392 results about "Hybrid neural network" patented technology

The invention provides a cellular network flow space-time prediction method and system based on a hybrid neural network, and belongs to the technical field of intelligent communication. The method adopts a layered deep neural network architecture, and comprises a data embedding layer, a space-time coding layer, a feature fusion layer and an output layer. The data embedding layer maps a historical traffic sequence, cross-domain external data and metadata into high-dimensional features; the space-time coding layer is used for respectively fusing one-dimensional causal convolution and a Mama neural network to extract multi-scale time features and densely connecting convolution and a multi-head attention mechanism to capture multi-scale space features through time and space modeling branches; the feature fusion layer realizes adaptive weighted fusion of spatial-temporal features, cross-domain features and metadata features by using a gating fusion mechanism; and the output layer performs linear transformation on the fusion features to generate a final prediction result. According to the method, the spatial-temporal dynamic capture of the service traffic is accurate, the prediction curve is highly fit with the true value, and the accurate prediction of the multi-service traffic of the cellular network is realized.

A hybrid neural network architecture is disclosed that integrates matrix multiplication-free (MatMul-free) transformation layers with spiking neural network (SNN) layers for efficient, low-power computation. The system includes an interface module configured to convert intermediate continuous-valued data from MatMul-free layers into a spike-compatible format using encoding techniques such as rate coding, phase coding, or threshold-based conversion. The SNN layers process the spike-encoded data in an event-driven manner, enabling sparse, temporal inference. Training is supported by a hybrid optimization strategy combining backpropagation in MatMul-free components with surrogate gradient descent or spike-timing-dependent plasticity (STDP) in SNN layers. The architecture reduces computational complexity, supports real-time adaptability, and enables deployment in energy-constrained environments such as edge devices and neuromorphic platforms. The system may be implemented in hardware, software, or a co-designed pipeline optimized for dynamic sensor data, control signals, or continuous inference tasks.

The invention belongs to the field of aero-engines and turbines, and provides a quick prediction method, system and device for a two-dimensional viscous compressible flow field of a gas compressor and a medium, and the method comprises the steps: obtaining CFD simulation data of a two-dimensional blade profile of the gas compressor under multiple working conditions, and carrying out the preprocessing to obtain a standardized training data set; on the basis of the data set, constructing a hybrid neural network model which takes a blade profile geometric parameter and an operation condition parameter as input and takes a two-dimensional viscous compressible flow field as output; combining mean square error, gradient loss and physical consistency loss based on RANS equation residual error to construct a total loss function, and performing end-to-end training on the model; during reasoning, to-be-predicted parameters are input into the trained model according to the same preprocessing mode, and then a flow field prediction result can be rapidly output. According to the method, the problems of long consumed time, low precision and poor consistency of traditional CFD simulation can be solved, millisecond-level flow field prediction is achieved, high precision and high physical consistency are achieved, and the method is suitable for efficient optimization design of the blade profile of the gas compressor.

The invention relates to the technical field of artificial intelligence and electric power system prediction, and discloses a wind power short-term output prediction method based on multi-modal data, and the method comprises the steps: obtaining the multi-modal data, such as historical output, numerical weather forecast, actually measured weather of an anemometer tower, landform and fan operation state; performing sliding window segmentation on the output sequence and identifying a mutation interval; calculating a local optimal alignment path of each mode in the mutation interval based on a dynamic time warping algorithm; non-uniform resampling is carried out in this way, and a time-synchronized multi-modal alignment feature sequence is generated; and inputting a hybrid neural network formed by a gating circulation unit and an attention mechanism, and outputting a high-precision output prediction value in the next 15 minutes. The system comprises corresponding function modules. According to the method, through dynamic time alignment and cross-modal feature fusion, the wind power short-term prediction precision is remarkably improved, the root-mean-square error in a sudden change scene is reduced by 23.7%, and reliable support is provided for power grid dispatching.

The invention discloses a rapid heat transfer simulation method and device based on a neural network, and relates to the technical field of physical simulation. The method comprises the steps that a hybrid neural network model is trained, the model learns operator mapping from an input function to a temperature or heat flow field, and meanwhile physical constraints such as a heat conduction partial differential equation are coded into a loss function; for a new simulation task, single forward inference is carried out by using the operator mapping, and an initial prediction result is rapidly generated; then, according to physical constraints of coding, calculating a physical residual error of initial prediction, and when the residual error exceeds a preset threshold value, executing a small amount of optimization iteration by taking the prediction as an initial value to carry out rapid local correction; the problems that a traditional numerical method is long in calculation time and an existing neural network method is insufficient in physical fidelity are solved, and high efficiency and high precision of heat transfer simulation are achieved.

The invention relates to an X-ray weld defect intelligent detection method based on weld line perception and reversible domain mapping, which comprises the following steps: collecting a gray X-ray image of a weld region, and carrying out joint labeling on a weld center line track and a defect mask to construct a standardized training data set; designing a weld line sensing hybrid neural network architecture; performing end-to-end training on the network by using a joint loss function formed by detection task loss and center line consistency constraint to realize collaborative optimization of defect category discrimination and spatial positioning; and deploying the trained model in an automatic welding quality detection system, performing forward reasoning on X-ray images acquired in real time, and continuously outputting defect detection results with accurate category, position and form information in combination with a post-processing strategy of weld direction clustering. According to the invention, stable and accurate identification of slender and weak-texture weld defects under the condition of bent welds is realized, and automation and reliability of nondestructive detection of welds are remarkably improved.

The invention discloses an IMU online adaptive calibration method and calibration device and an automatic driving system, and relates to the technical field of deep learning, automatic driving and navigation. Comprising the following steps: acquiring original measurement data from an IMU in real time; inputting the original measurement data into an error prediction model to output predicted IMU error parameters, wherein the error prediction model adopts a hybrid neural network architecture, is trained and updated based on an end-to-end framework containing a differentiable physical model, and is configured to execute offline training in an offline stage and execute adaptive updating in an online application stage; and correcting the original measurement data by using the IMU error parameter and executing dead reckoning to generate a calibrated motion state track. According to the invention, a static off-line calibration system is improved into a dynamic on-line adaptive system, so that the long-term robustness and reliability are remarkably improved. A differentiable physical model is introduced as a training bridge to realize an end-to-end physical model training normal form, so that the training precision can be improved and the training convergence can be accelerated.

The invention belongs to the technical field of deep neural networks, and discloses a short-term power load prediction method based on a hybrid neural network model, and the method comprises the steps: 1, obtaining an original power load time sequence; 2, performing adaptive decomposition on the original power load time sequence to obtain an intrinsic mode function (IMF) component; step 3, for the plurality of IMF components obtained after decomposition, fusing meteorological features and peak period features to obtain a fusion time sequence, then performing feature extraction and joint prediction, and constructing a multi-modal prediction model; and step 4, intelligently optimizing the key parameters of the multi-modal prediction model, outputting optimal hyper-parameters, and performing optimal parameter setting on the multi-modal prediction model to obtain a final prediction result. According to the method, the accuracy and robustness of prediction are effectively improved, and the adaptability of the model to non-stationary and multi-scale energy time sequences is effectively improved.

The invention relates to an intelligent safety equipment state monitoring method, system and equipment and a medium. According to the method, attitude data of a self-locking device toggle clip are collected in real time, the wearing compliance is judged in advance, and an alarm is triggered immediately for a violation state to realize risk interception; for a compliance state, based on the preprocessed time sequence data, extracting a multi-scale feature vector representing a short-term dynamic state and a long-term trend by using a multi-stage time convolutional network, and dynamically fusing a multi-expert prediction result through an expert hybrid neural network to output a future risk probability value; and finally, a graded early warning signal is generated according to a probability value matching risk grade, so that while the real-time performance of basic state monitoring is guaranteed, accurate prediction and active protection of potential risks in a compliance operation scene are realized, and a dual safety guarantee mechanism of'real-time interception + beforehand prevention 'is formed.

The invention relates to an atmospheric concentration field inversion method and system based on a space-time diagram network. The method comprises the following steps: acquiring atmospheric observation data of different platforms; the method comprises the following steps: standardizing and preprocessing atmospheric observation data of different platforms to generate space-time point cloud data in a unified format; mapping the time-space point cloud data into a graph network structure for describing an atmospheric physical transmission process; constructing a hybrid neural network architecture, taking the graph network structure as the input of the hybrid neural network architecture, and training the hybrid neural network architecture to obtain a prediction model; inputting real-time radar observation data into the prediction model to obtain a three-dimensional particulate matter concentration field prediction result; and analyzing the prediction result of the three-dimensional particulate matter concentration field, and evaluating the pollution cause of the current pollution event. According to the technical scheme, a high-fidelity and time-space continuous concentration field can be generated under the condition that observation data is limited, and the accuracy and reliability of an inversion result are greatly improved.

The invention discloses an unmanned aerial vehicle real-time task scheduling and resource management system, which belongs to the technical field of unmanned aerial vehicle computing resource management, and comprises a task load prediction module, a priority dynamic evaluation module, a resource elastic distribution module and a closed-loop feedback optimization module, the task load prediction module adopts an LSTM-GRU hybrid neural network to predict a resource demand peak value in advance and triggers a pre-scheduling signal; the priority dynamic evaluation module dynamically adjusts the task priority based on the multi-dimensional evaluation matrix; the resource elastic allocation module realizes elastic allocation through a reserved resource pool and a multi-level preemption mechanism; the closed-loop feedback optimization module monitors performance deviation and reversely adjusts parameters of each module, the four modules are deeply coupled to form a closed-loop cooperation mechanism, and intelligent task scheduling and dynamic resource optimization are achieved.

A lithium ion battery thermal management method based on physical perception and entropy cooperation multiple agents comprises the steps that a lithium ion battery electric-thermal coupling model containing data driving compensation is established, and a full-state space system equation describing the dynamic characteristics of a battery is established; constructing a TCN-Transformer hybrid neural network fused with a physical constraint mechanism, and carrying out online identification on parameters in the electric-thermal coupling model by adopting the network to obtain real-time parameters; based on the full-state space system equation and the real-time parameters, constructing a minimum error entropy adaptive extended Kalman filter optimized by an entropy cooperative multi-agent flexible Actor-Critic algorithm, and performing joint estimation on the state of charge of the battery and the temperature of the battery; and based on a joint estimation result, constructing a TD3 deep reinforcement learning control algorithm embedded with a microsecurity layer, and realizing the self-adaptive thermal management direct control of the lithium ion battery under rule guidance through the algorithm. According to the invention, accurate, effective and safe thermal management control of the lithium ion battery is realized.

The invention discloses an ultra-high voltage direct current transmission system fault intelligent analysis method and system based on a domestic artificial intelligence large model framework, and a storage medium, and belongs to the field of electric power system artificial intelligence fault diagnosis. According to the method, multi-source heterogeneous fault data are accessed and standardized and aligned, the processed data are input into a domestic artificial intelligence large model adapted by knowledge in the electric power field, and the model is utilized to perform multi-modal feature fusion and intelligent analysis, so that automatic identification and accurate positioning of fault types are realized; and an analysis report and a disposal strategy are automatically generated. The system comprises a data input module, an intelligent analysis module and a decision output module, and the intelligent analysis module performs feature extraction and recognition by adopting a CNN-Transform hybrid neural network and realizes fault positioning based on a knowledge graph and a graph neural network. According to the method, full-process automatic intelligent analysis from data access to decision output is realized, and the efficiency, the accuracy and the engineering landing capability of fault diagnosis are remarkably improved.

The invention relates to the field of data processing, and discloses a logistics report multi-source heterogeneous data cleaning method, device and equipment and a storage medium, and the method is used for cleaning logistics multi-source heterogeneous report data. The method comprises the steps of obtaining multi-source heterogeneous report data, and matching the multi-source heterogeneous report data through a preset feature matching rule to obtain target data; performing exception preliminary screening and exception processing on the target data to obtain processed data; building a conflict detection model based on a CNN-LSTM hybrid neural network, and carrying out the recognition and classification of the processed data through the conflict detection model, and obtaining different types of conflict data; calling a repair algorithm corresponding to the conflict type to repair different types of conflict data to obtain repaired data; and performing full-dimensional quality verification on the repaired data based on a pre-trained data quality evaluation model, and if the verification is passed, completing data cleaning.

The invention discloses a natural gas production yield prediction method and system based on deep learning, and the method comprises the steps: obtaining dynamic production data, geological parameters and environmental factor data, carrying out the preprocessing of the collected data, and carrying out the multi-scale decomposition through EMD (Empirical Mode Decomposition), and obtaining the multi-scale feature data; capturing instantaneous fluctuation features based on an LSTM (Long Short-Term Memory) network, extracting a periodic trend through a CNN (Convolutional Neural Network), modeling spatial relevance between gas wells by utilizing a GNN graph neural network, and introducing an attention mechanism to dynamically weight and fuse each branch feature to obtain a mixed neural network model; and inputting the multi-scale feature data into the target hybrid neural network model for prediction, and outputting a yield prediction result. And data can be comprehensively understood, so that the yield prediction precision is improved.

The invention relates to the technical field of underground engineering construction safety monitoring, and discloses an underground pipe-jacking tunnel excavation pavement settlement prediction system and a prediction method thereof.The prediction system comprises an acquisition terminal and a computing server and is provided with a data preprocessing module, a model building module and other modules. The core lies in that a CA-CNN-LSTM hybrid neural network is constructed, geological spatial features are extracted by using parallel convolutional neural network branches, construction time sequence features are extracted by using bidirectional long short-term memory network branches, and deep fusion of spatial and temporal features is realized through a Cross-Attention module and a dynamic gating unit. The prediction method relates to multi-source data cleaning enhancement, model training and double-index grading early warning. According to the method, the problem of heterogeneous feature coupling is effectively solved, the settlement prediction precision is effectively improved, hierarchical control signals can be generated according to the prediction result, and the construction safety is effectively guaranteed.

The invention discloses a recommendation model training and application method and device based on multi-platform behaviors and a medium, and relates to the technical field of user recommendations, the method comprises the following steps: constructing a hybrid neural network model, the hybrid neural network model comprises a feature embedding module, a multi-cavity convolution module, a multi-level attention mechanism module and a prediction module, the feature embedding module processes the multi-platform behavior sequence to obtain a feature embedding vector, the multi-cavity convolution module processes the feature embedding vector to obtain a multi-scale feature vector, and the multi-level attention mechanism module processes the multi-scale feature vector and a commodity feature vector to obtain a user interest vector; and the prediction module predicts a predicted click rate of the user to the candidate commodities based on the user interest vector and the commodity feature vector, trains the hybrid neural network model to obtain a recommendation model, and performs recommendation by using the recommendation model. The recommendation model with better performance can be obtained through training, and the recommendation performance is improved.

The invention provides a lithium battery residual life prediction method based on multi-frequency decomposition and a hybrid neural network, and relates to the technical field of lithium battery residual life prediction. According to the lithium battery residual life prediction method, a component decomposition mechanism is introduced, and proper prediction models are designed for different frequency components, so that local fluctuation and long-term trend in a battery degradation process are effectively captured, and the accuracy and robustness of residual life prediction are remarkably improved; proper prediction models are respectively designed according to complexity differences of different components, so that redundant calculation caused by uniformly using complex models on all components is avoided, and the balance between prediction precision and calculation efficiency is realized; based on a hyper-parameter optimization strategy of hierarchical reinforcement learning, an optimal parameter combination can be automatically and efficiently searched according to different battery data characteristics, and the defects that a traditional optimization method is high in calculation cost and poor in adaptability are overcome.

The invention relates to the technical field of intelligent operation and maintenance of industrial equipment, and discloses an industrial equipment predictive maintenance method based on an adaptive neural network. According to the method, equipment operation data and sensor ontology data are acquired through a multi-source sensor network, an equipment health index and sensor data confidence coefficient are calculated based on principal component analysis, and a comprehensive health feature vector is constructed in combination with change rate and correlation analysis; inputting the feature vector and the state identifier into a hybrid neural network model, extracting a time sequence feature through an attention mechanism, and outputting an equipment health state, a residual life prediction value and a prediction confidence coefficient in parallel; based on a multi-dimensional evaluation result, a hierarchical maintenance strategy is generated and executed through a decision of data confidence management; according to the invention, through dual health assessment of the equipment and the sensor, the attention mechanism enhanced prediction model and data-driven closed-loop optimization, the accuracy and reliability of fault prediction of the industrial equipment and the self-adaptive capability of the system are improved.

The invention relates to a cognitive load monitoring method based on multi-source physiological information fusion. The cognitive load monitoring method comprises the following steps that 1, physiological information data collection is conducted through a multi-source data collection module; 2, performing data processing through a data preprocessing module to obtain preprocessed electroencephalogram, eye movement and electrocardio data; 3, performing data layer fusion, feature extraction and feature layer fusion on the preprocessed data to obtain a feature layer fusion matrix; and step 4, constructing a CNN-BiLSTM-Transform hybrid neural network model, carrying out decision fusion, and carrying out iterative optimization on the model through evaluation. According to the method, the core basis problem of multi-source physiological information data fusion is solved, and a high-quality data basis is provided for subsequent feature extraction and model training; the accuracy and environmental adaptability of state evaluation are greatly improved; and the sensitivity and the recognition precision of the model to the dynamic change of the cognitive load are effectively improved.

The invention relates to the technical field of network communication, and provides a hierarchical reinforcement learning scheduling and routing method for a multi-domain TSN, and the technical scheme comprises the steps: collecting the global state information of the multi-domain TSN, and carrying out the coding processing of the global state information to generate comprehensive state representation; performing a cross-domain routing decision, and outputting an inter-domain path and a time delay budget of a cross-domain flow; executing intra-domain scheduling, determining a sending sequence and a specific path of a domain flow, and generating gating list configuration; through hierarchical coordination and strategy optimization, a cross-domain routing decision and intra-domain scheduling are updated based on reward feedback of an intra-domain scheduling result; and deploying the finally updated intra-domain scheduling to a switch of the multi-domain TSN network for execution, thereby realizing deterministic transmission of the time-triggered flow. According to the invention, through a layered agent architecture, a hybrid neural network coding mechanism and a cross-domain collaborative optimization strategy, challenges of complexity, expandability, dynamic adaptability and the like of a joint routing and scheduling problem in a multi-domain TSN environment are effectively solved.

The invention discloses an error compensation method for a parallel robot, the parallel robot is a planar 2-RRR parallel mechanism, and the method comprises the following steps: firstly, establishing a pose error model and sensitivity analysis, and defining observable and unobservable errors; then two-stage compensation is adopted, and the method is characterized in that in the first stage, an improved sparrow search algorithm (MSSA) is used for off-line calibration of observable errors, and in the second stage, a hybrid neural network model (HNN) is used for on-line calibration of unobservable residual errors; and finally, integrating errors of the two stages to finish compensation. According to the method, error compensation of the parallel robot can be better completed, the compensation precision is improved, the stability and the real-time performance are improved, and the tail end execution control precision of the robot is better improved.

The invention discloses a lake water quality prediction method and system based on a hybrid neural network, and a computer readable storage medium, and belongs to the field of environmental science engineering and deep learning. The method comprises the following steps: screening original water quality data, removing abnormal values, performing linear interpolation, dividing a training set and a test set, decomposing a sequence by using VMD, optimizing VMD parameters by using PSO, reconstructing a new sequence with noise removed, and finally performing prediction by using LSTM-KAN. Through verification of total phosphorus concentration data of four sections of the Dian Lake, comparison with LSTM, VMD-LSTM, VMD-LSTM-KAN and LSTM-KAN models is carried out, and a correlation coefficient (), a mean absolute error (MAE) and a root-mean-square error (RMSE) are selected to evaluate precision. The result shows that the PVLK model has the best performance in single-step and multi-step prediction, the total phosphorus concentration prediction of each section can be kept at 0.75 in 10-step prediction with the step length of 4 hours, the applicability to time sequence data containing abnormal values and high sampling frequency is good, and efficient prediction of lake water quality is effectively promoted.

The invention discloses a mine fire grading early warning method, system, device, medium and product, and the method comprises the steps: constructing an underground global map of a mine, and obtaining the multi-modal data of all local areas of the mine according to the underground global map; performing preprocessing and three-level data fusion on the multi-modal data, and determining a preliminary fire early warning result of the corresponding local area by adopting an edge hybrid neural network according to the fused multi-modal data; and according to the global multi-modal data of the mine, the initial fire early warning result and the real-time dynamic early warning threshold, performing fire risk prediction on the mine by adopting a central hybrid neural network to obtain a fire graded early warning result of the mine. The fire risk can be predicted more accurately based on multi-modal data fusion and a hybrid neural network; the early warning threshold value and strategy are automatically adjusted according to the real-time change of the mine environment by utilizing the technologies of online incremental learning, meta learning, reinforcement learning and the like, the parameters do not need to be manually adjusted, real-time calibration is realized, and the timeliness of early warning is improved.

The invention discloses an intelligent prediction method for river beach foundation settlement, and relates to the technical field of crossing of geotechnical engineering and artificial intelligence. During operation of a system, foundation state information is collected through multi-source monitoring equipment, a three-dimensional dynamic environment state space is constructed, and a prediction model is established based on a hybrid neural network; spatial and temporal features are extracted in combination with a bidirectional long-short-term memory network and a graph neural network, and prediction precision is optimized by adopting a multi-objective loss function and an adaptive learning mechanism. And meanwhile, a multi-region collaborative prediction and risk reconciliation mechanism is introduced, so that settlement trend coordination and safe avoidance are realized. According to the method, the foundation settlement behavior can be predicted in real time, the prediction strategy is dynamically adjusted, the prediction accuracy and engineering safety in a complex geological environment are remarkably improved, and an efficient solution is provided for river shoal alluvial layer foundation settlement.

The invention belongs to the technical field of water supply network leakage monitoring, and particularly relates to a water supply network leakage node and time window combined positioning method based on a hybrid neural network, which realizes spatial positioning of a water supply network leakage node and time positioning of a leakage time window. Comprising the following steps of 1, collecting node pressure and flow time sequence data of the water supply network in different scenes, and recording leakage nodes and leakage starting and ending time labels; 2, carrying out normalization preprocessing on the data, and constructing a standardized time sequence data set containing a multi-leakage scene mark; step 3, a dual-branch hybrid neural network model TCN-Transform-KAN is built, and a dual-branch hybrid neural network model TCN-Transform-KAN is built; 4, constructing a joint loss function of node classification loss and time regression loss; and step 5, training through a joint loss function constraint model, and outputting a leakage node and a center position and a continuous width of a leakage time window. According to the method, joint precise identification of the leakage nodes and the time range is realized, and technical support of time-space two-dimensional information is provided for intelligent operation and maintenance of a water supply pipe network.

The invention discloses a multi-source data driven irrigation district resource regulation and control method and system fused with a neural network, and relates to the technical field of irrigation district resource regulation and control, and the method comprises the steps: collecting irrigation district multi-source data, carrying out the cleaning, standardization and time-space dimension reconstruction processing, constructing a hybrid neural network regulation and control model fused with CNN spatial feature extraction and LSTM time sequence modeling, and carrying out the calculation of the hybrid neural network regulation and control model. A weighted mean square error is used as a loss function, model training is completed in combination with an Adam optimizer and hyper-parameter optimization, and finally optimal regulation and control parameters meeting resource constraints and crop requirements are output; according to the method, the defects that traditional regulation depends on artificial experience and the multi-source data fusion capability is weak are effectively overcome, data space-time correlation information is fully mined, scientificity and accuracy of irrigation area resource regulation are improved, water resource waste and fertilizer loss are reduced, accurate matching of crop supply and demand is achieved, the crop yield is guaranteed, the ecological environment is improved, and the economic benefit is increased. The method is suitable for multi-resource cooperative regulation and control of complex irrigation areas.

The invention discloses a photovoltaic access power distribution network fault diagnosis method based on fault identification, and particularly relates to the technical field of power grid fault diagnosis, and the method comprises the steps: synchronously obtaining multi-point electrical quantity monitoring data, photovoltaic system inverter operation data and environment data of a power distribution network side, and carrying out the abnormality elimination and delay compensation preprocessing; based on the switching frequency of the photovoltaic inverter and the line parameters of the power distribution network, adaptively selecting a wavelet basis function and a decomposition scale, and extracting a fusion fault feature vector; inputting the feature vectors into a two-channel hybrid neural network model for intelligent diagnosis, and outputting fault section positioning, type classification, severity level and occurrence time information; and finally generating a fault processing instruction according to the diagnosis result. According to the method, the problems of low diagnosis precision, slow response and the like caused by single data source, poor feature extraction self-adaption and simple model structure of the existing method are effectively solved, the accuracy and reliability of complex fault recognition are remarkably improved, and closed-loop automation of fault diagnosis and processing is realized.

The invention discloses a bridge structure health state evaluation and early warning system based on deep learning, and relates to the field of bridge structure health monitoring. The system comprises a data acquisition module, a preprocessing and feature engineering module, a deep fusion modeling module, a dynamic evaluation early warning module and an online updating module, and a closed-loop link is formed through a high-speed real-time data bus. The data acquisition module integrates multiple types of sensors to realize synchronous acquisition of multi-source heterogeneous data; the preprocessing module constructs a multi-dimensional feature vector through three-stage processing; the deep fusion modeling module adopts a double-flow hybrid neural network architecture, and constructs a health index calculation model of three-dimensional feature fusion in combination with an attention mechanism; the dynamic evaluation and early warning module realizes four-level health grading evaluation and three-level early warning based on the degradation knowledge graph; and the online updating module ensures the model precision through incremental learning. The system breaks through the limitation of a traditional fixed threshold value, structure abnormity can be predicted 3-6 months in advance, the false alarm rate is reduced by 40%, the missing report rate is smaller than or equal to 2%, and the intelligent level and safety of bridge operation and maintenance are improved.

The invention relates to the technical field of traffic services, in particular to a traffic service transport capacity demand prediction method and system, and the method comprises the steps: collecting historical operation data, track data, passenger flow counting data and external impact factor data for prediction, carrying out the preprocessing of the data, and dividing the data into a plurality of space attributions; constructing a time sequence feature, a periodic feature, a neighbor space feature and an instant available transport capacity feature set external factor feature based on the preprocessed data; generating short-term baseline prediction based on a seasonal autoregressive moving average model, constructing and training a space-time hybrid neural network model to extract spatial correlation and time dynamics of space attribution, outputting a space-time depth prediction result, and integrating the space-time depth prediction result with the short-term baseline prediction to obtain a demand prediction value of a future multi-step time period; and establishing an integer programming model comprising vehicle availability, shift constraints and transfer time constraints and solving the integer programming model to generate vehicle allocation or shift suggestions by taking the transport capacity level as the target and minimizing the scheduling cost.