48results about How to "Improve mission performance" patented technology

The invention discloses an unmanned aerial vehicle cluster control and navigation method based on MAPPO. The method comprises the following steps: extracting basic characteristic data of an unmanned aerial vehicle, calculating a comprehensive capability index of the unmanned aerial vehicle through dynamic characteristic analysis and standardization processing, and constructing a cluster isomerism matrix; calculating a communication quality index according to the real-time communication data, predicting link reliability data, constructing a dynamic communication graph, and generating an optimal communication topological structure; based on the cluster isomerism matrix and the optimal communication topological structure, calculating a task allocation fitness matrix, and constructing a hierarchical strategy model and a hierarchical value function; performing multi-dimensional analysis on the communication data, generating an information importance index, calculating transmission priority data, generating a bandwidth allocation matrix, and constructing a self-adaptive routing strategy; and calculating a strategy uncertainty index, constructing a robustness constraint condition, optimizing a strategy model and realizing distributed execution. According to the invention, the cooperative control efficiency and the system reliability of the heterogeneous unmanned aerial vehicle cluster are improved.

The invention relates to the technical field of unmanned aerial vehicles, in particular to an unmanned aerial vehicle task automatic planning and situation display system based on artificial intelligence. Comprising a task demand identification and priority dynamic adjustment module, a multi-task path optimization module, a multi-dimensional situation awareness and data fusion module, a three-dimensional situation display module, a task dynamic adjustment and feedback control module, a task data storage and historical analysis module and an emergency task response and environment interference adaptation module. By collecting task requirements and environment parameters in real time, the system can dynamically generate a task priority queue and optimize path planning based on priorities, it is ensured that the unmanned aerial vehicle has efficient response ability when executing tasks such as reconnaissance, monitoring, searching and rescue, the system combines a multi-source data fusion technology, the three-dimensional situation of the task environment is displayed in real time, and the system has good application prospects. Visual support is provided for decision making; the self-optimization capability and the adaptability to emergency situations lay a foundation for long-term intelligent development of unmanned aerial vehicle tasks.

The invention discloses an unmanned aerial vehicle airport multi-target constraint planning and site selection method and system and electronic equipment. The planning and site selection method comprises the following steps: S1, obtaining geographic information of a target site selection area; s2, determining performance parameters of the unmanned aerial vehicle, and obtaining a flight coverage range of the unmanned aerial vehicle in combination with geographic information; s3, constructing a dynamic task point model, and determining an unmanned aerial vehicle task point set through the dynamic task point model; s4, obtaining an optimization target of airport site selection, establishing a target function according to the optimization target, constructing a site selection optimization model through the target function, and setting constraint conditions for the site selection optimization model; and S5, performing multi-target optimization calculation based on the site selection optimization model by adopting a heuristic algorithm to obtain a target layout of the unmanned aerial vehicle airport. According to the technical scheme disclosed by the invention, the navigation path can be precisely planned, the accurate navigation of the unmanned aerial vehicle in a communication limited or silent environment is realized, and the navigation reliability and the operation capability of the unmanned aerial vehicle are improved.

The invention discloses an intelligent agent control method and device and an intelligent agent, and belongs to the field of intelligent agents. The intelligent agent control method comprises the steps of generating a target task execution plan based on a three-dimensional scene graph corresponding to a target scene and a target scene graph; controlling the intelligent agent based on the target task execution plan; wherein the three-dimensional scene graph is dynamically updated based on a feedback result of the intelligent agent in the process of executing the target task execution planning, and the target scene graph is determined based on the target task. According to the intelligent agent control method, the dynamic target task plan is generated by the three-dimensional scene graph dynamically updated based on the feedback result of the intelligent agent in the process of executing the target task plan, and the intelligent agent is controlled to execute the related task based on the dynamic target task plan. Target task planning is optimized in reverse based on the actual sensing condition of the intelligent agent in the task execution process, the unknown environment is better adapted, and the task execution effect in the unknown environment is improved.

The invention relates to the technical field of mechanical arm structures, and discloses a mechanical arm structure and an inspection robot, the mechanical arm structure comprises a mechanical arm body, the top end of the mechanical arm body is fixedly connected with a rotating block, the rotating block is internally and fixedly connected with a rotating motor, and the output end of the rotating motor is fixedly connected with a fixed block; the side, away from the rotating motor, of the fixing block is fixedly connected with a connecting block, the cross section of the connecting block is triangular, the outer wall of the connecting block is connected with a clamping assembly, a cleaning assembly and a pressing assembly, and a driving assembly is connected into the connecting block. According to the mechanical arm structure and the inspection robot, the clamping assembly, the cleaning assembly and the pressing assembly are integrated, and the rotating motor and the driving system are combined, so that the function diversification and the operation flexibility of the mechanical arm are realized, and the inspection robot can execute various complex inspection tasks according to environmental requirements.

The invention discloses an unmanned aerial vehicle flight path planning method capable of completely avoiding a no-fly zone, which belongs to the field of unmanned aerial vehicle flight path planning and comprises an application scene of the unmanned aerial vehicle flight path planning method in the no-fly zone, and the application scene of the unmanned aerial vehicle flight path planning method in the no-fly zone comprises a communication scene of an unmanned aerial vehicle and a terminal user in the no-fly zone. According to the invention, the unmanned aerial vehicle can be absolutely prevented from entering any no-fly zone in the flight process; a plurality of straight lines are adopted to approximate the actual flight path of the unmanned aerial vehicle, and the distance from the intersection point of two adjacent straight lines to the center point of the no-fly zone is calculated, so that the flight path of the unmanned aerial vehicle is strictly monitored and planned, and the accuracy of flight path planning and the flight efficiency and task execution effect of the unmanned aerial vehicle are improved; according to the method, the constraint of the adjacent time slots on the continuous trajectory is increased, and the safety and robustness of the system are further improved through approximation of multiple straight lines and accurate intersection calculation.

The invention discloses a distributed online adaptive task planning system for an unmanned aerial vehicle group, and relates to the technical field of unmanned aerial vehicle task planning. The system comprises a data acquisition and processing module, a task planning and distribution module, a path planning module, a load configuration module, a real-time monitoring and adjusting module, a cooperative control module, a real-time communication module and an optimization module, and generating a task planning scheme and performing task allocation by applying an online optimization algorithm and a self-adaptive technology. According to the invention, the task planning and distribution module integrates advanced algorithms and technologies to perform task planning and task distribution, so that each unmanned aerial vehicle can bear appropriate task load, an unmanned aerial vehicle / vehicle cluster can autonomously and efficiently complete diversified tasks in a complex environment, intelligent scheduling, path optimization and cooperative operation of the unmanned aerial vehicles / vehicles are realized, and the unmanned aerial vehicle / vehicle cluster is enabled to be more intelligent. And the self-organization capability, the dynamic adaptability and the collaborative operation efficiency of the unmanned aerial vehicle group are improved.

The invention discloses a space-time four-dimensional coordinated guidance method for a dive section of an unpowered high-speed aircraft. The space-time four-dimensional coordinated guidance method comprises the specific steps that S1, a guidance model for dive terminal time and angle control of the aircraft is constructed; s2, diving optimal guidance meeting angle and position constraints; s3, two-stage diving space-time coordination guidance meeting the time constraint is carried out; and S4, carrying out online robustness compensation on the environment and ontology deviation. According to the method, the contradiction between time and space control in a traditional method is solved; the robustness is enhanced by means of a kinetic equation reverse solution technology, and real-time deviation estimation and compensation are achieved; meanwhile, the time control precision is improved by utilizing a two-stage switching logic and numerical integration optimization method based on terminal time prediction; according to the method, the flexibility and adaptability of guidance are improved, the robustness of the system is enhanced, and the strike precision and the overall performance of the aircraft when the aircraft executes complex tasks are remarkably improved.

The invention discloses a vertical take-off and landing control method for a distributed power tandem wing unmanned aerial vehicle, and belongs to the technical field of unmanned aerial vehicle flight control. An inner ring power unit and an outer ring power unit of the unmanned aerial vehicle are arranged at the front edges of the front wing and the rear wing of the tandem wing unmanned aerial vehicle; lifting ailerons are symmetrically hinged to the rear edges of the two ends of the rear wing; the front wing and the rear wing are connected into a tandem wing with a space height difference through a connecting assembly; the control method comprises the steps of controlling jumping take-off, transition flight, cruise flight and vertical landing, and performing accelerator quantity distribution on the inner ring power unit and the outer ring power unit for jumping take-off, cruise flight and vertical landing; a continuous climbing transition mode is adopted in the transition flight stage, and the flight state is controlled to be converted into a cruise level flight state after transition is completed. According to the unmanned aerial vehicle, the functions of large-angle jump take-off and 90-degree autonomous vertical landing of the tandem wing unmanned aerial vehicle are achieved, distributed power of the unmanned aerial vehicle is controlled and distributed, the energy utilization rate is increased, and the endurance time is prolonged.

The invention provides an unmanned aerial vehicle data transmission system comprising a plurality of fixed signal stations used for sending and receiving wireless signals so as to establish communication connection with an unmanned aerial vehicle, and each fixed signal station is in wired communication connection with at least one other fixed signal station so as to form an ad hoc network data link; and a plurality of fixed control stations, each fixed control station is connected to the ad hoc network data chain, and the fixed control stations send control instructions to the unmanned aerial vehicle through the fixed control stations of the ad hoc network data chain and receive sensing data returned by the unmanned aerial vehicle. According to the method, the data link coverage range of the unmanned aerial vehicle is expanded, comprehensive signal coverage can be realized in various terrains and environments theoretically, the task execution capability and the working flexibility of the unmanned aerial vehicle are remarkably improved, and meanwhile, the efficiency loss caused by limitation of traditional point-to-point communication is reduced.

The invention relates to the technical field of power failure prevention, in particular to a power failure prevention method, system and device for an anti-radiation emergency armored car and a medium. According to the power failure prevention method based on real-time monitoring and intelligent management provided by the invention, the accuracy and timeliness of power supply state monitoring are improved by simultaneously monitoring power supply parameters in two dimensions of voltage and current; a time point recording mechanism is introduced, and a power-off control strategy based on the duration for the flameout state of the engine is adopted, so that the problem of abnormal power supply can be found in time, and the continuity and reliability of a nuclear radiation detection task are effectively guaranteed.

The invention relates to the technical field of artificial intelligence, in particular to an efficiency evaluation method and device for intelligent operation and maintenance of a cluster system. According to the method, the efficiency of the cluster system in different operation stages can be scientifically and accurately evaluated by constructing the reasonable efficiency index system, the network average efficiency calculation model, the task execution balance degree evaluation model and the elasticity evaluation model. According to the method, the deep reinforcement learning algorithm is introduced, the multi-dimensional situation information is collected, analyzed and predicted in real time, and the intelligent operation and maintenance decision-making efficiency of the cluster system can be effectively improved.

The present invention discloses a MAPPO-based UAV cluster control and navigation method, including extracting basic characteristic data of UAVs, analyzing dynamic characteristics and standardizing them, calculating comprehensive capability index of UAVs, and constructing cluster heterogeneity matrix; calculating communication quality index according to real-time communication data, predicting link reliability data, constructing dynamic communication graph, and generating optimal communication topology structure; calculating task allocation fitness matrix based on cluster heterogeneity matrix and optimal communication topology structure, constructing hierarchical strategy model and hierarchical value function; performing multi-dimensional analysis on communication data, generating information importance index, calculating transmission priority data, generating bandwidth allocation matrix, and constructing adaptive routing strategy; calculating strategy uncertainty index, constructing robustness constraint condition, optimizing strategy model and realizing distributed execution. The present invention improves the collaborative control efficiency and system reliability of heterogeneous UAV clusters.

The invention discloses a space plasma energy collection device and method based on spacecraft surface potential difference, and belongs to the technical field of spacecraft energy supply. According to the invention, each sensor probe is connected with each monitoring point and is used for monitoring the potential of the monitoring point; the input end of the processor is connected with the output end of the sensor probe; the output end of the processor is connected with the first input end of the voltage regulator; when the potential of a certain monitoring point reaches a discharge threshold value, the monitoring point is conducted to work, and the storage battery is charged by utilizing the potential difference between the potential of the monitoring point and the potential before the grounding reference point; the monitoring point and the grounding reference point are connected with the input end of the phase inverter; and the output end of the phase inverter is connected with the second input end of the voltage regulator and the storage battery. Accumulation of charges on the surface of a spacecraft can be reduced, the discharge frequency of the plasma contactor is reduced, and the structure is compact.

The invention discloses a method for planning a trajectory of a UAV that completely avoids a no-fly zone, belongs to the field of UAV trajectory planning, and includes application scenarios of the method for planning a trajectory of a UAV in a no-fly zone, wherein the application scenarios of the method for planning a trajectory of a UAV in a no-fly zone include communication scenarios between the UAV and a terminal user in a no-fly zone. The invention supports the UAV to absolutely avoid entering any no-fly zone during flight. Multiple straight lines are used to approximate the actual flight trajectory of the UAV, and the distance between the intersection of two adjacent straight lines and the center point of the no-fly zone is calculated to achieve strict monitoring and planning of the flight path of the UAV, thereby improving the accuracy of trajectory planning as well as the flight efficiency and task execution effect of the UAV. The invention increases the constraints on the continuous trajectory between adjacent time slots, and further improves the safety and robustness of the system through multiple straight line approximation and accurate intersection calculation.

The invention provides a task execution method, device and system based on a plurality of container arrangement clusters, and relates to the computer technology, the method comprises the following steps: obtaining a plurality of task execution requests; each task execution request comprises a mirror image file, indication information and a to-be-run instruction; classifying the to-be-run instruction based on the mirror image file and the indication information; selecting a to-be-run instruction of a target category, and selecting a target container arrangement cluster and a target working node for the to-be-run instruction of the target category; sending a mirror image file and indication information corresponding to the to-be-run instruction of the target category to a target container arrangement cluster, so that the target container arrangement cluster creates a target execution unit on a target working node based on the mirror image file and the indication information; and injecting the to-be-run instruction of the target category into the execution unit to enable the execution unit to run the to-be-run instruction of the category. Therefore, the task execution performance of the task execution system is improved, and the task execution efficiency and reliability are improved.

The application relates to a laser radar SLAM and vegetation feature library matching method and system of a mowing robot, and relates to the technical field of data processing. The method comprises the following steps: determining initial laser radar parameters based on laser radar test data of a plurality of sample vegetation; establishing a vegetation feature library based on the initial laser radar parameters and the laser radar test data of the plurality of sample vegetation; determining environmental influence factors based on the initial laser radar parameters; acquiring real-time environmental information of the mowing robot; adjusting the initial laser radar parameters based on the environmental influence factors and the real-time environmental information to generate current laser radar parameters; acquiring point clouds of vegetation to be matched based on the current laser radar parameters; determining a vegetation type based on the point clouds of the vegetation to be matched and the vegetation feature library; and updating a two-dimensional grid map or a three-dimensional point cloud map of an environment where the mowing robot is located based on the point clouds of the vegetation to be matched and the vegetation type, so that the intelligent level of the mowing robot is improved.

The present invention relates to the field of surveying and mapping technology, and specifically to a deep learning-based unmanned aerial vehicle (UAV) surveying and mapping system; the system comprises: an UAV surveying and mapping terminal, wherein the UAV surveying and mapping terminal has a built-in demand analysis module and a surveying and mapping optimization module; by real-time monitoring and comprehensive analysis of the data volume and transmission performance in the UAV surveying and mapping process, optimization instructions can be automatically generated, thereby improving the stability of data acquisition and the performance of network transmission, and providing effective support for subsequent surveying and mapping optimization; by reasonable priority sorting and transmission strategy, the efficiency and reliability of data transmission are improved, and the utilization of system resources is optimized; by quantitatively analyzing the degree of repeated surveying and mapping of each section in the flight path to determine the optimal degree of repeated surveying and mapping of each section, and optimizing each section accordingly; the flight path of the UAV is dynamically adjusted to avoid repeated surveying and mapping to the greatest extent, reduce repeated data transmission, and improve the effective collection and effective transmission of surveying and mapping data.

The invention discloses a robot joint module and a networked control system. The robot joint module comprises an actuating unit, a control unit and a relay unit, the control unit comprises a calculation module, a power driving module, a signal acquisition module, an I / O module, a power supply module, a wired communication module and a wireless communication module; the relay unit comprises a communication bridge module and a driving relay module. According to the invention, the problem of realizing system capability recovery through reconstruction of electrical and control networks when communication and driving faults occur is solved, and the working reliability of the robot system is improved.

The application belongs to the technical field of artificial intelligence and relates to a role large model personality and task linkage ability enhancement method and related products. According to user input natural language task instructions and personality setting information, a role large model generates a subtask sequence and an ability requirement set. For each subtask, if the subtask and the corresponding target ability match the personality setting information, the role large model calls the target ability to execute the subtask to obtain a subtask result. If the subtask and the target ability do not match, the subtask and the target ability are subjected to personality adaptation reconstruction before execution to obtain a subtask result. According to the personality setting information, the subtask result is subjected to personality polishing to obtain a personalized subtask result. The personalized subtask results are integrated in sequence into a complete task result, which is subjected to personality polishing again to obtain a personalized complete task result that is output to the user. The application can improve the task execution ability of the role large model while maintaining personality consistency.

The robot dog with the multiple mechanical arms comprises a robot dog body, a multi-mode moving module is integrated at the bottom of the robot dog body, a mechanical arm butt joint rail is arranged on the upper side of the edge of the robot dog body, and at least two standardized power connection butt joint bases are arranged on the mechanical arm butt joint rail. A mechanical arm connecting base is fixedly connected to the front face of the power connection butt joint base in an inserted mode, an operation mechanical arm module is arranged on the front side of the mechanical arm connecting base, a modularized tail end claw is detachably installed at the tail end of the operation mechanical arm module, and the operation mechanical arm module and the modularized tail end claw are driven by a servo motor or hydraulically to operate. And through the arranged mechanical arm butt joint rail and the power connection butt joint bases, different numbers of power connection butt joint bases can be installed according to task requirements, and the number of the operation mechanical arm modules is flexibly adjusted so as to adapt to the task requirements of different scenes. Multi-task synchronous execution is realized, and the rescue efficiency is greatly improved.

The invention provides a computing resource dynamic scheduling method and device, equipment and a medium. The method comprises the steps of obtaining a task feature vector of at least one target model training task; obtaining a computing device computing resource vector provided by a resource awareness twin network, wherein the resource awareness twin network is constructed based on computing capabilities of various computing devices; obtaining a candidate computing device set according to the similarity between the task feature vector and the computing device computing resource vector; determining a task allocation scheme for allocating the at least one target model training task to computing equipment in the candidate computing equipment set according to the completion time of the at least one target model training task and target computing resources of the candidate computing equipment set; and dynamically scheduling the at least one target model training task according to the task allocation scheme. According to the method, the overall utilization efficiency and the task execution efficiency of heterogeneous computing power resources are improved.

The present application discloses a method and device for unmanned aerial vehicle navigation, including: when the confidence of GNSS posture is in a high confidence interval, the GNSS posture at this time is used for positioning flight, and the visual map construction step is performed based on the posture of the image frame; when the confidence of GNSS posture is in a medium confidence interval, the credibility of the current GNSS posture is judged by the constructed visual map, if it is credible, the visual map is updated based on the GNSS posture to obtain the updated visual map, if it is untrustworthy, the GNSS posture is discarded, and the positioning flight is performed based on the posture of the current image frame obtained by the visual map; when the confidence of GNSS posture is in a failure interval, the visual map is used to match the feature points in the current image frame and the visual map to estimate the current posture of the unmanned aerial vehicle. The problem of how to achieve positioning and navigation by combining GNSS and visual information when the unmanned aerial vehicle cannot receive or trust GNSS signals is solved.

The present invention discloses a task allocation conflict detection method and device based on data analysis, the method comprising: obtaining a task allocation information set; the task allocation information set comprising: a task deployment table, task requirement information, human resource information, task rule information and work resource information; initializing the task allocation information set to obtain an initialized task information set; performing task allocation conflict detection processing on the initialized task information to obtain a task allocation conflict information set.

The present invention relates to a UAV visual navigation method and device based on deep learning, including: obtaining a visual image collected by a UAV visual system, extracting features from the visual image, and obtaining an image feature representation; performing type analysis on the image feature representation to obtain scene data, depth information, and continuous image frames; performing scene recognition on the scene data to obtain recognition results of obstacles and passable areas; performing optical flow estimation on the continuous image frames and the depth information through a preset optical flow algorithm to obtain a pose estimation result; obtaining the mission requirements of the UAV to obtain a navigation control strategy with environmental adaptability; optimizing the adaptive hybrid learning model based on the feedback information combined with a preset attention mechanism and reward shaping conditions to obtain the optimized adaptive hybrid learning model. The present invention can improve the adaptability of UAVs in complex and dynamic environments.

The embodiment of the invention provides a data acquisition method and device, computing equipment and a computer readable storage medium, and the method comprises the steps: responding to a claim settlement request, and determining a candidate mechanism identifier in a plurality of reference mechanism identifiers based on the similarity with a to-be-analyzed mechanism identifier carried by the claim settlement request; through a text reasoning model, based on the to-be-analyzed mechanism identifier and the candidate mechanism identifier, determining a target mechanism identifier representing the same mechanism with the to-be-analyzed mechanism identifier; and mechanism data corresponding to the target mechanism identifier is acquired, and the mechanism data is used for executing a task corresponding to the claim settlement request. The method can improve the processing effect of the task corresponding to the claim settlement request.

The invention provides a robot control method and device and a robot, and is applied to the technical field of robots. Performing target detection on the visual image to obtain a candidate image area, and screening masks in the candidate image area; determining three-dimensional observation information in the candidate image area based on the screened mask, and aligning the three-dimensional observation information with a CAD model matched with the target object to obtain pose information of the target object; according to the pose information and the identifier of the CAD model, determining a hand grabbing pose of the robot from a preset hand pose database; according to the method, the robot is subjected to hand posture control according to the hand grabbing posture, object posture estimation is conducted by combining the three-dimensional observation information and the CAD model, then the hand grabbing posture is determined, hand posture control is conducted on the robot, the precision and efficiency of the robot for executing body grabbing are improved, and then the task execution effect is improved.

The invention discloses a heterogeneous sensing resource runtime selection and efficiency updating method based on data multiplexing. The method comprises the following steps: registering available sensing resource equipment and a corresponding inference model and evaluation model to different task partition directories of master control computing equipment; initializing parameters; calculating and updating efficiency values of the sensing resources and the reasoning model; selecting the inference model and the sensing resource with the highest efficiency value under the corresponding partition according to the task request category to execute the task; dynamically adjusting the time period and the efficiency threshold value of the selected task according to whether the execution efficiency of the perceptual resource task is lower than the selection efficiency threshold value or not; dynamically adjusting a time period and a threshold value of efficiency updating; fault sensing resources are quickly removed, and newly-added sensing resources are accessed. According to the method, through data multiplexing, data from the same sensing resource supports different types of tasks, the utilization efficiency of the sensing resource and the data is improved, and energy consumption caused by sensing resource selection and efficiency updating is reduced.

The invention discloses a method for calibrating an optical axis and an inertial navigation axis of an optical system in an airborne environment, and belongs to the technical field of optical system calibration. The method comprises the following steps: designing an inertial navigation axis calibration device, wherein the inertial navigation axis calibration device comprises a point light source, a light emitting / returning mirror group and an image sensor; a first test light path between the inertial navigation axis and the inertial navigation axis calibration device is constructed, a point light source emergent light part of the inertial navigation axis calibration device is emitted to inertial navigation through a light emitting / returning mirror group and then is returned to an image sensor, and a first target image is obtained; a second test light path between the tested optical system and the inertial navigation calibration device is constructed, part of emergent light of a point light source of the inertial navigation calibration device is emitted to the tested optical system through the light emitting / returning lens group, and a second target image is obtained through a sensor of the tested optical system; and comparing the first target image with the second target image to obtain the error between the optical axis of the measured optical system and the inertial navigation axis. According to the invention, the problem that real-time detection cannot be carried out in an airborne environment is solved.

This invention relates to the field of computing power scheduling technology, specifically to an AI computing power scheduling method and system for private large-scale models. The invention performs attention encoding on multi-source heterogeneous data to obtain format recognition results and acquisition priority labels; generates a memory pressure pre-transmission signal based on the format recognition results; extracts structural features from high-quality data streams and generates computing power affinity encoding vectors through graph neural networks; constructs a state space combined with a secure computing power pre-occupancy mechanism, outputting a resource allocation action sequence; further performs memory oversubscription, core binding, NUMA node crossing, and network bandwidth conflict detection, and resolves conflicts and performs affinity rearrangement based on task priorities to generate a conflict-free instruction sequence. This invention can improve the utilization efficiency of heterogeneous computing power resources, reduce the probability of resource conflicts, and improve the scheduling stability and execution efficiency during the training and inference process of private large-scale models.