7 results about "Total correlation" patented technology

The invention discloses a hybrid neural network training method based on quantum correlation. The method comprises the following steps: performing feature extraction processing and quantum state coding processing on an obtained input data sample, and determining a quantum state data sample feature; inputting the quantum state data sample features into a target parameterized quantum circuit, and outputting quantum associated data and quantum feature data corresponding to the quantum state data sample features; determining first quantum total correlation data according to the quantum correlation data; inputting the quantum feature data into the target processing network, and determining a prediction result of the input data sample; determining a target loss function according to the first quantum total correlation data and a prediction result and a sample label of the input data sample; and training the hybrid neural network according to the target loss function. Thus, by introducing the first quantum total correlation data as the regular term of the target loss function, the redundancy correlation between the quantum state data can be constrained, and the training process is more stable.

The invention relates to the technical field of intelligent operation and maintenance data processing, in particular to an operation and maintenance information processing method and system for an information system. An operation and maintenance information processing method for an information system comprises the following steps: S1, acquiring time sequence index data, data point identification data and abnormal data point data, and constructing a sequence data set according to the time sequence index data and the data point identification data; and calculating the linear correlation strength of the sequence data set, and determining a first total correlation and a second total correlation according to the linear correlation strength. According to the invention, a multi-dimensional fault assessment system is constructed through intelligent acquisition and correlation analysis, and accurate monitoring of the operation and maintenance state is realized; based on an influence quantification and type identification dynamic optimization strategy, a historical library and real-time resource adaptive matching are combined, backspacing and compression are intelligently adjusted, a closed-loop decision mechanism is formed, and system reliability and edge adaptability are improved.

The invention discloses a multi-radar cooperative detection method and device, relates to the field of radar detection, and is used for improving the target tracking quality and reliability of multi-radar cooperative detection. According to the invention, false alarm filtering is carried out on multiple radar detection points based on signal quality evaluation, motion rationality discrimination and spatial-temporal distribution analysis; according to the effective detection information, measuring a comprehensive association distance between the detected motion state of the target and the predicted motion state in multiple dimensions; with the purpose of minimizing the total association cost, association of the detection points and the tracks is carried out; performing state fusion on different radar detection points associated with the same track to obtain fused detection information; performing multi-model motion prediction based on the fused detection information; and evaluating the detection quality of the target motion state and feeding back and adjusting the control parameters. According to the method, the false alarm rate and the target association error rate can be effectively reduced, and the target tracking accuracy and reliability in a complex electromagnetic environment and a dense target scene are improved.

The invention belongs to the field of power system first-aid repair scheduling, and provides a power distribution network rapid maintenance decision-making method and system considering incomplete fault information, and the method comprises the steps: carrying out the training process of an offline stage, carrying out the preprocessing of historical fault scene information, generating a fault scene set, calculating the similarity of any two fault scenes, and carrying out the clustering processing of the fault scenes; weak rules of all fault scenes are calculated to obtain an alternative weak rule set, and a global maintenance rule set is formed; obtaining current fault scene information, calculating the similarity between the current fault scene and the fault scene in the fault scene set, and on this basis, obtaining the neighborhood of the scene and calculating the total correlation degree; and finally, identifying an optimal rule as an optimal maintenance decision according to similarity statistics and the Bayesian theorem. According to the method, the influence of incomplete fault information of the large-scale power distribution network is fully considered, and the optimal rule in the global maintenance rule set obtained through training is recognized in a mode of combining offline training and online decision making.

The application provides a product data storage management method and system, and relates to data management. The method comprises the following steps: obtaining a test flow and a test state of a product; scoring the importance of each test item and test parameter through a knowledge base and expert experience to obtain the importance score of each test item and the importance score of each test parameter; obtaining an abnormality classification; obtaining a first score of each abnormality parameter according to the frequency and influence of abnormality; obtaining a total correlation score of each parameter of each upstream test item according to the correlation degree between the abnormality parameter of a downstream test item and the parameters of each upstream test item; constructing a data query index of the product; setting a test item displayed on a user interaction interface according to the importance score of the test item; determining the storage period of each parameter of each test item according to the parameter importance score, the first score and the total correlation score; and realizing fine management of test data.

The invention discloses a model decision mechanism analysis and optimization method for trajectory prediction of a graph neural network, and the method comprises the steps: constructing a multi-layer graph neural network model for trajectory prediction, inputting the input feature data into the multi-layer graph neural network model, and carrying out the trajectory prediction based on the model weight parameters; quantizing a total correlation value corresponding to the prediction track, decomposing the total correlation value layer by layer, and reversely spreading obtained correlation distribution from a model output layer to a model input layer to obtain correlation distribution of the input layer; and according to the correlation distribution of the input layer, quantitatively analyzing the influence of each input feature on a model output result, and adjusting an input feature weight parameter according to the output result deviation. According to the method, analysis and weight parameter adjustment are carried out based on correlation distribution of each layer of the model, accurate positioning of key elements influencing model decision is realized, the interpretability of the model is improved, a direction is provided for optimization and improvement of the model, and improvement of the reliability and robustness of the model in practical application is facilitated.

The present application relates to radar technology field, specifically to a kind of efficient ISAR translation envelope compensation method based on iterative updating, first construct the target echo complex matrix X of ISAR and envelope matrix then construct reference pulse number set, and calculate the total correlation level of pulse in set and other pulses, to determine reference pulse η, by calculating the correlation coefficient vector of reference pulse and other pulses and carrying out fitting processing, construct update weight coefficient vector E, based on this to reference pulse η as center to both ends are carried out pulse envelope offset estimation and compensation, to avoid the damage of abnormal pulse to reference envelope, improve environmental adaptability, by set overall pulse information to determine reference pulse iterative updating compensation, realize the fusion of local information and global information, improve the robustness of processing.The present application has the advantages of small amount of computation, easy to engineering implementation, effectively ensure the real-time updating of reference envelope and the efficient, robust realization of ISAR translation envelope compensation.