194results about How to "Guaranteed robustness" patented technology

The invention discloses a generative model redrawing attack-oriented image robust watermark traceability method and system and a computer readable storage medium, and belongs to the field of digital information security and artificial intelligence content governance. The method comprises the following steps: in a watermark embedding stage, performing error correction coding and digital signature processing on traceability information containing identity information and a timestamp to generate a load to be embedded, and embedding the load to be embedded and a synchronization template for geometric synchronization into a host image in a function separation manner by using a deep neural network; in the training stage, combined optimization is carried out on the watermark embedding and extracting process by introducing generative model redrawing and microsimulation of image distortion attack, so that the robustness of the watermark under a complex attack condition is improved; in the extraction and verification stage, under the condition that an original image is not needed, geometric synchronous correction and blind extraction of a watermark load are carried out on an image to be analyzed, and authenticity confirmation of traceability information is completed through error correction decoding and digital signature verification. By adopting the technical scheme of the invention, the problems of insufficient traceability information robustness, difficult source confirmation and incomplete evidence chain in a generative model redrawing scene in the prior art are solved, and verifiable traceability and credible evidence generation of the image source information are realized.

The invention provides an NoC low-delay data transmission method based on a dynamic routing algorithm, relates to the technical field of data transmission, and aims to solve the technical problems that an existing NoC routing algorithm cannot adapt to a link dynamic load, is lack of congestion trend prejudgment, is slow in path search convergence and is insufficient in QoS differential scheduling. The method comprises the following steps: constructing a dynamic sensing module containing a double-branch LSTM time sequence prediction model, collecting states such as link bandwidth and queue length in real time, and predicting a congestion trend in 50-100ms in the future; constructing a multi-objective evaluation function taking delay as a core, and dynamically adjusting the weight according to the QoS level; solving an optimal path by adopting an improved ant colony algorithm which introduces a congestion penalty and pruning strategy; during transmission, dynamic reselection is triggered through a pre-calculation + fast matching mechanism; and iteratively optimizing the prediction model based on incremental learning. According to the method, congestion is avoided in advance, the path search efficiency and the transmission reliability are improved, the delay is remarkably reduced compared with a traditional algorithm, and the method is adaptive to delay sensitive scenes such as a high-performance processor and an AI chip.

The invention discloses a photovoltaic power station minute-level power prediction method and system fused with a satellite cloud picture, and belongs to the technical field of data processing and prediction, and the method comprises the steps: obtaining a continuous satellite cloud picture sequence of an area where a target photovoltaic power station is located; obtaining cloud cluster individuals and time-space trajectory data, and extracting time-varying multi-dimensional vital sign parameters of the cloud cluster individuals; inputting the multi-dimensional vital sign parameters into a cloud cluster state evolution prediction model to obtain a cloud cluster state evolution trajectory; obtaining a three-dimensional digital model of the target photovoltaic power station, and generating a super-local irradiance distribution diagram sequence of the target photovoltaic power station based on the state evolution trajectory; and power station electrical characteristic parameters of the target photovoltaic power station are acquired, and a power prediction sequence of the target photovoltaic power station is calculated according to the super-local irradiance distribution diagram sequence. By adopting a means of tracking and predicting multi-dimensional vital sign evolution of cloud cluster individuals and combining a power station three-dimensional model to perform light and shadow simulation calculation, the accuracy of power prediction can be improved.

This invention relates to the field of computational simulation technology for polymer materials, and particularly to a simulation method for evaluating filler dispersion in polymer materials. The method includes: acquiring at least two simulation snapshots generated during the potential energy convergence phase of a molecular dynamics simulation; dividing the simulation box space into three-dimensional grid regions; for each three-dimensional grid region, extracting polymer chain segments with a distance greater than a preset cutoff radius from the filler phase surface as a background reference frame, and calculating their first displacement vector within the inter-frame time interval; calculating the second displacement vector for the filler phase within the inter-frame time interval; calculating a local stability index based on the difference between the first and second displacement vectors; and statistically analyzing the local stability indices of all three-dimensional grid regions to generate a global stability assessment result, determining whether the filler dispersion is in a metastable state. This invention can identify pseudo-steady-state risks in advance during the potential energy convergence phase, and the assessment results are not affected by the filler shape, making it suitable for highly filled systems.

The application provides a sea-rail intermodal port multi-device collaborative scheduling method and system, and belongs to the technical field of port scheduling. The method comprises the following steps: acquiring real-time operation scene information of a railway-yard operation section of a sea-rail intermodal port; constructing a dynamic expansion heterogeneous graph model with new tasks arriving, representing the complex correlation between devices and processes; establishing a discrete event-driven mechanism to trigger scheduling decisions when operation events are completed, synchronizing decisions with operation processes; constructing an event-driven Markov decision process, defining state space, action space, state transition function and reward function; using a heterogeneous graph neural network to deeply encode the state space; and using a proximal policy optimization algorithm to train a global collaborative scheduling strategy based on the state embedding vector and output device-operation allocation actions. The application solves the problems of missing structured information, decision timing mismatch and insufficient generalization ability of existing methods.

This invention relates to a constrained automated optimization method for semiconductor measurement, comprising: step 3) optimizing auxiliary parameters using a training set; step 4) selecting one or more second candidate parameter combinations to form a candidate parameter combination pool, and saving the regression coefficient set corresponding to each second candidate parameter combination; step 5) using the auxiliary parameters in the optimal second candidate parameter combination as independent variables, using the corresponding regression coefficient set as the initial coefficients of the equation, and the key parameters as dependent variables to establish a second multiple linear regression equation; performing the first-stage optimization and the second-stage optimization to select the optimal predicted values ​​and optimal regression coefficients of the key parameters; and step 6) using a test set to perform the first-stage optimization and the second-stage optimization to verify the optimal candidate parameter combination and its corresponding optimal regression coefficients. The two-stage optimization strategy fundamentally solves industry pain points such as difficulty in convergence, difficulty in reproducibility, strong subjectivity, and long processing time.

The application discloses a lightweight indoor 3D laser ground extraction algorithm method, system and medium. The method comprises the following steps: acquiring indoor 3D laser radar data; obtaining coordinate values based on a laser radar coordinate system according to the indoor 3D laser radar data; sending the coordinate values to a preset plane equation to obtain a ground plane normal vector; sending the ground plane normal vector to a first preset coordinate system to establish a constraint to obtain a comprehensive error value; and accumulating squares of values in the comprehensive error value to obtain a comprehensive error sum of squares; extracting a minimum value in the comprehensive error sum of squares, and setting a coordinate point corresponding to the minimum value as an optimal current pose. The application is suitable for various lasers, and by giving a preset threshold, whether it is 16 lines, 32 lines or 128 lines, the algorithm accuracy and robustness are ensured, and the accuracy of recognized ground points is improved.

The invention discloses a method for constructing, updating and retrieving an action memory library, which belongs to the technical field of computer vision and comprises the following steps of: constructing a training data source with time sequence diversity; initializing an action memory library containing a plurality of learnable prototype matrixes and a double-flow interaction network; memory bank evolution is executed, an action prototype is retrieved by utilizing a query stream, a current memory state is dynamically generated by combining a memory state updating gate mechanism with a historical state, and dynamic memory is injected into a feature space by utilizing memory driving graph convolution; synchronously updating parameters based on multi-target loss, and driving a memory bank to evolve into optimal structured prior; and finally, performing structured reasoning on the to-be-detected sequence by using the optimal memory bank. According to the method, structured priori is constructed by mining a spatio-temporal topology mode of a human body action sequence, and hierarchical memory evolution and double-flow depth interaction are combined, so that the problem of depth ambiguity in a monocular vision task is effectively solved, geometric structure distortion is corrected, and the accuracy of action posture estimation is remarkably improved.

The application relates to the technical field of hybrid solid state disks, and discloses a hybrid solid state disk dynamic optimization method and device based on a large language model, which systematically classifies the parameter space of a hybrid solid state disk, constructs a parameter sensitivity matrix based on the classified parameters, determines the influence difference parameters of the parameters on performance under different workloads, collects system state information of the hybrid solid state disk in real time, generates an optimized parameter configuration scheme based on a large language model according to the system state information, the influence difference parameters and historical configuration information. The application can realize real-time perception of work load changes and dynamic focusing on the most sensitive parameter combination under the current load, realize collaborative reasoning and adaptive adjustment among multiple parameters, significantly reduce the write amplification rate and operation delay, and solve the problem that the response efficiency, system stability and energy consumption performance of the hybrid SSD are poor under a dynamic work load in the prior art.

The application relates to the technical field of traffic site selection, and discloses a road network space site selection method and system based on semantic uncertainty, which comprises the following steps: generating Beta distribution shape parameters corresponding to OD flow according to an OD flow data set and a research area POI data set; constructing a hierarchical binary tree decision environment based on city road network topology data, an OD trajectory data set and the Beta distribution shape parameters, and generating flow composition fingerprint features corresponding to each tree node; determining a state space, an action space and a marginal contribution reward mechanism based on sequence priority of a plurality of intelligent agents based on the above processing; initializing double-head Critic network and Actor network of the plurality of intelligent agents, generating experience trajectories through interaction between the intelligent agents and the environment, calculating a risk adjustment advantage function of uncertainty perception, updating network parameters based on Gaussian negative logarithm likelihood loss and PPO truncation agent loss, and outputting a final facility deployment road section set after iterative training until convergence.

This application relates to an event-driven method and computing device for recognizing revenue from intra-group transactions. The method includes: acquiring a transaction order and determining its type; if the transaction order is an intra-group transaction order, acquiring its triggering event; if the triggering event is a key business completion event, generating a cost recognition voucher for the purchaser of the transaction order and a revenue recognition voucher for the seller of the transaction order; wherein the key business completion event represents a business operation event indicating that the relevant performance obligations have been fulfilled. This achieves a fundamental shift from employee-based identification and triggering to system-based perception and response, improving the timeliness and accuracy of financial processing, thereby reducing operating costs and increasing processing efficiency.

The invention relates to the technical field of thermal management of electronic equipment, and discloses a precise temperature control method for components on a circuit board, and the method comprises the steps: executing multi-dimensional state synchronous sampling through a vertical conduction structure of a heat conduction through hole and a shielding shell, and obtaining the shell temperature, the environment temperature and the load current; processing current data by using amplitude domain judgment filtering and calculating instantaneous thermal power, substituting the instantaneous thermal power into a discretized thermal resistance-thermal capacity observer model to calculate the virtual junction temperature of a component to be subjected to temperature control in real time, and calculating the total thermal regulation power by adopting a strategy of combining virtual junction temperature feedback and power feedforward for control logic; and a target driving current is directly solved based on a thermodynamic reverse physical model, and a semiconductor chilling plate or a heating film is driven to work. According to the method, physical conduction lag is compensated through estimation of the observer, and the dynamic response speed and the steady-state control precision aiming at the load step are remarkably improved in cooperation with thermal resistance parameter self-adaptive calibration.

The application particularly relates to a distributed data acquisition instrument channel consistency calibration method and system, and belongs to the technical field of data processing.The method comprises the following steps: synchronously collecting the temperature and humidity and vibration parameters of each node, calculating the environmental difference degree through normalization and weighting, comparing and screening the homogeneous working condition interval with a preset threshold to serve as a calibration candidate set; performing interval mean and standard deviation standardization on the original response amplitude in the candidate set to obtain a normalized response amplitude; measuring the cross-channel response similarity according to the normalized amplitude mean, generating an adaptive calibration weight coefficient through accumulation normalization; combining the weight coefficient with a step factor related to the historical maximum error interval to iteratively correct the response correction value of each node until the drift error meets the accuracy requirement.The application can improve the consistency and robustness of multi-channel high-precision calibration.

The invention relates to the technical field of data transmission, and particularly provides a streaming Token double-path transmission method, system and device and a storage medium, and the method comprises the steps: responding to a streaming request, completing the initialization of a system and a session, building a bypass channel from a model to a client, and creating a session context in a control flow engine; identifying the type of each Token generated by the model, and shunting according to the type: data Token routing to a bypass channel and control Token routing to a control flow path processed by a control flow engine; data bypass transmission and control flow service processing are executed in parallel, and the states of the data bypass transmission and the control flow service processing are ensured to be consistent through a state synchronization mechanism; and the dual-path operation state is continuously monitored, and dynamic path adjustment or abnormity recovery is carried out when abnormity is detected. According to the method, the data flow and the control flow are decoupled, so that the transmission delay of the streaming Token is remarkably reduced while the state consistency is ensured.

The invention provides an electric vehicle power controller parameter adaptive setting method and system, and relates to the technical field of electric vehicle control, and the method comprises the steps: obtaining a motor torque feedback signal, carrying out the wavelet decomposition, extracting a working condition identifier, determining a phase margin and an amplitude margin offset based on Fourier transform, and obtaining a motor torque feedback signal; parameters are adjusted in combination with a gradient projection algorithm and a primal-dual interior point algorithm, and an optimal parameter vector is obtained through a multi-target particle swarm optimization algorithm. According to the invention, on-line adaptive adjustment of the control parameters is realized, and the power performance and the energy efficiency performance of the electric vehicle are improved.

This invention discloses an adaptive inertial navigation solution method and system based on motion feature recognition, comprising: extracting multidimensional features reflecting the motion state of the carrier based on angular velocity and specific force data collected by the inertial measurement unit, and constructing a feature matrix; performing dimensionality reduction processing on the feature matrix to obtain low-dimensional feature vectors; performing cluster analysis based on a clustering algorithm that initializes cluster centers using prior samples to identify the motion state of the carrier; and adaptively selecting an appropriate combined navigation filtering strategy for navigation solution. This invention effectively suppresses the accumulation of inertial navigation errors, improving the autonomous navigation accuracy and long-term stability of the system in scenarios of satellite signal rejection or visual failure; it can finely distinguish different motion modes such as hovering, uniform motion, and transitional states, avoiding the shortcomings of traditional threshold methods that are susceptible to noise interference and have poor adaptability; by introducing continuous frame mode voting and physical anti-jump mechanisms, it further eliminates single-frame misjudgments, ensuring the continuity and robustness of state recognition.

The invention discloses a blood oxygen signal denoising method and system based on adaptive filtering, and the method comprises the steps: synchronously collecting an original blood oxygen signal and a three-axis acceleration signal through a sensor module, and obtaining an original blood oxygen signal sequence and a three-axis acceleration signal sequence which are aligned in time; the three-axis acceleration signals are fused into one-dimensional motion artifact reference noise signals used for representing the limb motion state of the user; the method comprises the following steps: dynamically calculating an adaptive step length factor sequence through a nonlinear mapping function based on multilevel energy analysis of a motion artifact reference noise signal; constructing a time-varying step size adaptive filter, and calculating noise components in the original blood oxygen signal by using the motion artifact reference noise signal and an adaptive step size factor; and subtracting the calculation noise signal from the original blood oxygen signal, and reconstructing to obtain a pure blood oxygen signal without motion artifacts. According to the method, accurate calculation and efficient suppression of motion artifacts can be achieved, and the purification quality and processing efficiency of blood oxygen signals are remarkably improved.

The invention relates to a port storage yard container storage time prediction method and device based on a self-attention mechanism, and the method comprises the steps: calculating the correlation degree between each characteristic variable and the storage time through gray correlation degree analysis; feature variables with high correlation are screened out based on the correlation degree to serve as prediction input, a complex dependency relationship and long program column features among modeling variables of the multi-head self-attention network are introduced, nonlinear mapping and fitting are achieved in combination with a full-connection network, and therefore the accuracy of a prediction result is effectively improved; in the model training process, a loss function with measurement monotonous consistency is designed, a constraint model keeps monotonous sequence consistency of output results on the premise that prediction precision is not lost, and the container turnover and secondary transfer risks caused by prediction sequence deviation in storage yard operation are reduced. And the model prediction performance is comprehensively evaluated from two dimensions of prediction precision and monotone consistency. According to the method, prediction precision and application practicability are both considered, and reliable time information can be provided for storage optimization of the container in the storage yard.

The application discloses a kind of multi-source information mixed combination navigation method under visual assistance, gyro angular velocity and accelerometer linear acceleration information are provided by inertial navigation system;Speed measuring equipment obtains the real-time speed information of mobile object;Global navigation satellite system provides position information and second pulse signal;Visual sensor provides the image information required for navigation;Strapdown inertial navigation algorithm update module carries out navigation solution;Kalman filtering navigation algorithm module estimates the state variable of system in real time;Fault detection and elimination module detects whether the system fails;Image feature detection and tracking module realizes the extraction tracking of image key frame;Sliding window graph optimization algorithm module updates covariance matrix;Loop detection module gives similarity score through image information to judge whether there is loop;Navigation result output module outputs correct navigation information.The problem that existing navigation mode cannot position and orient for a long time is solved.

The invention provides a medical image classification model training method and system, a medical image classification method, equipment and a medium, and the method comprises the steps: carrying out the time-sequence feature extraction of a time sequence of a resting-state functional medical image through an encoder, so as to obtain the multi-scale features of the resting-state functional medical image; performing feature extraction processing on the metabolic network of the functional state medical image to obtain features of the functional state medical image; processing the multi-scale features and the features of the functional state medical image by using an adapter to obtain feature pairs of the resting state functional medical image and the functional metabolism medical image; performing constraint processing on each feature pair by using comparative learning to obtain an aligned feature pair; and training the initial model by using the aligned features, and performing group-level modal distillation processing on the initial model to obtain a medical image classification model. According to the medical image classification model training method, the flexibility and robustness of cross-modal feature fusion can be improved.

The application discloses an internet hospital hierarchical diagnosis and treatment and resource scheduling method and system, the method comprises the following steps: in response to the diagnosis and treatment request sent by the user, obtaining the diagnosis and treatment data containing the patient's complaint information; inputting the diagnosis and treatment data into the disease assessment model for feature extraction, and outputting a feature vector representing the patient's disease state; the feature vector at least includes a first dimension parameter representing the emergency degree of the disease and a second dimension parameter representing the complexity of the disease; based on the feature vector, determine the target resource node in the hierarchical medical resource database; according to the target resource node, generate a resource scheduling instruction, and establish a communication link between the user terminal and the corresponding server of the target resource node. The application introduces a two-dimensional evaluation system of emergency degree and complexity, combines a dynamic threshold adjustment mechanism based on supply and demand prediction, and significantly improves the utilization efficiency of medical resources and the accuracy of diagnosis matching.

The invention discloses a double-side voltage synchronous sampling switch device for loop closing operation of a power distribution network and a use method of the double-side voltage synchronous sampling switch device, and relates to the technical field of power system power distribution network operation control. Extracting a robust baseline and robust dispersion of the annular phase angle difference sequence based on the phase angle; calculating a single-phase peak cluster saliency and a linkage amount by combining a double-side voltage amplitude ratio, and generating a continuous pollution weight; carrying out weighted aggregation to obtain a credible phase angle difference and a credible pressure difference; extrapolating a credible phase angle difference in combination with the frequency difference, constructing a cost function, and solving the optimal switching-on delay; and finally, determining the closing feasibility according to the cost function minimum value and the continuous pollution weight, and generating a closing permission instruction. According to the method, abnormal phase angle fluctuation caused by high-humidity dirt can be effectively identified and inhibited, accidental closed loop locking and failure caused by misjudgment of measurement abrupt change as system asynchronism are avoided, and the safety consistency of closed loop operation of the power distribution network is remarkably improved.

The application discloses a kind of based on low height angle optimization PPP-RTK positioning method and system, the method utilizes the average visible height angle of entire observation period in region as the attribute of satellite, all satellites in region are divided into high angle satellite group and low angle satellite group, in clock difference estimation stage, for high angle satellite group, take conventional estimation strategy, and for low angle satellite group, then take enhanced estimation strategy, improve the clock difference estimation precision of low height angle time, to improve user positioning precision, also can avoid the strategy jitter caused by satellite instantaneous height angle frequent change, guarantee the stability and robustness of clock difference estimation.

The present disclosure provides a DFT resource configuration prediction method and device, electronic equipment and storage medium, belonging to the technical field of electronic design automation. The method comprises the following steps: determining a reference device density based on a process node of a target chip; determining an effective digital logic layout area based on a chip planning area of the target chip and a predicted non-digital logic function area; correcting the effective digital logic layout area according to a utilization factor corresponding to a physical design constraint of the target chip to obtain an available standard cell layout capacity; generating a timing element quantity prediction interval of the target chip through a logic depth mapping model according to the standard cell layout capacity and the reference device density; and generating a design for test (DFT) architecture configuration parameter of the target chip according to the timing element quantity prediction interval. The chip register quantity can be quickly and accurately predicted and the DFT architecture configuration parameter can be generated at an early stage of chip design.

The application provides a multi-round reasoning network traffic prediction method and device based on size model cooperation, which comprises the following steps: inputting historical traffic data into at least two time series prediction models, obtaining prediction results and calculating comprehensive confidence based on prediction errors, and generating a prediction effect ranking; inputting historical data, prediction results and confidence into a large language model, and evaluating and ranking the prediction effect of each model by the large language model; comparing the two rankings, and if they are consistent, calibrating and optimizing the large language model, and if they are inconsistent, adopting an independent adjustment strategy to enhance the sensitivity of the prediction results of the model with the best ranking, and obtaining target prediction results. Through the two-way comparison of the independent ranking of multiple models and the evaluation ranking of the large language model, the large language model can be isolated when the rankings are inconsistent, and an independent adjustment strategy is adopted, thereby effectively avoiding the direct impact of the reasoning deviation of the large language model on the prediction results, and improving the robustness and accuracy of network traffic prediction.

The application discloses a frame structure design method of an AFDM system based on adaptive asymmetric guard bands, solves the problem that the frame structure design of the AFDM system under a high dynamic channel is difficult to simultaneously consider the bit error rate of a receiving end and spectral efficiency, and belongs to the technical field of information and communication.The application comprises the following steps: performing frame-level reliability determination according to a demodulation result vector of a current frame, when a corresponding threshold is satisfied, the current frame accesses an adaptive updating process; on the current frame that accesses, candidate pairs of a time delay index and a Doppler index are obtained, an effective path set of the current frame is screened out, a maximum Doppler spread priori and a Doppler energy directionality priori are extracted from the effective path set, a total guard band width of a next frame and a smooth direction factor are determined, asymmetric distribution of the lengths of guard bands on the left and right sides of a pilot is carried out according to the total guard band width of the next frame, the guard bands are relatively widened on the side of a Doppler energy concentration and are relatively shortened on the side of energy sparseness, and a next frame adaptive asymmetric guard band and a data area layout are generated.

The invention relates to a multiband infrared adversarial sample generation method and device, equipment and a medium. The method comprises the following steps: constructing a three-dimensional model based on target real data, rendering to-be-optimized confrontation textures to the surface of the model by adopting a neural rendering technology, generating a confrontation texture three-dimensional model, processing the confrontation texture three-dimensional model by a differentiable neural renderer to obtain a two-dimensional rendered graph, and converting the two-dimensional rendered graph into grey-scale maps corresponding to short-wave, medium-wave and long-wave infrared rays. A mask is generated through semantic segmentation, a random background is superposed, an optimized sample matched with the real radiation characteristics of each wave band is generated in combination with style migration, and the target detection network feeds back iterative optimization confrontation textures and is finally arranged on the target surface to realize multi-band infrared interference detection. The method can realize full-band effective interference, is strong in robustness and high in concealment, is adaptive to various detection systems, and improves the target safety protection capability.

The present application relates to a kind of multi-unmanned aerial vehicle narrow airspace cooperative path planning method based on distributed optimization, comprising the following steps: establishing the three-dimensional kinematic model of unmanned aerial vehicle, and defining narrow airspace boundary constraint;According to task demand, generate reference path for each unmanned aerial vehicle;Design distributed optimization controller based on local information;With reference path as tracking reference, use distributed optimization controller to solve control input vector in real time, to control the actual flight path of unmanned aerial vehicle.The method of the present application can realize completely distributed control, eliminate single point failure risk, and can ensure that unmanned aerial vehicle is collision-free and maintains desired formation, realizes speed consistency, improves cooperative efficiency.

The application discloses an extensible laser mapping method based on prior topological information and belongs to the technical field of multi-sensor fusion of unmanned vehicles. The method comprises the following steps: S1, multi-sensor data acquisition; S2, data preprocessing; S3, radar odometry; S3.1, IMU pre-integration; S3.2, radar point cloud matching; S3.3, pose fusion; S3.4, map construction; S4, map updating and expansion based on topological information; S4.1, dynamic loading module; S4.2, map association; S4.3, map updating evaluation and map expansion and fusion. The method realizes an incremental expansion, globally consistent and resource-efficient laser mapping system by tightly coupling multi-sensor data, embedding prior topological constraints and implementing dynamic resource management, and provides key mapping technical support for long-term autonomous navigation in a dynamic environment when the operation area is expanded.

The invention provides a subway station peak facility state intelligent optimization method and device, and relates to traffic safety. The method comprises the following steps: collecting historical passenger flow data; constructing a prediction model based on a GRU (Gating Recirculation Unit); optimizing the layer size and the discard ratio of the prediction model by adopting a BKA (Black-Wing Proteurative Optimization Algorithm) to obtain a final prediction model; according to the final prediction model, predicting the passenger waiting times in the target time period, and obtaining the real-time operation data of the subway station based on the predicted passenger waiting times; according to the real-time operation data, establishing a ticket entrance opening time robust model; and improving a genetic algorithm GA by using a Bengalger optimization algorithm SBTO, and solving the ticket entrance opening time robust model by using the improved genetic algorithm GA to obtain an optimal ticket entrance opening time scheme. According to the method, the prediction precision of the model is improved, the operation efficiency of the subway station is improved through robust optimization, and intelligent support is provided for peak management of the subway station.