247results about How to "Improve interpretability" patented technology

This invention provides a method for hypertension risk assessment, relating to the field of epigenetic detection technology. It includes: performing event alignment and residual modeling on a qualified signal set to establish a base probability at the locus level and forming baseline features through neighborhood consistency screening; constructing a tunneling sensitization model and contextual attention, fusing discriminative and generative evidence to obtain enhanced features; quantifying IGF2BP3 and aligning and fusing it with the enhanced features; constructing an IGF2BP3-mediated metabolic network based on this, performing time-series modeling and stability assessment, and extracting final-state features; fusing the final-state risk vector with multi-gene scores to complete adaptive grading and compliant report generation. This method is robust, interpretable, and easy to deploy.

The application belongs to the technical field of artificial intelligence, and specifically discloses a power customer service model optimization method and system based on artificial intelligence, which obtains a support set and a query set for large model small sample learning, uses the support set and the query set to perform meta-training on a large model, retrieves an auxiliary set to perform inference test on the pre-trained large model for power business scene problem instances, and then performs reinforcement learning on the pre-trained large model based on a comprehensive strategy reward value according to the inference steps and the inference result obtained through the test, so as to obtain a strategy optimized large model to infer and optimize the solution to the power business scene problem in actual application. The application uses small sample learning technology to realize fine-tuning of the inference ability of the large model, and uses comprehensive strategy feedback to perform reinforcement learning, thereby improving the explainability of the inference process of the large model and the accuracy of the inference result, and providing more efficient decision support for power customer service question answering of the large model.

The current transformer error prediction method based on dynamic feature graphs and spatiotemporal graph convolutional networks belongs to the interdisciplinary field of power big data and artificial intelligence. It addresses the problems of existing methods neglecting multi-source data topological relationships, insufficient spatiotemporal feature mining, and weak noise resistance. This method first collects historical error data, load current data, and ambient temperature and humidity data of the current transformer. After preprocessing such as denoising with a Savitzky-Golay filter, each physical quantity is defined as a graph node. A dynamic adjacency matrix is ​​constructed by combining Pearson correlation coefficients and adaptive node embedding. Then, an ST-GCN model with alternating stacked "time-graph-time" convolutions is built to extract spatiotemporal coupling features. Finally, the model is trained using the Huber loss function and the Adam optimizer to output the error prediction value. This invention achieves explicit modeling of the dynamic coupling relationship of multi-source variables, improving the prediction accuracy and robustness under non-stationary operating conditions, and is suitable for current transformer condition monitoring and metering performance evaluation in smart grids.

This invention provides an intelligent identification and quantification method for microscopic residual oil, belonging to the field of microscopic residual oil analysis. The method includes preprocessing visual modal micro-CT images and textual modal geological description text, and inputting them into a multimodal pyramid pooling visual model for deep cross-modal fusion and semantic recognition to generate a high-precision semantic segmentation map of residual oil. The semantic segmentation map is then input into a task scheduler, where dynamic scheduling and collaborative intelligent agents execute seepage simulation and potential prediction to obtain dynamic simulation results. Based on the current residual oil state, a decision-making intelligent agent uses closed-loop optimization to obtain the optimal strategy and its verification simulation results. A structured report is generated based on the semantic segmentation map, dynamic simulation results, optimal strategy, and its verification simulation results. This invention solves the problems of fragmented identification-modeling-simulation-decision-making processes, low automation, and unquantifiable results in microscopic residual oil analysis.

The application discloses a specific field pipeline task type dialogue system based on an LLM, which comprises an inquiry subsystem and an answering subsystem, wherein: the inquiry subsystem parses and guides a user to supplement key information of a question according to user input task text, the answering subsystem summarizes key information of a dialogue task, and performs vector matching through a built-in vertical field local vector database to generate a final answer to the dialogue task. The application uses a large language model to realize natural language understanding, dialogue state tracking and natural language generation modules in a task type dialogue system pipeline, simultaneously realizes a rule matching-based strategy learning module, understands and replies to user input text, and combines the content of a local knowledge base, so that the whole process is more interpretable, the illusion problem of the large language model is reduced, only a small amount of sample prompt learning is needed, a large amount of data training or fine tuning is not needed, and the workload during field migration is reduced.

The application discloses a quantitative evaluation method for fusion degree of new and old asphalt in hot recycled asphalt mixture, and belongs to the field of recycled asphalt, and comprises the following steps: crushing and sieving the hot recycled asphalt mixture to a preset particle size range to obtain a to-be-tested sample; performing solvent extraction treatment on the to-be-tested sample to obtain an asphalt solution; and performing selective separation treatment on the asphalt solution under controlled conditions, wherein the controlled conditions comprise synergistic control of solvent type, extraction time and extraction temperature, so that unfused asphalt components at the interface between new and old asphalt and having low bonding strength are preferentially selectively dissolved out. Through the selective separation treatment under the controlled conditions, the unfused asphalt components at the interface between new and old asphalt and having low bonding strength are preferentially dissolved out, and the fused asphalt components with structural intergrowth are retained, so that the different fused asphalt components can be effectively distinguished without damaging the original structure.

This invention provides a vehicle simulation speed correction method and system, relating to the field of vehicle intelligent dynamics modeling technology. The correction method includes the following steps: S1: Input and process simulation data output from the vehicle dynamics simulation model and corresponding real vehicle test data; S2: Construct a dynamic graph structure representing the interaction between state variables based on a preset physical coupling relationship of the vehicle powertrain; S3: Input the simulation data into a GCN and perform graph convolution operations under the constraints of the dynamic graph structure to extract graph embedding feature sequences representing the spatial dependencies between state variables; S4: Input the graph embedding feature sequences into a TCN and perform temporal convolution operations to learn the dynamic evolution law of state variables in the time dimension and output the correction amount of the vehicle simulation speed; S5: Output the result. Based on this, this invention solves the problem that existing correction methods have various limitations in practical applications.

This application discloses a method and system for determining pollution causes based on the correlation between atmospheric diffusion conditions and emission contributions, relating to the field of atmospheric pollution cause identification. The method includes: aligning meteorological, emission, and receptor observations to a unified spatiotemporal reference to obtain multi-source baseline data; calculating a diffusion stability index characterizing vertical mixing and horizontal transport using meteorological monitoring, generating time-varying diffusion condition weights; generating a time-response diffusion kernel from the emission source to the receptor point using a diffusion model under unit emission conditions, and constructing diffusion-contribution correlation information with observed concentrations to characterize the temporal coupling strength; establishing a diffusion-contribution dynamic optimization model, using diffusion condition weights to constrain the coupling relationship, and iteratively solving for time-varying contribution coefficients based on the convergence target of simulated and observed concentrations, outputting the dynamic contribution of each source and the pollution cause. Thus, time-varying diffusion conditions are explicitly introduced into the source contribution inversion framework, achieving stable separation and reliable attribution of source contributions.

This invention discloses a method and system for full life-cycle management of roads and bridges, relating to the field of road and bridge technology. The method includes acquiring the BIM model of the bridge during the design phase and structural state data and environmental-load data for each stage of the bridge's life-cycle, and constructing a unified model for structural performance degradation. A performance function is generated through multi-factor temporal coupling processing. The performance function is then spatiotemporally aligned and fused with the BIM model, structural state data, and environmental-load data for each stage of the bridge's life-cycle to form a digital structural ontology. Structural state data is extracted from the digital structural ontology, and a dynamic knowledge graph with semantic reasoning capabilities is constructed to output the structural health status. Based on the performance function and structural health status, the failure probability trend is determined and high-risk components are identified, taking into account the bridge's road network level, design capacity, and preset weights for traffic interruption losses, public safety risks, and ecological environment disturbances.

The invention belongs to the technical field of groundwater reserve prediction, and particularly relates to a physical logic driven groundwater reserve intelligent prediction method and system. Comprising the following steps: performing complementation and physical verification on multi-source heterogeneous observation data related to groundwater reserves in a target area to obtain a physical verification complete data set conforming to physical logic; performing space-time decoupling causal contribution analysis on the physical verification complete data set to generate a causal contribution matrix; according to the causal contribution matrix, constructing and training a physical constraint emergence type space-time prediction model; applying a physical anchoring adversarial migration strategy to adapt the trained emergence type prediction model to the target new region to obtain a region adaptive prediction model; and inputting the future scene conditions into the regional adaptation prediction model for simulation, and generating a groundwater reserve prediction result. According to the method, intelligent prediction with high precision, high robustness and physical interpretability on groundwater reserves can be realized, and scientific decision support is provided for water resource management.

The application discloses a LIBS element quantitative analysis method based on double-branch feature fusion and electronic equipment, the method comprises the following steps: preprocessing the collected LIBS spectrum signal, including baseline correction and spectrum resampling; inputting the spectrum signal data after preprocessing into a double-branch feature extraction network, extracting local features through a CNN branch, and extracting global features through an MLP branch in parallel; performing adaptive weighted fusion on the local features and the global features through a gating fusion module to obtain fusion features; inputting the fusion features into a WMA-MLP model, the WMA-MLP model is based on MLP, integrates a multi-head self-attention mechanism and a residual module, is used for modeling the global dependency relationship between features, and outputs a final element quantitative analysis result. Through the parallelly arranged CNN and MLP branch feature extraction structures, more comprehensive spectrum feature representation can be obtained, and the accuracy and robustness of the spectrum analysis model are improved.

The application provides a kind of intelligent structured medical record generation method and system based on multimodal doctor-patient interaction, which comprises: real-time acquisition of dialogue voice and transcription into text sequence, while recognizing visual attention entity by listening to mouse operation in electronic medical record system.Based on the history of visual attention entity and text sequence, a logical demonstration track is constructed, and an implicit reward function is derived from it using a reverse reinforcement learning algorithm. Use the function to calculate the action reward value of each combination of visual attention entity and dialogue text, select the highest value combination as the optimal alignment strategy to determine the timing causal relationship. Map the entity and text to the medical knowledge graph, extract the shortest semantic path as the implicit clinical reasoning chain, and generate structured electronic medical record. The application deduces the diagnosis and treatment decision logic from the doctor's multimodal behavior through reverse reinforcement learning, solving the technical problem that traditional methods cannot establish the internal causal relationship between the doctor's visual attention focus and spoken content.

The invention discloses a crop irrigation short-term prediction method and system based on fuzzy control, and relates to the technical field of intelligent agriculture and irrigation control, and the method comprises the steps: collecting weather, multi-depth soil moisture content and field time sequence data of a crop growth stage, and obtaining irrigation feedback; processing the data, and inputting a time sequence prediction model to output water demand prediction values of one to three days and confidence obtained by mapping historical verification errors; fuzzifying the water demand predicted value, the real-time root zone moisture content and the growth stage index, performing weighted deduction and fuzzification based on a fuzzy rule base with a rule dynamic weight, and outputting planned irrigation duration or irrigation amount; and calculating the actual water consumption and updating the dynamic weight of the rule in the post-irrigation evaluation time window according to the water balance and the root zone soil moisture reserve variation. Through confidence risk suppression and post-irrigation closed-loop self-tuning, the stability of short-term prediction driving irrigation decision is improved, the over-irrigation and over-seepage risks are reduced, and the utilization efficiency of irrigation water is improved.

The application relates to a chronic kidney disease prediction method based on clinical information graph representation, belongs to the technical field of computer-aided diagnosis of chronic kidney disease (CKD), and aims to solve the problem of missed diagnosis caused by the fact that the kidney function index of early CKD is not obvious. The incidence of CKD is high, the early symptoms are hidden, and the disease is easy to be missed, thus developing into end-stage renal disease. Therefore, the application provides a chronic kidney disease prediction method based on clinical information graph representation. The method first extracts fundus image and clinical index features; the clinical index is fused into a clinical index joint feature through text embedding and numerical feature fusion, and the joint feature is fused with the fundus image feature through cross-modal attention; the similarity between subjects is calculated based on the fused feature, and a subject relationship graph is constructed by using an adaptive dynamic threshold mechanism; finally, a hybrid graph neural network is used for graph representation learning, pathological similarity between subjects is mined, and accurate prediction of chronic kidney disease is realized. The application can improve the detection rate of early CKD and is applied to non-invasive early screening and risk early warning.

The application provides an engine cylinder pressure calculation method and an engine cylinder pressure virtual sensing system. The method comprises the following steps: establishing a hybrid control combustion model of an engine, inputting first parameters for calculation into the hybrid control combustion model, and performing the following calculation in the hybrid control combustion model: calculating in-cylinder temperature and pressure of the engine in a first stage according to the first parameters, calculating instantaneous in-cylinder local turbulent kinetic energy density in a second stage, calculating a burned gas dilution factor according to the first parameters, the burned gas dilution factor reflecting an obstruction effect of inert gas in a combustion chamber of the engine on mixing of active components, calculating a heat release rate after an ignition delay period in a third stage according to the first parameters, a fuel kinetic energy change rate and the burned gas dilution factor, and calculating a curve of engine cylinder pressure changing with a crank angle according to the heat release rate by using a thermodynamic formula.

This invention discloses a physical twin modeling method for cross-condition industrial control systems. It constructs a data-physical fusion physical twin model applied to an anomaly detection framework for industrial control systems. This framework uses the physical twin model as its core, expressing industrial physical processes through a system of differential equations and introducing physical constraints using PINN. Simultaneously, it continuously calibrates the model using real-time operational data to achieve dynamic updates. Furthermore, it achieves adaptive capabilities across conditions, stages, and scenarios through parameter sharing and transfer learning. The trained physical twin model is compared with real-time data, and the existence of anomalies is determined by analyzing the deviation between predicted and observed values. This framework organically combines the high fitting properties of data-driven approaches with the interpretability of physical-driven approaches, providing industrial control systems with high-precision, interpretable, and transferable anomaly detection capabilities.

The application provides a self-evolving intelligent fund analysis method, relates to the field of data processing, and comprises the following steps: receiving a natural language analysis instruction, combining a large language model decomposition task, extracting account fund flow characteristics and matching with historical cases, and generating a weighted execution instruction; identifying a transfer path based on a fund correlation graph, extracting a priority path subgraph structure according to a user identifier, and forming a directional analysis model; performing node risk scoring on the fund correlation graph, outputting a risk account identifier and an analysis report, and realizing intelligent and accurate fund risk identification.

The application provides a medical question and answer method and system based on cooperative dual-source retrieval enhancement generation, which comprises the following steps: according to an input original medical question and candidate items, performing task self-adaptive query rewriting through a large language model to generate differential diagnosis keywords, entity enhanced queries and hypothetical medical abstracts; for the differential diagnosis keywords, performing an iterative network retrieval with a reflection mechanism to obtain network candidate evidence; for the entity enhanced queries and the hypothetical medical abstracts, performing a hybrid retrieval in a local medical knowledge base to obtain local candidate evidence; aggregating and deduplicating the network candidate evidence and the local candidate evidence, obtaining the aggregated heterogeneous evidence, performing deep semantic correlation scoring through a cross-encoder model, screening out a target evidence set, and splicing the original medical question as context input into a generative large language model to generate a final medical answer. The application improves the accuracy, timeliness and explainability of the medical question and answer system.

This invention discloses a method and apparatus for detecting harmful Chinese memes that combines relative intent reasoning and knowledge enhancement. It generates an expanded set of meme semantic keywords by training memes, and for each keyword, it searches a set of background knowledge documents in a Chinese harmful semantic knowledge base. A re-ranking model constructs prompt words for each candidate document in the background knowledge document set and its corresponding query. Based on the meme content of the trained memes, it optimizes and updates a multimodal detection model to obtain a classification multimodal detection model. The meme to be detected is input into the classification multimodal detection model to obtain the corresponding label and reason, where the label is either harmful or harmless. This invention integrates an external cultural knowledge base and possesses multi-faceted dialectical reasoning capabilities for the detection and interpretation of harmful Chinese memes, thereby improving the accuracy, robustness, and interpretability of the detection.

The application discloses a transient stability prevention control measure generation method and a related system, including the following steps: in a predicted fault scenario, a machine learning model is used to predict the transient stability of the current operation mode of a power system, the input of the machine learning model including various electrical characteristics related to the transient stability of the power system; the operation mode predicted as unstable is taken as an explained sample, a model explanation method is used to determine the influence degree of each electrical characteristic in the explained sample on the transient stability; candidate control units are determined according to the influence degree, and the candidate control units are adjusted, and the machine learning model is used to predict the transient stability of the adjusted operation mode until an operation mode predicted as stable is obtained, so that transient stability prevention control measures are generated. The method improves the explainability of the data-driven transient stability analysis model, the overall scheme requires a short time, and is suitable for online application.

This specification provides a data processing method and apparatus, wherein the data processing method includes: preprocessing initial audio data to obtain audio data, and inputting the audio data into an audio recognition model to obtain initial text containing prosody identifiers; determining at least one text unit corresponding to the initial text, and determining time information corresponding to each of the at least one text unit based on the audio data; updating the initial text in the prosody identifier dimension based on the time information corresponding to each of the at least one text unit to obtain target text corresponding to the audio data, wherein the target text and the audio data are used to train an audio generation model.

This disclosure provides a text data processing method, apparatus, device, storage medium, and program product, relating to the field of big data processing. The method includes: acquiring incremental text data and historical clustering data; the incremental text data includes at least one text data item; performing incremental clustering processing on the text data items based on historical clusters in the historical clustering data, matching historical clusters or creating new clusters for each text data item, and generating corresponding cluster topics for clusters that meet preset conditions, thereby obtaining incremental clustering results; and updating historical clustering data according to the incremental clustering results. This method can effectively integrate newly added text data, dynamically adjust the clustering results as data is updated, avoid the high overhead of full re-clustering, and enhance the interpretability of clustering results by automatically generating cluster topics. It achieves efficient dynamic clustering of incremental text data, improving the automation and intelligence level of data management.

The application discloses a kind of multi-micronet robust game optimization scheduling method and system based on dynamic auction official algorithm, and relates to energy system scheduling and optimization technical field.The method constructs system physical model and double-layer robust optimization model based on Stackelberg master-slave game, upper layer maximizes system operator's profit to formulate price signal, lower layer maximizes the worst scenario income of multi-micronet alliance to optimize resource scheduling, combined with multiple constraint conditions, Stackelberg equilibrium is solved iteratively using dynamic auction official algorithm.The application converges to global optimal solution quickly through three-dimensional hybrid driving price updating mechanism, realizes system economic benefit maximization, operation robust and reliable, and has good explainability and practical application value.

The application discloses a semiconductor thin film deposition process intelligent recommendation system and method based on a knowledge graph, belongs to the technical field of semiconductor manufacturing processes, and comprises the following: a semiconductor thin film deposition process knowledge graph construction module; a prompt word template generation module: used for receiving process requirements and constraint conditions input by a user, performing semantic analysis on the process requirements and constraint conditions, filling the analysis results into a preset prompt word template, and outputting standardized instructions meeting the calling requirements of a large language model; and a large language model reasoning module: used for receiving the standardized instructions, performing semantic retrieval and logical reasoning on the standardized instructions within the constraint range of the semiconductor thin film deposition process knowledge graph, generating a plurality of candidate thin film deposition process schemes meeting process constraint conditions based on entity relationships and causal associations in the semiconductor thin film deposition process knowledge graph; and the application constructs a complete process recommendation solution and exhibits significant advantages in multiple aspects.

The application discloses a ground TT&C station fault diagnosis system and relates to the technical field of space TT&C.The application aims at the problems that the existing fault diagnosis of the ground TT&C station under the environment of low signal-to-noise ratio, multi-node concurrency and strong interference depends on fixed threshold, is difficult to depict the space-time correlation of faults and is highly sensitive to the quality of spectral data, etc., and the application is characterized in that: a plurality of monitoring branches are arranged before and after a plurality of key devices of a link, and multi-modal data such as spectrum, temperature, voltage and environmental parameters are collected, so that the diagnosis system no longer depends on single spectral information, and other observation quantities can be used to maintain the discrimination of the device state when the signal is seriously attenuated or interrupted for a short time; in cooperation with a data preprocessing module, the missing value compensation and targeted normalization or interval scaling are performed on the data from different sources, so that the characteristics of the same node at each time point are kept continuous and dimensionally unified, and the numerical stability of the training and reasoning process of the graph attention network is significantly improved.

This invention provides a hot-rolled strip quality prediction method and system based on PINN modeling and cross-modal graph learning, belonging to the fields of industrial intelligent manufacturing and hot-rolled strip quality monitoring. The method first acquires historical production process variable data, strip quality data, and sensor monitoring data; it then constructs a temperature field model of the hot-rolled strip water-cooling process based on PINN, and uses the trained temperature field model to solve the continuous temperature field during the laminar cooling stage, extracting temperature field features; it constructs modal subgraphs for both temperature field features and sensor features, building a cross-modal full graph structure; a preliminary prediction model is built based on GNN, using the cross-modal full graph structure as the model input layer structure, and N base learners based on a residual learning strategy are introduced into the output layer. The trained preliminary prediction model and the N base learners constitute the hot-rolled strip quality prediction model. This invention achieves a deep integration of physical mechanisms and deep learning, significantly improving the accuracy and robustness of hot-rolled strip quality prediction.

The present application relates to the technical field of electrical system design, and particularly relates to a spacecraft electrical system interconnection component path planning optimization method and system. The method comprises the following steps: obtaining interconnection component data; constructing adjacency relationship according to the interconnection component data to obtain component adjacency data; selecting a beam splitting point according to the component adjacency data to obtain beam splitting point data; estimating point port cost of the beam splitting point data to obtain point port cost data; generating a control linkage graph according to the component adjacency data to obtain control linkage graph data; generating an interpretation field according to the control linkage graph data to obtain interpretation field data; determining the splitting benefit of the point port cost data and the interpretation field data to obtain splitting benefit data; and evaluating the hard constraint of the splitting benefit data to obtain beam splitting point feasibility evaluation data. The present application realizes the constraint of the beam splitting point position and the control semantics, effectively avoids the cross-stage control reversal and redundancy failure problems.

This invention provides a fact-verification method and system based on retrieval enhancement and contrastive argument synthesis, belonging to the field of large language model technology. The method involves: acquiring the statement to be verified; encoding the statement to be verified and statements in the training set using a pre-trained embedding model and calculating cosine similarity; selecting the k statements with the highest similarity as context examples; initially screening a portion of the document set; then re-ranking the initially screened documents using a dense retrieval model, selecting the m most relevant documents; and using a large language model to extract both supporting and rebuttal arguments from the selected m most relevant documents. These extracted arguments are then combined with the statement to be verified and the context examples to guide the large language model in predicting and interpreting the veracity of the statement to be verified. This invention provides the model with rich semantic information through context examples, enhancing its adaptability in different domains, generating high-quality explanations, and improving the transparency of the results.