75 results about "Network completion" patented technology

The invention discloses a reinforcement learning method and system for source network load storage collaborative multi-scene optimization, and the method comprises the steps: constructing a power distribution network optimization model with the minimum cost, and converting a mixed integer nonlinear problem into a solvable mixed integer second-order cone optimization problem; converting a mixed integer second-order cone optimization problem into a reinforcement learning decision model, and performing multi-round assignment on boundary condition parameters by using different operation scene data to form a multi-scene training task set; designing a reinforcement learning decision model multi-scene training loss function, combining the reinforcement learning decision model to construct a strategy neural network, an evaluation network neural network and a scene representation embedded neural network, and completing reinforcement learning decision adaptive to multi-scene optimization; and training a power distribution network multi-scene optimization decision model based on the multi-scene training task set, the multi-scene training reinforcement learning loss function and the neural network structure, and deploying the power distribution network multi-scene optimization decision model to an actual system to complete reinforcement learning decision model application oriented to source network load storage collaborative multi-scene optimization.

The invention belongs to the technical field of power system optimization scheduling, and provides a source network load storage intelligent collaborative optimization method, which comprises the following steps of: deploying sensors at four ends of a source network load storage respectively, collecting in real time by utilizing a cloud data center, enabling data of the four ends to be consistent in time sequence through a PTP protocol, constructing a topological graph according to parameters and data, and establishing a source network load storage intelligent collaborative optimization system. Selecting a model in a digital twinning environment for simulation; dividing independent agents at four ends of a source network load storage, setting observation data, an execution space and excitation feedback, forming an excitation item by economy, stability and environmental protection, interactively circulating actual data, a prediction instruction and an excitation value, recording into a sequence, inputting the sequence into a strategy network, and calculating and outputting logarithmic probability gradient to update the parameters of the strategy network; and the intelligent agent completes interactive circulation according to the strategy network, generates a local scheduling instruction, aggregates the instruction to perform weighted calculation, generates a global scheduling scheme, issues the global scheduling scheme to execution equipment, updates parameters by using an average deviation calculated by a deviation vector, resolves the global scheduling scheme and issues the global scheduling scheme to form a closed-loop mechanism.

A system for completing tasks using the web is disclosed. The system is programmed to receive a user query for completing a task in natural language. The system is programmed to generate from the user query a first plan having a first sequence of website action steps using a large language model. Each website action step specifies a website and includes a request for performing an action on a website in natural language or network or software protocol language. To execute a website action step, the system is programmed to generate from a corresponding request a current plan having a current sequence of function action steps using a large language model. Each function action step specifies a function in the website's application programming interface and includes values for parameters of the function. To execute a function action step, the system is programmed to call the function with the parameter value.

The invention provides a key component fault prediction method and system based on multi-source data fusion, and the method comprises the steps: S1, collecting key component data of a screen scarifier, constructing a multi-source heterogeneous sensor network, and completing the time-space alignment and quality optimization of multi-dimensional monitoring data; s2, a dynamic topological graph structure is constructed based on the physical connection relation of the components, the fault propagation path and strength are quantified, and multi-level feature representation covering the local state and the overall health degree is formed; s3, designing a hybrid prediction model fusing space-time analysis and memory modeling, and completing accurate description of an equipment degradation trend and early warning of potential faults through a self-adaptive feature integration mechanism; and S4, combining real-time prediction errors and historical operation and maintenance knowledge to dynamically optimize the hybrid prediction model, establishing a data-driven and knowledge-guided dual learning framework, and completing self-adaptive continuous learning along with equipment aging. The method breaks through the limitation of static state of a traditional prediction model, and the accuracy and reliability of a prediction result are remarkably improved.

The invention belongs to the technical field of legged robot control, and provides a quadruped robot robust motion control method and system based on joint learning, and the method comprises the steps: obtaining the body observation information and privileged observation information of a quadruped robot; extracting key features in the obtained privileged observation information by adopting a privileged encoder to obtain a first potential feature vector; obtaining a second potential feature vector matched with the first potential feature vector based on the obtained historical ontology observation information and an adaptive network; constructing a joint loss function according to the obtained first potential feature vector and the second potential feature vector, and performing strategy network updating training by taking the minimum joint loss function as a target; and robust motion control of the quadruped robot is completed according to the trained strategy network.

The invention discloses an underground water condition detection method and system based on water flow numerical model simulation, and relates to the data processing technology, and the method comprises the steps: building a terrain model containing a target region; determining a proportional relation in the terrain model, and cutting in multiple target directions of the terrain model based on the proportional relation; determining a plurality of representative sites according to the sections, collecting formation data of the representative sites, and mapping the collected formation data to the corresponding sections; fitting is carried out on corresponding sections based on stratums of the mapped representative sites, so that stratigraphic distribution is constructed; constructing a data sequence according to the acquired associated monitoring data of the underground water of the plurality of acquisition points and the rainfall data of the target area; and complementing the groundwater condition data of each section of the island terrain model by using a preset complementing network model according to the data sequence. Groundwater condition detection is realized in combination with a stratum structure model and a network completion mode, and a new method for underground water condition detection is provided.

The invention provides a neural architecture search-based multi-modal automatic modeling and fusion method and device, and relates to the technical field of computer science. The method comprises the following steps: acquiring multi-modal data and a task type; automatically generating a corresponding optimal unit architecture for each modal data based on neural architecture search, and performing feature extraction on each modal data; according to a high-dimensional feature of each modal data, analyzing a dependency relationship between modals of a feature level to generate a dynamically updated correlation thermodynamic diagram, obtaining a specific fusion strategy according to a fusion strategy decision maker, and constructing an optimal fusion network architecture; and an output layer and a loss function are automatically adjusted according to task types, a network meeting task requirements is obtained, and different downstream tasks are completed. According to the method, end-to-end joint optimization of single-mode feature extraction and a multi-mode fusion strategy can be realized, the manual intervention cost and GPU resource consumption are remarkably reduced, and meanwhile, the task performance and the model generalization ability in a complex scene are improved.

A VCN process may receive information associated with a value chain network. A VCN process may provide the information to a set of Artificial Intelligence (AI)-based learning models, wherein at least one member of the set of AI-based learning models is trained to classify at least one of: an operating state, a fault condition, an operating flow, or a behavior of the value chain network and at least one member of the set of AI-based learning models is trained on the training data set to determine, upon receiving the classification of the at least one of: the operating state, the fault condition, the operating flow, or the behavior, a task to be completed for the value chain network. A VCN process may configure a robotic process automation system to execute the task to facilitate an improvement in the value chain network.

The invention provides a cluster wireless ad hoc network method for port intelligent operation equipment, and belongs to the technical field of wireless communication. The method comprises the steps that firstly, port communication system layered architecture initialization is carried out, port intelligent operation equipment neighbor discovery is achieved, and meanwhile neighbor equipment access information packets and channel quality index information are recorded; secondly, performing cluster head node election through a maximum score priority rule to form a local cluster structure consisting of cluster head nodes and terminal equipment, and introducing a cluster wireless ad hoc network to complete a judgment mechanism after the cluster head node election of the intelligent operation equipment is completed; and finally, performing multi-path construction, standby path caching and a main and standby path seamless switching mechanism. The method supports dynamic election of cluster head nodes, supports multi-path cache, can realize seamless communication switching, and solves the problems of slow remote control response speed and high failure rate of port intelligent operation equipment.

The invention provides an FPVA biochip layout method considering reagent type difference in unit multiplexing, which comprises the following steps of: modeling a biochip layout task as a Markov decision process: according to the size of an input FPVA chip, the dependency relationship of each operation in a biochemical reaction sequence diagram, the reagent type contained in each operation and a binding scheduling scheme, determining the FPVA biochip layout task; initializing an FPVA mesh model and defining a state space, an action space and a reward function; wherein the reward function is set according to the total unit multiplexing complexity of the current layout scheme, the legality of the current component layout and the distance between the component and the center point of the FPVA chip; and finally, completing the FPVA biochip layout through the dual deep Q network based on a Markov decision process. By designing a reward function, the unit multiplexing complexity is minimized, and the distance between components containing the same type of reagents is reduced, so that the purposes of minimizing the cross contamination degree of a final layout scheme, the total length of a fluid transportation path and the bioassay completion time are achieved.

The invention provides a power equipment remote cooperative operation and maintenance method based on a satellite flash technology, and the method comprises the steps: deploying a satellite flash positioning tag on power equipment, building a satellite flash communication network based on a satellite flash gateway, and completing the three-dimensional coordinate calibration and networking of the power equipment; each satellite flash positioning tag transmits positioning data of the power equipment where the satellite flash positioning tag is located to the satellite flash gateway in real time; the handheld terminal obtains positioning data of the operation robot and the power equipment from the star flash gateway, generates a dynamic navigation path according to task requirements, guides the robot to move to an operation position, and issues an operation instruction to the operation robot; and the operation robot receives and executes the operation instruction, and feeds back an execution result in real time. The invention further discloses a corresponding method. By implementing the method, the positioning precision in remote operation and maintenance of the power equipment can be improved, the operation time delay can be reduced, and multi-equipment coordination can be realized, so that the safety and efficiency of power operation and maintenance are improved.

The invention relates to the technical field of structure health monitoring, and discloses a building structure health real-time monitoring method and system based on multi-sensor fusion, and the method comprises the steps: enabling the system to operate in a low-power-consumption passive monitoring mode, awakening sensor nodes in a target region according to the needs when a preset condition is met, and enabling the sensor nodes to be in a real-time state; and self-organizing to form a local diagnosis network and electing a dominant node. And then, the leading node coordinates the network to complete excitation-response type active diagnosis, cross validation and information extraction are carried out on original data locally, and only a generated structured diagnosis abstract is uploaded. And finally, the central processing unit fuses the abstract, the historical baseline and causal confidence analysis, performs weighted calculation and then outputs an evaluation result. Through a dynamic and static combined monitoring mode and distributed intelligent processing of a network edge, the contradiction between high-precision diagnosis and low-power-consumption operation is effectively solved, the communication overhead is reduced, and the real-time performance and reliability of an evaluation result are improved.

The invention discloses a container exit-oriented robot scheduling method and system, and relates to the technical field of intelligent scheduling, and the method comprises the steps: collecting multi-source data through an Internet of Things sensor network, and inputting the multi-source data into a dual-depth Q network after time-space alignment to complete task pre-distribution; performing rolling time domain path optimization based on a three-dimensional space-time coordinate system, and resolving track conflicts in combination with a virtual guide point strategy; the weight of a multi-target cost function is dynamically adjusted by using a long-short-term memory network, and adaptive optimization of indexes such as transportation efficiency and energy consumption is realized; for an abnormal event, a bee colony optimization algorithm is adopted to generate a local conflict-free path, and a container door orientation optimization strategy is integrated to reduce energy consumption, so that the collaborative scheduling efficiency of the container terminal AGV cluster is improved, and the dynamic environment adaptability is enhanced.

The invention discloses a global nerve drawing method and system based on a programmable rasterization engine, and belongs to the technical field of computer graphics, and the method comprises the steps: at the programmable rasterization engine, analyzing a rasterization descriptor according to a rasterization instruction, and extracting vector microoperation and control parameters; maintaining a task state machine according to the parameters and distributing a control signal, selecting an execution entry from a vector kernel table according to the control signal, and instantiating an operation into a parallel vector thread; in a vector thread execution process, tracking data dependence of a vector register and a synchronization state of a direct memory access unit, executing vector loading / storage operation so as to carry data between the register and an on-chip shared memory according to the data dependence and the synchronization state, and dynamically scheduling vector micro-operation to an execution component so as to complete rasterization calculation; and outputting a result to the neural rendering network to complete global neural rendering. According to the method, the multi-representation neural rendering load can be uniformly and efficiently supported on the AI accelerator, the memory access overhead is remarkably reduced, and the calculation efficiency is improved.

The application relates to the technical field of fault diagnosis, and provides a high-frequency transformer fault diagnosis method and system, which comprises the following steps: collecting a target signal with a time stamp and extracting corresponding features, simultaneously relying on a transformer structure, material parameters and physical rules to build a digital twin model, simulating insulation and structure degradation equivalent working conditions, solving multi-physical field data and generating multi-physical field mechanism samples; then, the mechanism samples and field measured data are fused through a generative adversarial network to expand the fault sample data set and solve the sample scarcity problem; in the running stage, the digital twin model is updated in real time through parameter online inversion, a feature dynamic graph representing multi-physical coupling is constructed by combining the parameter deviation of internal mechanism degradation and the multi-source features of external working conditions, finally, the feature aggregation and time sequence reasoning are completed through the graph neural network and the time sequence neural network trained offline, the fault probability is output, and the optimal diagnosis result is determined; thereby, the fault recognition accuracy and the robustness in the running stage are improved.

This application relates to the field of flight control technology, specifically to a high-maneuverability flight control method for fixed-wing aircraft based on curriculum-based reinforcement learning. It constructs a closed-loop learning system comprising a curriculum scheduler, an agent, a simulation environment, and an experience buffer. The agent has a policy network. The state space of the aircraft and the normalized control surface and throttle action space output by the policy network are defined. A curriculum difficulty measurement model is established, quantifying task difficulty through weighted state deviation and envelope penalty terms. Based on the current curriculum level and this model, a safe and difficulty-matched training task set is dynamically generated. In the simulation environment, the policy network is iteratively updated through a two-layer loop training process. After completing all courses, a high-maneuverability flight control strategy is obtained. This strategy is deployed to the flight control system to achieve high-maneuverability flight control based on real-time state. This achieves efficient, safe, and adaptive flight control strategy training and deployment.

This invention discloses a traffic flow prediction method and system based on Transformer, belonging to the fields of intelligent transportation and deep learning technology. Addressing the technical problems of existing traffic flow prediction models, such as difficulty in simultaneously considering long-term and short-term dependencies, inability of static road network topology to characterize dynamic spatial heterogeneity, and poor modeling performance of spatiotemporal feature coupling, this invention proposes a multi-timescale adaptive graph attention Transformer model. This method first reconstructs the original traffic data at low, medium, and high time scales, and then aggregates spatiotemporal features through a temporal convolutional network and a compressed excitation network. Next, an adaptive data graph generation module learns node embedding vectors to generate an adaptive adjacency matrix that integrates static topology and dynamic associations. Finally, an encoder incorporating temporal one-dimensional convolutional multi-head attention and spatial graph attention, and a decoder integrating causal convolution and temporally gated convolution, are constructed to achieve high-precision multi-step prediction of traffic flow. This invention effectively captures the spatiotemporal dependencies of traffic flow, with prediction accuracy and generalization superior to mainstream models, and can be widely applied to urban intelligent traffic management, dynamic path planning, and traffic congestion mitigation scenarios.

The invention provides a strategy migration method and device based on multi-resolution simulation. The method comprises the following steps: constructing a first simulation environment and a second simulation environment in a target scene; in the first simulation environment, training the hierarchical agent based on the simulation data to obtain a teacher strategy; and based on the teacher strategy, guiding the student strategy network in the second simulation environment to perform initialization training to obtain an initial strategy of the target scene, and in the second simulation environment, performing adjustment training on the initial strategy to obtain a final strategy of the target scene. Large-scale training is carried out through high calculation efficiency of the first simulation environment to obtain a teacher strategy, and a student strategy network in the second simulation environment is guided to complete initialization and subsequent adjustment training, so that smooth transition from high-efficiency coarse-grained exploration to high-fidelity fine migration is realized; and when the intelligent agent is trained in the simulation environment, the accuracy of the generation strategy is improved while the intelligent agent training efficiency is guaranteed.

In order to solve the problems of long design time, low efficiency, high calculation cost and low prediction accuracy caused by information loss in the prior art, a resonance super surface design method and device based on spectrum perception improvement are provided.The method comprises the following steps: designing a GLSAT forward prediction network based on spectrum perception improvement;training and optimizing the GLSAT forward prediction network; designing a DNN reverse design network; cascading the DNN reverse design network and the trained GLSAT forward prediction network to obtain a cascaded reverse design network; inputting the ideal spectrum pretreated by GSSG into the cascaded reverse design network, training and optimizing the DNN; and completing the design of the resonance super surface by using the optimized DNN reverse design network.The method has the characteristics of short time consumption, high efficiency and low calculation cost while improving the design prediction accuracy.

The invention discloses a data security transmission method and device, equipment and a storage medium, and relates to the technical field of information security, and the method comprises the steps: obtaining a data transmission parameter generated based on a ring signature in a block chain network, generating a target key pair of each transaction party, generating corresponding hidden address information based on the data transmission parameter and a public key of a receiver, and transmitting the hidden address information to the block chain network; encrypting the to-be-transmitted data according to the waiting party public key, the hidden address information and the transmitting party key pair; generating data supervision information of the to-be-transmitted data and identity supervision information of the receiver public key based on the supervisor public key; data transmission content is constructed according to the hidden address information, the to-be-transmitted data ciphertext, the data supervision information and the identity supervision information, the supervision party verifies the data transmission content based on the supervision party private key and the supervision list, and after it is judged that the identities of the data transmission party and the data receiving party are correct, the data transmission party completes data transmission through the block chain network. Privacy protection and supervision auditing of data can be balanced, and recovery and supervision of transmission data are realized.

The invention provides a global intelligent remote 5G joint sharing system and method. The system comprises a remote access and session management module which is configured to complete access, identity authentication and authority distribution of remote medical participants through a 5G network; the multi-modal data transmission and processing module performs multi-modal transmission, coding and decoding processing and synchronous management; the intelligent analysis and personalized content generation module is used for generating personalized interaction content and auxiliary prompt and risk early warning information; the cooperative arrangement and terminal control module is used for executing corresponding meeting control and control operation on the remote medical terminal; and data transmission and cooperative processing are carried out through the 5G access network and the edge computing node. Thus, in combination with 5G network characteristics and intelligent analysis capability of artificial intelligence, fusion processing can be performed on multi-modal medical data such as medical images, medical record data and vital sign monitoring data, and stable and efficient collaborative services are realized.

The invention provides an open domain information extraction algorithm based on a graph neural network and relation discrimination. The method comprises the following steps of: extracting an entity relationship in a text and finishing construction of an entity relationship graph; performing text embedding representation and context coding on the input text; fusing the entity relationship to perform graph information embedding; and decoding is completed by using a double affine network, and an SPO triple extraction result is obtained. The invention aims to better solve the problem of open domain information extraction. According to the algorithm, technologies such as a pre-training model and a graph neural network are combined, meanwhile, the thought of relation judgment is combined, entity relation information is embedded into the network in a topological graph structure, and therefore the model can better utilize dependency knowledge in a text. Therefore, the problem that a traditional open domain information extraction method is limited in effect when facing complex texts and the defect that text dependent information cannot be fully utilized are overcome. According to the method, the open domain information extraction requirement in a complex scene can be met, and a technical basis can be provided for downstream tasks such as a question answering system and knowledge graph construction.

The application relates to the technical field of flight control, in particular to a fixed-wing aircraft high-maneuver flight control method based on course reinforcement learning, a closed-loop learning system containing a course scheduler, an intelligent agent, a simulation environment and an experience buffer is constructed, and the intelligent agent has a strategy network; a state space of an aircraft and a normalized rudder and throttle action space output by the strategy network are defined. A course difficulty measurement model is established, the task difficulty is quantified by weighting the state deviation and the envelope penalty term; based on the current course level and the model, a training task set that is difficult to match and safe is dynamically generated. In the simulation environment, the strategy network is iteratively updated through double-loop training, and a high-maneuver flight control strategy is obtained after all courses are completed; the strategy is deployed to the flight control system to realize high-maneuver flight control based on real-time state. Efficient, safe and adaptive flight control strategy training and deployment are realized.

The invention discloses a dialogue-level text emotion recognition method based on graph pooling representation learning, which aims at the problem of weakening remote statement information during graph construction, uses a pre-training model to extract context-independent statement features, and uses a graph network to complete modeling of context information. Most previous models based on a graph network method adopt a time window method during graph construction, so that long-distance statement information in a long dialogue is weakened. In order to solve the problem, a graph pooling layer and an anti-pooling layer are adopted, a graph network is constructed from a fully connected graph to screen edges and nodes, and dependence of long-distance statements is better utilized when context information is modeled. According to the method provided by the invention, a contrast experiment is carried out on IEMOCAP and MELD data sets, the method is obviously improved compared with a baseline method, ablation research is carried out on the method provided by the invention, and the effectiveness of each module is proved.

The application discloses an aspect-level sentiment analysis method and device based on a cross-modal syntax-visual graph convolutional network, containing a cross-modal graph structure construction module and a type-sensitive graph convolutional network for updating features between modes. Feature vector representations of pictures and texts are obtained through a pre-training model, a graph structure representation of the text is obtained through a syntax analysis method, a new dependency relationship is constructed to integrate the picture feature vector into the graph to form a new integrated cross-modal graph structure, a type-sensitive graph convolutional network is used to update the feature vector corresponding to the aspect word, and a full-connection neural network is used for final aspect-level sentiment detection of the group of comments. The pre-training model is used to obtain basic feature representations containing prior knowledge, syntax analysis and construction of the new cross-modal graph structure are used for fine-grained division and combination of the cross-modal comments, a graph network is used for fine-grained fusion, and finally, sentiment classification detection is completed.

The application discloses an infrared weak small target detection method based on line-by-line detection, and solves the problems of high detection delay and large resource consumption caused by global image caching in the prior art. The method skips the traditional "read-out-caching-global detection" process and performs real-time detection while reading out the infrared image data line by line. The method comprises the following steps: performing first-order and second-order differential calculation on a single-line image vector, extracting and fusing the intra-line spatial features; inputting the continuous multi-line features into an inter-line fusion module based on a self-attention mechanism in parallel, and restoring the high-dimensional features of the target; then, completing target detection through a U-Net network, and adapting the dimension through a line expansion and line compression module; and in the training, adaptively selecting a loss function according to whether the current line block contains a target, so as to improve the training efficiency. The application realizes parallel processing of detection and data reading, significantly improves the timeliness of detection, and reduces the demand for cache and computing resources of edge devices.

The invention discloses a key task seamless switching method based on a Mesh network, and belongs to the technical field of key task communication, a server and a client of a key task establish a group call session, when a current bearer network fails, both the server and the client are switched to the Mesh network from the current bearer network, and the server and the client are switched to the Mesh network from the current bearer network when the current bearer network fails. The method specifically comprises the steps of S1, session pre-storage, S2, fault triggering, S3, session recovery request, S4, server session verification, S5, client session verification and S6, server decision making, smooth switching and service continuity are realized on the basis of various heterogeneous network bearers for 3GPP standard key task communication, the integrity of service context information in a switched network is ensured during network switching, and the service continuity is ensured. The method comprises the following steps of: receiving data of a network to be switched, synchronizing the data to the network to which the network is to be switched, reconstructing a service session before a network switching critical point, applying for corresponding resources, and after service context information is completely established, completing network switching, releasing service context information of a standby switching network, and switching a target network to complete service recovery.

The present application relates to the technical field of property transaction, in particular to an intelligent contract automatic execution and supervision system for the whole process of property transaction, comprising: an intelligent contract generation unit; a contract automatic execution unit, which maps the intellectual property right transfer and transaction settlement process agreed in the standardized intelligent contract into a multi-stage execution phase sequence of corresponding performance nodes, and each execution phase is subject to the necessary condition of performance risk assessment; a contract supervision and verification unit; and an intellectual property right ownership verification unit.The present application builds a whole-process intelligent contract automatic execution and supervision system, uses a time sequence attention mechanism and an improved lightweight residual time sequence convolution network to complete dynamic risk assessment, combines hierarchical execution and reversible rollback to realize abnormal disposal, forms an execution, supervision and ownership verification closed-loop architecture, can monitor and dispose transaction abnormalities in real time, ensures stable transaction execution, realizes whole-link ownership tracing and compliance supervision.

The invention discloses a deep learning assisted waveform index modulation single carrier communication method, and belongs to the technical field of wireless communication. According to the method, the frequency spectrum efficiency and the system performance are improved by fully utilizing the waveform freedom degree of the single-carrier system. Cooperative transmission of index information and symbol information is realized by jointly optimizing a constellation mapping set, a shaping filter bank and a Bi-LSTM detection network of a receiving end. Specifically, a sending end divides information bits into symbol bits and index bits, the index bits dynamically select a constellation mapping set and a shaping filter, and the symbol bits generate a time domain waveform through the selected constellation and filter. And after a receiving end adopts frequency domain equalization and matched filtering, joint detection of indexes and symbols is completed through a Bi-LSTM network. According to the method, bit mutual information can be achieved through end-to-end training optimization, the signal power and the spectrum template are constrained at the same time, a high-performance and low-complexity index modulation implementation method is provided for a single-carrier system, and the method is suitable for a future high-spectrum-efficiency communication scene.

A method and system provide the ability to interpolate fluids. At least two keyframes are produced, for a physics based fluid simulation. The keyframes are within a continuous-time framework and separated by a defined interval. Each keyframe includes one or more fluid elements having a corresponding state. Data is prepared utilizing a pre-trained transformer-based network by: (i) handling a tokenization process in a physics-adapted context; and (ii) generating temporal embeddings for states of the one or more fluid elements. Based on the prepared data, a time-continuous density is prepared for substeps between the two keyframes using a density network.