1112 results about "Feature learning" patented technology

The invention relates to the technical field of drainage basin water scheduling, and discloses a drainage basin water regulation and control optimization method based on ecological element changes. The method comprises the following steps: deploying a drainage basin monitoring system, and collecting ecological element real-time data such as a hydrological parameter sequence and a remote sensing image; after the data is cleaned and converted, hydrological trend features and spatial distribution features are extracted by adopting a feature learning model, and the hydrological trend features and the spatial distribution features are fused into unified ecological representation through a cross-modal alignment mechanism; inputting the unified ecological representation into a physically constrained neural network prediction model, and outputting a water regimen dynamic prediction value; and finally, based on the predicted value, a water resource regulation and control instruction is generated and executed by using a multi-objective decision algorithm so as to optimize the watershed water circulation process. According to the method, feature extraction comprehensiveness is improved through multi-source data fusion and cross-modal analysis, prediction reliability is enhanced in combination with physical constraints, reasonable allocation of water resources is achieved by means of multi-target decision, the ecological condition of a drainage basin can be improved, and the water utilization efficiency is improved.

The invention discloses a modeling method based on shield tunneling data feature analysis and parameter relevance, and relates to the field of tunnel engineering data processing. The method comprises the steps that shield tunneling time sequence parameters are obtained, and a non-uniform time sequence is resampled into a space-aligned standardized footage domain sequence through state cleaning and coordinate domain transformation; by means of mixed variable rejection and lagging correlation analysis, environment common cause interference is stripped, physical response delay among parameters is recognized, and a time-delay directed correlation graph model is constructed; and inputting the footage domain sequence and the graph model into a graph neural network, performing feature learning by using a time delay compensation aggregation mechanism, and outputting a key parameter influence degree set with symbols based on a prediction gradient. According to the method, the problem of data space-time dislocation caused by propelling speed fluctuation and the problem of parameter relevance misjudgment caused by physical response lag are solved, and accurate identification and explanation of shield tunneling key parameters are achieved.

The invention discloses a hybrid expert and KAN-based cyclic attention network time sequence prediction method. The method comprises the following steps: S100, inputting time sequence data needing to be predicted; s200, constructing a graph structure by using an attention mechanism, learning basic correlation characteristics among variables of the input time sequence data through an adaptive and learnable graph convolutional network, and then performing global averaging and maximum pooling on the basic correlation characteristics along a time dimension to obtain complementary time domain statistical information, so as to provide effective time-space correlation characteristics for the follow-up process; s300, after feature learning is completed, collaborative modeling of the KAN and an attention mechanism is brought into full play, rapid and efficient time sequence modeling is carried out on data by adopting a cyclic attention network embedded based on the KAN, and a foundation is laid for subsequent time sequence prediction; and S400, establishing a hybrid KAN expert-based time sequence prediction network, and adaptively fusing differentiation prediction results by a gating mechanism. The time sequence prediction method is designed from the three aspects of feature learning, time sequence modeling and time sequence prediction.

The invention discloses a CAD automatic modeling method and system based on parameter driving. The method comprises the steps that model parameter information needing modeling is obtained; generating a CAD (Computer Aided Design) model on the basis of parameter analysis, geometric construction and surface recognition schemes; performing reverse reconstruction of the three-dimensional CAD model based on topology analysis, geometric feature extraction, pattern recognition and parameter relation inference; modeling, optimization and feedback evolution are carried out based on model analysis, feature learning and generation of parametric modeling rules; and according to the obtained data information, CAD automatic modeling based on parameter driving is completed. According to the method, CAD automatic modeling based on parameter driving is achieved, the reliability is higher, the accuracy is better, and the efficiency is higher.

The embodiment of the invention provides an equipment fault prediction management method and device based on a large model, and the method and device achieve the accurate evaluation of a state through the innovative design of a multi-modal feature fusion model, data integration and feature learning. A fault diagnosis system is constructed, and a reliable root cause analysis mechanism is established in combination with semantic analysis and case retrieval. Maintenance guidance is introduced, and the feasibility of a maintenance scheme is ensured through experience precipitation and priority evaluation. According to the method, the defects of the traditional technology in the aspects of feature extraction, fault diagnosis, maintenance guidance and the like are effectively overcome, and technical guarantee is provided for equipment management.

The invention provides a blood pressure dynamic monitoring system integrating overall risk and local anomaly detection, and relates to the field of electric digital data processing. Comprising a multi-source blood pressure data fusion and acquisition module, a dual-path feature learning and abnormity pre-detection module, a local and overall interactive abnormity accurate identification module and a dynamic risk assessment and intelligent early warning decision module, and the multi-source blood pressure data fusion and acquisition module is used for acquiring blood pressure and context data; the dual-path feature learning and anomaly pre-detection module is used for extracting time sequence features and performing preliminary anomaly screening, and the local overall interactive anomaly accurate identification module is used for detecting and analyzing local anomaly. The dynamic risk assessment and intelligent early warning decision module is used for comprehensively assessing the risk, forming a feedback optimization mechanism and outputting a personalized early warning decision; the system can accurately identify abnormal blood pressure, realizes accurate assessment and prediction of risks, and provides effective support for clinical decision and personalized health management.

The invention relates to the technical field of model analysis, in particular to an imaging flow cytometry cell detection method based on an improved model. The method comprises the following steps: introducing a cell sample to be detected into an imaging flow cytometry system integrated with a micro-fluidic chip for continuous image acquisition to generate an initial cell image sequence; an automatic digital focusing algorithm is applied to the initial cell image sequence, and a cell image frame set with the optimal focal plane is screened out; inputting the cell image frame set into a preset PA-YOLO improved model for multi-dimensional extraction and fusion, and generating a multi-scale cell characteristic spectrum; carrying out refined feature learning and cell target positioning and classification on the multi-scale cell feature spectrum, and outputting a cell detection result; and carrying out validity verification on the cell detection result, and carrying out comparative analysis in combination with an imaging flow cytometry system to generate a cell detection report. According to the method, the imaging quality and the detection accuracy of cell images with different depths can be remarkably improved.

The invention belongs to the field of environment monitoring and computer vision, and particularly relates to a water body color recognition regression method and system based on space-time causality and manifold learning, and the method mainly comprises the steps: carrying out the detection of a current target water body video sequence, extracting a water body region, carrying out the high-dimensional feature dimension reduction of the water body region, and obtaining a water body color recognition result; and performing feature extraction on the water body region through a space-time causal feature learning model, fusing the flow shape learning features and the space-time causal features to obtain fused features, and outputting a finally predicted water body color value. According to the method, end-to-end assembly line design of preprocessing-segmentation-feature modeling-regression is adopted, manual intervention is not needed from video input to color prediction, and through cascade cooperation of five core modules (video preprocessing, water body segmentation, manifold learning, time sequence causal modeling and color recognition), the real-time performance of the system is improved. Full-link automation from environmental interference suppression, feature extraction to result output is realized, information loss of intermediate links is avoided, and recognition efficiency and robustness are improved.

The invention provides a dense overlapping target detection method based on a wavelet enhancement sparse hybrid expert model. The method comprises the following steps: firstly, extracting multi-layer features through a backbone network to capture multi-scale spatial representation; secondly, discrete wavelet transform is introduced to each level of features, spatial features are decomposed into a frequency domain, collaborative modeling of frequency domain and spatial domain features is realized, the reservation capability of detail and texture information is improved, a lightweight dynamic hypergraph aggregation module is introduced into the deepest layer of features, a hyperedge structure is adaptively learned, and the feature fusion is realized; modeling a high-order incidence relation in a local area in an explicit manner; and thirdly, in the decoding process, candidate queries are screened and reweighted through an IoU perception query selection mechanism, and a dynamic routing mechanism of sparse hybrid experts is introduced, so that query self-adaptive specialized representation learning is realized, and the target detection precision and reliability in a complex scene are effectively improved.

The invention relates to a robust unmanned aerial vehicle detection method based on dynamic feature fusion and context attention, and belongs to the technical field of image processing. Aiming at the problems of small target feature loss, semantic gap, background noise interference and the like caused by a fixed convolution kernel scale, one-way feature fusion and a static attention mechanism in an existing unmanned aerial vehicle aerial image target detection method, the method comprises the following steps: constructing a detection model comprising a backbone network, a neck network and a detection head network; a feature rearrangement and extraction module is designed in the backbone network to enhance feature learning, an enhanced double-flow feature fusion pyramid is designed in the neck network to optimize multi-scale feature fusion, and a dynamic multi-scale context attention mechanism is designed in the detection head network to suppress irrelevant background noise. The method effectively improves the accuracy and robustness of small target detection, and achieves a clearer and more stable detection effect in a complex environment.

The invention discloses a large-model-driven cross-modal power insulator defect detection method and system, and the method comprises the steps: constructing an image-text dual-channel feature learning frame based on a pre-trained vision-language large model, introducing fine-grained semantic description as prior information, and carrying out the detection of the defects of a cross-modal power insulator. Feature enhancement and space focusing under semantic guidance are realized through a cross-modal attention fusion mechanism, and the robustness and discrimination ability of the model in a complex scene are further improved in combination with a multi-scale pyramid structure and a channel re-calibration mechanism. The method can effectively solve the practical problems of difficult recognition of tiny targets, strong environmental interference, limited sample data and the like.

The invention belongs to the technical field of image processing and deep learning, and particularly relates to an image defogging method based on dynamic wavelet prior and double-domain learning. Aiming at the requirements of all-weather clear imaging in the fields of intelligent traffic systems, safety monitoring and the like, and in order to overcome the defect that a static convolution kernel adopted by a traditional defogging method is difficult to adapt to different haze degradation, the invention provides a method for dynamically generating a convolution kernel by using haze priori contained in a multi-scale wavelet LL sub-band; and an efficient, robust and accurate image defogging model is constructed. According to the invention, based on a multi-scale U-shaped coding-decoding architecture, a dynamic wavelet depth separable convolution module DyWConv is embedded in front of each level of a coder to realize content adaptive feature extraction, and a double-domain feature learning module SPAFormer Block cooperatively utilizing Fourier domain global modulation and wavelet domain multi-scale decomposition is designed. And double-domain features are fully fused through an adaptive gating fusion mechanism, and finally a clear image is reconstructed and output step by step. According to the method, a method for explicitly encoding frequency domain degradation prior into dynamic convolution kernel parameters is innovatively provided, the complementary advantages of Fourier transform and wavelet transform are cooperatively utilized, spatial non-uniform haze can be effectively removed, image details can be recovered, leading performance is achieved in a synthetic data set and a real scene, and the method has a wide application prospect.

The invention relates to the technical field of artificial intelligence, and discloses an event prediction method and system based on multi-modal fusion, and the method comprises the steps: constructing a knowledge graph encoder, and converting domain expert knowledge into learnable vector representation; constructing a multi-granularity feature extraction network, and extracting features from different microcosmic, mesoscopic and macroscopic scales; realizing a knowledge-guided attention mechanism, and dynamically adjusting feature scale importance; constructing a prototype learning module, and establishing prototype representation of the abnormal category; a knowledge migration mechanism is constructed, and the generalization ability of the model to novel anomalies is enhanced; and multi-granularity abnormal event detection and early warning are realized, and a detection result and interpretable analysis are output. According to the method, through combination of knowledge guidance and multi-scale feature learning, efficient identification and early warning of social abnormal events under the condition of data scarcity are realized, and the method is suitable for the fields of public place safety monitoring, urban traffic safety management, large-scale activity safety guarantee and the like.

The invention provides a power grid mountain fire prediction method based on causal driving and a space-time diagram convolutional network, and belongs to the technical field of mountain fire prediction. A dynamic feature encoder and a static feature encoder are designed to extract high-dimensional spatial-temporal features of meteorological time sequence information, geographic space environment and power transmission line distribution multi-source heterogeneous data, a causal discovery algorithm is adopted to construct a dynamically evolved causal graph topology, a real causal driven relationship between variables is identified, a causal intensity matrix is decoupled into positive and negative adjacent matrixes, and the dynamic evolved multi-source heterogeneous data is obtained. And designing a causal constrained graph convolution module to aggregate and propagate high-order information, and finally realizing accurate prediction of the power grid forest fire. According to the method, the PCMCI causal discovery algorithm is introduced, so that the real causal driven relationship among multivariate time sequence factors is effectively identified; a causal GCN layer in the CSTGCN model extracts spatial dependence features by using an adjacent matrix constrained by a causal structure, the CSTGCN model embeds causal structure information into a spatio-temporal feature learning framework, and higher prediction precision and generalization performance are achieved.

The invention discloses a network security attack path prediction system based on a graph neural network, and relates to the technical field of network security, and the system comprises a data collection module which collects network full-link time sequence security data, and outputs standardized time sequence data through time sequence alignment and abnormal noise reduction processing; the time sequence diagram construction module is used for constructing a dynamic attack graph containing nodes and time sequence edges; the feature learning module introduces a time sequence attention mechanism, calculates a time-space fusion attention coefficient based on a graph attention network framework, and outputs a node embedding vector; the reasoning and pruning module is used for generating attack paths based on node embedding vectors and outputting a high-value attack path set; and the analysis decision module is used for carrying out importance sorting on all nodes on the high-value attack path, determining a path core risk point and generating a key node decision basis of the attack path. According to the method, the problem that the traditional technology cannot accurately capture the attack behavior time sequence dependence is solved, and the high-precision prediction of the attack path is realized.

The invention discloses an explainable appearance feature learning and quantitative representation method for aerodynamic parameter prediction, and belongs to the technical field of aerodynamics and artificial intelligence crossing. The method comprises the following steps: constructing a collaborative network architecture comprising an aerodynamic prediction module, an airfoil concept learning module and a quantitative distillation agent module; while high-precision and high-efficiency aerodynamic parameter prediction is realized, a prediction result can be decomposed into the sum of quantitative contributions of different airfoil profile concepts, so that a direct explainable basis is provided for the prediction result, and expert users are assisted in understanding and verifying the prediction process of the model.

The invention discloses a global dynamic perception space-time traffic flow prediction method based on graph packet representation learning, and belongs to the technical field of traffic flow prediction, and the method comprises the following steps: S1, traffic data input, S2, traffic graph packet construction, S3, graph packet initial feature extraction, S4, time sequence feature extraction, S5, spatial feature extraction, and S6, traffic flow prediction and output. Through a space-time modeling technology of graph packet representation learning and global dynamic perception, space-time characteristic elements of a traffic road network can be comprehensively covered, traditional traffic indexes such as flow and speed are concerned, elements such as road network topological association and cross-regional multi-hop association are also included, a dynamic dependency relationship between a time sequence and a spatial dimension is deeply mined, and a real-time dynamic perception effect is achieved. Therefore, the prediction result can reflect the real evolution law of the traffic flow more accurately, and a more scientific basis is provided for traffic management and decision making.

The invention discloses a household energy storage dynamic optimization method and system combined with load feature learning, which are used for reducing prediction errors caused by sudden loads and improving response efficiency and economical efficiency of a household energy storage system. The method comprises the following steps: calculating a prediction error between real-time total load data and a basic prediction result, identifying and locking one or more high-power electric appliances causing the prediction error as target electric appliances, and decomposing and extracting historical operation state data of the target electric appliances from the total load data; constructing a time sequence feature vector based on the historical operation state data, and training by using a recurrent neural network model to extract the sudden load feature of the target electric appliance; fusing the optimized sudden load characteristics with a basic prediction result to generate a total load prediction curve; and based on the total load prediction curve, combining the residual capacity and the operation health state of the household energy storage system, and adopting a multi-objective optimization method to dynamically generate a charging and discharging strategy.

A system for learning optimal driving behavior for autonomous vehicles comprises a deep neural network, a first stage training module, and a second stage training module. The deep neural network comprises a feature learning network configured to receive sensor data from a vehicle as input and output spatial temporal feature embeddings and a decision action network configured to receive the spatial temporal feature embeddings as input and output an optimal driving policy for the vehicle. The first training stage module is configured to, during a first training stage, train the feature learning network using object detection loss. The second stage training module is configured to, during a second training stage, train the decision action network using reinforcement learning.

The invention discloses a hardware tool defect online detection system based on AI image recognition, and belongs to the technical field of image generation type adversarial networks. A YOLOv8 detection framework model is trained, a CBAM attention module is specifically added into a YOLOv8-S detection model, feature extraction of a hardware tool rare defect area is enhanced, feature distribution of hardware tool rare defect samples generated by a GAN is fused in a network bottleneck layer, the feature learning ability of the YOLOv8-S detection model for hardware tool rare defects is improved, and the hardware tool rare defect feature extraction method based on the CBAM attention module is obtained. The detection effect on the rare defects of the hardware tool is enhanced, and the problem of model performance bottleneck caused by scarcity of rare defect samples of the hardware tool is solved.

The invention provides a Mini led display screen dead pixel detection and positioning method, device and system based on deep learning, and the method comprises the steps: applying a preset driving signal sequence to a pixel array of a Mini led display screen line by line, obtaining the brightness feedback data of each pixel in different driving stages, obtaining a brightness response sequence, and carrying out the detection and positioning of the dead pixel of the Mini led display screen. Carrying out spatial recombination processing on the brightness response sequence according to a physical arrangement structure of the pixel array to generate a brightness response matrix, carrying out local region feature difference analysis on the brightness response matrix, identifying a brightness response abnormal region, generating a candidate dead pixel position set, and obtaining a candidate dead pixel position set; and calling a pre-trained dead pixel recognition network to carry out feature learning processing on the candidate dead pixel position set, outputting a dead pixel probability corresponding to each candidate dead pixel position, screening the candidate dead pixel position set based on the candidate dead pixel probability, removing candidate positions with the dead pixel probability lower than a preset condition, and generating a dead pixel positioning result set. According to the invention, the accuracy and reliability of detecting and positioning the defective pixels of the Mini LED display screen can be effectively improved.

The invention provides a multi-source data fusion driven electric bus battery SOC prediction method. According to the method, firstly, data preprocessing and feature extraction are performed on collected electric bus operation data, then, for the problem that a traditional model is insufficient in complex time sequence data feature extraction capability, on the basis of a TFT model, the time sequence feature learning advantage of an LSTM network is fused, an improved TFT model is constructed, and the time sequence feature extraction capability of the LSTM network is improved. The combined modeling capability of the model for short-term change and long-term dependence features is enhanced, and the improved TFT model can select and process extracted multi-dimensional data features. A prediction result of the improved TFT model is compared with an existing typical prediction model, and the prediction performance of the improved TFT model is superior to that of the existing prediction model on a plurality of evaluation indexes. The electric bus battery SOC prediction method provided by the invention has prediction stability and accuracy under complex charging and discharging working conditions, and provides technical support for energy consumption management and intelligent control of the electric bus.

The invention relates to a geological disaster change detection method and system based on an improved twin U-Net and a central surrounding double-flow network. According to the method, an improved twin U-Net network is constructed as a feature extraction trunk, and a deformable convolution module is integrated in an encoder to adaptively adjust a receptive field; a central surrounding double-flow network is embedded in a decoding path, detail features are extracted through a central flow path, and a global context is obtained through a surrounding flow path; a feature fusion module is designed to realize double-path feature deep fusion, and a gating attention unit is adopted to calibrate feature response; introducing a contrast feature learning mechanism to reinforce feature space clustering characteristics; multi-modal degradation enhancement training is implemented to improve the robustness of the model; and an objective function containing difference perception loss and feature comparison loss is minimized through end-to-end joint optimization. The method can accurately identify geological disaster change areas such as landslide, debris flow and land subsidence, and has the advantages of high detection precision and strong anti-interference capability.

The invention discloses a multi-modal data adaptive denoising and missing reconstruction method and system, and relates to the technical field of point data denoising and reconstruction, and the method comprises the steps: obtaining to-be-processed multi-modal original data and a modal missing mask; performing unsupervised denoising on image data in the multi-modal original data to obtain a denoised image, and further obtaining multi-modal data; inputting the multi-modal data into a double-flow encoder for processing to obtain a multi-modal embedded vector of cross-modal alignment; the method comprises the following steps of: performing mapping and adding position embedding on a modal embedding vector to obtain each modal coding feature, determining a missing modal based on a modal missing mask, inputting an available modal coding feature into a retrieval enhanced expert model based on prototype memory to perform missing reconstruction to obtain a multi-modal joint representation, and mapping the multi-modal joint representation to a task output space through a full connection layer. Through introduction of unsupervised denoising, double-flow coding alignment and modal knowledge expert hybrid reconstruction, robust representation learning and information complementation under the condition that noise and modal missing exist in multi-modal data are realized.

The invention relates to the technical field of city data representation. The invention discloses an urban space general representation learning method and device based on multi-modal spatio-temporal data fusion, a terminal and a storage medium, which can improve the applicability of urban general representation and enable the urban general representation to be applicable to diversified urban analysis tasks. The method comprises the following steps: acquiring multiple modal spatio-temporal data of each space unit in a target city, and setting a corresponding view for each modal spatio-temporal data; on the basis of each modal spatio-temporal data of each space unit, generating a representation of each space unit in a single view under a view corresponding to each modal spatio-temporal data; generating a multi-view fusion representation corresponding to each space unit based on the representation of each space unit in a single view under the views corresponding to all modal spatio-temporal data, and forming a multi-view fusion representation matrix by the multi-view fusion representations corresponding to all the space units; and performing global aggregation processing on the multi-view fusion representation matrix to obtain urban general representation.

The invention provides an extreme rainfall prediction method and device based on deep learning and electronic equipment, and the method comprises the steps: collecting multi-source data, carrying out the preprocessing of the multi-source data, and constructing a multi-source heterogeneous fusion data set; inputting the multi-source heterogeneous fusion data set into a multi-scale feature extraction network, and extracting space-time evolution features of the rainfall process; generating a synthetic extreme rainfall sample through a generative adversarial network, carrying out mixed training on the synthetic extreme rainfall sample and a real sample, and strengthening extreme rainfall feature learning through a weighted loss function; a channel attention and space attention mechanism is introduced, the feature weight is dynamically adjusted, and the time sequence dependency relationship of the rainfall process is captured. According to the invention, accurate prediction and early warning of the extreme rainfall event are realized, and the problem of poor extreme rainfall prediction accuracy in the prior art is solved.

The invention provides a medical image visualization analysis system based on deep learning, and relates to the field of artificial intelligence. The objective of the invention is to overcome the defects of an existing system in the aspects of deep learning model interpretability, clinical interactivity and multi-modal data fusion. The system comprises an image data acquisition and standardization module, a deep feature extraction and representation learning module, a multi-task intelligent analysis module, an interpretability analysis module, a multi-dimensional visualization and interaction module, a clinical knowledge fusion and feedback learning module and a system management and integration module. Through the system, the diagnosis efficiency and accuracy can be improved, the trust of doctors is enhanced, the obstacle of a traditional black box model is overcome, and a new man-machine cooperation intelligent diagnosis normal form is constructed.

The invention discloses a photovoltaic array fault diagnosis and positioning method and system based on digital twinning and deep learning. The method comprises the following steps: constructing a photovoltaic array digital twin, synchronously collecting multi-source monitoring data and converting the multi-source monitoring data into a time-frequency spectrum; constructing a twin network model, carrying out feature learning through a triple loss function, and extracting high-discrimination depth features; an adversarial transfer learning mechanism is introduced, cross-working-condition domain invariant feature extraction is realized through adversarial training of a feature generation unit and a domain discriminator, and the diagnosis robustness is improved; a multi-source domain generalization strategy is adopted, domain invariant features and domain private features are extracted through a double-branch network, difference constraints are applied, adaptive fusion is carried out, and a generalization diagnosis model oriented to unknown working conditions is constructed; and integrating the model to a digital twinborn body to realize accurate positioning and visualization of a fault component. According to the method, the problems of low fault diagnosis precision and poor generalization ability of the photovoltaic array in data scarcity, variable working conditions and unknown environments are solved.

The invention discloses a dam safety detection method and system based on multi-source detection data, and relates to the technical field of dam safety detection, and the method comprises the steps: cooperatively collecting dam multi-source data through multiple platforms and multiple sensors; the collected multi-source data are preprocessed; constructing a multi-scale teacher network, and performing high-precision feature learning and risk quantification by using labeled multi-source data; teacher network knowledge migration is carried out through a knowledge distillation technology, and a lightweight student network is trained in combination with cross-domain pseudo data; and deploying the trained lightweight student network to an unmanned aerial vehicle or a robot dog, and carrying out dam safety real-time detection. Through integration of multi-source data acquisition, cross-domain mutual training of teachers and students, lightweight model deployment and automatic early warning decision, pain points of single data, difficulty in cross-domain adaptation, insufficient precision, deployment limitation and low efficiency are solved, the system can be directly deployed on an unmanned aerial vehicle or a robot dog, dependence on cloud computing power is not needed, data acquisition and analysis time delay is reduced, and the system can be widely applied to unmanned aerial vehicles or robot dogs. And emergency scenes are quickly responded.

The invention provides an industrial cognitive base system based on multi-modal comparative learning and an execution method, and the industrial cognitive base system based on multi-modal comparative learning is characterized in that a system architecture comprising a knowledge graph management module and a cross-modal coding fusion module is constructed; industrial field knowledge is dynamically injected into a multi-modal feature learning process in the form of a structured sub-graph, and a comparative learning mechanism introducing industrial semantic constraints is combined, so that a model obtained by final training can realize semantic alignment of multi-modal data; it can be ensured that generated unified semantic representation and reasoning results strictly conform to predefined industrial logic rules, and therefore the defect that in the prior art, a pure data driving method may generate results violating industrial common knowledge is effectively overcome; and the output reliability, the decision credibility and the practical application value of the industrial cognitive system in key tasks such as fault diagnosis and state monitoring are remarkably improved.