406 results about "Sequence model" patented technology

The invention discloses a welded pipe conveying abnormity prediction method and system based on large model reasoning, and aims to solve the problems that multi-source data is difficult to align, cross-station false correlation is caused, prediction lacks executable positioning and time sequence, and linkage control reliability is insufficient. Event alignment is carried out by taking a controller edge signal and an encoder zero position as time anchor points, a production line topology semantic graph containing time delay, capacity and interlocking attributes is constructed, and topology reachability and physical time delay constraints are applied in a self-attention long sequence model to carry out multi-step rolling prediction. And outputting a risk probability, refining the risk probability to spatial positioning of a roller way section or a shaft and the minimum executable intervention time, and generating a risk interval in combination with uncertainty estimation and calibration so as to drive an upstream beat self-adaptive speed reduction, shunting or stopping strategy. The technical effects of improving accuracy and interpretability, reducing false alarm and missing alarm, ensuring that linkage can be executed in advance and meeting edge time delay budget are achieved.

The invention discloses a composite structure damage form monitoring method and system based on deep learning, and the method comprises the following steps: collecting multi-source monitoring data of a composite structure in a loaded state, and carrying out the preprocessing; reconstructing a damage evolution trajectory in a high-dimensional phase space by adopting a delay coordinate embedding method, and executing dimension reduction to generate a chaotic dynamics low-dimensional trajectory; extracting singular attractor features, and generating a singular attractor feature set; carrying out sequence modeling through an improved Linformer damage identification network, and generating a prediction vector; training an improved Linformer damage identification network based on the prediction vector, and introducing nonlinear dynamic constraints to generate a damage identification network of the nonlinear dynamic constraints; and performing damage form classification and damage evolution prediction. According to the method, dynamics and deep learning are fused, composite structure damage monitoring is achieved, and the method has the advantages of being high in accuracy, high in stability and reliable in early warning.

The invention relates to the technical field of electric power engineering purchase demand prediction, in particular to an electric power engineering purchase demand prediction system based on machine learning, and the system comprises the steps: obtaining historical purchase data, construction progress information and electric power engineering design parameters, carrying out the standard stage division and time alignment, and constructing a stage sequence model reflecting the material use rhythm; and a coupling factor matrix is generated based on the material co-occurrence frequency and the stage position relationship, and the modeling capability of the model for the material cooperation relationship is enhanced. And the stage time sequence features, the coupling information and the structured engineering parameter vectors are fused and input into a regression prediction model, so that accurate mapping of material demands and multi-dimensional engineering features is realized, and the purchase prediction precision in a target period is improved. A deviation sequence is constructed based on historical prediction errors, and error correction is performed through a feedforward neural network, so that prediction accuracy and response capability are effectively improved, and resource waste and construction delay are reduced.

The invention discloses an airport hub construction intelligent regulation and control method and system based on BIM + 5G, and relates to the technical field of construction scheduling optimization, and the method comprises the steps: collecting on-site real-time data to construct a path construction sequence model, constructing a dynamic construction state set with consistent space and time sequence through component coding and semantic attribute mapping, and constructing a dynamic construction state set with consistent time sequence; and constructing a minimum disturbance optimization algorithm of four-dimensional disturbance cost based on the construction disturbance mapping graph, outputting procedure sequence adjustment, resource rearrangement, path decoupling and environment avoidance intervention suggestions, and tracking a construction response state in real time based on an intervention execution feedback mechanism. According to the method, unified modeling of construction plans, resource allocation and green indexes is achieved by constructing a ternary structure and a construction disturbance mapping graph, the field state is dynamically collected in combination with a 5G and edge sensing system, process abnormity and resource conflicts are accurately recognized, an efficient and executable regulation and control strategy is generated based on a multi-target disturbance optimization algorithm, and the efficiency and the reliability of the system are improved. And the intelligence, responsiveness and energy-saving level of the construction process are obviously improved.

The invention provides an intelligent customer service dynamic intention recognition system based on a semantic analysis model, and the system collects the multi-dimensional data of a user through a data collection module, constructs a user portrait through a feature extraction module, extracts the portrait features, carries out the semantic analysis of multiple rounds of historical dialogue data, and extracts preference features. And constructing a dynamic user-entity association graph to extract GNN node features. Multi-modal data of a user is analyzed through an emotion recognition module, emotion features are recognized, portrait features, preference features and emotion features are fused and analyzed based on an MLP model to obtain a final emotion state, and the portrait features, the preference features, GNN node features and the emotion features are fused through an intention recognition module to obtain a final emotion state. According to the method, dynamic intention analysis is carried out on the basis of the Transform sequence-to-sequence model, the current intention of the user is recognized, the recognition accuracy is high, the dynamically changing intention is adjusted in real time, and more accurate and targeted answers or services are provided for the user.

The invention discloses a railway bogie bearing fault diagnosis method and equipment based on a feature extraction network, and relates to the field of intelligent diagnosis and maintenance guarantee of urban rail trains. The method comprises the following steps: data acquisition: acquiring vibration signals of a bogie axle box bearing under the same rotation speed and load combination; performing data preprocessing: performing frequency spectrum adaptive decomposition on each section of signal based on improved empirical wavelet transform, introducing a Fisher score and frequency spectrum entropy joint scoring mechanism, screening out key modal signals, and then performing image mapping to reconstruct a diagnosis sequence; model construction: introducing a topology perception attention mechanism module on the basis of the lightweight convolutional neural network, and constructing a fault diagnosis model; model training: training the fault diagnosis model to obtain an optimized fault diagnosis model; and diagnosis result output: using the optimized fault diagnosis model to diagnose the fault signal of the bogie axle box bearing, and outputting the diagnosis result. The method can improve the accuracy of fault diagnosis.

The invention provides an industrial equipment residual service life prediction method based on a DMM-JA model, and relates to the technical field of industrial equipment predictive maintenance, and the DMM-JA model comprises a dynamic bimodal fusion and multi-scale feature extraction module DBF-MSFEModule, an LSTM-Mama mixed sequence modeling module, a jump perception attention module and an output layer. The method comprises the following steps: preprocessing bimodal sensing data of industrial equipment; the preprocessed bimodal sensing data is processed through a dynamic bimodal fusion and multi-scale feature extraction module DBF-MSFEModule, and multi-scale fusion features are obtained; the multi-scale fusion features are input into an LSTM-Mamba mixed sequence modeling module, and joint time sequence features are obtained; the joint time sequence features are corrected through a jump perception attention module, and robustness features are obtained; and inputting the robustness characteristics into an output layer to obtain an RUL prediction result of the industrial equipment. Key features in the equipment degradation process can be accurately captured, and the accuracy of residual service life prediction is improved.

The invention relates to the technical field of monitoring behavior analysis, and discloses an intelligent behavior analysis system and method for a monitoring camera. The system obtains video stream data through a video data acquisition module, extracts a video frame sequence and a timestamp, identifies a moving target and a position coordinate, associates behavior segments and calculates an active value, and generates a target behavior active set; a behavior sequence modeling module extracts active values and coordinates, sorts adjacent behavior segments and marks consistent and conflict sections to obtain a behavior consistency partition marking set; the environment mapping module obtains consistent section target behaviors, extracts illumination and shielding time sequences, evaluates the influence intensity of environment interference on the behaviors in combination with behavior types, and generates an overlay analysis result; the exception screening module recognizes exceptional points with response values larger than the average reference value and located in the conflict section, and a behavior exceptional point set is formed; and the risk output module obtains abnormal point information, marks diffusion risk point locations, and generates behavior detection and risk early warning results.

The invention relates to an event-based lip reading method and device, equipment and a storage medium. The method comprises the following steps: collecting an original event stream of a lip sequence image through an event camera; therefore, the brightness change of each pixel can be asynchronously recorded with microsecond-level time resolution, and ultra-low delay, high dynamic range and sparse data representation are realized. Converting the original event stream into a frame-shaped event tensor based on a voxel representation method to obtain a voxelized event body; performing spatial feature extraction on the voxelized event body through a front-end network in an event-based lip reading model to obtain multi-scale spatial features; performing time-dependent modeling on the multi-scale spatial features through a rear-end sequence model in an event-based lip reading model to obtain a sequence code; therefore, space and time features can be fused, and the accuracy and stability of sequence coding are improved. And determining the lip reading recognition content according to the sequence code. Therefore, stable recognition of lip movement can be realized, and lip reading accuracy is improved.

The invention discloses a multi-source heterogeneous engineering monitoring data fusion method based on deep learning, and relates to the technical field of engineering monitoring and artificial intelligence crossing. Collecting multi-source heterogeneous time series data; inputting the multi-source heterogeneous time sequence data into a dynamic space-time alignment module, performing time sequence alignment to obtain alignment data, and extracting time sequence characteristics of the alignment data through a sequence-to-sequence model; constructing a sensor topological graph through an adjacent matrix based on the sensor data, determining node features of the multi-source heterogeneous time series data through a statistical feature method, and inputting the sensor topological graph and the node features into a graph convolutional network for processing to obtain spatial features; inputting the structural crack image sequence data into the residual convolutional network to extract visual semantic features; and dynamically fusing the time sequence features, the spatial features and the visual semantic features through a space-time cross attention mechanism to obtain fused features. According to the method, the compatibility of engineering monitoring data fusion can be improved.

The invention discloses a wearable fatigue monitoring and feedback method based on deep learning. The method comprises the steps that a heart rate variability signal, a gamma wave band electroencephalogram signal and body movement posture information of a user are collected in real time; denoising, normalizing and synchronously fusing the acquired multi-mode signals; extracting a fatigue state representation vector in real time by adopting a Mamba linear state space sequence model; estimating a user fatigue index in real time based on a lightweight full-connection neural network decoder and constructing an individual fatigue threshold dynamic model; calculating a phase synchronization index of the electroencephalogram signal in real time; and generating and outputting an individualized 40Hz gamma wave band sensory nerve stimulation feedback signal in real time based on the fatigue index and the phase synchronization index. According to the invention, high-robustness fatigue identification and low-delay feedback adjustment in a complex motion noise environment are realized.

The invention provides a user behavior analysis and interest recommendation method which fuses time-assisted features, introduces comparative learning and uses meta-learning optimization. According to the method, a user interaction sequence is enhanced by using a time homogenization strategy, and optimization is performed by using contrast learning, so that a next interested article is recommended for a user. The specific process comprises the following steps: preprocessing original interaction data, and performing time-level data enhancement on a user interaction behavior sequence, so that the original interaction sequence becomes a more uniform sequence; and then constructing a comparative learning framework, performing comparative training on the original sequence and the enhanced sequence, and optimizing a comparative learning result by constructing a positive and negative sample pair, combining the original loss and the comparative loss and adopting a meta-learning mechanism. Moreover, a multi-head self-attention mechanism is introduced to model a context relationship between nodes, user behavior features in time evolution are captured, and the discrimination capability of feature representation is improved. According to the method, on the basis of fusing time information, contrast learning and meta learning, the sequence modeling capability in a data sparse environment is effectively improved, and accurate description of user interest dynamics is realized.

The invention relates to a big data security management system, and provides a security management system covering a whole life cycle aiming at the problems of untrusted source end, link leakage, centralized storage of single points, extensive authorization, disposal lag and the like. The system comprises an acquisition security module, a transmission encryption module, a storage security module, an access control module, a threat detection module, a situation awareness module, a traceability module and an emergency response module. A session basis is generated through multi-factor authentication, a dynamic key is derived according to time slices, and data fragments are encrypted one by one; distributed storage is adopted, and metadata is chained to be non-tampered; implementing fine-grained authorization by taking the attribute as a core; utilizing a sequence model linkage knowledge graph to identify abnormity and calculate a risk index; and when the threshold is exceeded, automatically reconstructing a log chain and executing isolation, alarm or recording. According to the system, trusted acquisition, confidential transmission, traceable storage, fine authorization and automatic disposal are realized, and the safety and compliance are remarkably improved.

The invention discloses a landslide image instance segmentation method based on a dual adaptation mechanism, relates to the technical field of image processing, and ensures that a model can have a better effect and numerical stability on data from different sources through a complete process from multi-source data collection to data normalization processing. The designed model is based on a multi-scale state alignment mechanism, in the forward process of the model, exponential moving average fusion is carried out on feature information in the same level, transmission fusion is carried out on feature information of different levels, feature robustness is enhanced, and error accumulation caused by local deviation is reduced. A meta-context incremental learning mechanism is designed, and input data are dynamically converted into a series of key value vector sequences. In the reasoning process of the model, data distribution different from a training domain is dynamically recognized, a gradient descent process is implicitly executed according to the characteristics of data, and model parameters are finely adjusted, so that the characterization capability is greatly improved, and efficient and robust instance segmentation is realized.

The invention discloses a bill recognition method and system based on artificial intelligence image enhancement, and relates to the field of image recognition. The method comprises the following steps: S1, extracting multi-dimensional quality features based on an original image of a bill and calculating a scene consistency factor; s2, adjusting a global enhancement weight according to a scene consistency factor, adjusting a local gain in combination with a detail fidelity factor, and performing adaptive enhancement on the original image to generate an enhanced image; s3, establishing an optical flow field model to analyze geometric deformation of the enhanced image, and performing adaptive correction in combination with local deformation rigidity to generate a corrected image; and S4, analyzing gradient features and character confidence of the corrected image, extracting a candidate character region, and performing context recognition by adopting a sequence model to obtain a text field set. Scene complexity is quantified through multi-dimensional quality features, and detail fidelity self-adaptive enhancement and optical flow deformation correction of high-frequency character distinguishing are combined, so that bill character definition and recognition accuracy are remarkably improved.

The invention relates to the technical field of computer processing, in particular to a Kafka cluster configuration parameter adjusting method and device, equipment and a medium, and is used for relieving problems caused by a fixed change means of code modification. The method comprises the following steps: according to related historical data of a Kafka cluster, predicting a first predicted load index in a future time period through a first time sequence model, and predicting a second predicted load index in the future time period through a second time sequence model; weighting the first predicted load indicator; weighting the second predicted load index; determining a predicted load index according to the weighted first predicted load index and the weighted second predicted load index; determining a plurality of first candidate adjustment actions according to the predicted load index; selecting a first candidate adjustment action from the plurality of first candidate adjustment actions as a first adjustment action; and adjusting a partition allocation index and a copy allocation index of the Kafka cluster configuration parameters by using the first adjustment action.

Aiming at the problems of low efficiency, strong subjectivity, insufficient modeling ability, complex hyperspectral data noise, weak waxiness spectrum difference and the like of a traditional rice grain waxiness discrimination method, the invention provides a hyperspectral rice grain waxiness discrimination method based on multi-branch collaborative modeling. The method comprises the following steps: S1, acquiring glutinous and non-glutinous rice grain images by using a hyperspectral imaging system to obtain spectral data; s2, preprocessing spectral data through a combined method of SG smoothing, an asymmetric weighted penalty least square method and multivariate scatter correction to reduce noise and interference; s3, constructing a multi-branch modeling architecture which comprises a CNN local feature extraction module, an SRU spectrum sequence modeling module and a HorNet global high-order modeling module, and outputting a discrimination result through a classifier after multi-branch features are subjected to fusion and pooling attention weighting; and S4, evaluating the performance. According to the method, high-precision lossless discrimination of the waxiness is realized, and a technical support is provided for germplasm screening and quality evaluation in rice breeding.

The invention discloses a drug target binding affinity prediction method based on multi-scale feature fusion, and the method comprises the steps: introducing the multi-scale structural features of atoms, atomic groups and molecular levels in the expression of drug molecules, and combining a graph neural network and a sequence modeling module as a feature extractor; and deep interaction and fusion between different scale features of the protein and the drug are realized by using a multi-head bilinear cross attention mechanism, so that potential binding site information is effectively captured, and the accuracy and interpretation capability of affinity prediction are improved. The method not only overcomes the problems that a traditional machine learning method depends on manpower and is low in efficiency in feature construction, but also solves the technical bottlenecks that an existing deep learning method is only limited to local neighborhood information and cannot model global structural features, and the interaction relationship modeling capability is insufficient due to direct splicing of drugs and protein representation.

The application discloses a power equipment defect detection method, system and device based on multi-modal alignment, and particularly relates to the technical field of power equipment detection, and the technical points are as follows: the method comprises the following steps: in the collected visible light image, infrared image and ultrasonic signal, the spatial feature difference and the time feature difference between every two modal data are obtained; the spatial feature difference and the time feature difference are subjected to logic regression processing, the alignable coefficient between every two modal data is obtained, the to-be-aligned data combination for power equipment defect detection is determined based on the alignable coefficient; the to-be-aligned data combination is subjected to data alignment, the aligned modal data is obtained, and according to the aligned modal data, the time sequence model is used for defect detection analysis on the power equipment, so that the power equipment defect detection result is obtained.

The invention relates to the technical field of computer vision and deep learning, and discloses an image classification system and method fusing a space attention mechanism and long and short-term memory network sequence modeling, which combines the feature extraction capability of a convolutional neural network and space attention and time sequence attention mechanisms. And processing the spatial position sequence by using a long-short-term memory network. Firstly, advanced spatial features of an image are extracted through a feature extraction module by adopting a pre-trained convolutional neural network with a frozen weight, then a spatial attention module is introduced, a spatial attention graph is generated through channel dimension statistics, and important region features are enhanced. After the spatial attention is weighted, a dual-path feature is utilized, one path enters a feature transformation module, and a convolutional layer is used for reducing dimensionality and enhancing feature expression ability. Then, a sequence modeling module is carried out, the spatial features are flattened into a position sequence, a spatial position dependency relationship is modeled by adopting a bidirectional long-short-term memory network, and learnable attention vector dynamic aggregation key position features are introduced; the other path retains spatial global features. And outputting the dual-path features to a feature fusion module, extracting spatial global features and sequence aggregation features in parallel, and designing a gating mechanism to adaptively fuse the dual-path features. And finally, entering a classification module, and realizing end-to-end image classification based on fusion features.

The invention provides a residual life lightweight prediction method and system based on feature decoupling and sparse optimization, and the method comprises the steps: efficiently extracting equipment degradation features through a double-branch architecture: capturing a time sequence degradation law through a GRU network, and generating a deep feature h1 in combination with an attention mechanism, one branch directly extracts a key feature h2 from an original sensor signal through a sparse attention mechanism, the other branch directly extracts a key feature h2 from the original sensor signal through a sparse attention mechanism, then two feature vectors are fused and processed through a lightweight decoder, and finally a residual life prediction value is output. The balance between precision and calculation efficiency is realized through feature decoupling and sparse optimization, and the method is particularly suitable for deployment in an edge calculation environment with limited resources.

The invention provides a phytoplankton chromatography sequence identification method and a phytoplankton chromatography sequence model building method, and belongs to the technical field of image enhancement identification. The method comprises the following steps: firstly, acquiring microscopic chromatography sequence data of phytoplankton, performing view field extraction and serialization recombination, and constructing a three-dimensional data set; then, constructing a three-dimensional recognition model containing physical perception and a sequence aggregation mechanism, extracting single-frame semantic features by the model by adopting a parameter-shared twin network, and introducing a physical definition prior module to calculate a space-frequency domain quality score of a slice; secondly, designing a deep perception sequence aggregation module, and adaptively aggregating key features of a high signal-to-noise ratio by taking definition scores as gating signals and combining spatial context information between slices; and finally, training and optimizing the model based on the image-level weak supervision label to obtain an optimal model. According to the method, the problems of information truncation and out-of-focus noise interference caused by extremely shallow depth of field of high-power microscopic imaging are solved, and full-depth-of-field stereoscopic perception can be realized under the condition that frame-by-frame fine labeling is not needed.

The invention belongs to the technical field of artificial intelligence and multi-modal emotion calculation, and discloses a multi-modal emotion recognition method based on a Mama state space model and cross-modal self-distillation. Through the organic combination of the efficient sequence modeling capability of the Mamba state space model and the knowledge sharing mechanism of cross-modal self-distillation, the advantages of the state space model in the aspects of time sequence modeling and calculation efficiency are fully played, and meanwhile, the limitation of a single model architecture is made up through a cross-modal attention mechanism; the technical bottlenecks of an existing multi-modal emotion recognition method in the aspects of long sequence processing efficiency, cross-modal information fusion and knowledge transfer sufficiency are effectively solved, and an efficient and reliable technical solution is provided for further development and practical application of the multi-modal emotion recognition technology.

The invention discloses a neural signal-based wounded rehabilitation effect evaluation model construction method, and relates to the technical field of medical rehabilitation, and the method comprises the steps: synchronously collecting electroencephalogram, myoelectricity and motion parameters, extracting multi-source features to construct feature vectors corresponding to training actions, and according to the feature vector data of multiple training actions of a wounded, constructing a neural signal-based wounded rehabilitation effect evaluation model; and capturing dynamic changes of neural signals and action control ability in the training action sequence based on the sequence model, and constructing an evaluation model. According to the invention, time-dependent learning is carried out on the corresponding characteristics of each training action by using the sequence model, dynamic changes of neural signals and action control ability along with the training actions can be captured, an improvement trend in a rehabilitation process is embodied, and an evaluation model aiming at an action training effect is constructed; the action indexes output by the evaluation model can reflect the instant effect and gradual improvement condition of each action in training, and help to identify rehabilitation inflection points and training bottlenecks.

The invention discloses a sewage plant water quality prediction and management operation method, a computer program product and an electronic product, and belongs to the technical field of water quality prediction.The method comprises the steps that a water quality prediction model is constructed, the water quality prediction model comprises a GeoMAN model and a time sequence model which are cascaded, and the GeoMAN model comprises an HRT dislocation compensation module; the dynamic alignment module is used for carrying out dynamic alignment treatment on the effluent quality data of each treatment unit according to the actual hydraulic retention time of each process section treatment unit; and after training the water quality prediction model, performing water quality prediction based on the real-time data, and outputting a water quality prediction result. The spatial features of the data are captured through the GeoMAN model, and the time features of the data are captured through the time sequence model, so that the spatial and time complexity of the water quality data and the process control parameter data is comprehensively understood, and the water quality prediction precision is improved. And the HRT dislocation compensation module is used for carrying out time alignment treatment on the effluent quality data of the treatment units in different process sections, so that the accuracy of water quality prediction can be further improved.

The invention discloses a lightweight lip language recognition method and device. The method comprises the steps of obtaining a to-be-recognized original video; performing data preprocessing on the obtained original video through face tracking detection and face feature point detection technologies to obtain a lip video sequence and a compressed representation of the lip video sequence; expanding a training data set for the obtained lip video sequence data by using a data enhancement algorithm; carrying out feature extraction on the lip movement video sequence by using a 3DConv + 2DResnet convolutional neural network according to the enhanced training data set; performing sequence modeling on a feature vector obtained by performing feature extraction on the lip movement video sequence by using a Conformer time sequence modeling network to obtain a lip language recognition model; performing visual speech decoding on the lip language recognition model obtained through sequence modeling by using a classification network to obtain language characters corresponding to the to-be-recognized video; and carrying out lightweight processing on the lip language recognition model structure by using a model compression algorithm. The lip language recognition performance is improved, and the actual application requirement is met.

The invention relates to the technical field of knowledge tracking, and discloses a graph enhanced double-memory collaborative knowledge tracking model based on an ACT-R cognitive architecture. Comprising a static knowledge structure coding module based on hypergraph projection, a batch-level dynamic learning track construction and coding module, a cross-graph gating fusion mechanism, a sequence modeling module and an expert hybrid prediction module. According to the method, long-term stable structured semantic association between concepts in declarative memory is modeled through a static knowledge structure diagram, a dynamic learning trajectory diagram based on batch reconstruction is designed to accurately capture an evolution rule of a behavior sequence in programmed memory, and on the basis, a cross-diagram gating fusion mechanism and a hybrid expert mechanism are introduced, so that the evolution rule of the behavior sequence in the programmed memory is accurately captured. And self-adaptive fusion and multi-path decision of double-graph features are realized.

The application provides a wind farm equipment health degree evaluation method and system based on multi-source data fusion and adaptive dynamic modeling, comprising: collecting multi-source data, preprocessing the data, using a timestamp as an index, fusing data of different sources by using a data fusion technology, intercepting sequence data within a fixed time period p, reducing the dimension of the sequence data by using PCA, and extracting main components; training a dynamic LSTM sequence model, calculating a sliding step q according to a deviation index, setting p = p + q, returning to training; and outputting a prediction result. An adaptive dynamic LSTM model is constructed by using multi-source data and a sliding time window with variable width, real-time discrimination of fan faults is realized by combining a dynamic threshold algorithm, data-driven equipment health degree evaluation and degradation trend prediction are supported, an attention mechanism is added, important moments in the time sequence are captured, the convergence speed is accelerated, the accuracy of the model is improved, and power generation optimization and ecological compatibility decision-making are supported.

The invention provides a self-adaptive weighted hybrid modeling method and system for data drift sensing. The method comprises the following steps: acquiring multi-dimensional time series data and performing feature interaction analysis and screening to generate a fusion factor set; inputting the fusion factor set into a mixed structure comprising at least one ensemble learning model and at least one sequence model for training; based on the verification set and the dynamic evaluation indexes, determining fusion weights of all models in the mixed structure, and constructing a dynamic weighted mixed model; and setting a data distribution drift detection mechanism, comparing the distribution difference between a current data window and a historical reference data set, and automatically starting a retraining process of the hybrid structure to update the model when the difference reaches a trigger condition. According to the method, through cooperation of data drift perception and an adaptive weighting mechanism, the problems that an existing hybrid model is rigid in fusion strategy and lags behind a retraining mechanism are solved, and the robustness and long-term effectiveness of the model in a non-stationary data stream are remarkably improved.

The invention discloses a road and bridge construction progress prediction method and system based on unmanned aerial vehicle vision, and belongs to the technical field of construction monitoring. Image and material approach data are collected through an unmanned aerial vehicle, construction components are recognized by combining target detection and a three-dimensional reconstruction technology, visual engineering quantity estimation and material consumption time sequence data are fused, and multi-source collaborative analysis and anomaly detection are achieved. The construction progress is automatically judged, the completion percentage is calculated, deviation root causes are actively identified through a dynamic risk feature library, and future completion time and risk early warning are predicted in combination with a time sequence model. Compared with a traditional method, the accuracy and real-time performance of construction progress monitoring are improved, full-process automatic management from data collection to intelligent early warning is achieved, the manual checking cost is effectively reduced, and the engineering risk prevention and control capacity is improved.