297 results about "Metacomputing" patented technology

The invention provides an elevator energy recovery application method and system based on a super capacitor, and relates to the technical field of energy recovery, and the method comprises the steps: collecting elevator operation data and super capacitor state data, constructing a state space and an action space, and constructing a reward function according to peak-valley electricity price economic benefits, super capacitor life loss and energy utilization efficiency; training a deep reinforcement learning model based on the double-Q network to obtain an energy recovery optimization strategy; when the elevator is braked, regenerative braking energy is stored in the super capacitor; when the elevator is driven, the stored energy is released for traction; the optimal storage allocation scheme of the multiple elevator rooms is calculated through the virtual energy storage unit, hierarchical control is implemented based on an optimization strategy and the allocation scheme, the upper layer adopts fuzzy self-adaptive weight to determine the energy storage priority, and the lower layer adopts a distributed algorithm to calculate the energy allocation coefficient and the target power; and the charge-discharge rate is updated according to the health state of the supercapacitor.

The invention discloses a power distribution cabinet bus temperature change analysis and early warning method and system, and relates to the technical field of power distribution cabinets, and the method comprises the steps: collecting the temperature data of a three-phase bus of a power distribution cabinet, dividing a temperature sequence into time-space units, and calculating the spatial characteristics and time characteristics of each time-space unit; performing frequency domain analysis on the temperature data of the space-time unit, decomposing the temperature data to obtain a low-frequency component, an intermediate-frequency component and a high-frequency component, extracting sub-components of different time scales, and calculating a time coupling feature, a space coupling feature and a physical coupling feature; fault-free data is collected to establish a normal mode library, historical fault data is collected to establish a fault mode library, the deviation degree of features and a baseline is calculated in real time, and marking and feedback updating are carried out on potential new faults; on the basis of distributed monitoring, temperature abnormal points are positioned, reconstruction errors of all scales are calculated, the abnormal origin scale is determined, and graded early warning is triggered, so that the problem that a traditional monitoring method is insufficient in fault recognition capability is solved.

The invention relates to the technical field of hydraulic engineering and geotechnical engineering, in particular to a parameter updating method and device for a rock-fill dam deformation analysis model.The method comprises the steps that a plurality of Duncan-Zhang E-B model parameter combinations of an indoor triaxial experiment of a target rock-fill dam project are obtained to construct a copula multivariate joint distribution model; sampling a parameter combination sample of the target rock-fill dam project, and inputting the parameter combination sample into the finite element deformation analysis model to generate a finite element calculation sample; training a pre-constructed deep neural network by using the parameter combination sample and the finite element calculation sample to generate a deep learning agent model; and fusing the copula multivariate joint distribution model, the deep learning agent model and an NSGA-III multi-objective optimization algorithm to search for final model parameters for updating the rock-fill dam deformation analysis model generation by generation. Therefore, the problem that an existing dam project parameter updating method neglects complex correlation existing between different parameters is solved.

The invention discloses a control system and method for an automatic beer production line, relates to the technical field of automatic control, is used for solving the problems that state judgment is lagged and abnormal trend is difficult to identify in time, identifies electronic tags of key equipment through an RFID reader-writer, imports models and serial numbers into a database, and automatically associates operation parameters. The distributed sensors collect temperature, pressure, flow and vibration data in real time, the analysis unit calculates the operation deviation and the health degree, a multi-condition control strategy is configured through the programmable logic controller based on the health degree, a time window and a task priority algorithm are combined, real-time data and historical data are compared to recognize abnormity, and the real-time data and the historical data are analyzed. And a fuzzy control algorithm is fused to generate an adjustment instruction, the equipment state is automatically adjusted, time sequence modeling is utilized to analyze whole-process data, a start-stop strategy is formulated, energy distribution and process optimization are performed, and intelligent scheduling and self-adaptive optimization of the beer production line are realized.

Provided in the embodiments of the present application are a server system, a resource scheduling method for a server system, and a chip and a chiplet. The server system comprises: a multivariate computing resource pool, a data storage resource pool, a switching module and a control module, wherein the multivariate computing resource pool comprises a general computing resource pool and a heterogeneous computing resource pool, the general computing resource pool comprising a group of general computing units, and the heterogeneous computing resource pool comprising a group of heterogeneous computing units, and the multivariate computing resource pool is connected to the data storage resource pool via the switching module by means of a cache consistency bus inside the server system; the data storage resource pool comprises data storage resources allowed to be shared by multivariate computing resources in the multivariate computing resource pool; and the control module is configured to perform computing power scheduling on the multivariate computing resources in the multivariate computing resource pool, and dynamically allocate the data storage resources in the data storage resource pool to the multivariate computing resources in the multivariate computing resource pool.

The invention relates to the technical field of intelligent supply chain management and multi-modal artificial intelligence crossing, and discloses a multi-modal large model-based supply chain toughness evaluation system, which comprises a data acquisition module, a multi-modal processing module, a toughness evaluation module, a cloud edge collaboration module and a decision support module, the data acquisition module is used for acquiring supply chain texts, images and time sequence heterogeneous data; the multi-modal processing module is used for realizing dynamic feature fusion through term matrix enhanced Transform; the toughness evaluation module quantifies toughness scores based on a three-level index system and a fuzzy matter element calculation engine; the cloud edge collaboration module realizes edge end real-time reasoning through a lightweight model; the decision support module optimizes node reinforcement and path planning based on network topology and real-time weather data. According to the method, the problems of insufficient data isomerism processing, single evaluation dimension and lagging real-time response in the prior art are solved.

The embodiment of the invention provides a BIM-based bridge model intelligent construction method and system, and belongs to the field of BIM-based bridge model intelligent construction. The method comprises the steps that bridge model construction related data are collected and preprocessed, and a component library is established according to the preprocessed bridge model construction related data; defining geometric linkage rules and topological connection logic of components in the component library, and generating an initial BIM bridge model through a generative adversarial network; performing lightweight finite element calculation on the initial BIM bridge model to obtain an optimized BIM bridge model; and performing standard conflict detection and conflict parameter adjustment on the optimized BIM bridge model to obtain a final bridge model. According to the method, the parameter relation matrix and the dynamic linkage mechanism are constructed, the generative adversarial network is introduced to generate the complex bridge type, and the graph convolutional network is introduced into the generative adversarial network, so that reasonable bridge type topology and dynamic parameter linkage are guaranteed.

The invention relates to a server-side satellite precision product anomaly detection method and system and a terminal, specifically, a previous time window of a current epoch is obtained, the time window comprises a plurality of epochs, the residual mean value and the root-mean-square of each satellite in each epoch in the time window are calculated, and the residual mean value and the root-mean-square of each satellite in each epoch in the time window are calculated; normalizing the residual mean value and the root-mean-square to obtain the characteristics of the satellite in epochs; acquiring features of all the satellites in all epochs in the previous time window to form a feature set, performing outlier detection on the feature set by adopting an LOF method, and determining the LOF score of the satellite at the current epoch moment by utilizing the features of the current epoch of the satellite and the feature set after removing the satellite of which the LOF score is greater than a preset value from the feature set; the LOF scores of the satellites are sent to the PPP user side, when the PPP user side carries out ambiguity fixing, all the satellites are tried to be fixed firstly, if fixing fails, the satellite with the maximum LOF value is removed from the ambiguity set, fixing is tried again, and continuous circulation is carried out till the ambiguity is fixed successfully.

The invention discloses a multi-disaster-type disaster internet-of-things time sequence adaptive anomaly detection method and system, and relates to the field of internet of things, and the method comprises the steps: S1, constructing an anomaly detection model; s2, acquiring a training data set; s3, training an anomaly detection model; s4, acquiring to-be-detected data; s5, analyzing the reconstruction structure of the to-be-detected data; s6, analyzing an abnormal score; the emergency disaster early warning system comprises an acquisition unit, a storage unit, a calculation unit and an early warning unit. A multi-scale time convolutional network and a self-adaptive spectrum feature module are fused, the characteristics of time sequence data in a time domain and a frequency domain are deeply mined, a gating memory mechanism is introduced, normal time-frequency features in the data can be accurately captured and enhanced, and the recognition capability is improved; by adding the radial basis function layer, the detection capability of the model on tiny anomalies is remarkably improved, so that the model can detect tiny abnormal changes more accurately.

The invention discloses a hydrodynamic analysis method for a floating structure with multi-body and multi-coupling characteristics, which comprises the following steps of: acquiring geometric characteristics of each floating body, and dividing boundary element computational grids; wave excitation force, additional mass and radiation damping of each floating body are calculated, and a mass matrix and a rigidity matrix of each floating body are obtained; determining a connection mode and a coupling factor of the floating body; according to a connection mode and coupling characteristics of the system, establishing displacement connection conditions among the floating bodies; constructing a motion constraint matrix according to the displacement continuity condition and the motion constraint relation; determining a frequency domain motion equation of the multi-floating-body system according to a Lagrange multiplier method; and solving the motion equation to obtain the motion response of the multi-floating-body system, and further systematically analyzing the wave energy capture power according to the coupling characteristics of the system. According to the analysis method, the problem that the motion response of each degree of freedom cannot be accurately described when the hydrodynamic response characteristics of the complex multi-body system are obtained through an existing simulation method is solved.

The invention belongs to the field of hardware neural network design, and particularly discloses a CMOS-based attractor recurrent neural network system and an implementation method thereof.A neuron calculation module receives an external input high-level signal and generates an initial synaptic high-level signal; integration and activation calculation are carried out based on the current output by the synaptic calculation module, and whether a synaptic high-level signal is generated or not is determined according to an activation state; the synaptic storage module writes the non-zero weight into a synaptic weight SRAM (Static Random Access Memory) array; the synaptic calculation module performs product operation on the synaptic high-level signal and the synaptic weight in the synaptic weight SRAM array, and outputs current; when the STDP updating module is in a neuron training mode, the STDP updating module updates the synaptic weight SRAM array according to the synaptic high-level signal; and the group state module outputs a neuron sequence number corresponding to the maximum activation frequency. A cyclic connection structure of the attractor neural network is formed, and the real-time cognitive behavior requirement of the brain is met.

The invention relates to the technical field of building energy consumption analysis, and discloses a building energy consumption and carbon neutralization analysis system based on big data analysis, which comprises an acquisition unit, an analysis unit, a calculation unit, a deficit unit, a judgment unit and a carbon neutralization management unit, and is used for generating a carbon neutralization management instruction of a target object according to a carbon neutralization urgency index, performance degradation of a target object along with increase of service life is captured, an analysis unit combines real-time carbon emission data of a power grid, generates a dynamic carbon track slope and reflects a carbon emission change trend in real time, a calculation unit performs segment processing on operation duration based on big data, and a complete-cycle carbon track vector is constructed. A healthy deficit degree and a stepped urgency index are generated through big data analysis of the deficit unit and the judgment unit, and finally a management instruction is output by the management unit, so that the problem of insufficient tracking of full-life-cycle energy consumption attenuation and carbon emission increment of a target object by traditional analysis is effectively solved, and the accuracy of long-term carbon neutralization planning is greatly improved.

The invention relates to the technical field of point cloud semantic segmentation, and discloses a point cloud segmentation method and system based on point cloud serialization and a Mama network. Specifically, a geometric distance matrix and a semantic distance matrix are calculated for a plurality of representative point units obtained by sampling on an original point cloud, and a path sequence is generated based on a fused comprehensive distance matrix. And traversing the path sequence by adopting a bidirectional scanning strategy to generate a forward sequence and a reverse sequence, extracting features from the forward sequence and the reverse sequence by utilizing a Mama network, cascading or fusing the extracted features, and finally performing classification based on the fused features. According to the method, better long-range dependency and local context modeling effects can be obtained, the comprehensive expression capability of detail features and global context in sparse and irregular scenes is enhanced, and the calculation and memory overhead required for global dependency capture of large-scale point clouds is greatly reduced.

The invention provides a mixed modeling method for the pumping and printing process of 3D printing concrete. According to the method, intelligent aggregate is introduced, real-time monitoring and data collection of particle motion behaviors in the 3D concrete pumping and printing process are achieved, and comparative analysis is conducted on collected multi-dimensional dynamic data of particle speed, acceleration and collision frequency and discrete element-computational fluid mechanics (DEM-CFD) coupling simulation results; therefore, efficient inversion and accurate calibration of input parameters of the DEM-CFD coupling simulation model are achieved, and compared with unobservability and uncertainty of the particle flow state in the traditional concrete pumping process, the method can achieve transparency in the particle movement process; the parameter inversion method provided by the invention is based on fitting optimization of actually measured particle motion data and a numerical simulation result, so that manual intervention and experimental overhead are remarkably reduced.

The invention, which relates to the technical field of mine geology and urban safety, discloses a refined collaborative prediction method and system for subsidence of a global goaf, comprising the steps of establishing a hierarchical storage system, and correcting a working face calculation boundary; extracting the maximum influence radius of all the working faces and four-dimensional coordinates of the mining boundary of each working face, and calculating to obtain a global prediction range to generate a global surface prediction grid; point set insertion and Delaunay triangulation are adopted to obtain an effective triangular unit, a local influence domain range of the effective triangular unit is calculated, and an influenced sub-grid is screened from the global surface prediction grid; and calculating a predicted point movement deformation value based on a parallel computing architecture, accumulating the predicted point movement deformation value to a global result array through a coordinate index mechanism, synthesizing a deformation component in a specific direction, generating a visual result, and performing statistical analysis. According to the method, the problems of poor adaptability and low efficiency of a traditional scheme are effectively solved, and the prediction precision and efficiency are remarkably improved.

The invention discloses a machine tool dynamic characteristic finite element calculation method based on bolting joint surface parametric modeling, and belongs to the technical field of bolting joint surface and machine tool overall modeling and simulation. According to the method, a gantry machine tool reduced scale model is taken as an example, contact parameter calculation is conducted on bolting joint surfaces based on the fractal theory, then finite element parametric modeling is conducted on the bolting joint surfaces through an abaqus basic connector unit, and dynamic performance analysis of the whole machine is completed. Compared with a test method, the accuracy of the method is proved. According to the method, more accurate and more efficient calculation of the dynamic performance of the machine tool is realized, and a theoretical basis is provided for subsequently exploring the influence of each combination surface on the dynamic performance of the whole machine.

The invention relates to the technical field of rolling mill automatic control, in particular to a special rolling mill controller without a PLC framework and an integrated control system, and aims to solve the problems that in rolling of a rolling mill, the sampling rate difference of pressure and thickness sensors and material plastic deformation accumulation cause the time sequence dislocation of a rolling force peak value and a thickness valley value, the collaborative utilization of data by a thickness control algorithm is influenced, and the cost is reduced. In order to solve the problems that rolling force, thickness data and roller pulse signals are collected through a special controller, and a rolling length observation benchmark is converted; the dynamic phase compensation unit calculates a plastic deformation cumulant, after triggering compensation, a basic phase difference is established based on a sampling rate difference, a compensation coefficient is dynamically increased, a compensation data segment is generated through linear interpolation, and a time sequence is forcibly aligned; and the integrated control unit inputs the alignment data into a thickness control algorithm, the rolling force serves as feedforward, the thickness data serves as feedback, a precise roller gap adjusting instruction is generated, and the material thickness rolling precision is improved.

The invention discloses a 3D modeling calculation auxiliary system based on natural language driving, in the system, a basic information input unit is used for receiving a modeling request input by a user, and the modeling request comprises description of specific scene modeling requirements of the user in a natural language form; the semantic recognition unit performs semantic recognition on natural language input information to extract structural modeling calculation parameters, and performs standardized output; a corresponding modeling tool flow is called through a 3D modeling unit according to the semantic recognition result of the large model, and parametric modeling is carried out; the finite element calculation unit can perform data interaction with the 3D modeling unit, geometric information established by the 3D modeling unit is input into finite element calculation software, calculation parameter structures serve as other parameters to be input into the finite element calculation software, corresponding finite element models are established according to the calculation parameter structures, and solving calculation is performed. According to the system, a hybrid architecture of natural language semantic link parameterized workflow is adopted, so that the learning cost of parameterized modeling, finite element calculation and special calculation requirements can be effectively reduced, and the efficiency of modeling and calculation is improved.

The invention belongs to the technical field of optoelectronic equipment testing, and particularly relates to a method for testing irradiation intensity of an area light source device, which comprises the following steps of: firstly, dividing the area light source device into a plurality of test areas, and numbering each area; secondly, acquiring the irradiation intensity of each test area in real time by using a high-precision optical sensor array; then, the collected analog signals are converted into digital signals through an A / D conversion circuit, and the digital signals are transmitted to a data processing unit; then, the data processing unit calculates the irradiation intensity value of each area, and analyzes the uniformity and consistency of the overall irradiation intensity distribution; and finally, displaying a test result through an upper computer control system, and generating an irradiation intensity distribution diagram. Through the high-precision sensor array and the data processing technology, the irradiation intensity distribution of the area light source device can be rapidly and accurately tested, a reliable basis is provided for light source performance optimization, and the method has the advantages of being efficient, accurate and high in practicability.

A method may include: receiving a prompt and generated text from the LLM; computing an original token probability distribution for each token in the prompt and in the generated text; receiving a token position probability distribution for each token position in the generated text from the LLM; identifying keywords in the prompt; perturbing embedding vectors for the keywords used by the LLM by adding noise to the embedding vectors; computing a perturbed probability distribution for the perturbed embedding vectors by providing the perturbed embedding vectors as an input to a neural network used by the LLM, wherein the neural network returns a perturbed token probability distribution; evaluating a divergence between the original token probability distribution and the perturbed token probability distribution; identifying semantically meaningful tokens in the generated text; calculating a mean of divergences for the semantically meaningful tokens; and classifying the LLM based on the mean of divergences.

The invention relates to a real-time path planning method and system based on conflict resolution, equipment and a medium. The method comprises the following steps: preprocessing original observation data to obtain an evidence unit of each sensor; calculating a total conflict amount and a sensor pair contribution matrix based on an evidence unit, updating the credibility of each sensor and selecting a fusion operator to obtain fusion evidence and a conflict degree after fusion; a short-term time sequence is generated according to the fusion evidence and the conflict degree after fusion, the expected occupation probability and the expected conflict trend of each grid unit are predicted, and a time sequence conflict map is obtained; according to the time sequence conflict map, conflict-level adjacent grid unit aggregation fuzziness and area-level conflict characteristics are carried out; and inputting the fusion evidence, the conflict fluctuation degree, the expected stable time and the region-level conflict feature as cost and constraint into a real-time local path planner to obtain a control decision. By adopting the method, the problem of path decision oscillation caused by multi-source sensor conflicts can be relieved.

The invention discloses a seismic simulation method and system fusing an implicit iterative graph network and a spectral element method. The method comprises the following steps: acquiring acceleration data of a small time step node; performing grid sparse division and interpolation mapping processing, and projecting to a low-dimensional manifold to obtain large-time-step node acceleration data and a low-order unit calculation force matrix of the current frame; obtaining a physical relation sparse matrix I, a physical relation sparse matrix II and sparse matrix topological relation information of the physical relation sparse matrix I and the physical relation sparse matrix II through a processing method of Jacobian iteration and projection to a low-dimensional manifold; obtaining a seismic simulated diagram neural network model; and predicting by using the seismic simulation diagram neural network model, and predicting the acceleration by using the large time step of the sparse unit nodes projected back to the target area by using the projection matrix. According to the method provided by the invention, the mapping between the high-order unit small time step and the low-order unit large time step is established through the fitting capability of the neural network, so that the calculation time is greatly shortened, and the earthquake simulation is quickly carried out.

Systems and methods to receive a text sentence and an identification of a class, calculate a bias direction with respect to the class for an embedding model used by an artificial intelligence model to be analyzed, calculate a protected gradient score for each token in the text sentence, aggregate the protected gradient scores of the tokens to form a sentence-level protected gradient score for the sentence, determine a fairness indicator of the AI model based on the sentence-level protected gradient score, and responsive to the fairness indicator being below a threshold, prevent deployment of the AI model. Intelligent detection of bias of an AI model with respect to a protected class is determined via a protected gradient score for the protected class at a token level of a text sentence.

The invention relates to the technical field of network-on-chip data storage, and discloses a network-on-chip data consistency system, which comprises an intra-core synchronization unit, a computing core and an inter-core synchronization unit, according to the system, methods of atomic operation and synchronous lock are adopted, the memory access authority of an on-chip core and the memory access authority of a local core are managed in a unified mode, and ordered read-write of shared storage is achieved. For the local core, memory access of the on-chip core is controlled by initializing an inter-core lock, scheduling of a thread bundle is controlled by initializing an in-core lock, and a producer-consumer data mode of the local core and the on-chip core is realized; and for the on-chip core, the integrity of the continuously written data is ensured through the inter-core lock according to the memory access authority obtained by the atomic operation. The core of the application is to avoid the out-of-order problem of inter-core read-write operation through intra-core locking, inter-core locking and atomic operation, ensure the synchronization of data in shared storage while ensuring efficient execution of each core, and improve the overall parallel performance of the architecture.

The invention is suitable for the technical field of flood monitoring, and provides a flood inundation area dynamic monitoring method based on time sequence feature enhancement, and the method comprises the steps: fusing optical images, synthetic aperture radar images and surface temperature time sequence data before and after a flood event; calculating the water existence probability of each time point pixel by pixel through a random forest model by using the fused time sequence features; pixel volatility is calculated based on the water body probability time sequence, and the flood process is divided into a dramatic change period, a slow change period and a stationary period; a three-dimensional convolutional neural network is used for mining a spatio-temporal evolution law of submerging and water recession in a dramatic change period, a time sequence and spectrum collaborative interpretation model is used for separating progressive water body and vegetation interference in a slow change period, dimensionality reduction is performed on a water body probability sequence in a stationary period, and a lightweight time sequence convolutional network is used for simulating a water recession rate; and optimizing the identification result of each stage, and finally generating a dynamic submerging map. The accuracy of monitoring the submerged area is effectively improved, and the real-time response efficiency of monitoring is improved.

The invention provides an energy-saving unloading method and device for meta-computing tasks, equipment and a storage medium. The method comprises the following steps: adding white noise into a task signal to form a sending signal; capacity is calculated according to signals received by the main channel and the eavesdropping channel, and then the secrecy rate and the traversal value and the lower limit of the secrecy rate are determined; calculating power consumption and delay between a plurality of meta-calculation users and the node through traversing the secrecy rate; constructing a bipartite graph to determine an unloading path; and calculating the optimal power of each path according to the lower limit of the traversal secrecy rate. According to the method, task signals and white noise are combined, the secrecy rate is calculated through channel information, power consumption, delay and unloading paths are determined according to the secrecy rate, the optimal power is calculated, and safe, efficient and energy-saving unloading of calculation tasks is achieved. The whole process aims at optimizing the unloading strategy of the meta-computing task, improving the system efficiency and reducing the energy consumption.

The invention discloses an offset automatic calibration method for defect source analysis, and relates to the technical field of wafer detection and defect source analysis, and the method comprises the steps: determining the geometric center position of a wafer according to a theoretical coordinate matrix and an actual coordinate matrix of a chip unit in a wafer detection image, and extracting a candidate chip unit; calculating comprehensive stability scores of the candidate chip units, screening out an optimal reference chip unit, and calculating coordinate deviation values of the optimal reference chip unit in the actual coordinate matrix and the theoretical coordinate matrix; performing weighted average on the coordinate deviation value through a symmetry constraint condition to obtain a local calibration offset, and expanding the local calibration offset into a global offset calibration parameter; and applying the global offset calibration parameters to defect positioning calculation in an AOI detection process, and comparing a detection result with a preset verification chip unit to analyze calibration precision so as to complete offset automatic calibration. According to the invention, the consistency and reproduction precision of defect positioning are improved.

The invention relates to the technical field of electrical equipment manufacturing, and discloses a high-low voltage electrical cabinet continuous processing abnormity rapid response system. The system comprises a processing data acquisition module for acquiring sensor signals, process parameters and other elements and establishing a time sequence processing data unit; the edge calculation preprocessing module retrieves preprocessing rule terms according to the time sequence data unit and calculates a preprocessing complexity factor to judge data quality; the abnormal feature extraction module is used for calculating a cross-time-period feature association closeness index based on the multi-time-series data unit and constructing a multi-time-period feature association map; and the response strategy generation module is used for judging and generating a comprehensive abnormal response prompt in combination with the map and the data preprocessing state. In addition, the system can also be provided with an abnormal data storage module and an edge node cooperation module, abnormal data storage and cross-node cooperation analysis are realized, processing abnormity can be quickly responded, and the processing efficiency and accuracy are improved.

The invention relates to a system for determining the compaction of a flowable, solidifiable building material, in particular concrete, during the production of a structural element with the building material, wherein the flowable building material is introduced into a formwork system (10), comprising: at least three sensors (S1, S2, S3) to determine an amplitude (A S1 , A S2 , A S3 ) a vibration introduced into the flowable building material by a vibration unit (VE), wherein the sensors (S1, S2, S3) are arranged on the formwork system (10), at least one computing unit that is set up to perform calculations based on the amplitudes (A) determined by the at least three sensors (S1, S2, S3). S1 , A S2 , A S3) to calculate possible positions (D1, D2, D3) of the vibration unit (VE) relative to the respective sensors (S1, S2, S3) and, based on this, to triangulate the current position of the vibration unit (VE) relative to the formwork system (10), a documentation unit for documenting the dwell time of the vibration unit (VE) in the respective current positions of the vibration unit (VE) relative to the formwork system (10), the computing unit determines a calculated compaction value in the respective current positions based on the respective current positions of the vibration unit (VE) and the documented dwell time in the respective current positions.

Embodiments described herein provide a training framework for generative NLP models that operate on previously learnt knowledge from pretrained large language models. Specifically, to train an NLP model to generate a response to a user utterance (e.g., “resolve login issue”), document embeddings of support IT documents encoded by a pretrained LLM are fed to an NLP decoder together with a training dialogue (e.g., a dialogue between the chat agent on how to “resolve login issue”). The NLP decoder can thus be trained by a causal language modeling loss computed based on the predicted next token and the ground-truth token from the training dialogue.