54results about How to "Good explainability" patented technology

The invention provides a multi-process-mode intelligent switching method and system for vehicle-mounted sludge treatment, and the method comprises the steps: carrying out the time-space alignment and standardization of multi-dimensional sensing data such as sludge characteristics, environmental parameters and equipment states, carrying out the real-time reasoning of a process switching critical value through combining with a lightweight incremental regression model, generating a self-adaptive threshold value band with a confidence interval, and carrying out the real-time reasoning of the process switching critical value. Dynamic monitoring and mode switching pre-judgment of actual operation parameters are realized; in the switching process, a weighted fusion progressive transition control strategy is adopted, the transfer rate is adaptively adjusted according to real-time feedback, the operation stability is guaranteed, and the intelligence, reliability and switching smoothness of process switching of the vehicle-mounted sludge treatment system are remarkably improved.

The invention discloses an enhanced generation query method and system based on multi-granularity retrieval and reinforcement learning strategies. According to the method, semantic analysis is carried out on an original query by utilizing a large language model, the original query is decomposed into a plurality of sub-problems, and each sub-problem generates a query vector; in the multi-granularity vector knowledge base, respectively executing vector similarity retrieval on the query vector of each sub-problem to obtain a related document fragment vector of each sub-problem under each granularity, and further generating a multi-granularity vector set of each sub-problem; weighting the multi-granularity vector set of each sub-problem to obtain a comprehensive context information block, determining the weight of the multi-granularity vector set of each sub-problem through a multi-arm reinforcement learning strategy, and inputting the comprehensive context information block into a large language model to output a query result. According to the method, the retrieval accuracy is improved through semantic decomposition and multi-granularity knowledge collaborative retrieval, results are intelligently integrated through a self-learning mechanism, and lightweight self-adaptive optimization is achieved.

The invention discloses an action quality evaluation method based on video and skeleton bimodal fusion, and the method comprises the steps: inputting a video frame sequence containing a complete action process into a fragmentation network, and generating continuous fragments; secondly, inputting each video clip into a video modal feature extraction network and a human body skeleton feature extraction network, respectively obtaining a video modal feature and a skeleton modal feature, fusing the video modal feature and the skeleton modal feature through a cross-modal attention mechanism, and generating a fused clip-level feature representation; and finally, inputting each segment-level feature representation into a multi-layer Transform encoder, obtaining a global time sequence feature of the action through a self-attention mechanism, inputting the global time sequence feature into a multi-layer perceptron (MLP) regression network, and generating an action quality score. According to the method, a prediction result is ensured to be more stable in numerical value and more accord with human perception in semantics, and the accuracy and consistency of action quality evaluation are remarkably improved.

The present application relates to a multi-modal large model-based industry electricity abnormality intelligent analysis method and system, comprising the following steps: S1: obtaining power system multi-source data and preprocessing to obtain standardized time series data set; S2: based on the standardized time series data set, feature extraction is carried out, and a feature data set is constructed; S3: based on the feature data set, an industry electricity abnormality knowledge graph is constructed in combination with a multi-level correlation relationship network; S4: a multi-modal abnormality detection model is constructed by fusing time series Transformer, graph neural network and knowledge embedding module, and is trained based on the feature data set and the power knowledge graph; S5: the trained multi-modal model is deployed on a distributed stream processing platform to realize real-time abnormality detection and multi-dimensional intelligent analysis; S6: the real-time abnormality detection and multi-dimensional intelligent analysis results are visualized. The present application can realize all-round intelligent monitoring of industry electricity abnormalities facing complex industry actual demands.

The present application relates to the technical field of short-circuit current calculation, and particularly relates to a flexible DC system short-circuit current calculation method based on a physical information neural network, comprising: setting a model input feature vector, the model input feature vector being used to represent a system operating state before a fault and fault information, and performing data preprocessing on the model input feature vector; constructing a hybrid driving calculation model, the hybrid driving calculation model comprising a physical calculation module and a neural network module, and being coupled based on a preset fusion architecture; performing end-to-end training and optimization on the hybrid driving calculation model by using a preset composite loss function; and performing flexible DC system short-circuit current calculation based on the optimized hybrid driving calculation model, so that the problems of poor precision, speed and convergence, poor interpretability and weak generalization ability in the prior art are solved.

The application discloses a machine learning algorithm for screening of metal elements in a diatomic catalyst full cycle based on product yield prediction and application thereof, and the algorithm comprises the following steps: (1) collecting attribute features of metal elements in the diatomic catalyst as a feature space 1; (2) preliminarily screening the attribute features according to comprehensive scores of feature importance to obtain a feature space 2; (3) performing feature operation processing on the feature space 2 to generate a multivariate descriptor space matrix; (4) performing dimension reduction processing on the multivariate descriptor space matrix by combining a gradient boosting regression model with a recursive feature elimination algorithm; (5) further screening to obtain an interpretable descriptor by using a least absolute shrinkage and selection operator; and (6) applying the interpretable descriptor to a random forest regression model to screen a catalyst for a propane dehydrogenation reaction to prepare propylene. The application can solve the problem that a traditional descriptor is difficult to directly predict a catalytic product yield.

The application discloses a new correlation image recognition and machine learning method based on fuzzy monotony, and belongs to the technical field of artificial intelligence of pattern recognition and machine learning; the method is defined as FMMCA, which evaluates local fuzzy monotone correlation by comparing row vectors and column vectors of an image matrix pair by pair, then the local correlations are weighted and accumulated, and finally the global fuzzy monotone correlation between images is obtained. The application directly uses fuzzy monotone correlation analysis to replace classical correlation analysis for multi-view research, so that the problems existing in classical correlation analysis do not exist, and the fuzzy monotone method feature does not need to be measured statically by distance, but can be measured dynamically by interval change, so that the influence of some noise is offset, the performance is improved, a new fuzzy monotone machine learning method is formed, the optimization of a focus loss function is not needed, the parameters are few, the robustness is good, and the computing power is small.

The application is a kind of automation testing and application fluency testing method based on control matching, relating to the field of Android end automation testing.The application method comprises: setting a test script containing behavior simulation and fluency determination in advance, reading the corresponding test script after the Android end receives a test command; acquiring interface control structure positioning target control, executing simulation operation; starting timing after each simulation operation, and acquiring all control elements of the current interface when reaching the set time threshold, judging whether the interface is loaded or not in combination with the interface structure information recorded in the fluency determination script; obtaining application fluency determination result according to the interface loading condition at different time thresholds and returning to the server end.The application does not need to acquire ROOT permission, does not need to connect PC end for testing, is safer, is more convenient to use, reduces the adaptation cost, and is higher in testing accuracy and testing speed.

This invention discloses a method, apparatus, device, and storage medium for depth extraction of the focus stack weight domain. The method includes: acquiring a multi-focal plane image sequence; calculating focus sharpness pixel-by-pixel and normalizing it along the focal plane to generate an initial weight map; using the original image as a structural reference to perform edge-preserving filtering on the initial weight map to obtain an optimized weight map; constructing a weight-focal plane curve based on the optimized weight map, and sequentially performing three levels of logical judgment: the first level removes textureless regions based on the maximum weight value; the second level removes high-reflectivity false peaks based on waveform flatness; and the third level filters candidate focal plane positions based on peak focus sharpness; and performing sub-pixel fitting on the selected candidate focal plane positions to convert them into physical depth and generate a 3D point cloud. This invention effectively solves problems such as false points on textureless backgrounds, false peaks on highly reflective surfaces, edge breaks in weak textures, and limited depth accuracy through edge-preserving filtering to optimize the weight map, three-level progressive judgment, and sub-pixel fitting.

The invention discloses a credible drug target prediction method based on function space regularization, and relates to the technical field of drug research and development, and the method comprises the steps: collecting interaction data and feature data of drug molecules and EGFR mutants; constructing a drug similarity graph and a target similarity graph; defining a bilinear prediction function and a loss function; a function space regularization term based on double-graph convolution is introduced, and the regularization term forms a regularization term of the smoothness of the constraint prediction function in the joint similarity space through Laplacian matrix construction of the joint drug graph and the target graph; combining the loss function with the regularization item to construct an objective function, and performing optimization solution through a gradient descent method; and finally, predicting the interaction between the new drug and the EGFR mutant by using the optimized model. By introducing a mechanism based on double-graph convolution function space regularization, combined similarity information of a pharmaceutical chemical structure and a target sequence can be fused in drug target prediction of non-small cell lung cancer EGFR mutants.

The present application relates to power system scheduling and new energy output prediction technical field, especially to a kind of control method of distributed wind, light, storage output prediction model with scene adaptive capacity, including empirical mode decomposition, high-frequency IMF component wavelet threshold denoising, signal reconstruction, hybrid prediction modeling, energy storage system power regulation strategy.The present application introduces empirical mode decomposition and wavelet threshold denoising method, and the leading component in wind light power is extracted and high-frequency disturbance is suppressed;In the aspect of prediction modeling, the long short-term memory network LSTM is used to capture time series trend, and the KAN network is introduced to nonlinearly map high-dimensional features, to build a series-parallel double-channel collaborative prediction architecture, to realize full-scale modeling from trend to disturbance;At the system control level, combined with the prediction error feedback mechanism, the energy storage system is guided to carry out dynamic charging and discharging response, effectively suppresses wind and light output fluctuation, improves the controllability and actual availability of prediction results.

The invention belongs to the technical field of artificial intelligence and medical speech analysis, and discloses a voice state intelligent classification method based on speech spectrum features and a reinforcement learning optimization mechanism. The method comprises the following steps: carrying out data preprocessing, acoustic feature extraction and feature optimization on a tested voice sample, and mapping the preprocessed tested voice sample into a two-dimensional Mel spectrogram; realizing voice state classification by using a multi-scale feature extraction structure comprising a first convolution branch, a second convolution branch and a third convolution branch and a bidirectional time sequence feature learning structure; meanwhile, a reinforcement learning optimization mechanism is introduced, feature selection, model structure configuration, training hyper-parameters and an updating strategy are subjected to self-adaptive optimization, confidence coefficient calibration, uncertainty judgment and interpretability result generation are combined, and a final voice state judgment result, calibrated confidence coefficient and an acoustic attention area are output. The voice state classification accuracy, stability and interpretability can be improved.

This invention discloses a graph mining-based questionnaire analysis method, system, electronic device, and medium, relating to the field of data processing. The method includes: acquiring questionnaire data to be analyzed; establishing a questionnaire matrix based on the questionnaire data; cleaning the questionnaire matrix to obtain a scoring matrix; calculating the similarity between different attribute scores based on the scoring matrix to establish a core attribute adjacency matrix; using attributes in the core attribute adjacency matrix as nodes in the network to establish a core attribute network; performing graph mining on the core attribute network to obtain a minimal core attribute network; and analyzing the influence of different core attributes based on the minimal core attribute network to obtain the questionnaire analysis results. This invention can reduce the influence of extreme data in questionnaires and improve the interpretability of the data in questionnaires.

The application provides a kind of fusion voiceprint feature dental case generation method and system, through high sensitivity multi-channel pickup and edge real-time processing, realize multi-device acoustic fingerprint extraction, and based on spectrum analysis, identify device mechanical state transition;Combined with natural language processing and standard medical terminology mapping, the operation log of medical record is standardized coded, and is normalized to clinical path logic;Adopt cross-modal graph attention mechanism to realize the accurate space-time synchronization of voiceprint event and medical record event, then encrypt the corrected data binding, generate unique, anti-fake fusion data, finally through block chain distributed evidence guarantee non-tamperable, the application significantly improves diagnosis and treatment data synchronization, credible evidence and identity anti-fake ability.

The application discloses a kind of based on ranking learning's SDN data plane low-rate DDoS attack mitigation method, belong to network security field.The method includes: based on OpenvSwitch switch, polling SDN switch flow table and extracting flow table entry, form original data;Feature sextet and identification of extracting flow table entry are combined with source IP address etc.Information, mark correlation label and query ID for flow table entry;Adopt integrated learning XGBoost method, based on Pairwise, establish flow table entry ranking learning model, and deploy on SDN switch;Attack mitigation system on switch real-time monitoring whether flow table overflow caused by DDoS attack has occurred;If attack occurs, ranking learning model predicts the ranking score of each flow table entry, and rearranges flow table according to ranking score descending order, sets attack detection threshold, finally traverses flow table entry from top to bottom, decides which flow table entry should be deleted.The application has high detection rate for data plane low-rate DDoS attack, low false alarm rate and miss rate, strong adaptability, rapid and effective mitigation.

The invention relates to the technical field of scheme detection, in particular to a road design scheme detection scoring method based on a multi-modal large model, and provides the following scheme: analyzing demand information, vector linear data and three-dimensional topographic data in a road design scheme to generate structured project parameters; dynamically distributing the structured project parameters to corresponding data analysis modules based on a routing condition graph, and generating an intermediate result used for detecting scores and a retrieval condition used for standardizing retrieval; and further combining road design specification knowledge graph retrieval to obtain a specification constraint set matched with the project condition, comparing the intermediate result with the specification constraint set, determining the compliance state of each index, and generating a detection score. Through multi-module collaborative analysis and knowledge graph constraint matching, automation, interpretable detection and quantitative evaluation of a road design scheme are realized.

The invention relates to an aviation structure impact load non-negative sparse coding algorithm expansion identification method and system. The method comprises the following steps: acquiring a monitored aviation structure and making a data set comprising an impact load force signal and a response signal; constructing a non-negative sparse coding model based on an impact load dynamic model and non-negative sparse prior; constructing an iterative convergence threshold algorithm for solving the non-negative sparse coding model; constructing a depth algorithm expansion network based on the iterative convergence threshold algorithm; training the depth algorithm expansion network by using the data set to obtain a network model with optimal parameters; and performing an impact load identification test by using the network model with the optimal parameters, and outputting an impact load to be identified. The method has the advantages that the calculation speed is high, parameters do not need to be manually set, and good noise immunity is shown; the neural network provided by the invention has better interpretability in structural design.

The invention provides an image fusion method and device, and belongs to the technical field of image fusion, and the method comprises the steps: importing a plurality of original high-resolution hyperspectral images, and carrying out the down-sampling analysis of the original high-resolution hyperspectral images, and obtaining an original low-space hyperspectral image and an original high-space multispectral image; performing fusion analysis on the original low-space hyperspectral image and the original high-space multispectral image through a training model to obtain a target high-resolution hyperspectral image; and updating and analyzing the training model according to the target high-resolution hyperspectral image and the original high-resolution hyperspectral image to obtain an image fusion model. According to the method, the fidelity of the space and the spectrum of the fused image is kept, the calculation requirement is reduced while the spectrum and the space information are recovered, the calculation complexity is also reduced, better algorithm interpretability is achieved, and therefore a more accurate fusion result is obtained.

The invention relates to the technical field of medical artificial intelligence and disease risk prediction, and discloses a pancreatic cancer risk prediction model optimization method based on metabolic syndrome data, and the method comprises the steps: a training stage: obtaining historical clinical data of five physiological abnormalities of a sample individual, carrying out the grading assignment of the historical clinical data based on a metabolic syndrome diagnosis standard, and carrying out the prediction of a pancreatic cancer risk prediction model; in the application stage, the clinical data of a target individual are subjected to same preprocessing and feature engineering and input into the risk prediction model, and the risk prediction model is obtained by combining a pancreatic cancer diagnosis tag and training a logistic regression model through a maximum likelihood estimation method. And calculating to obtain a pancreatic cancer risk assessment result of the target individual. According to the method, data modeling is carried out by innovatively integrating metabolic syndrome data and diabetes disease course information, accurate and early individualized prediction of the pancreatic cancer risk is realized, and an effective tool is provided for clinical screening and intervention.

The invention discloses a highway abnormal parking identification method and identification system based on Internet of Vehicles big data, and the method comprises the following steps: obtaining vehicle data reported by a current vehicle, and judging whether the current vehicle is located in a main road driving interval of a highway; judging starting and ending points of the parking event according to the vehicle speed, the gear, the flameout state and the displacement distance; based on spatio-temporal context analysis of vehicle data, if it is judged that the vehicle is in a congestion state, the judgment that the vehicle is abnormally stopped is excluded; on the basis of GIS geographic information system data of the expressway, if a vehicle is obtained in a specific functional area, the judgment that the vehicle is abnormally stopped is eliminated; the determined parking event vehicle is the abnormal parking vehicle on the expressway. According to the method, the state of the vehicle is fused with multi-source information such as the high-precision map and the surrounding traffic flow, a multi-check mechanism is constructed, the detection accuracy is remarkably improved, misinformation and missing report are greatly reduced, abnormal parking and normal traffic phenomena are effectively distinguished, and practicability is high.

This invention discloses a deep learning-based method for self-tuning controller parameters in a motor servo system. The method includes: constructing a hybrid model combining a mechanistic model and feedforward neural network residual compensation; performing multi-objective optimization based on this model to obtain a set of initial offline controller parameter values; and constructing a parameter adjustment model based on a Long Short-Term Memory (LSTM) network during online operation. This model takes the system state sequence as input and outputs the percentage change in parameter offset relative to the offline initial values, achieving online, constrained, and small-range dynamic tuning of the controller parameters. This invention improves the reliability of performance prediction through the hybrid model and ensures engineering safety through a constrained online adjustment mechanism, thereby significantly improving the tracking accuracy, stability, and adaptability of the servo system under complex and variable operating conditions.

The application is suitable for the field of artificial intelligence and formal verification technology, and provides a conversion system and method from natural language to linear temporal logic based on self-learning, which comprises a text conversion module, an equivalence checking module, a rule induction module and a rule library management module; the text conversion module is used for generating a predicted formula according to a basic rule and a retrieval rule; the equivalence checking module is used for verifying the equivalence of the predicted formula and a target formula and extracting a failure sample; the rule induction module is used for analyzing the failure sample and automatically inducing a new conversion rule; and the rule library management module is used for storing the rules and retrieving based on semantic similarity. The system can automatically learn from conversion errors and continuously optimize the rule library. The application overcomes the dependence on artificial rules or labeled data in the prior art, realizes unsupervised adaptive learning, and significantly improves the accuracy and field adaptability of the conversion from natural language to linear temporal logic.

The invention relates to a bridge data reconstruction method based on multi-scale space-time fusion and uncertainty perception, belongs to the technical field of civil engineering structure health monitoring and artificial intelligence data processing, and aims to solve the problems that space-time feature mining lacks dynamic adaptability, multi-scale feature fusion is insufficient and the like in the prior art. The method comprises the following steps: collecting time sequence data of a multi-dimensional sensor and generating a structured data set; extracting multi-scale time sequence features by using a parallel causal convolutional network; constructing a dual-path time sequence coding architecture comprising an LSTM main path and a GRU auxiliary path; after hierarchical attention and adaptive time sequence fusion, sensor spatial correlation dynamic modeling is introduced to generate high-dimensional enhanced features; a dual-branch network is utilized to predict the uncertainty of the mean value and the heterovariance, end-to-end optimization training is carried out through a mixed loss function, and a trained model is used for outputting reconstructed bridge data with an uncertainty confidence interval. According to the method, the precision and reliability of bridge monitoring data reconstruction can be remarkably improved.

The application discloses an echo cancellation method based on a room impulse response estimation model and belongs to the technical field of acoustic signal processing, and comprises the following steps: step S1, a room impulse response estimation model is built; step S2, the room impulse response estimation model is trained, evaluated and verified; step S3, double-speech detection is performed; and step S4, according to the detection result, no speech, single-speech echo speech and double-speech echo speech can be obtained, and echo cancellation is performed on the different paths. The application mainly aims at processing echo problems in the network audio call and video conference environment, can weaken and eliminate the echo problems, improves the voice and video call quality, and is suitable for a wide range of applications.

The application provides a robot joint friction model parameter identification method, device and equipment. The method provided by the application comprises the following steps: a speed-related friction model, a temperature-related friction model and a load-related friction model are established to constitute a robot joint friction model; a plurality of groups of friction characteristic data of a robot joint in different working states are acquired, and initial value ranges of various parameters in the model are set based on expert knowledge and prior physical knowledge to constitute an initial parameter search space; an error square sum of model prediction is taken as a fitness function, the initial parameter search space is optimized based on the plurality of groups of friction characteristic data by using a genetic algorithm to obtain a candidate parameter search space; and an optimization function with the optimization objectives of minimizing a prediction error and minimizing a parameter physical deviation is constructed for the robot joint friction model.

The application discloses a kind of low-voltage alternating current system insulation defect early warning method and system based on incremental admittance, the residual voltage and current of each line in substation station low-voltage alternating current system are collected, residual voltage and current mutation variable, Fourier value are structured, and starting threshold trigger detection is set;When meeting the starting condition of insulation state early warning detection, controllable voltage disturbance is applied to substation station low-voltage alternating current system by neutral point voltage regulating device;Synchronous acquisition line residual voltage and each feeder residual current before and after disturbance, extract equivalent ground conductance component and equivalent ground admittance component;According to equivalent ground conductance component and equivalent ground admittance component, the leakage current constant of each feeder is calculated, and insulation state quantitative characteristic quantity is constructed;The insulation state quantitative characteristic quantity is compared with preset criterion, and the insulation state type of each feeder is identified according to the comparison result, and is respectively determined as normal state, recoverable insulation defect or unrecoverable insulation defect.

The invention discloses a fish body length measuring method and system. The method is realized based on geometric measurement and statistical model fusion. According to the method, stable and reliable fish body length measurement under the conditions of complex fish body postures and segmentation errors is realized by constructing a two-channel length estimation method of a geometric measurement and area statistics model and combining a shape quality index-based adaptive fusion mechanism. Specifically, geometric measurement length is obtained through fish body contour extraction and skeleton analysis, a statistical length estimation model is constructed based on fish body area and shape features, and area estimation length is obtained. In order to improve the stability of the system in a complex scene, a self-adaptive fusion mechanism based on a shape quality index is further constructed, indexes such as fish body contour integrity, contour smoothness and a region filling rate are evaluated, and a measurement credibility weight is calculated; and carrying out adaptive fusion on the geometric measurement result and the area estimation result according to the weight to obtain a final fish body length estimation value.

The invention discloses a multi-element granite buried hill reservoir quantitative evaluation method which comprises the following steps: acquiring and counting age, structure and fluid related data of a granite buried hill in a target area; performing normalization processing on the related data; and based on the normalized data, obtaining a quantitative evaluation result of the development degree of the granite buried hill reservoir through comprehensive calculation. According to the method, all the influence factors are subjected to full-process parameterization, and a quantitative calculation method which is easy and convenient to operate, transparent in process and capable of being repeatedly achieved is constructed. The method not only can identify the maximum risk factor influencing the development of the buried hill reservoir, but also breaks through the limitation of a traditional'black box 'prediction model dependent on statistical association, so that the prediction process has better interpretability. Reliability and universality of granite buried hill reservoir prediction are improved, intuitive quantitative evaluation and comprehensive queuing of different buried hill target reservoir development degrees are realized, and a scientific basis is provided for decision and deployment of granite buried hill oil-gas exploration.

The invention discloses a multi-fidelity model construction method based on hyperplane correction, and belongs to the technical field of multi-fidelity agent models. According to the invention, the multi-fidelity model approximately replaces an expensive high-fidelity model. The core of the method is that a hyperplane correction mechanism is introduced to explicitly learn and correct a feature mapping relationship between models with different fidelity in an input space, so that the dependence on the number of high-fidelity samples is remarkably reduced, and meanwhile, the low-fidelity model and the high-fidelity model are accurately corrected; and finally, a multi-fidelity agent model with high prediction precision and strong generalization ability is constructed. In the face of new design points or working conditions with strong nonlinearity and multi-modal characteristics, the model constructed by the method shows better generalization performance and prediction robustness, and the risk of optimization decision errors caused by model mismatch is reduced; the method provides a visual quantitative basis for the main source of the difference between the high-fidelity model and the low-fidelity model, and has better interpretability.

The invention discloses an automatic prompt optimization method and system for retrieval guidance, equipment and a medium. The method comprises the following steps: acquiring a to-be-identified customer dialogue; retrieving a plurality of annotated dialogue examples similar to customer dialogue semantics from a pre-constructed insurance field corpus; generating a plurality of candidate prompts in combination with the basic prompt template according to the example dialogue texts and the labeled intention labels in the plurality of labeled dialogue examples; performing reasoning enhancement processing on the plurality of candidate prompts to obtain a plurality of optimized candidate prompts; and evaluating the intention recognition accuracy of each optimized candidate prompt by utilizing a verification data set containing a real intention label, and selecting an optimal prompt from the plurality of optimized candidate prompts. According to the method, the problems that manual engineering prompting depends on expert knowledge, coverage is incomplete, stability is poor and maintenance cost is high are solved, the defect that traditional model fine adjustment depends on large-scale annotation data and computing resources is overcome, and intention recognition accuracy and adaptability in the insurance field are improved.