67results about How to "Interpretable" patented technology

The invention relates to an interpretable event prediction method based on a graph attention network and a hokes process. The method comprises the following steps of: constructing a time prediction process as a sub-graph extension reasoning process by adopting a graph attention mechanism and a Horkes process in an event prediction time sequence point process, and taking a to-be-predicted node as a sub-graph base point; the attention value and a conditional intensity function value obtained in the Horkes process are combined to serve as the basis of sub-graph search expansion and node embedding vector updating, the event development context is deduced, and future events are deduced with the event development context serving as a forward condition. The method not only pays attention to the relative time interval information between the events and further improves the accuracy of the prediction result, but also can give a clear reasoning path, so that the prediction of the events has interpretability, the prediction accuracy of the urban safety events is improved, and the adverse effects of the safety events are reduced.

The invention discloses a table data exploration type visual analysis system and method based on a large language model. The method comprises the steps that original table data are analyzed, and a data model is established; automatically excavating insights and generating three different forms of the insights; user dominant exploratory data analysis and data story generation are carried out in the interaction module; an insights engine in the calculation module searches insights with structural association through an insights graph; and the big language model in the reasoning module performs insight recommendation through semantic association recommendation. According to the method, the structural table data can be modeled, so that data analysis and insight mining are performed. Potential insights in a table are automatically mined, so that a user gets rid of repeated and dense analysis operations, and interactivity, transparency and interpretability of an exploratory data analysis process are enhanced by fusing human intuition and insights, computing power of an external actuator and intelligence of a large language model, thereby promoting efficient and intuitive data analysis.

The invention relates to the technical field of hydroelectric generators, in particular to a three-dimensional flow field load reconstruction method based on reinforcement learning of a gated circulation unit, which comprises the following steps of: establishing a Python-based finite element secondary development platform, and realizing finite element simulation and obtaining a finite element simulation result; fusing a plurality of simulation characteristic indexes extracted from the finite element simulation result by using a multi-scale model fusion algorithm to obtain a finite element simulation result evaluation index, and optimizing simulation parameters; constructing an initial three-dimensional flow field load reconstruction model based on reinforcement learning of a gated circulation unit, and training the initial three-dimensional flow field load reconstruction model by using the optimized simulation parameters; and based on the trained three-dimensional flow field load reconstruction model, obtaining a reconstructed three-dimensional flow field load. By means of the reconstruction method, automatic simulation of finite element software and full-automatic optimization iteration of a parameter decision and reinforcement learning reconstruction network in the three-dimensional flow field load reconstruction process of the hydroelectric generating set component can be achieved.

The invention relates to an image-text matching analysis method, system and device based on a large model and multiple modes. The method comprises the following steps: 1) image-text analysis: analyzing an image into an image and a text; 2) carrying out fine-tuning multi-mode large model processing on the image, and carrying out word segmentation processing on the characters; 3) constructing a bisection weight; 4) performing Few-shot cue word construction on the processed characters and the constructed equally-divided weight; 5) performing training / fine tuning on the characters and the pictures in the step 4 and the step 2 to serve as a large oracle model; according to the multi-mode image-text matching method, cooperative work of the multi-mode large model and the large language model is introduced, the advantages of the multi-mode large model and the large language model in semantic comprehension, context analysis and dynamic adjustment capacity are comprehensively utilized, a comprehensive image-text matching framework is established, and the accuracy, flexibility and interpretability of matching can be improved.

The invention discloses a physical information neural network hyperspectral band selection method for target identification, which uses physical information SDI as prior information to be combined with a channel attention mechanism added in a deep neural network to guide band selection, and evaluates the advantages and disadvantages of band weights through a reconstruction module after the weights are generated. Wave band selection is carried out accordingly; according to the hyperspectral image target recognition waveband selection method based on the physical information neural network, compared with a traditional method, the process is simplified and the precision is improved through end-to-end learning, the utilization of spatial context information can be enhanced by adding an attention mechanism, and the model has interpretability by adding prior SDI information.

The invention belongs to the technical field of intelligent diagnosis of four-stroke diesel engines, and particularly relates to a PSO-RF intelligent diagnosis method and system for thermal faults of a marine diesel engine. The method comprises the following steps: designing a diesel engine simulation model by using simulation software; constructing a data set and dividing the data set; performing noise processing and normalization processing on the data in the data set; extracting characteristics which can best represent diesel engine faults in the data set; a random forest fault diagnosis model is constructed, parameter optimization is performed on the random forest model by using a particle swarm optimization algorithm, a PSO-RF intelligent diagnosis model is finally obtained, and fault diagnosis is performed by using the PSO-RF intelligent diagnosis model. The system comprises a diesel engine simulation unit, a data acquisition unit, a data processing unit, a feature extraction unit and a fault diagnosis unit. According to the method, automatic fault positioning and diagnosis can be realized, sailors are helped to quickly and accurately identify problems, the dependence on expert experience is reduced, and the diagnosis precision is remarkably improved.

The invention relates to a medical image report automatic generation method and system. The method comprises the following steps: extracting local image features and global image features related to a focus in a medical image; fusing the local image features and the global image features to obtain fused image features, and inputting the fused image features into an image diagnosis classifier to obtain a medical image related disease diagnosis classification result; wherein the image diagnosis classifier is obtained by pre-training an original classifier based on a multi-dimensional image fusion feature, and the multi-dimensional image fusion feature is obtained by fusing a local image feature and a global image feature which are extracted from a sample medical image and related to a focus; querying a pre-constructed medical field knowledge graph based on the disease diagnosis classification result to obtain a corresponding first medical field knowledge code; and fusing the local image features, the global image features and the first medical domain knowledge codes to obtain target text features, and inputting the target text features into a medical big language model to automatically generate an image diagnosis report.

The invention discloses an internet-based online gynecological disease intelligent consultation system, and relates to the field of data processing, and the system comprises the steps: generating a public and private key pair of a patient through employing an RSA asymmetric encryption method; a BERT model based on deep learning is adopted to process the decrypted inquiry information, processing is carried out on local equipment, and structured patient condition description is obtained; the encrypted patient condition description is uploaded to a server, a system private key is used for decryption on the server, diagnosis is carried out by using a reasoning method based on a knowledge graph according to the patient condition description, and a disease candidate list is obtained; transmitting the encrypted candidate list to local equipment of the patient, decrypting by using a private key of the patient, and reasoning by adopting a CART decision tree according to the disease candidate list to obtain a final diagnosis result; aiming at the low precision of online gynecological disease consultation diagnosis in the prior art, asymmetric encryption, knowledge graph reasoning, a CART decision tree and the like are adopted, so that the diagnosis precision is improved while the privacy of a patient is protected.

This invention discloses a method, apparatus, electronic device, and storage medium for classifying license plate images. The method includes: in response to an image classification instruction for a target license plate image, determining image feature data of the target license plate image; inputting the image feature data into a pre-trained license plate classification model; and outputting an execution result for classifying the target license plate image, the execution result including at least two classification items and the probability of each classification item, wherein the at least two classification items are distributed in a tree-like structure. The technical solution of this invention achieves more accurate classification of license plate images, and the classification process is interpretable.

The present application relates to a kind of preparation methods of essence.The problems that traditional process parameter adjustment relies on artificial experience and is difficult to realize cross batch self-adaption, multi-modal feature extraction, parameter-function-stage three-dimensional label matrix generation, graph neural network behavior atlas construction and expert strategy distribution constraint dual-channel strategy learning method are proposed.By process parameter and derived feature multidimensional fusion, time series causal strength analysis and action instruction simulation verification, optimization process parameter combination with explainable function synergistic relationship can be output.The technical scheme effectively realizes the automatic optimization of process control decision and the rapid self-adaption of model under new raw material batch, improves product consistency, quality stability and control energy efficiency.

The invention discloses a front door adjustment-based harmful modal cause detection depolarization method, which comprises the following steps of: constructing a structural causal model, creating a back door path between a modal cause and a harmful label by using unobserved confounding factors, and eliminating harmful modal cause detection through front door adjustment; the front door adjustment then blocks the back door path by introducing reasoning as an intermediary between the memetic and the tag. The method comprises the following steps: decomposing the causal effect of a medal factor on a harmful tag into the influence of the medal factor on an intermediary, quantifying by adopting a multi-modal large-scale language model with different beam searches to calculate the probability of each reasoning sequence and the influence of the intermediary on the tag, and performing effective approximation through a normalized weighted geometric averaging method; finally, post-interpretation is synthesized from the inference sequence and the predictive tag using a large language model. According to the scheme, the most advanced performance is achieved while information interpretation is provided.

This application relates to the field of data processing technology, and in particular to a data fusion method and system based on neural pathways. The method includes the following steps: First, acquiring electroencephalogram (EEG) data and eye-tracking (EMT) data, aligning them using timestamps, and preprocessing them separately; then, extracting EEG features based on the preprocessed EEG data; extracting EMT features based on the preprocessed EMT data; next, fusion of features based on the EEG and EMT features by downsampling the data; finally, evaluating the synergy of the fused features. This application constructs a complete evaluation system by building a framework from a stimulus-visual paradigm to a signal acquisition platform and analysis methods related to neural structures. The time delay between the most relevant points in the time domain of eye-tracking and EEG features is used as an indicator of brain-eye synergy. The difference between these indicators has interpretability based on physiological structures, constituting a novel evaluation indicator for brain-eye synergy when assessing dynamic visual acuity.

The present invention proposes a method, device and equipment for identifying hypertension based on pulse cycle characteristics. The method is: the pulse data of the sample personnel are pre-processed, period-divided and processed with sliding windows in sequence and labeled, and then four cycle-to-cycle network features with greater contribution to the classification of hypertension diseases are selected as the pulse cycle-to-cycle network feature vector, and then the pulse data of the sample personnel after sliding window processing is input into the multi-scale permutation entropy algorithm, and the multi-scale permutation entropy value is extracted. Finally, the two features are fused to form a hypertension pulse feature vector, which is input into the classification model, and supervised and trained with the labels corresponding to the sample personnel to obtain a trained classification model. The present invention proves the effectiveness of the cycle-to-cycle features of the pulse signal by mining the cycle-to-cycle sensitive features, and also improves the accuracy of hypertension disease classification; the network features are used to make the mined cycle-to-cycle sensitive features interpretable.

The application provides a three-dimensional part retrieval method based on graph similarity search, relates to the field of computer graphics, and solves the technical problem that the prior art cannot fully utilize the geometry and design semantic information of CAD, and it is difficult to capture and display the topological relationship, resulting in low retrieval efficiency. The method comprises the following steps: obtaining part three-dimensional data of a computer-aided design (CAD) model, and constructing a training data set; constructing an edge-face connection graph based on the part three-dimensional data; calculating a graph edit distance (GED) matrix of all edge-face connection graphs in the training data set as a supervision signal; training a ranking model based on the GED matrix, mapping the edge-face connection graph to a hidden space, and constructing a part vector database according to the output graph-level embedding vector; the ranking model is constructed based on a graph attention network; inputting a CAD part to be queried into the trained ranking model to obtain a feature vector, performing nearest neighbor search in the part vector database, and returning a similar part result.

The application discloses a kind of robust direction of arrival estimation method based on ADMM-Net, by expanding ADMM algorithm into model-driven deep network ADMM-Net to improve DOA estimation accuracy, speed up DOA estimation speed and have robustness to array disturbance.Firstly, the sparse transformation of source and array received multi-shot data is carried out by space over-complete dictionary, and the DOA estimation is converted into compressed sensing sparse recovery problem;Then, ADMM algorithm is expanded, and model-driven deep network ADMM-Net with explainability is formed;ADMM-Net is used to reconstruct source power spectrum and carry out DOA estimation.The present application can learn the hyperparameters in iterative algorithm and array disturbance, solve the problem that existing compressed sensing DOA estimation method based on calculation speed is slow, and fails under the condition of array disturbance, realize fast, accurate, explainable and robust DOA estimation.

The invention discloses a cardiac magnetic resonance report generation method and system based on a multi-dimensional tissue feature knowledge graph. The method comprises the following steps: performing motion correction and multi-view feature extraction on a cardiac magnetic resonance multi-sequence image to obtain an original numerical attribute of an instance node; constructing a dynamic knowledge graph containing tissue characteristics and medical priori, and generating a target reasoning chain with a clinical basis in combination with neural symbol reasoning; then, in a controlled generation stage, an inference chain is injected into a prefix tree, in a decoding step, mask physical shielding logic conflict description is utilized, and a pointer network is synchronously started to directly copy original numerical value attributes into a text sequence; and finally, executing consistency logic verification through a text restoration technology, and identifying and intercepting factual errors. Through the atlas constraint and numerical unvarnished transmission mechanism, the factual illusion problem of a generative model is effectively solved, and the accuracy and reliability of a heart diagnosis report are remarkably improved.

The invention relates to the technical field of ultrasonic image analysis, and discloses an image detection method and system for prolonging the second birth process of double-fetal vaginal delivery and SPPH risk based on ultrasonic Speckle-tracing, and the method comprises the steps: obtaining an ultrasonic two-dimensional gray-scale image sequence; marking a conjoint analysis area between the cervical wall and the front and rear walls of the vagina; constructing a joint area displacement field; constructing a joint region strain trajectory; extracting parturition process characteristic indexes; and executing risk assessment through a fuzzy rule. The technical problem that in the prior art, risk judgment depends on clinical experience, and especially under the condition that continuous stretching occurs on the cervix uteri and vagina combined wall in the second birth process of double-fetal vaginal delivery, early recognition of the evolution process of the tissue from reversible stretching to irreversible instability is difficult to achieve is solved. Due to the fact that joint area displacement field modeling and logarithmic nonlinear strain analysis are introduced, quantitative characterization and continuous risk description of the mechanical behavior of the tissue in the labor process are achieved.

The invention relates to the technical field of injection molding, in particular to an intelligent control system and method based on injection mold opening of an injection mold, and aims to solve the problems that an existing injection mold opening control system depends on a single signal in safety judgment, lacks multi-physical field fusion and fault confidence evaluation, is difficult to distinguish disturbance and real faults, and is low in reliability. The interlocking logic is extensive, and the grading response based on the abnormal grade cannot be realized. Multi-source preprocessing data are deeply fused through the mold opening state monitoring and interlocking module, accurate recognition and causal traceability of faults are achieved by quantifying physical interpretable indexes such as temperature field uniformity, force sudden change, vibration spectrum and micro-strain and combining an evidence combination and credibility updating mechanism, disturbance and real abnormity are effectively distinguished, and the fault diagnosis accuracy is improved. And according to the hierarchical interlocking of the abnormal levels, the mold opening safety permission is output only when the whole channel is normal and no high-confidence fault exists, so that the safety and the intelligent level of the mold opening process are remarkably improved.

The invention discloses a deep learning sea clutter identification method based on a constant false alarm convolution kernel. Firstly, multi-frame radar signals are converted into graph data, sea clutters are preliminarily suppressed by using spatial-temporal characteristics of the graph data, then, a detection sliding window convolutional neural network model is constructed by taking a classical constant false alarm detection sliding window model in radar signal processing as a convolution kernel of the convolutional neural network model, and the detection sliding window convolutional neural network model is used for detecting the sea clutters. And feature extraction and classification are carried out on graph data nodes by using the provided detection sliding window convolutional neural network model, so that target identification under the sea clutter background is realized. Simulation experiments prove that the detection probability can reach 98% when the signal-to-clutter ratio is 9dB, the method is superior to traditional CFAR detection, a convolution kernel is designed through a detection method similar to a constant false alarm rate according to the working principle of a detection sliding window in CFAR, a network model is more interpretable, meanwhile, more data features are integrated, the parameter scale is smaller, and the operation speed is higher.

The present application relates to the technical field of aluminum electrolysis, and provides an aluminum electrolysis cell damage inversion method based on a machine learning algorithm, comprising the following steps: obtaining current data of an aluminum electrolysis cell through a fiber-optic current sensor, and obtaining temperature data of the aluminum electrolysis cell; performing discharge area division according to a discharger of the aluminum electrolysis cell; performing area division on the current data and the temperature data, and performing feature extraction on the current data and the temperature data; establishing a physical information forward model based on a neural network, and associating and mapping a damage state vector with current feature data and temperature feature data through the physical information forward model; constructing a cathode joint loss function and an anode joint loss function; respectively performing iterative solution on the cathode joint loss function and the anode joint loss function through an optimization algorithm, outputting an optimal damage state vector, and generating a damage space distribution result of the electrolysis cell. The present application can realize online inversion and spatial distribution visualization of anode and cathode damage states.

The present application provides a method for predicting the efficacy of umbilical cord blood natural killer cell treatment of neuroblastoma, belonging to the technical field of medical prediction and evaluation. The present application solves the clinical need for early and accurate prediction of the efficacy of immunotherapy for neuroblastoma. The method comprises: obtaining peripheral blood test data of a neuroblastoma patient before multiple reinfusion treatment nodes of umbilical cord blood natural killer cell treatment; based on the processed node-level peripheral blood test data, at least one dynamic linkage feature reflecting the dynamic linkage relationship of the indicators in the patient's body is calculated; at least one static feature index is obtained; the dynamic linkage feature and the static feature index are input into a pre-trained efficacy prediction model to obtain the response probability of the patient to umbilical cord blood natural killer cell immunotherapy. The present application accurately predicts the efficacy of umbilical cord blood natural killer cell treatment of neuroblastoma by analyzing the peripheral blood test data of the patient at multiple time points and constructing dynamic linkage features.

The invention discloses a sludge age self-adaptive regulation and control method and system based on dissolved oxygen trend recognition. Aiming at the problem of impact response lag caused by difficulty in COD on-line measurement, the method comprises the following steps: under the condition of stable water inlet and aeration, filtering and sliding window feature extraction are performed on an on-line DO signal to obtain a regression slope representing the change of DO along with time, and COD rising impact is reliably identified in combination with an ORP persistence criterion. A mechanism and machine learning fusion model is adopted to determine an identification threshold value and a target sludge age SRT, the sludge discharge amount is rapidly adjusted after impact triggering, the treatment capacity is synchronously matched, and aeration is used for assisting in maintaining DO stable. The system integrates an online monitoring module, a DO recognizable interval judgment module, a feature calculation module, a load recognition module, an SRT decision module and a sludge discharge execution module. According to the method, rapid impact recognition and closed-loop regulation and control can be realized without an online COD instrument, and the impact resistance and the operation stability of the process are remarkably improved.

The application discloses a kind of multi-agent learning method, device and equipment based on dynamic weighted average field, method includes: according to the action of dynamic weighted average action corresponding to agent selects action;Current position information of agent is obtained Calculation aggregation factor, the weight of agent is dynamically updated using improved PSO algorithm in combination with aggregation factor;According to the action update dynamic weighted average action with weight, execute joint action to obtain next state, according to the reward update action and state, and according to the weight and reward update individual weight optimal value and group weight optimal value;Experience replay buffer is sampled from experience data including current state, joint action, reward, next state and dynamic weighted average action and updates the Q function of agent, according to the updated Q function and specified learning rate updates the target network of agent.The application can guide multi-agent to tend to group performance optimization, and also ensure that individual agent has sufficient exploration ability.

The invention discloses a geological disaster dynamic early warning method driven by multi-source data and an intelligent model, and belongs to the technical field of geological disaster early warning. The method comprises the following steps: acquiring multi-source heterogeneous geological monitoring data; performing space-time alignment and semantic standardization processing, and constructing a standardized data set; constructing a multi-modal deep fusion neural network model, extracting features through a parallel feature extraction sub-network, and carrying out feature fusion by adopting a cross-modal attention mechanism based on a historical accuracy rate dynamic adjustment weight; disaster risk prediction is carried out based on fusion features, and online model updating is realized through a sliding time window mechanism; and generating an early warning report according to the risk level. According to the invention, an early warning effect feedback closed loop mechanism is also established, and model iterative optimization is realized through postmortem verification of returned data. According to the method, a complete closed loop from data perception to model self-optimization is constructed, and the problems of poor semantic consistency, static feature fusion and insufficient early warning real-time performance of multi-source data are solved.

The application relates to the technical field of data processing, and provides a business processing method, device and equipment under a parallel architecture of a new system and an old system. The method comprises the following steps: receiving a processing request; according to a shunting configuration, shunting the processing request to a target system for processing; the target system comprises at least one of a new system and an old system; when the new system fails to process the processing request, identifying a failure category of the processing failure according to a white-box model; when the white-box model fails to identify the failure category, identifying the failure category of the processing failure according to a black-box model; and according to the identified failure category, deciding to re-shunt the processing request. The embodiment of the application can reduce the processing burden and invalid processing cost of the old system under the parallel architecture of the new system and the old system.

The invention discloses a hierarchical alignment method and system for a multi-modal perception data set of a soft manipulator, and the method achieves the coverage of a whole process from data quality evaluation to feature space alignment to semantic consistency optimization through the three-stage design of single-modal noise evaluation, feature-level adaptive alignment and semantic-level constraint alignment. The limitation that a traditional method only pays attention to single-level alignment is broken through, and the model can gradually adapt to the complex characteristics of touch multi-mode perception. Secondly, an adaptive temperature function is designed based on modal pair comprehensive reliability, and a strict monotone mapping relation between temperature coefficients and modal noise characteristics is established, so that a high-noise modal pair automatically obtains a relatively large temperature coefficient to enhance robustness, and a low-noise modal pair automatically obtains a relatively small temperature coefficient to improve alignment precision; the technical problem that the fixed temperature coefficient cannot adapt to the touch multi-mode heterogeneous noise characteristic is fundamentally solved, and end-to-end sensitivity adjustment is achieved through the global learnable self-adaptive adjustment coefficient.

This invention discloses a big data-based EVA membrane product production management system and method, belonging to the field of industrial big data analysis and production management technology. It includes: acquiring multi-source heterogeneous production data from various stages of the EVA membrane product production process; performing data cleaning and standardization to obtain a standard production dataset; extracting key production features from the dataset and constructing a multi-dimensional feature vector space; using a density-based clustering-based anomaly detection algorithm to perform pattern recognition on the feature vector space, generating a set of abnormal production events; using an association rule mining algorithm to extract association rules between abnormal events and production parameters, obtaining a root cause parameter set; constructing a production parameter adjustment decision tree; and generating production optimization strategies based on the decision tree. This invention also provides a corresponding system that realizes automatic root cause analysis and parameter optimization feedback for EVA membrane product production anomalies.

This invention provides a method and system for precise source tracing of nitrogen compound pollution sources based on multimodal data fusion and graph neural networks, belonging to the field of pollution source tracing technology. This invention acquires and quantifies the multimodal fingerprint features of each typical pollution source to form a source feature matrix, obtains multimodal spectral data from downstream monitoring points to obtain actual monitored downstream fingerprint vectors, and constructs a spatial topology map based on the spatial topology between monitoring points. The source feature matrix and the spatial topology map are input into a graph neural network to obtain contribution weight vectors, and the predicted downstream fingerprint vectors are reconstructed accordingly. A composite loss function including spectral reconstruction loss, physical constraint loss, and sparsity constraint is constructed, and convergent contribution weights satisfying non-negativity and sum-to-one constraints are obtained through iterative optimization. This invention can output a physically meaningful and interpretable quantitative contribution rate of pollution sources for precise source tracing analysis in nitrogen compound pollution scenarios.

This disclosure provides an interpretable air combat decision-making method and apparatus. The method includes: acquiring current situation information from an air combat environment; wherein, the air combat environment contains a first fighter jet and a second fighter jet hostile to the first fighter jet; determining, based on the current situation information, rules corresponding to the current situation information from a rule base corresponding to the first fighter jet, and constructing a matching set corresponding to the first fighter jet based on the rules; obtaining the predicted behavior of the second fighter jet through a behavior prediction model of the second fighter jet, determining alternative maneuvers for the first fighter jet based on the predicted behavior and the matching set, and constructing a set of maneuvers corresponding to the first fighter jet based on the alternative maneuvers and the matching set; and controlling the first fighter jet to execute the maneuvers corresponding to the set of maneuvers based on the set of maneuvers. By applying this disclosure, not only can the adversarial performance of air combat decision-making be improved, but also air combat decision-making rules with generalizability and interpretability can be obtained.

The present invention proposes a Chinese legal adversarial text generation method based on an explainable perturbation strategy, which belongs to the technical field of adversarial text generation. It includes: extracting legal knowledge from legal text; scoring and ranking the importance of each word in the text based on legal knowledge; selecting the perturbation strategy to be executed for the text to be perturbed; and executing the perturbation strategy to generate legal adversarial text. The present invention solves the technical problem of the lack of adversarial text generation methods for legal crime classification tasks in the prior art. The present invention first extracts the main features of each type of crime as prior knowledge, and then combines the prior knowledge with two perturbation strategies of strengthening similar labels and weakening original labels, and generates adversarial text by strengthening the unique features of labels similar to the original labels or weakening the unique features of the original labels themselves, which makes the generation process of adversarial text explainable.