67 results about "Cyclic network" patented technology

The invention relates to a traffic flow prediction method based on a multimode dynamic memory graph convolutional network, and belongs to the technical field of intelligent traffic. An existing graph convolutional neural network method is insufficient in space-time correlation characteristic extraction, insufficient in periodic characteristic consideration and inaccurate in dynamic evolution correlation capture. In order to solve the problem, firstly, a time sequence feature extraction module and a bidirectional memory loop network module are established, and captured time sequence features are fused to form a comprehensive feature vector; then, in a dynamic graph convolution module, through a fusion mode of a diffusion graph convolution neural network, an attention mechanism and a typical traffic mode, capturing spatial correlation between nodes, and generating a new dynamic adjacency matrix; the dynamic adjacency matrix can reflect the node connection relation changing along with time. According to the method, the adjacent matrix is dynamically updated through the multimode dynamic memory graph convolutional network, emergencies and abnormal conditions can be better dealt with, and the accuracy and robustness of traffic flow prediction are effectively improved.

The invention discloses a charging and moisture regaining equipment temperature prediction method based on a time sequence fusion network model, and relates to the field of production process control, and the method comprises the steps: collecting time sequence data in a charging and moisture regaining equipment production environment, and carrying out the preprocessing; dividing the data set into a training set, a verification set and a test set, and injecting Gaussian noise into the training set; a prediction model for predicting the outlet temperature is constructed, and the prediction model is a time sequence fusion network model and comprises a residual TCN time sequence convolutional network, an SK-Net multi-scale attention network and a BiLSTM bidirectional circulation network; pre-training the prediction model by using the training set; utilizing the trained prediction model to predict the outlet temperature of the feeding and moisture regaining equipment; and evaluating a prediction result, and if an evaluation index is greater than a threshold value, starting an incremental training process to re-train the prediction model. According to the invention, through a multi-module combined deep learning model, the prediction accuracy of the outlet temperature of the charging and moisture regaining equipment can be improved in a complex and changeable industrial environment.

The invention relates to the technical field of storm surge disaster monitoring and intelligent forecasting, and provides a regional collaborative storm surge water increase forecasting method of a coupling graph attention and gated circulation network, which comprises the following steps of: obtaining storm surge water increase residual field data based on a mixed wind field driving two-dimensional hydrodynamic model formed by typhoon parameterization and reanalysis data fusion, registering the water-increasing residual field data and the observation time sequence in time and space, and taking the registered data as physical prior input; constructing a dynamic edge weight graph structure based on a space-time causal correlation analysis method, and adaptively updating the dependency relationship among multiple observation stations; physical priori and key features are used as node input, a graph attention mechanism of GATv2 and a time sequence feature extraction capability of GRU are fused, and time-space coupling modeling is carried out on storm surge water increase; and carrying out multi-station joint prediction through regional cooperation, carrying out model evaluation and result optimization, and outputting a storm surge water increase prediction sequence with 15-minute time steps and 15-4-hour advance.

The invention discloses a complex product process route flexible planning method based on a deep cycle Q network. The method is suitable for efficient process route optimization in a dynamic manufacturing environment. The method comprises the following steps: firstly, modeling process characteristics, processing operation, resources and constraint relationships of a part, and constructing a state space and reward mechanism based on a partial observable Markov decision process; and then constructing a long short-term memory network to extract time sequence characteristics in a processing state, and constructing a DRQN on the basis to realize an optimal decision of processing operation. In order to adapt to variable process targets and environments, a self-adaptive reward function adjustment strategy is introduced, and dynamic optimization is realized by adjusting weights such as processing quality, time and energy consumption; and meanwhile, a selective forgetting mechanism is constructed, and model parameters are suppressed and updated by utilizing a Fisher information matrix, so that efficient forgetting of outdated process knowledge is realized, and the migration ability and generalization performance of the model are improved.

The invention provides a multi-step time sequence prediction method of a quantum bidirectional circulation network based on causal convolution. According to the quantum causal convolutional network based on time sequence perception, the quantum causal convolutional network comprises a quantum causal convolutional layer with multiple layers of time sequence dimensions and a residual block formed by a quantum extrusion excitation mechanism layer, longer time dependence can be captured in a shallow network, and by means of the advantages of quantum calculation, the time sequence perception can be improved. The long-time dependency relationship, the causal relationship and the time sequence relevance among the data can be quickly captured and processed; through a quantum bidirectional circulation network guided by an attention mechanism, time sequence state information and time sequence feature information are fused to enhance the expression of time sequence information, the capability of processing a complex time-varying relationship is effectively improved, and a multi-head attention mechanism is utilized to guide updating of hidden layer important information of a bidirectional circulation unit, so that the time-varying relationship processing efficiency is improved. And long-time-span important information is captured and integrated. Compared with the prior art, the performance of multi-step time sequence prediction is effectively improved.

The invention provides a potential predictability embedded deep learning industrial process soft measurement method and device, and relates to the technical field of deep learning industrial process soft measurement. The method comprises the following steps: designing a predictability regularization item according to a probability dynamic model of potential features constructed by a variational cycle network, a time graph constructed by a training data set and a time lag constant, constructing a target-related auto-encoder embedded by a probability potential predictability model, constructing a supervision depth network embedded by the probability potential predictability model, and constructing a target-related auto-encoder embedded by the probability potential predictability model. And obtaining a quality variable prediction value of the industrial production process. According to the method, firstly, a potential feature prediction design regularization item of probability distribution is designed, and network parameters are optimized through a joint training model to ensure overall convergence of the network. Secondly, an improved variational graph recurrent neural network is provided, randomness is brought into an advanced potential space, modeling is carried out on potential predictability of a supervised deep network, and additional graph structure information is added to help to analyze time dependence;

The invention provides a modeling prediction method and device based on a dynamic knowledge graph, equipment and a medium, and the method comprises the steps: S1, taking the data of a fishing boat and a sea area as nodes, representing the migration and interaction between the fishing boat and the sea area through edges, and introducing a time attribute to represent a dynamic characteristic, thereby constructing the dynamic knowledge graph of the fishing boat and the sea area; and S2, constructing a prediction model in combination with the gated loop network GRU and the relational graph attention network rGAT, and processing updating of node representation and graph-level representation in the dynamic knowledge graph by using the constructed prediction model. The method has the advantages that the dynamic change of the fishing boat and the sea area can be accurately modeled in the time dimension, so that the migration mode and the group activity behavior of the fishing boat in the sea area can be accurately captured; meanwhile, on the basis of updated node representation and graph level representation, better fishing vessel migration can be analyzed and predicted, early warning can be performed on abnormal fishing behaviors, and the like, so that intelligent decision support can be well provided.

The invention discloses a supply chain order business automatic processing system and a method thereof. The system comprises a dynamic intelligent center, an elastic process engine, a green loop network and a trusted security layer, the dynamic intelligent center dynamically optimizes decisions through federated learning and generative AI, and the elastic process engine realizes flexible recombination of processes based on micro-service and a low-code platform. The green circulation network utilizes a block chain to track carbon footprints and manage reverse resources, and the credible security layer adopts quantum encryption and dynamic authority to guarantee data credibility. The corresponding method comprises the steps of combining a multi-party data training model and generating a multi-scheme, performing a micro-service splitting and low-code configuration process, driving intelligent contract transaction by a carbon label, and performing quantum encryption evidence storage and anomaly detection. Through the edge-cloud collaborative architecture, the scheme synchronously improves the efficiency, flexibility and sustainability, solves the problems of insufficient intelligence, framework stiffness and security defects in the prior art, and is suitable for complex scenes such as e-commerce and manufacturing.

The invention relates to a video content description method. The method comprises the following steps: S1, acquiring a video stream; s2, based on a video stream adjacent frame motion feature extraction algorithm and a clustering algorithm, extracting a preset number of key frames from the video stream; s3, extracting regional features from the preset number of key frames by adopting a target detection algorithm of regional feature extraction; s4, according to the regional features, calculating corresponding regional attention features based on a self-attention mechanism algorithm; s5, extracting video frame features of each frame by adopting a classification algorithm and a video stream adjacent frame motion feature extraction algorithm according to the video stream; and S6, according to the regional attention features and the video frame features, decoding by using a cyclic network algorithm to obtain a description text of the video content.

The application relates to a traffic flow prediction method based on a multi-mode dynamic memory graph convolution network, and belongs to the technical field of intelligent transportation. At present, the time-space correlation characteristics of the graph convolution neural network method are insufficient, the periodic characteristics are not fully considered, and the dynamic evolution correlation is not accurately captured. In order to solve the above problems, firstly, a time sequence feature extraction module and a bidirectional memory cycle network module are established, and the captured time sequence features are fused to form a comprehensive feature vector. Then, in the dynamic graph convolution module, the spatial correlation between nodes is captured through the fusion mode of the diffusion graph convolution neural network, the attention mechanism and the typical traffic mode, and a new dynamic adjacency matrix is generated. The dynamic adjacency matrix can reflect the node connection relationship changing with time. The application dynamically updates the adjacency matrix through the multi-mode dynamic memory graph convolution network, better responds to emergencies and abnormal situations, and effectively improves the accuracy and robustness of traffic flow prediction.

The invention provides a traditional Chinese medicine data classification processing method and system based on a graph loop network, and a medium. The method comprises the following steps: collecting traditional Chinese medicine data; preprocessing the collected traditional Chinese medicine data; constructing a graph loop network; and inputting the preprocessed traditional Chinese medicine data into the graph loop network to obtain a traditional Chinese medicine data classification result. In the application, the first-level graph convolution layer focuses on local feature extraction and enhancement of the perception ability of the model, and the high-level graph convolution layer integrates global information, captures detail information around nodes and enhances the understanding ability of the model. And the feedback mechanism layer adaptively adjusts the mechanism, monitors data in real time, adjusts network parameters, and reduces the over-fitting risk. The traditional Chinese medicine knowledge is fused into the graph structure construction process, multi-dimensional features can be extracted from the traditional Chinese medicine data, local and global information can be comprehensively captured, the classification and screening performance of the unstructured traditional Chinese medicine data is improved, and rapid data selection is provided for new medicine research and development or medicine compatibility and the like aiming at diseases.

The invention provides a service logic vulnerability detection method and device, equipment, a medium and a program product, and relates to the technical field of network security. The method comprises the following steps: constructing a business process state machine through a key entity obtained by performing natural language analysis on a business document; the business process state machine is used for describing state changes of the business process in different stages; obtaining an operation sequence generated based on the interaction operation through a flow agent deployed in a network environment; carrying out modeling and feature extraction on the operation sequence based on a hierarchical attention circulation network to obtain operation behavior features; comparing the operation behavior characteristics with a business process state machine to obtain a vulnerability detection result; the vulnerability detection result comprises permission violation and / or flow abnormality.

The invention discloses a water quality index BOD5 monitoring method for efficiently extracting data feature information based on width learning. According to the method, under the inspiration of a loop network structure, recursive cascading is carried out on feature windows of a width learning system to extract time correlation features of process data. On this basis, a non-Gaussian feature extraction window is added, non-Gaussian features of sewage data are extracted by using an over-complete independent component analysis method, the ability of the model to extract different features of the data is enriched, and the accuracy of the model is improved. The method comprises an off-line training stage and an on-line monitoring stage. Besides, aiming at the condition that the model precision is insufficient in a severe environment and an operation condition in the sewage treatment process, the model provided by the invention can be reconstructed in an increment mode, the whole network does not need to be retrained from the beginning, and the network training can be completed in a very short time; therefore, the water quality index of the sewage can be rapidly monitored in time.

The invention relates to the technical field of intelligent peripherals, in particular to a user-oriented interaction information expansion method, which comprises the following steps that: each keycap acquires displacement and pressure, generates an event packet, uploads the event packet through an optical wireless link, and obtains a clock and supplies power; constructing a keycap graph by a master controller, obtaining an intention vector through graph convolution and pooling, inputting the intention vector into a cyclic network to generate a pixel matrix and a phase matrix, and downlink after Bernoulli sparse coding; performing key cap side orthogonal matching pursuit decoding and four-round gradient optimization, and outputting a differential pixel set, a differential phase set and a synchronization mark; the master control refreshes the digital micromirror array and the ultrasonic phased array in a differential manner, locks a frame-level time sequence through a synchronization pulse, compensates link delay and hardware drift in real time based on Kalman filtering, and keeps an optical-acoustic alignment error to be less than 100 microseconds; degradation monitoring enables a link to be automatically switched to a pre-carved character layer and to be quickly recovered when the link is abnormal, and a keyboard interaction information expansion scheme with low delay and high consistency is provided.

According to an update procedure for a network of interconnected nodes, a node keeps track of markers received on the node's incoming channels, transmits a marker on each outgoing channel (if any exist), and updates its node configuration from an old configuration to a new configuration. In one version of the update procedure, the node updates its configuration and transmits markers after receiving its initial incoming marker. In another version for acyclic networks, the node updates its configuration and transmits markers after receiving its final incoming marker. The node determines how to handle (i.e., process, queue, or drop) each incoming data packet based on (i) whether or not it has updated its node configuration yet and (ii) whether or not it has received a marker on the corresponding incoming channel yet. A controller initiates the update process at a subset of nodes that ensures eventual completion of the update process.

The invention discloses a short video recommendation method and system based on a hypercycle network and multi-task training, and the method comprises the steps: firstly constructing a super node in a hypercycle neural network for integrating a plurality of user interaction behaviors, thereby carrying out the modeling of the semantic association of different user interaction behaviors; inputting the obtained historical behavior sequences of the user and a target short video into a trained short video recommendation model, and capturing a sequence dependency relationship among the historical behavior sequences by a hyper-recurrent neural network based on super-node design, modeling directed connection between the click node and the favorite node through a high-speed neural network; then calculating user preferences in different interactive behavior spaces based on an addition attention mechanism of the target short video; and finally, fusing the user preferences in different interactive behavior spaces, and outputting the probability of clicking the target short video by the user to finish the recommendation of the short video.

The application discloses a deep learning-based time-space fusion early warning method for deep rock burst, and constructs a dual-branch Transformer-CNN time-space feature fusion early warning model. The model captures local mutation features in a microseismic sequence through a parallel multi-scale CNN network branch, combines a Transformer network branch to extract global evolution features, realizes comprehensive mining of rock burst precursor information and identification of key precursor information, and further enhances the dynamic adaptability of the early warning model to complex microseismic sequences by using a self-adaptive gating fusion mechanism, thereby enhancing the adaptability of the early warning model in a complex monitoring environment. In addition, a learnable position coding is introduced to replace a traditional fixed coding mode, so that the early warning model can more flexibly capture time sequence dependence in mining stress evolution, and the problems of gradient vanishing and low calculation efficiency of a traditional cyclic network in long sequence modeling are overcome. Finally, the classification accuracy of the rock burst danger level and the early warning reliability are effectively improved.

An intelligent relay protection hidden danger detection method and system, the method includes data collection, data preprocessing, fault hidden danger prediction, fault classification and positioning, and relay protection hidden danger detection. The present invention relates to the field of relay protection hidden danger detection technology, and specifically refers to an intelligent relay protection hidden danger detection method and system. This solution adopts a fine-grained relay protection hidden danger detection method from hidden danger detection to classification and positioning, and provides intelligent detection demand support by first detecting and predicting the probability of fault occurrence and the trend change of the probability of occurrence, and confirms the type and spatiotemporal location of the fault by further performing classification and positioning, thereby improving the overall efficiency, intelligence, and detection dimension of relay protection hidden danger detection; adopts a time-series convolutional bidirectional gated recurrent network combined with feature reconstruction optimization to predict fault hidden dangers; adopts a convolutional bidirectional gated recurrent fusion network combined with a position perception graph to classify and locate faults.

The invention discloses a reservoir porosity prediction method based on a one-dimensional convolutional gated recurrent network, and relates to the technical field of porosity prediction. The method comprises an offline part and an online part; wherein the offline part comprises: obtaining a first seismic attribute set related to reservoir parameters at a well point in a target work area; obtaining an optimal attribute combination from the first seismic attribute set; taking the optimal attribute combination as input, taking the porosity corresponding to the optimal attribute combination as output, and training a one-dimensional convolutional gated recurrent network model; the online part comprises: obtaining a second seismic attribute set related to target reservoir parameters in a well-free area in the target work area; screening out a final attribute combination from the second seismic attribute set according to the optimal attribute combination; inputting the final attribute combination into a trained one-dimensional convolutional gated recurrent network model, and outputting the target reservoir porosity; and the method improves the prediction accuracy of reservoir porosity and expands the prediction range.

The embodiment of the invention provides a logistics congestion early warning method and system based on space-time coupling and track matching, and belongs to the technical field of logistics management. The early warning method comprises the following steps: acquiring real-time distribution data of smoke box logistics trolleys; constructing and training a distribution prediction network model; inputting the real-time distribution data into the prediction network model to obtain real-time prediction distribution data; acquiring real-time track data of the trolley according to the real-time prediction distribution data; and calculating a track congestion index according to the real-time track data and carrying out congestion early warning. According to the method, the distribution prediction model is constructed through the graph convolution network and the circulation network, accurate matching of the trolley track is realized through combination of depth-first search and the Hungary algorithm, the congestion index is calculated through historical track data so as to realize congestion early warning without scheduling dependence, and dependence on manual judgment is remarkably reduced.

The application provides a power distribution network carrying capacity evaluation method and system fusing GNN and deep learning, in the method, a graph neural network model of the power distribution network is established based on power distribution network topological structure data and historical parameter data; global topological electrical coupling features in the graph neural network model are extracted based on a graph attention network and a gated recurrent network; future power data predicted by an LSTM is taken as input of power flow calculation in the graph neural network model, so that node state parameter features are obtained; the node state parameter features and the global topological electrical coupling features are taken as input, a PPO algorithm is used to adaptively adjust an operation strategy, and control actions of power distribution network equipment are output; the equipment actions output by the PPO algorithm are fed back to the graph neural network, power flow is recalculated and node state parameters are updated; and a power distribution network carrying capacity evaluation result is obtained based on the updated node parameter state, so that the efficiency and reliability of power distribution network carrying capacity evaluation are improved.

The invention belongs to the technical field of aircraft control, and particularly relates to a path planning method combining a double-confrontation technology and a recursion mechanism. According to the method, an optimal Q value function is calculated through a Bellman equation, and a deep learning Q network model DQN which continuously trains and optimizes parameters and replaces a traditional Q value table with a network output value is established; constructing a deep recursion Q network model DRQN containing a cyclic network; in combination with a double-adversarial technology, a double-adversarial deep recursion Q network model D3RQN which combines a value function and a reward value function and introduces a target network parameter learning rate is established, and path planning of the four-rotor aircraft in a semi-known environment is realized. According to the method, on the premise of a semi-known environment, the burden of the calculation amount is relieved, the performance of the POMDP of the incomplete observable Markov decision process is improved, the training speed is increased, meanwhile, the problem that value estimation is too optimistic is solved, the system stability and accuracy are improved, and the path planning performance of the quadrotor unmanned aerial vehicle for reconnaissance and transportation is improved.

The application discloses a robot intelligent control method and system based on a causal mechanism, and relates to the technical field of intelligent robots, which comprises the following steps: inputting an operation task and an environment state of a target robot at a current time into a causal structure model to determine each candidate action for pushing a target object from a current time position to a next time position; inputting the candidate action into a graph recurrent network to obtain a predicted position of the target object under the candidate action at the next time; screening an execution action of the target robot at the next time according to the predicted position of the target object under each candidate action and a target position of the target object; and updating the predicted position corresponding to the execution action of the target robot at the next time to the current time position until the target object reaches the target position. The application can effectively improve the intelligent control capability of the robot by determining the action to be executed according to the predicted position of the target object under the execution action and the target position based on the causal structure model and the graph recurrent network.

This invention relates to the field of storm surge disaster monitoring and intelligent forecasting technology, and provides a regional collaborative storm surge rise forecasting method using coupled graph attention and gated cyclic networks. The method includes: using a hybrid wind field-driven two-dimensional hydrodynamic model formed by fusing typhoon parameterization and reanalysis data to obtain storm surge rise residual field data; registering the rise residual field data with the observation time series in time and space, and using it as a physical prior input; constructing a dynamic edge-weighted graph structure based on a spatiotemporal causal correlation analysis method to adaptively update the dependencies between multiple observation stations; using the physical prior and key features as node inputs, fusing the graph attention mechanism of GATv2 and the temporal feature extraction capability of GRU to perform spatiotemporal coupled modeling of storm surge rise; conducting multi-site joint prediction through regional collaboration, performing model evaluation and result optimization, and outputting storm surge rise forecast sequences with a 15-minute time step and a lead time of 15 minutes to 4 hours.

The invention provides a short-term power load prediction method and system based on a KAN cycle network, and the method comprises the steps: carrying out the feature enhancement operation of historical multi-time scale variables of a to-be-predicted region based on a Pearson's correlation coefficient analysis method, so as to determine the correlation between each variable in the historical multi-time scale variables and a short-term power load, obtaining a historical vector sequence; training a preset KAN loop initial network based on the historical vector sequence to obtain a target KAN loop network; obtaining a current multi-time scale variable, and constructing a current vector sequence; and taking the current vector sequence as the input of the target KAN cycle network, so that the target KAN cycle network outputs a short-term power load prediction result. According to the short-term power load prediction method provided by the invention, the short-term power load prediction in different scenes can be realized by utilizing the feature enhancement and the KAN circulation network, the prediction precision is improved, and the power supply and demand balance and the operation stability of a regional power grid are effectively guaranteed.

The invention discloses an equipment state prediction method based on a parallelization loop network. The method comprises the following steps: S1, improving a GRU network architecture; s2, extracting an input time sequence deep feature mode by using a GELU feedforward network; s3, adding a noise module to realize consistency of parallel training and sequence prediction; s4, improving the efficiency and the stability of the model by using RMSNorm; s5, capturing long-period causal time sequence information by adopting causal convolution; s6, the modules are deepened, stacked and combined through residual connection; s7, performing optimization by using an ADAMW optimizer, superposition dynamics and prediction loss; s8, inputting historical and future control quantity information, outputting differential historical and future prediction, and calculating mean square error improvement performance; and S9, carrying out equipment state early warning according to a prediction result. Compared with a traditional LSTM model, the method has the advantages that the prediction error is reduced, the single-sequence reasoning time is greatly shortened, and a powerful solution is provided for accurate and efficient prediction of the state of the industrial equipment.

The invention discloses an industrial data prediction method and device based on a robust cyclic random configuration network, and aims to solve the problem that the modeling precision and robustness are insufficient when an existing model processes industrial data which is uncertain in input order, strong in time sequence and contains noise and abnormal values. Through combination of a robust learning mechanism and a recursive structure, order identification is not needed, reserve pool nodes are randomly generated based on a supervision mechanism, a network structure is constructed, and the global approximation capability of any nonlinear mapping function is ensured. The method not only inherits the efficient modeling capability of a cyclic network for time sequence characteristics, but also significantly enhances the stability and generalization capability of the model in a complex interference environment, and is suitable for accurate prediction and intelligent decision support scenes in a complex dynamic industrial process.

The invention relates to the technical field of boiler fault diagnosis, in particular to a digital fault early warning method for a through-flow boiler and a remote operation and maintenance platform. Comprising the following steps that boiler operation parameters are collected through a distributed sensor network, and a full-dimension monitoring system is constructed; performing noise reduction, normalization and missing value interpolation on the original data at an edge computing node, constructing a working condition feature library in combination with historical operation data, and matching a current working condition type through a dynamic time warping algorithm; a space-time attention mechanism is adopted to carry out feature fusion on multi-source data, a sensor space association graph is constructed through a graph convolutional network, time sequence features are captured in combination with a gated loop network structure, and the feature weight of each sensor is adaptively adjusted based on data dynamic features. According to the method and the device, noise reduction, normalization and missing value processing are performed on the acquired data on the edge side in real time, so that transmission delay caused by centralized cloud processing of the data is avoided, and the timeliness and the efficiency of data processing are improved.

The present invention discloses a method for sentiment analysis of multiple interlocutors in heterogeneous graph structure based on topic semantic enhancement, including: performing sentiment word embedding operation on input dialogue to convert it from human language into vector representation with sentiment; constructing a syntactic dependency graph according to dependency syntactic relationship, wherein nodes are words in the dialogue, inputting the syntactic dependency graph into a graph convolutional neural network to update node information, obtaining semantically enhanced word vectors and corresponding sentence representation vectors; constructing a topic extraction model to extract the topic of each dialogue sentence to obtain topic-enhanced sentence representations; clustering according to topic similarity, constructing a dialogue subgraph according to sentence topic information and temporal relationship, constructing a heterogeneous dialogue graph, wherein nodes are sentence representations of each sentence, and updating the graph nodes using a graph recurrent network; and obtaining classification results. The present invention fully considers the interactive information between interlocutors and improves the prediction accuracy of sentiment analysis of multi-interlocutors dialogue.

The invention relates to the technical field of transformer gas detection, and discloses an intelligent gas leakage detection system and method for a transformer, and the method comprises the steps: collecting multi-source operation data, such as gas pressure, oil level and temperature, carrying out the standardization processing, and building a unified feature matrix to eliminate the dimensional difference. Then extracting space-time correlation characteristics of gas leakage, and constructing a supervisible training sample in combination with historical records; a deep learning model is utilized to fuse a convolutional network and a cyclic network structure, local change features and global dynamic features are captured respectively, and common working conditions and extreme working conditions are distinguished through a multi-branch classifier. And angle subareas are divided according to the wind direction of the whole field, a correction curve between historical gas concentration and actual gas concentration is established, and space subarea correction is achieved. And finally, obtaining optimal gas leakage state evaluation through weighted fusion of the branch prediction value and the comprehensive correction value. According to the invention, an intelligent detection principle of data driving, space-time coupling and working condition self-adaption is effectively realized.