9 results about "Post-hoc analysis" patented technology

The invention discloses an intelligent management system and method for expressway comprehensive services, and belongs to the technical field of intelligent management. A first construction module obtains multi-source heterogeneous data and preprocesses the multi-source heterogeneous data, a feature engineering module carries out feature engineering processing on the processed multi-source heterogeneous data to obtain industry features, and a second construction module carries out feature engineering processing on the industry features; the second construction module inputs the industry characteristics into the trained service model to obtain a service result, the large model module obtains a vertical industry large model based on a large language model, and the engine construction module constructs five service decision engines based on the multi-source heterogeneous data, the industry characteristics, the service result and the vertical industry large model. According to the method, multi-source heterogeneous data can be deeply fused, a unified industry data base can be constructed, post-analysis is changed into pre-prediction and in-process intervention, the perspectiveness and predictability of management are improved, the whole process from analysis to execution can be automatically completed by constructing a decision closed loop, and fine and automatic intelligent management is achieved.

This invention discloses a method and system for tracing data in the separation process of radiopharmaceutical preparation, including: data reliability processing: performing anti-interference repair and verification on the collected data, and outputting repaired and verified data; correlation map construction: performing correlation analysis on multi-source heterogeneous data in the separation process, generating a correlation map defined with data entities as nodes; process deduction and inverse inference: solving the symbolic differential equations describing the dynamics of the separation process, and outputting deduction data and parameter deviation information; intelligent traceability application: based on the deduction data and parameter deviation information, performing data retrieval and attribution analysis on the separation process, and outputting full-link traceability results. This invention transforms the traditional discrete, passive, and manual-dependent data monitoring and post-analysis mode into an active, continuous, and automated intelligent analysis paradigm. The system can autonomously complete the repair of original data and the correlation of multi-source heterogeneous data.

The invention discloses a power distribution network operation state evaluation and fault early warning method and system based on data fusion and deep learning. The method comprises the following steps: collecting and preprocessing multi-source heterogeneous data to obtain preprocessed data; based on the preprocessed data, high-dimensional feature extraction and fusion are carried out to obtain a fusion feature vector; performing operation state evaluation and risk quantification based on the fusion feature vector to obtain a risk index; and based on the risk index, carrying out fault early warning and positioning. According to the application, a data island is broken through, effective fusion and collaborative analysis of electrical, physical, chemical, environmental and other multi-dimensional data are realized through the deep learning model, and perception is more comprehensive; unsupervised or self-supervised learning is utilized, unknown and weak fault precursor features can be found, transformation from post analysis to pre-warning is achieved, and a pre-warning time window is greatly advanced.

The invention provides a full-stack data performance optimization method based on user behavior analysis, and the method comprises the steps: obtaining front-end user behavior data, rear-end service data and infrastructure data, and carrying out the integration to generate an initial full-stack data set; the current situation that data are isolated in a traditional method is broken through. And performing data standardization on the initial full-stack data set according to a preset index system and a health degree scoring algorithm to generate an intermediate full-stack data set. And the problems of disordered data formats and incompatible indexes are eliminated. And performing real-time streaming processing on the intermediate full-stack data set to generate a target full-stack data set. The full-stack performance fluctuation can be captured in real time, and post analysis is avoided. And associating the data in the target full-stack data set through a full-stack data association analysis engine, and performing anomaly detection on the associated data to generate alarm data and optimization suggestion data. The problem root is accurately positioned in combination with anomaly detection, and the optimization suggestion covering the whole stack is output, so that the problem that the single-level optimization effect is poor is solved.

The application discloses a model training detection method and an electronic device, relates to the technical field of model training detection, and comprises the following steps: obtaining training task execution data of a target model and resource state data of a target operation entity where the target model is located; determining a current training trend of the target model and comparing the current training trend with an expected training trend of the target model; determining a training resource support degree of the target operation entity to the target model and comparing the training resource support degree with a resource pressure intensity of the target model; in response to the fact that there is a deviation risk in training and there is no imbalance risk in resource supply, generating a training correction parameter to correct a training process of the target model and obtaining a training correction result; and in response to the fact that the training correction result represents training correction failure, determining that the target model training is abnormal. The application can realize efficient detection of abnormalities in the model training process, and at least solves the problems of scattered detection information, lagging behind in post-analysis positioning and misjudgment of recoverable risks as abnormalities in related technologies, thereby reducing resource waste.

The invention relates to a method and a system for predicting a disinfection by-product generation risk based on fusion of a graph neural network and a transformation network. The method comprises the following steps: performing non-targeted analysis on a source water sample of a target water body through a high-resolution mass spectrum before and after disinfection to obtain an organic matter molecular formula list; constructing a molecular conversion network based on molecular formula differences, and identifying a key conversion path through topology analysis; converting the chemical structure of the key path related molecule into a molecular map; training a predictive map neural network model in combination with the molecular map and disinfection reaction condition data; and performing risk prediction on new organic matter molecules or reaction conditions by using the trained model to obtain a prediction result of the generation risk of the disinfection by-product. By adopting the method, a closed loop from panoramic conversion path identification to individual molecular risk accurate prediction can be realized, and the problem that a traditional method can only carry out post analysis and cannot carry out pre-warning in advance is effectively solved.

The invention relates to the technical field of a silicon steel normalizing process, in particular to an in-situ analysis method for the silicon steel normalizing process. The method comprises the following steps: processing a silicon steel sample into a thin sheet and polishing; placing the sample in a high-temperature laser confocal microscope, and setting temperature, atmosphere and scanning parameters; in the whole stages of heating, heat preservation and cooling in the normalizing process, the microscope is used for carrying out real-time and in-situ observation on the microstructure of the sample, and image data are collected at high frequency; and finally carrying out quantitative analysis on the image data. According to the method, dynamic and continuous in-situ observation of grain evolution and second-phase particle behaviors in the silicon steel normalizing process is realized for the first time, the limitation of information loss of a traditional post analysis method is overcome, and the instantaneous influence of process parameters on a microstructure can be accurately revealed; and direct and accurate data support is provided for optimizing the normalizing process to improve the magnetic performance of the silicon steel.

According to the front-end project code quality detection method and the program product provided by the invention, the target code is analyzed from the technical architecture dimension, the coding specification dimension, the maintainability dimension and the security dimension, the scoring index is obtained, and the code quality can be comprehensively and meticulously evaluated. By means of the multi-dimensional evaluation mode, problems of codes in different aspects can be found more accurately, richer quality information is provided for developers, and the developers are helped to carry out code optimization and improvement in a targeted mode. The front-end project code quality detection method can play a role in the project research and development process, changes the traditional post analysis strategy, prepositions the quality detection link, effectively solves the hysteresis problem in the prior art, can find and intercept the problem in the early stage, and prevents the quality risk from being exposed in a delayed manner and even brought into the production environment.

The invention discloses an oil-gas exploration-oriented intelligent research platform and a well location deployment risk early warning method, and the platform constructs a'data-knowledge-decision 'three-layer intelligent architecture, and fuses multi-source heterogeneous data through metadata-driven data governance and a multi-modal knowledge graph. The core lies in that a geological engineering integrated digital twinborn environment is adopted, a risk early warning model of a deep learning method based on a multi-scale feature fusion network and physical mechanism constraints is operated, and risk quantitative evaluation of the whole life cycle of well location deployment is achieved. According to the method, an early warning-diagnosis-optimization intelligent decision closed loop is innovatively introduced, the model is continuously evolved through an online incremental learning mechanism, normal form transformation from passive response to active early warning and from post-event analysis to beforehand optimization is finally achieved, and the exploration success rate and operation safety are remarkably improved.