14results about How to "Avoid performance bottlenecks" patented technology

The application discloses an Internet of Things mass service bill processing system, method and device and a storage medium, and comprises an application software service layer, a platform environment service layer and an infrastructure service layer.The application software service layer is used for collecting first service bills, and performing format analysis, deduplication, merging and distribution processing on the first service bills to obtain first bill files and first bill messages, and then uploading the first bill files and the first bill messages.The platform environment service layer is deployed with a distributed file system, a distributed coordination service, a data database and a streaming processing framework, is used for receiving the first bill files and the first bill messages, and performing full-amount synchronization on the data database according to the first bill files and incremental synchronization on the data database according to the first bill messages.The infrastructure service layer comprises a bottom cloud host, and is used for providing a computing resource pool, a storage resource pool and a network resource pool.The application improves the processing efficiency of service bills, optimizes the use perception of users, and can be widely applied to the technical field of data processing.

The application discloses a user behavior analysis method, which can deeply understand the operation mode and behavior characteristics of users by modeling the user behavior sequence using an RNN model, thereby optimizing the user interaction experience; the ARIMA model is used to model the page performance data, which can accurately capture the time sequence characteristics of the page performance, identify potential performance bottlenecks and change trends; the RNN model is introduced into the ARIMA model, the page performance data is taken as an input of the sequence model, and the user behavior data is analyzed together, so that comprehensive analysis of the user behavior and the page performance can be realized, and the prediction accuracy and practicability of the model are improved; the trained model parameters are used to predict the optimization scheme and optimization cost of a future time point, so that the possible performance problems and optimization requirements in the future can be predicted, the optimization scheme can be prepared and implemented in advance, and potential performance bottlenecks can be avoided.

This application provides a user mining method and apparatus, electronic device, program product, and storage medium. The method includes: acquiring user data through public API interfaces of multiple social networking platforms and storing the acquired user data in an intelligent computing performance-enhanced file storage service deployed on a cloud computing platform; using a data processing cluster on the cloud computing platform to read the user data from the intelligent computing performance-enhanced file storage service, reconstructing features based on the user data, and storing the reconstructed feature data back to the intelligent computing performance-enhanced file storage service; constructing and training a reinforcement learning model on the cloud computing platform based on the reconstructed feature data; storing historical experience data of the agent in the intelligent computing performance-enhanced file storage service during the training process; deploying the trained reinforcement learning model to a production environment and using the reinforcement learning model to perform user mining based on real-time acquired user data.

The application discloses a data read-write processing method and an electronic device in a storage system, relates to the technical field of computer storage, and realizes the asynchronization and isolation of read-write operations by constructing a three-layer resource processing architecture, decoupling host read-write requests into independent controllable processing stages. The first resource focuses on recording metadata, the second resource serves as a temporary carrier for data transfer, and the third resource serves as a persistent cache management center. This architecture enables the write operation to return a response to the host as soon as the data reaches the second resource, greatly reducing the write delay; the read operation realizes non-blocking access through the concurrent control mechanism of the third resource. Overall, the response speed and throughput of the storage system are improved, and the performance bottleneck of mutual blocking of read and write due to competition for the same data resource in the related architecture is solved.

The application belongs to the technical field of databases, and specifically discloses a multi-mode time series data processing method and system. The method comprises the following steps: establishing a row storage table and a column storage table according to a table creation request, and establishing an associated mapping relationship between the row storage table and the column storage table; receiving time series data to be inserted and extracting label data and index data, determining a target data node based on label data hash calculation, writing the label data into the row storage table, and writing the index data into a target time series data block of the column storage table; performing partition pruning and scanning on the stored time series data based on a screening condition, and generating a query result by parallel execution of split query tasks and merging of results based on cross-mode query requirements; determining the affected time series data block based on an operation request, and updating the label data and the index data or deleting invalid data according to the type of the operation request. The application can realize low-redundancy storage, high-concurrency read-write and fast query analysis of time series data.

This application discloses a page update method, electronic device, and computer program product. The method includes: in response to detecting an update event triggered for a target page, obtaining a first DOM tree corresponding to the target page after the trigger and a second DOM tree corresponding to the target page before the trigger; determining nodes to be updated by comparing the first DOM tree and the second DOM tree, wherein the nodes to be updated include at least one of the following: at least one modified node, at least one node to be destroyed, and at least one node to be created; selecting a target number of nodes matching the current resource indicators from the nodes to be updated as first update nodes according to priority order; and updating the first update nodes in response to detecting an update timing, wherein the update timing is determined based on the current resource indicators.

The application discloses a TC4 titanium alloy and a preparation method and application thereof, and relates to the technical field of titanium alloys. The application provides a TC4 titanium alloy, wherein the TC4 titanium alloy is sequentially provided with a first layer, a second layer and a third layer along a thickness direction; the first layer is a coarse crystal structure, the second layer is a nano twin crystal structure, and the third layer is a nano sheet structure, so that a continuous gradient microstructure is formed, and the synergistic strengthening effect of high strength, high crack propagation resistance of the surface layer of the titanium alloy and high toughness of the core can be considered.

The application discloses a kind of high-efficiency casual neural network conversion methods, client and server side run preprocessing process in advance and obtain a set of matrix for encryption, client will generate matrix as key and use one-time encryption technology to encrypt original input data for prediction, send encrypted data to server side to calculate linear layer, server side will add Laplace noise to the result after linear layer calculation is completed and send back to client, then complete nonlinear function calculation in client, and the result is passed through noisy model Purifier, finally the result of noisy completion is returned to server side in plaintext form, server side gets plaintext and completes subsequent calculation of resnet101 network, finally completes prediction work.The method of the application can be applied to Resnet101 and other complex neural network models, and the privacy protection purpose is realized, which can greatly reduce the time required for prediction compared with the previous solution, and reduce the additional communication and additional calculation amount introduced by the encryption scheme.

This invention provides a method for automatic synchronization of multiple databases under an AI platform, belonging to the field of automatic database synchronization technology. This invention initializes the multi-database synchronization topology and deploys a change data capture module to monitor data changes in real time. The captured change records are converted into unified format synchronization messages and input into an intelligent scheduling optimization model for dependency analysis and transaction reordering based on dynamic programming. Data conversion rules are dynamically generated based on mode version differences, and Merkle tree hierarchical structure is used to verify data consistency. After conflict detection and circular dependency handling are performed at the target node, multi-level adaptive compression transmission and data writing are completed. Finally, periodic full reconciliation tasks ensure eventual consistency among the multiple databases, solving the technical problem of low transaction scheduling efficiency leading to synchronization performance bottlenecks during automatic multi-database synchronization.

This invention relates to the fields of computer vision and edge computing technology, and discloses a multi-computing-power collaborative video structured analysis method. The method includes: acquiring multiple raw video streams and synchronously decoding them; using gradient analysis to assess the quality of the decoded raw video streams and selecting a combination of preprocessing algorithms based on quality feature vectors; establishing a data transmission channel using an asynchronous communication protocol and a multi-level caching system; performing content complexity analysis and obtaining scheduling decisions through model matching based on the analysis; distributing the model and performing pipelined inference on the deployed model; fusing the pipelined inference results and verifying the consistency of the fused results; and optimizing system configuration parameters based on the video structured analysis results to achieve video structured analysis. This invention improves the real-time performance, accuracy, and system stability of video analysis.

This application discloses an adaptive multi-granularity text embedding representation learning method based on gradient optimization. The method includes: text encoding, multi-granularity embedding generation, difficulty-aware dynamic sampling, sliding window negative sample management, adaptive loss optimization, and model training and validation. It aims to address the performance degradation caused by existing embedding vector dimensionality compression techniques. Through a difficulty-aware dynamic sampling mechanism, a Beta distribution is introduced to achieve progressive learning, significantly improving low-dimensional performance; sliding window negative sample management improves sample quality and training efficiency; and a gradient-oriented update strategy avoids gradient conflicts and optimizes training stability. Furthermore, this invention supports incremental dimensionality expansion, reducing training costs while maintaining consistency across multi-granularity embeddings. This method significantly improves performance in low-dimensional environments, maintains lossless performance in high-dimensional environments, significantly improves training efficiency, greatly reduces storage costs, and exhibits strong generalization ability, making it suitable for data from various domains.

This invention relates to a custom time-delay message processing method based on a distributed message queue, belonging to the field of distributed message queue technology. The method includes: a client receiving an arbitrary delay duration specified by the user, calculating a target Unix timestamp based on the client's current system time, and sending a message carrying that timestamp to a message broker layer; the message broker layer storing the message in a delay queue and inserting it into the corresponding time slot of a hierarchical circular time wheel according to the target Unix timestamp; the hierarchical circular time wheel consists of multiple levels of circular arrays, connected by an overflow mechanism; the message broker layer advances a global pointer at fixed time intervals, scans the messages to be scheduled in the time slots, and transfers expired messages from the delay queue to a regular queue for consumer consumption. This achieves flexible support for arbitrary delay durations and achieves task scheduling with O(1) time complexity through the hierarchical circular time wheel.

The present disclosure relates to a blockchain cross-chain method, device, storage medium and electronic equipment. The method is applied to a first cross-chain service node in a first blockchain network. The method comprises the following steps: receiving a cross-chain request of a second blockchain network, wherein the cross-chain request comprises an access address of a first digital identity blockchain network and digital identity information registered by the second blockchain network in the first digital identity blockchain network; accessing the first digital identity blockchain network based on the access address and verifying the digital identity information of the second blockchain network; and performing a cross-chain operation corresponding to the cross-chain request if the verification is passed.

This invention relates to a business rule optimization method based on the linkage of model analysis and rule performance monitoring, belonging to the field of information technology. The method includes: constructing a mapping library of business rules and database entity relationships; synchronously monitoring rule execution performance and database access metrics; locating deep bottlenecks from rule logic to data access through correlation analysis; generating targeted optimization suggestions and assessing their impact; and finally implementing optimization and performing closed-loop verification. This invention overcomes the limitations of traditional optimization methods that only focus on the rules themselves, achieving linkage analysis and collaborative optimization between the business logic layer and the data persistence layer. It can proactively identify system bottlenecks, accurately propose improvement measures, significantly reduce the risk of rule iteration, and significantly improve the performance and maintainability of complex enterprise systems.