommendation module. The present invention adopts clustering algorithms, support vector machines, and random forest algorithms to accurately assess learner abilities, K-means clustering and hierarchical analysis method to achieve matching between teaching content and abilities, Dijkstra algorithm and dynamic programming to design personalized learning paths, Q learning and policy gradient method to optimize push strategies, long short-term memory networks and seasonal decomposition time series to predict learning needs, and utilizes item response theory, Markov decision process and Bayesian network to improve feedback adjustment accuracy. Deep Q network and Monte Carlo tree search are used to optimize personalized recommendations and enhance teaching efficiency and results.

The invention provides a graph neural network training method based on depolarization topological attribute consciousness. The method comprises the steps of obtaining a relation network graph used for training a graph neural network; identifying sensitive attribute features by using an integral gradient method according to the set baseline feature matrix and the attribute feature matrix of the relation network diagram; calculating unbiased attribute similarity among the business object nodes according to the identified sensitive attribute characteristics; reconstructing the topological structure of the relation network diagram according to the unbiased attribute similarity between the business object nodes; inputting the reconstructed relation network graph into a graph neural network model, and estimating mutual information of the sensitive attribute characteristics and a prediction result according to the prediction result of the graph neural network model; and constructing a final loss function of the graph neural network model by combining the task loss of the graph neural network model and the mutual information of the sensitive attribute characteristics and the prediction result to update the parameters of the graph neural network model. According to the method, topology disturbance and attribute poisoning attacks can be effectively resisted, the prediction accuracy and stability of the graph neural network in an adversarial environment are guaranteed, and reliable model defense capability is provided for a key service scene.

The present application provides a method and an apparatus for determining the maximum sum rate of a cooperative rate splitting multiple access system. The method includes: optimizing a preliminary optimization vector by using the Newton steepest descent gradient method to obtain a target optimization vector, and optimizing a preliminary sum rate by using the Newton steepest descent gradient method to obtain a target sum rate; when the iteration end condition is not satisfied, replacing the preliminary sum rate with the target sum rate and replacing the preliminary optimization vector with the target optimization vector, where the target difference is the difference between the target sum rate of the current optimization process and the target sum rate of the previous optimization process; when the iteration end condition is satisfied, the obtained target sum rate is the maximum sum rate, and the iteration end condition is that the target difference is less than a preset value and the target optimization vector simultaneously satisfies the sum rate constraint condition, the common rate constraint condition, and the total power constraint condition. This method solves the problem of poor communication quality of long-distance users in the rate splitting system in the prior art.