1042 results about "Target distribution" patented technology

The invention provides an intelligent power distribution load prediction and adaptive scheduling method, and aims to improve the response capability and operation efficiency of a power distribution system to load fluctuation. According to the method, historical load data, real-time meteorological information and user power consumption behavior data of a target power distribution area are collected, a multi-dimensional input data set is established, a short-term load prediction model is constructed based on the data set, and prediction of future load levels of multiple feeder lines and key nodes is achieved. According to a prediction result, potential voltage deviation, load imbalance and equipment overload risks are identified, and a dynamic scheduling control strategy is generated by adopting a rolling horizon optimization method in combination with a power grid topological structure and operation constraints; the strategy comprises control measures such as transformer tap position adjustment and capacitor bank switching; through issuing a scheduling strategy to an equipment control unit, active adjustment is completed before load change, voltage stabilization, load balancing and energy optimization are realized, and intelligent and adaptive operation guarantee is provided for a power distribution system.

The invention discloses an intelligent target allocation method and system based on deep reinforcement learning, particularly relates to the field of multi-agent systems, and is used for solving the real-time decision problem under multi-dimensional constraints in a dynamic confrontation simulation scene. Time, space and resource constraints are decomposed into independent models, and the decision complexity of a multi-agent system in a dynamic confrontation scene is reduced; by means of adaptive Monte Carlo sampling, prediction perception path planning and dynamic threshold value resource allocation, diversified high-quality candidate schemes are quickly generated; when the environment changes, the multi-layer influence propagation network and the spatial topological relation graph are used for positioning an affected agent, and scheme fragments are generated through local adjustment, so that global re-calculation is avoided, the overhead is reduced, and real-time response is ensured; meanwhile, an efficient and reliable optimal solution is screened out through resource delay coupling and path efficiency comprehensive evaluation in combination with multi-objective optimization and task consistency verification.

The invention provides an intelligent distribution decision-making method and system for production materials, and the method comprises the steps: firstly obtaining the material demand information, including material types, demand priorities and demand time windows, of all production nodes in a preset period of a target factory, and then combining the real-time traffic and storage layout data of an area where the target factory is located; the method comprises the following steps: determining dynamic environment parameters including a road traffic state and a storage node capacity, then generating an initial distribution path set according to a demand priority and a time window, inputting the initial distribution path set and the dynamic environment parameters into a pre-trained path optimization model after each path is associated with a transportation tool and a material distribution scheme, and outputting a target delivery path and a dynamic scheduling instruction used for adjusting the loading capacity and the driving speed of the transportation tool, and finally controlling the transportation tool to deliver materials according to the target delivery path and the dynamic scheduling instruction, and re-planning the path on the way based on the real-time dynamic environment parameters to realize intelligent and efficient delivery.

The invention provides a power distribution network state evaluation method and system based on artificial intelligence. The method comprises the following steps: acquiring operation state data of each power distribution network device in real time; performing feature extraction on the operation state data of the power distribution network equipment according to a preset operation state feature extraction model to obtain equipment multimode operation features; according to a particle swarm optimization algorithm and an adaptive gradient information strategy optimization algorithm, carrying out weighted fusion on the equipment multimode operation characteristics of the power distribution network equipment to obtain equipment state fusion characteristics; and constructing a power distribution network state topological graph based on the target power distribution network topology according to the equipment state fusion features of the power distribution network equipment, and analyzing the power distribution network state topological graph according to a preset graph neural network model to obtain a power distribution network operation state evaluation result. According to the invention, based on the self-adaptive dynamic fusion analysis of the multi-source data characteristics of the power distribution network equipment in combination with the topological correlation analysis between the equipment, the state evaluation comprehensiveness and accuracy of the power distribution network are effectively improved.

The invention relates to the technical field of path planning, in particular to a path planning method and system for an inspection robot, and provides the following scheme: executing an inspection task through an initial electronic map, collecting first image data and pose information, recognizing an inspection target, completing image registration, and generating second image data; inverting and calibrating the optimal observation direction based on the structural sensitivity distribution of the local visual features and the illumination variation trend, and determining the shooting position and angle to form a path node; a multi-objective optimization model is adopted, the path length, the view angle continuity and the illumination interference are comprehensively considered, and an optimal inspection path is generated; the method is suitable for outdoor inspection scenes with complex structures and non-uniform target distribution.

The present application discloses a data management method, apparatus and system for a distributed storage system, and an electronic device. The method comprises: traversing all storage paths in a target distributed storage cluster, and acquiring file storage information of the storage paths during traversal; on the basis of the file storage information, calculating a redundancy policy score of each storage path, and on the basis of the redundancy policy score and a first redundancy policy score threshold, determining a storage policy adapted to the storage path; when it is determined that the storage policy adapted to the storage path is an erasure code storage policy, determining a data recovery mode of the storage path on the basis of the architecture information of the storage path; and when it is determined that the storage policy adapted to the storage path is a multi-copy storage policy, determining the number of copies of the storage path on the basis of a service importance degree of the storage path and the redundancy policy score.

The invention discloses a laser cutting parameter adjusting system based on machine learning, and relates to the technical field of laser cutting. According to the method, multiple laser cutting parameters in the laser cutting process are collected, a short-term state index and a long-term state index are calculated based on the cutting parameters, and a laser cutting instant index and a laser cutting long-term index are generated respectively; based on the laser cutting instant index and the laser cutting long-term index, optimized laser cutting parameters are generated, and working parameters of the laser cutting machine are adjusted in real time; the adjustment result and the related data are sent back to the data storage module for updating the historical database; a real-time distribution curve is generated and compared with a target distribution curve, deviation is calculated, and a parameter optimization target curve is generated; and according to the optimized sub-targets, laser cutting parameters are adjusted in real time, and the stability and high efficiency of the laser cutting process are ensured. According to the system, accurate control over the laser cutting path is achieved through multi-layer and non-uniform control unit division and time-space correlation model construction.

The invention provides an information resource distribution method and system applied to a dynamic network environment, and the method comprises the steps: firstly obtaining a real-time state feature set of a target distribution network, determining a resource distribution constraint condition set in a current network environment based on the real-time state feature set, carrying out the priority analysis of a real-time resource request sequence of a target user group, and obtaining a real-time resource distribution constraint condition set; obtaining resource timeliness grade distribution and a request path optimization feature set, generating a dynamic distribution strategy parameter set matched with a resource distribution constraint condition, triggering a resource redirection operation, and enabling the resource request response rate of the target user group to be improved to a preset threshold interval; and the information resource distribution efficiency and stability in the dynamic network are improved.

The invention discloses a UUV cluster dynamic task planning method based on a multi-target genetic algorithm, a program, equipment and a storage medium, and belongs to the field of underwater multi-UUV cooperative detection of multiple targets. According to the method, firstly, for path planning of regional task points, chromosome representation is completed through sequential coding, chromosomes are selected and ranked randomly, and the chromosome with the highest fitness value is selected from each group; then, in each group of the current population, selecting an optimal parent individual to carry out crossover and mutation operations so as to improve the quality of offspring individuals, and carrying out updating to obtain a next-generation population; and finally, obtaining a plurality of optimal individuals, and outputting a task planning path result. Dividing the search area into a plurality of task sub-areas, and completing the dynamic task planning of the UUV cluster according to the target distribution condition of the sub-areas and the condition of each UUV. According to the method, the search strategy can be adjusted in real time according to the real-time detection result and the environment change, so that the target distribution non-uniformity and the cluster efficiency difference are effectively relieved.

The invention discloses an unmanned aerial vehicle inspection control system based on binocular vision, and relates to the technical field of unmanned aerial vehicle inspection, a binocular vision acquisition module is used for acquiring a to-be-inspected target image, detecting shielding and degree, and extracting a target feature and a distribution rule when an unmanned aerial vehicle inspects according to a preset route; the information acquisition module is used for acquiring a target position, peripheral complementary target distribution, unmanned aerial vehicle pose and kinematics parameters when it is detected that the shielding degree of the target point exceeds a set value. Complete coverage of the two types of target sub-regions can be ensured, the flight trajectory curvature of the second type of target sub-regions can be minimum, and stable flight is ensured.

The invention provides a distribution agent ad hoc network method and system based on dynamic node collaboration, and the method comprises the steps: firstly obtaining a network state data set of all distribution nodes in a target network region, including real-time communication load, link stability and resource occupation characteristics, generating dynamic collaboration topology information based on the network state data set; comprising a node cooperation relation topological graph and a communication path weight table, a distribution agent strategy set is generated according to dynamic cooperation topological information, path optimization processing is carried out on a target distribution task according to an initial path distribution strategy, a dynamic load adjustment rule and a path switching trigger condition, and an optimized distribution path set is obtained; and deploying to the corresponding distribution node to execute dynamic cooperative distribution, and updating the network state data set according to the distribution real-time feedback data, so that the distribution efficiency and reliability can be improved.

The invention relates to the technical field of power system fault diagnosis, and discloses a distribution network line fault detection data processing method, system and device and a medium, and the method comprises the steps: obtaining load data of a target distribution line detection point, and carrying out the first preprocessing of the load data, and obtaining a fusion feature vector; performing second optimization operation on the fusion feature vector to obtain an optimized fusion feature vector; calculating a potential feature matrix according to the optimized fusion feature vector, and mapping the potential feature matrix to a low-dimensional space to obtain a low-dimensional sample set; presetting an adaptive label propagation algorithm, and performing fault category judgment on the low-dimensional sample set based on the adaptive label propagation algorithm; and storing a judgment result in a relational database. The problems that feature extraction is not accurate in a high-noise environment, and a classification model is insufficient in new fault expansion capacity are effectively solved, and fault signal processing robustness is improved.

The invention relates to the technical field of time synchronization, and discloses a hybrid time service method based on Beidou satellite navigation and a TSN, which comprises the following steps: acquiring network topology structure information and historical time service error data of each synchronization node in a target distributed system, and constructing a dynamic time delay state set for jitter identification; based on the dynamic time delay state set, a current global time reference is extracted through a time calibration algorithm, a time difference reference vector sequence is generated, and initial synchronization adjustment is carried out on the TSN main node; constructing an inter-node clock convergence function curve according to the slave node response stability, and judging whether the synchronous network is in a low jitter interval or not based on the time offset rate and the time drift trend change; in combination with the node time state after directional offset suppression, a multi-channel redundant time synchronization mechanism in a time window is applied, a weight adjustment coefficient is extracted, and an unbalanced calibration factor is introduced; and according to a regression correction result, evaluating a time service stability index in real time. The method has the advantage of improving the time service stability.

The invention relates to a check point reading and writing method and device in distributed training, a storage medium and a program product. The method comprises the steps that in response to check point file generation and the number of check point files in a memory of a training node cluster corresponding to a target distributed training task does not reach N, an idle target memory is determined in the memory of the training node cluster, and the check point files are written into the target memory, the memory of the training node cluster comprises the memory of each training node in the training node cluster; and in response to a fault of any training node in the training node cluster, reading the target check point file from the memory of the training node cluster according to the metadata of the target check point file corresponding to the training node, and recovering training based on the target check point file. According to the method and the device, the check point file can be quickly read from the memory of the training node for recovery when the training node fails.

The invention discloses a distributed power grid power generation equipment optimization scheduling method and system, and the method comprises the steps: firstly obtaining historical operation state data and real-time operation state data of a distributed power generation network, then taking one piece of distributed power generation equipment as target distributed power generation equipment, and determining an initial output range and an actual output range of the target distributed power generation equipment; and finally, determining the actual output range of the plurality of distributed power generation devices, constructing an output optimization scheduling model, determining an optimization objective function and constraint conditions, solving the output optimization scheduling model, outputting an optimal scheduling scheme, and adjusting the output of the distributed power generation devices. According to the method, the target distributed power generation equipment is evaluated and scored by integrating historical and real-time data so as to accurately determine the actual output range of the equipment, and the scheduling scheme is ensured to better fit the actual operation condition by constructing the optimization model and determining the appropriate target and constraint, so that the operation efficiency and stability of a power grid are improved.

The invention provides a real-time human body detection method and system based on a microwave radar, and is applied to the field of signal data processing. By constructing a sparse reconstruction model and combining multi-dimensional feature decomposition, the method effectively relieves the recognition difficulty caused by target signal aliasing, continuously outputs and receives radar signals, extracts distance, Doppler and angle features, constructs target distribution point clouds in a three-dimensional feature space, further utilizes angle continuity judgment and point cloud density analysis, and improves the recognition accuracy of target distribution point clouds. According to the method, the number and distribution of overlapped targets are accurately estimated, a Kalman filter is introduced for dynamic tracking for a scene in which a continuous area is not formed, continuous separation and pseudo-color image reconstruction of multiple targets are realized, and thus the resolution precision and target positioning capability of a radar in a multi-person close-range aggregation scene are remarkably improved.

The invention provides an scRNA-seq data clustering method, system and device based on ZINB distribution and graph attention. The scRNA-seq data clustering system mainly comprises three core modules: a ZINB auto-encoder, which is used for modeling scRNA-seq data based on zero-expansion negative binomial distribution, generating robust potential representation through a denoising auto-encoder, and accurately capturing sparsity, excessive discreteness and shedding events of gene expression; the residual image attention auto-encoder is used for constructing a cell relation graph by using a Pearson correlation coefficient, dynamically learning a neighborhood weight in combination with a multi-head attention mechanism, retaining original features through residual connection, and relieving the excessive smoothness problem of image convolution; and a deep clustering model is self-optimized: target distribution and soft label distribution are minimized through KL divergence, and end-to-end joint optimization of embedded learning and clustering is realized.

The invention provides an insulator defect detection method based on mixed attention and multi-scale features. By constructing a multi-scale feature extraction module and combining a bottom-up residual neural network and a top-down reverse residual neural network, the ability of the model to extract different scale features is improved. An introduced multi-head mixed attention mechanism enhances the attention of the model to details and overall information through the combination of local attention and global attention. An improved Transform framework is combined with a multi-head mixed attention and Hungary bisection matching algorithm, so that the target distribution and detection precision is optimized, and the small defect recognition capability of the model under a complex background is improved. According to the method, high-precision detection is realized at low calculation cost, and particularly, the method has a remarkable detection effect and practical application value when being used for coping with multiple targets, complex backgrounds and small defects.

A distribution network site three-dimensional reconstruction method, system and device and a readable storage medium, and relates to the technical field of three-dimensional reconstruction, the method comprises the steps of obtaining a target unmanned aerial vehicle aerial image sequence, the target unmanned aerial vehicle aerial image sequence comprising multi-view images obtained by performing multi-angle shooting on a tower in a target distribution network site; extracting key points of each image in the aerial image sequence of the target unmanned aerial vehicle based on a key point detection algorithm; based on the extracted key points, cross-image matching is carried out through distance measurement and context information of the key points, so that corresponding points of the same object in different images are determined; based on each matched point pair and a preset camera parameter, generating a sparse point cloud by using a three-dimensional reconstruction algorithm; and generating a dense point cloud based on the sparse point cloud and the depth information of the image so as to realize three-dimensional reconstruction of the target distribution network site. According to the invention, details and features in a complex scene can be effectively captured, so that high-quality three-dimensional reconstruction is realized.

The invention provides a post-earthquake emergency material distribution optimization method considering distribution fairness and demand urgency, and relates to the technical field of emergency logistics multi-target distribution optimization under the background of urban earthquake disasters, and the method comprises the following specific steps: S1, determining a basic hypothesis: considering the actual scene of disaster rescue operation, and making the basic hypothesis for the emergency material distribution problem; s2, establishing an emergency material distribution fairness model; s3, establishing a disaster point emergency material demand emergency degree evaluation index system; s4, establishing a multi-target path optimization model; and S5, solving the established multi-target model by using an improved mixed multivariate universe optimization algorithm. The mixed multivariate universe algorithm based on differential evolution is provided, higher convergence precision and better robustness and uniformity are achieved, the earthquake disaster emergency logistics distribution path optimization accuracy is improved, material demand urgency meeting and distribution fairness are guaranteed, the distribution cost is reduced, and the distribution efficiency is improved. And the disaster victim satisfaction degree and the demand emergency adaptation degree are improved.

The invention relates to the technical field of electric vehicle data analysis, and discloses an electric vehicle charging demand prediction and distribution network optimization method based on data fusion. According to the method, related data of the electric vehicle are collected through a multi-source data acquisition module, feature fusion is performed by using a convolutional neural network, and charging demand parameters are predicted through a long-short-term memory network model. And constructing a multi-target distribution network optimization model, performing global optimization by adopting a genetic algorithm and introducing adaptive crossover and mutation probability, and outputting an optimal distribution network scheme. A layered charging control model is established, a strategy layer performs global planning, a scheduling layer performs local scheduling, and an execution layer realizes charging control based on fuzzy control. In addition, a differential privacy technology is adopted to protect user data privacy, a path is planned based on an A * algorithm of real-time road resistance, a prediction result is corrected in combination with weather factors, accurate prediction of the charging demand of the electric vehicle and distribution network optimization are achieved, and the resource utilization efficiency and the distribution network stability are improved.

The invention discloses a cross-platform content distribution intelligent adaptation method and system. The method comprises the following steps: acquiring multi-modal original input content; performing feature extraction on the multi-modal original input content to obtain a content feature group; obtaining platform feature parameters of the target distribution platform; obtaining a feature matching degree between the content feature group and the platform feature parameter; generating a content transformation scheme according to the platform feature parameters and the feature matching degree; and according to the content transformation scheme, carrying out transformation adjustment on the multi-modal original input content to obtain transformation content. According to the technical scheme provided by the invention, the original content is subjected to intelligent transformation adaptive to each platform; various content forms are supported, and the application scene is wide; quality evaluation and optimization are carried out on the transformation content, and the quality and the effect are improved; a publishing strategy is formulated, and content value maximization is achieved; fast access of a new platform is supported, a feature library is continuously optimized and updated, and a content modification scheme is dynamically adjusted; and the multi-platform operation efficiency is obviously improved.

The invention provides a panoramic image processing method and device suitable for an unmanned aerial vehicle and a panoramic unmanned aerial vehicle. The method comprises the following steps: acquiring multi-view images in real time through a multi-camera array of an unmanned aerial vehicle, acquiring pose data by using an inertial sensor, converting each view image into a first projection image and a second projection image, and performing dynamic target detection and recognition on the second projection image to obtain a dynamic target, geometric scene features and target distribution. Differentiated processing areas are divided, specifically, a high-precision algorithm is adopted for the first processing area, a low-precision simplified algorithm is adopted for the second processing area, and computing power focusing and processing capacity optimization are achieved. In the splicing stage, a dynamic target historical trajectory optimization splicing algorithm is combined, and processing strategies are adapted to different areas, so that the quality of the key areas is guaranteed, and the process of the non-key areas is simplified to reduce the calculation load. According to the scheme, the problem of real-time panoramic splicing under the condition that unmanned aerial vehicle hardware resources are limited is effectively solved, and the processing real-time performance and efficiency are remarkably improved.

The invention provides an unmanned aerial vehicle cluster dynamic hunting method for a complex three-dimensional scene. The method comprises the steps that the state of an escaper unmanned aerial vehicle, the state of each hunting unmanned aerial vehicle and the environment state are obtained respectively; sub-targets are generated around the escaper unmanned aerial vehicle through a Fibonacci ball algorithm, and the sub-targets are dynamically adjusted; based on the dynamically adjusted sub-targets and the state of each trapper unmanned aerial vehicle, optimal sub-target distribution is carried out on each trapper unmanned aerial vehicle through an improved market auction algorithm, wherein the improved market auction algorithm considers path cost and angle cost in a cost function; constructing a dynamic target control obstacle constraint based on the environment state, the state of the escaper unmanned aerial vehicle and the state of each trapper unmanned aerial vehicle; an escaper position change factor and a self-adaptive attenuation rate are introduced into the dynamic target control obstacle constraint; and constructing a distributed MPC optimization problem based on optimal sub-target distribution and a dynamic target control barrier function to perform trajectory optimization and motion control. According to the invention, the efficiency and the safety are ensured at the same time.

The invention discloses a link tracking method and device, equipment and a storage medium, and relates to the field of distributed technologies. The method comprises the following steps: determining a candidate service label of at least one candidate service scene according to service data of the at least one candidate service scene in a target distributed system; determining a target service label from the candidate service labels according to a target service scene corresponding to the target service request under the condition that the target service request of the user is identified; determining a target dyeing label of the target service request according to the target service label; and dyeing the link track of the target service request according to the target dyeing label so as to track the target service request. According to the technical scheme, the dyeing labels are self-defined in the link tracking of the distributed system, so that the problems of difficulty in link carding, difficulty in quickly judging the fault influence range and the like caused by multiple nodes and long links of the distributed system are solved, and the reliability and the accuracy of distributed link tracking can be improved.

The invention relates to the technical field of power distribution networks, and discloses a power distribution network scheduling method and system based on deep reinforcement learning, and the method comprises the steps: carrying out the clustering and partitioning of a target power distribution network, and obtaining a plurality of power distribution network sub-regions; performing numerical calculation on the operation state monitoring data to obtain state vectors of the corresponding power distribution network sub-regions; training a pre-constructed first deep reinforcement learning model to obtain a second deep reinforcement learning model corresponding to the power distribution network sub-region; evaluating each second deep reinforcement learning model to obtain an evaluation result corresponding to the second deep reinforcement learning model; according to the global aggregation parameters, updating each second deep reinforcement learning model to obtain a third deep reinforcement learning model corresponding to the power distribution network sub-region; and inputting each real-time state vector into a corresponding third deep reinforcement learning model for strategy mapping to obtain a scheduling strategy of each power distribution network sub-region. According to the invention, the dispatching reliability and real-time performance of the power distribution network are improved.

An MHCl binding peptide design method based on evolutionary information and a Transform neural network algorithm relates to the field of protein design, and comprises the following steps: S1, extracting evolutionary information features of alleles of MHCII molecules and binding core sequences of binding peptides corresponding to the alleles, S2, establishing a neural network model based on fusion of a convolution module and a Transform module, and S3, establishing a neural network model based on fusion of the convolution module and the Transform module, the method comprises the following steps: S1, extracting two frequency characteristic tensors from S11 and S12, taking the two frequency characteristic tensors extracted in S11 and S12 as double inputs, and finally obtaining probability distribution of 20 amino acids at each position of each sequence, and S3, according to an output result of a neural network model, carrying out random sampling according to the probability, and generating a binding core sequence of MHCII-peptide meeting target distribution. According to the method, evolutionary information such as sequence position amino acid frequency (first-order conservative analysis) and combined frequency (second-order conservative analysis) of amino acid pairs is introduced to design a new short peptide sequence, the problem that short peptides cannot be designed based on structures is solved, and the reliability of short peptide sequence design based on evolutionary information is provided.

The invention discloses a remote sensing target detection large model construction method based on context feature deep learning, and the method comprises the following steps: S1, dividing an original data set according to a strategy, forming a training set, a verification set and a test set, and guaranteeing that each subset has representativeness in sample number, scene diversity and target distribution; s2, constructing a remote sensing target detection large model by adopting a deep learning method based on context features, and fusing multi-level features and introducing a query optimization mechanism; and S3, performing model testing and performance evaluation, selecting optimal model parameters, performing comprehensive verification on the test set, and evaluating the comprehensive performance of the model in the aspects of detection precision, speed and robustness. According to the method, the pre-trained DINOv2 backbone network, the context feature enhancement module and the intensive supervision strategy are introduced, so that rich context information can be captured in a multi-scale and complex scene, and the detection precision is greatly improved.

The embodiment of the invention discloses an intelligent unmanned aerial vehicle food material delivery method and system, and the method comprises the following steps: obtaining to-be-delivered order information and parameter information of an unmanned aerial vehicle, determining a target delivery strategy according to the to-be-delivered order information and the parameter information of the unmanned aerial vehicle, and carrying out the food material delivery according to the target delivery strategy. Real-time environment information images of the unmanned aerial vehicle are obtained in real time in the distribution process; and comparing the real-time environment information image with an initial environment information image to obtain incremental image information, and adjusting a target distribution strategy or controlling an unmanned aerial vehicle to perform intelligent obstacle avoidance according to the incremental image information. According to the method, global distribution path planning and local intelligent obstacle avoidance are combined, manual frequent intervention and adjustment are not needed, the autonomy and efficiency of unmanned aerial vehicle distribution are improved, the method can adapt to complex scenes, the environment adaptability is higher, the safety of the unmanned aerial vehicle can be guaranteed, and food distribution can be efficiently completed.

Certain aspects of the present disclosure provide techniques and apparatus for generating a response to a query input in a generative artificial intelligence model. An example method generally includes receiving a plurality of sets of tokens generated based on an input prompt and a first generative artificial intelligence model, each set of tokens in the plurality of sets of tokens corresponding to a candidate response to the input prompt; selecting, using a second generative artificial intelligence model and recursive adjustment of a target distribution associated with the received plurality of sets of tokens, a set of tokens from the plurality of sets of tokens; and outputting the selected set of tokens as a response to the input prompt.