14 results about "Incremental computing" patented technology

Methods, systems, and devices for storing and retrieving checkpoint values using delta computation and compression are described. Techniques described herein may enable a memory system may compute deltas between initial checkpoint values and updated checkpoint values associated with values of weights of an artificial intelligence (AI) model. The memory system may compress and store the deltas. In some examples, the memory system may perform delta computation and compression simultaneously while receiving the updated checkpoint values (e.g., without storing the updated checkpoint values in a volatile memory of the memory system in accordance with a first operation mode) or after receiving the updated checkpoint values (e.g., after storing the updated checkpoint values in a volatile memory of the memory system in accordance with a second operation mode). In some examples, the memory system may switch between the first operation mode and the second operation mode.

The present application relates to the technical field of computer model generation, and is a concrete parameter optimization model generation method and system, specifically comprising: establishing a digital twin evolution model fusing prestressed vector field and heat conduction gradient; extracting effective axial force prediction value and interface debonding displacement increment changing with fire duration based on the digital twin evolution model, calculating and obtaining dynamic debonding sensitivity coefficient, synergy factor and topological redundancy response coefficient; generating structure topological robustness maintenance rate; online evaluating deviation tolerance of maintenance rate generated by the current model and target maintenance rate, and adjusting concrete aggregate gradation parameter combination. The present application solves the problem in the prior art that the dynamic collapse risk of the effective contact network of aggregate cannot be characterized, it is difficult to give a gradation optimization boundary with thermal and mechanical robustness, and a technical blind area exists in the safe operation and maintenance of the structure.

The application provides a vehicle multi-sensor fusion positioning method and device based on extended Kalman filtering, equipment and medium, comprising: caching DR data to a first data buffer and caching VO data to a second data buffer; performing relative pose increment calculation on the DR data and the VO data in the cache area to determine the relative pose increments of the DR data and the VO data at adjacent time points; adaptively adjusting a VO observation noise covariance matrix based on the relative pose increments; and performing fusion processing on the adjusted VO observation noise covariance matrix and DR observation data based on sequential update extended Kalman filtering to determine the final fusion positioning result of the vehicle. Through the closed-loop fusion architecture of "relative pose increment calculation-adaptive observation noise adjustment-sequential update EKF", the positioning accuracy, robustness, real-time performance and engineering reliability are synergistically improved.

Embodiments of the present invention provide a method, apparatus, electronic device, medium, and chip for determining data association increments. The method includes: acquiring current state fingerprints corresponding to multiple data tables in a target database; determining a state change feature vector corresponding to each data table based on the current state fingerprints and local state data; determining at least one first-type data table and at least one second-type data table; determining a detailed data feature set based on the at least one first-type data table; acquiring a historical key data feature set corresponding to each second-type data table from locally cached data; determining a candidate association set based on the detailed data feature set and the historical key data feature set; and determining the association increment result corresponding to the first-type data table based on the candidate association set. Through the solution of the present invention, accurate perception and incremental calculation of data changes are achieved, thereby significantly improving the efficiency of association discovery.

A system includes at least one hardware processor and at least one memory storing instructions that cause the at least one hardware processor to perform operations, such as configuring, at a first computing node, a feature registry based on a plurality of feature definitions received via a user interface associated with an application programming interface. The operations include obtaining, at a second computing node, decoded raw data comprising streaming data and batch data, and performing incremental computation of features using a dependency graph comprising dynamic table objects and using configuration data from the feature registry. The operations include maintaining, as a feature store, a low-latency serving storage for computed feature values and a versioned historical storage. The operations include updating the low-latency serving storage using an event-driven task, controlling retrieval by a computing node based on a retrieval application programming interface, and backfilling historic results using a version-aware query.

This invention discloses a knowledge graph inference optimization system and method based on real-time gradient-sensitive pruning and incremental computation. The system includes a dynamic pruning module, an incremental inference module, and a dual-loop feedback control module. It adopts an FPGA-GPU heterogeneous architecture, quantifies the importance of nodes and edges through a real-time gradient-sensitive analysis algorithm, and generates a dynamic pruning mask. A dual-channel graph attention-gated network is constructed to achieve adaptive extraction of topological features, and historical inference results are reused using an incremental computation mechanism. The dual-loop feedback control mechanism dynamically adjusts the pruning rate to balance inference accuracy and latency. This invention effectively reduces redundant computation, improves the real-time performance of dynamic graph inference, achieves a stable pruning rate of 47.6% ± 2.1%, an end-to-end inference latency as low as 9.1 ms, and an F1 retention rate of 98.2%, demonstrating significant application value in fields such as medical diagnosis and financial risk control.

This invention discloses a hybrid decoding method for laser seekers, comprising the following core modules: a high-speed main channel, an auxiliary verification channel, and a joint decision unit. The main channel rapidly scans the signal stream using a digital matched filter, triggering a local window upon detecting a candidate frame. The auxiliary channel performs incremental sliding autocorrelation analysis within the window, efficiently verifying the signal structure integrity using an incremental recursive formula. The joint decision unit determines the valid signal based on a dual-dimensional threshold of the matched peak value and the autocorrelation peak value. This invention significantly improves decoding robustness under non-ideal channel conditions, while ensuring real-time performance through local window constraints and incremental calculations, making it suitable for high-interference scenarios such as laser guidance.

This invention discloses a method for processing error data of smart energy meters based on edge computing, belonging to the technical field of error data processing methods. The invention includes: S1, calculating the mean and standard deviation of the total energy increment sequence within a sliding window, defining a first-order autoregressive hysteresis correction term, setting an adaptive threshold, defining the state machine system state, and finding the computer increment sequence during inactive periods by constructing an inactive period criterion; S2, storing the filtered effective data increments, calculating the energy increment of each sub-meter, defining a continuous zero increment detection function, collecting effective rearranged data points by setting the meter failure detection window and parameters, and constructing an observation matrix X and an observation vector Y; S3, using an improved genetic optimization algorithm framework, calculating eigenvalues ​​and condition numbers, and iteratively obtaining optimal and distinct individuals; S4, determining the optimal regularization parameters using the L-curve method, and calculating the error coefficients of each energy meter using improved Tikhonov regularization.

The application discloses an industrial safety management and control system and method based on a risk field and distributed autonomy. It relates to the technical field of industrial safety intelligent control, spatial intelligent modeling and complex system autonomous control, and comprises a multi-source sensing and collecting unit, a spatial risk field construction unit, a risk standardization processing unit, a nonlinear risk coupling calculation unit, a risk potential energy conversion unit, an autonomous evolution control unit, a distributed collaborative stability unit and a real-time incremental calculation and computing power scheduling unit. The application maps multi-source risks to a voxelized spatial domain by constructing a three-dimensional spatial risk field; introduces a nonlinear coupling matrix to describe the enhancement or inhibition relationship between risks, calculates the total risk field; constructs a risk potential energy function, triggers control actions based on the gradient descent control law; determines the system stability through the system total energy function and its time derivative; solidifies parameters and maintains local autonomy when communication is interrupted; and adopts voxel-level incremental updating and dynamic allocation of computing power.

This invention discloses a method, system, terminal, and medium for predicting and analyzing urban gas pipeline network failures and gas supply. The invention proposes the concept of a pipe segment, abstracting and aggregating pipe sections, fittings, and valves in the gas pipeline network into specific data structure nodes to construct a pipeline network relationship graph. When pipeline network data changes or failure simulations are performed, a local graph update strategy based on a large language model is adopted. Utilizing the semantic understanding of the graph node attributes and connection relationships of the large model, incremental calculations or corrections are performed only on affected local nodes and edges, rather than reconstructing the entire graph. Simultaneously, combining GIS data and logical rules, algorithms are used to automatically clean and correct abnormal data regarding valve opening / closing logic and gas-carrying status of pipe sections. This effectively solves the problem of excessively long calculation times for full-scale calculations of pipeline networks with millions of nodes. It enables rapid completion of valve closure analysis, single-pipe gas supply determination, and failure impact range assessment even with frequent pipeline network updates, meeting the real-time requirements of emergency command.

The application discloses a modal identification method based on an unsupervised optimization covariance stochastic subspace method, and comprises the following steps: arranging vibration sensors on a to-be-measured structure with noise interference or weak excitation modes, and acquiring structural dynamic response data; calculating a covariance matrix and constructing a toply matrix; calculating an extended observable matrix and a controllable matrix based on the weighted toply matrix; defining a toply matrix row block number and a model order parameter range, and iteratively calculating a parameter optimization index; calculating a cumulative contribution rate based on a singular entropy increment of the toply matrix and determining a critical model order; calculating a cumulative parameter optimization index in an interval from a minimum model order to the critical model order under different toply matrix row block numbers, and determining an optimal parameter combination through an average minimum value; substituting the optimal parameter combination into a covariance stochastic subspace algorithm to identify a system matrix and calculate structural modal parameters; and automatically identifying each order physical mode from candidate modes by using a DBSCAN clustering method. The application further discloses a system and an electronic device.

This invention discloses a streaming tag computation method based on event-driven and metadata-driven approaches, addressing issues such as poor real-time performance in IPTV tag management, high rule modification costs, untraceable status, and difficulties in tag system expansion. This invention constructs a real-time data stream access platform and a tag rule metadata repository, decoupling rules from computation tasks; it generates real-time computation tasks based on metadata-driven approaches and completes incremental tag computation through event-driven mechanisms; it supports dynamic hot updates of rules, establishes a closed-loop management system for the entire tag lifecycle, and designs a unified data model and hybrid storage architecture for multiple tag types. This invention pioneers a collaborative mechanism between event-driven streaming incremental computation and metadata-driven hot updates, achieving tag synchronization latency ≤1 second, a 90% improvement in rule modification efficiency, and 100% traceability of status transitions, providing IPTV operations with real-time, flexible, and controllable tag capabilities.

The application discloses a key screening method, system and device for continuous variable quantum key distribution, and relates to the technical field of quantum communication security. The method comprises the following steps: initializing a model and saving training statistics; calculating an abnormal score for training data block by block and filling a sliding window; calculating a dynamic threshold value in a training stage; completing initial screening in the training stage; performing incremental LoOP calculation on each new key block; dynamically updating the threshold value of the sliding window and making a screening decision; and the method adopts incremental LoOP, so that each new key block only needs to perform kNN query once to obtain an abnormal score, thereby avoiding the problem of repeated training of the model and realizing real-time detection capability at the level of quantum key blocks.

This invention relates to an incremental calculation method for a bidirectional skip list of the Jacobian matrix in FPGA power flow, comprising: designing a skip list of non-zero elements of the Jacobian matrix; reading binary power grid branch data sequences, concatenating the read binary power grid branch data sequences using a shift register to obtain binary power grid branch data; and dynamically incrementally calculating the non-zero element structure of the Jacobian matrix based on the binary power grid branch data and the skip list of non-zero elements of the Jacobian matrix. This invention can improve the calculation efficiency of the Jacobian matrix.