138 results about "Adaptive selection" patented technology

The invention discloses an Internet of Vehicles CAN bus intrusion detection method based on noise perception active learning, and belongs to the technical field of Internet of Vehicles safety and machine learning. The invention aims to solve the technical problems of false label noise interference, high manual labeling cost, high attack missing report rate caused by class imbalance and the like. The core of the method is to execute a noise sensing mixed query strategy in an iterative loop: firstly, generating a pseudo tag through clustering and correcting by using an integrated noise detector; secondly, calculating uncertainty scores and noise probabilities of the samples, fusing the uncertainty scores and the noise probabilities to obtain a comprehensive score, and preferentially selecting the samples with high uncertainty and low noise probabilities; and then adaptively selecting a sampling strategy according to the model performance and applying category balance constraint. The query batch is used to iteratively update the model while dynamically adjusting the classification threshold to reduce the missing report rate. According to the method, the influence of pseudo label noise can be effectively suppressed, and the attack detection precision and generalization capability are remarkably improved with extremely low labeling cost.

The invention discloses a security guarding method and system based on a communication-guide-remote fusion technology, and the method achieves the self-adaptive selection and redundant transmission of communication links in a weak network, shielding and other environments through the construction of a multi-link communication system, an indoor and outdoor integrated high-precision positioning mechanism and a multi-mode intelligent sensing cooperation mechanism. Cross-scene continuous positioning is realized by combining multi-source information such as GNSS, inertial navigation, geomagnetism and remote sensing image backbone maps, and inertial navigation, video, sound and environment sensing data are fused to generate an abnormal behavior recognition result, so that the whole-process and whole-scene safety guarding capability for all personnel is formed, and the accuracy of abnormal behavior recognition is improved. The problems of easy communication interruption, easy positioning misalignment and unreliable abnormity identification in the prior art are effectively solved, and the method is suitable for a plurality of application scenes such as crowd safety guarding and the like.

The invention discloses a self-adaptive text extraction method and system based on artificial intelligence, and the method comprises the steps: carrying out the analysis of the document structure entropy of an example document set, quantifying the noise density, geometric distortion degree and background complexity of the example document set, and carrying out the self-adaptive selection of a preprocessing assembly line intensity grade according to the above; dynamically configuring image preprocessing parameters and AI recognition model parameters, and generating a recognition engine instance to output a preliminary recognition text; after regularized coarse screening extraction is carried out based on key field description, a multi-candidate generation strategy is started for low-confidence-coefficient candidate text fragments, a multi-person cooperative verification process is triggered for lower-confidence-coefficient fragments, finally all the fragments are processed through a text standardization module, and structured text extraction information is output. According to the method, accurate adaptation of processing intensity is achieved through document quality quantitative evaluation, the extraction accuracy and system robustness of complex heterogeneous documents are effectively improved through a multi-level confidence coefficient verification mechanism, and the identification error risk caused by image quality fluctuation or rule solidification is reduced.

The invention relates to a self-adaptive selection restoration method for defocus blurred image restoration, and belongs to the technical field of image restoration. The method comprises the following steps of: establishing a multi-scale branch which consists of three coding blocks and decoding blocks and is used for processing images with different resolutions; in each branch, shallow layer features of an input image with the corresponding resolution are extracted through a convolutional layer, and then image reconstruction is carried out from the shallow layer features by self-adaptive selection modules in corresponding coding blocks and decoding blocks; wherein each self-adaptive selection module comprises a self-adaptive double-branch fractional order module and a double-path fusion strategy based on gating reweighting, so that image reconstruction and feature fusion among different branches are carried out respectively, and finally a reconstructed image is output. Compared with the prior art, the method provided by the invention can realize a more efficient restoration effect on the compressed blurred image with lower model parameter quantity and calculation complexity.

The invention discloses a self-adaptive multi-feature fusion hybrid retrieval sorting method and system, and belongs to the technical field of information retrieval. The method comprises the steps that after user query is received, a mixed retrieval process and an intention recognition process are executed in parallel; performing multi-dimensional feature extraction on the candidate documents obtained by the mixed retrieval, wherein the multi-dimensional feature extraction comprises semantic correlation features, keyword matching features, document authority features and timeliness features; and according to the identified query type, adaptively selecting a fusion weight, and carrying out weighted fusion on the multi-dimensional feature vector to calculate a final score and sort the final score. According to the method, the problem that weight distribution is rigid in traditional mixed retrieval is solved through an intention self-adaptive dynamic weight mechanism, meanwhile, by introducing multi-dimensional service features, the ranking result not only ensures the correlation, but also meets the quality requirement under a service scene, and the accuracy and practicability of a retrieval system are remarkably improved.

The invention relates to the technical field of computer vision and target detection, and discloses a spatial faint target detection method based on semantic large model assisted reasoning, which comprises the following steps: preprocessing an image to be detected; by improving the primary detection of a YOLOv10 double-detection-head frame, the frame supports a user to controllably or adaptively select a detection mode; extracting a semantic generation prompt from the initial result, and inputting a large language model to obtain a potential omission category; key categories are extracted and mapped into IDs, and secondary directional detection is carried out; and fusing the two results and outputting. According to the method, the YOLO and the large model are combined to propose a YOLO-large model detection framework, compared with a single YOLO algorithm, the method can better complete detection of space dark weak tiny targets or low-pixel targets, meanwhile, the problems of missing detection and error detection in target detection are effectively solved, the defects of an existing YOLOv10 technology are overcome, and the optimal balance of precision and speed is achieved.

The invention discloses a sleep emotion state recognition-oriented multi-mode sensing signal analysis method and system, and relates to the technical field of sleep emotion data processing. According to the scheme, a progressive data processing method is constructed, and emotion-sleep state recognition and self-adaptive intelligent feedback regulation and control are achieved. The core process comprises the steps of ensuring the time sequence consistency of original data through multi-mode signal transmission time sequence interference analysis, adaptive selection of fixed clock calibration alignment and multi-mode signal dynamic resynchronization, then extracting and fusing deep features of sleep emotion multi-mode signals by using a deep network, and further completing emotion-sleep state recognition. And finally, sleep emotion collaborative intelligent regulation and control are driven according to a sleep-emotion state recognition result, and a physiological safety monitoring mechanism is introduced, so that the reliability and safety of the regulation and control process are ensured, and the effect of improving the emotion-sleep state recognition and intelligent feedback regulation and control accuracy is achieved.

The invention discloses a non-intrusive power user identification method based on self-supervised comparative learning, and relates to the technical field of non-intrusive power monitoring and user identification, and the method comprises the steps: input data are electricity utilization power time sequence data recorded by equipment such as a household intelligent electric meter; inputting the data into a double-layer contrast learning framework based on clustering guidance, and completing power user identification in combination with an adaptive time-frequency feature enhancement mechanism; the method comprises the following steps: constructing a double-layer contrast learning module, introducing a CEEMDAN algorithm to decompose IMF components of power consumption data of different dates, mapping distribution characteristics of the IMF components into selection probabilities of convolution kernels with different widths by a training network, completing adaptive selection of the convolution kernels by means of a Gumbel-softmax mechanism, removing accidental fluctuation of data through convolution in a time domain, and obtaining a time domain; and finally, fusing features of different dates through an attention module, mapping the features into probability distribution through a full connection layer, and classifying the probability distribution, thereby finally realizing accurate identification of non-intrusive power consumers.

The invention provides an efficient single-stage super-resolution method and a remote sensing image super-resolution system for RGB remote sensing images, and the method integrates Fourier position coding guided geometric node modeling, hypergraph high-order relation modeling and a Mama long-range dependence modeling mechanism based on a state space model. And consistent modeling of a multi-scale structure is realized while the linear calculation complexity is kept. In the method, FOPE is introduced as explicit geometric priori to guide adaptive screening of key nodes on global, regional and local scales; then, an FOPE enhanced hypergraph network is constructed, and high-order space association among multiple nodes is modeled through hyperedge; according to the method, a two-stage Mama aggregation module is further designed, and intra-node feature consistency modeling and inter-node long-range spatial dependency modeling are completed with linear complexity; and finally, a high-resolution RGB remote sensing image is generated through a lightweight direct up-sampling module.

The invention discloses a multi-robot path planning method in a complex environment based on MQAHA, and belongs to the technical field of multi-robot path planning. According to the method, firstly, an optimal point set initialization strategy based on the number theory is adopted, an initial population which is evenly distributed is constructed, and diversity and coverage of a search starting point are remarkably improved; secondly, proposing an elite-guided variation migration foraging strategy, fusing direction guidance of a current optimal solution and a disturbance mechanism of a random individual, and effectively enhancing local development capability and global escape performance; furthermore, a learning-driven strategy self-adaptive selection mechanism is introduced, the foraging behavior probability is dynamically adjusted according to the population state, efficient balance of exploration and development is achieved, and therefore the premature convergence problem of the algorithm is systematically relieved. According to the method, the multi-robot cooperative motion path can be quickly generated in a complex obstacle environment, static and dynamic obstacles are effectively avoided while the total path length is reduced, and the motion efficiency and operation safety of a multi-robot system are remarkably improved.

A multi-element microcouple adaptive selective corrosion dynamic expansion simulation method, belonging to the field of electro-corrosion simulation technology, is disclosed. The method includes: constructing a geometric model of compound particles and meshing the model; constructing the compound particle transport control equations and establishing the coupling relationship between potential, particle concentration, current density, boundary coverage, and reaction rate; adaptively selecting the anode and cathode among the multi-element microcouples of the compound particles and performing fixed-value processing on the geometric model; setting boundary conditions and initial values, including setting insulating boundaries, electrode surface boundaries, and initial particle concentrations; re-meshing the mesh, calculating the movement velocity of the corrosion interface, and constraining and controlling boundaries that do not corrode or undergo geometric deformation to complete the construction of the finite element model; solving the finite element model and outputting relevant physicochemical parameters during the corrosion process. This application clarifies the driving mechanism of pitting corrosion under open-circuit conditions and the existence state of corrosion products.

The invention relates to a large model compression method based on continuous layer pruning and endpoint tuning, and the method comprises the steps: firstly introducing a learnable continuous interval soft mask, and building a differentiable hierarchical mask mechanism in a model in cooperation with a residual bypass; secondly, by minimizing the KL divergence between output distributions before and after pruning, the optimal pruning starting point and length are automatically learned, and adaptive selection of continuous layer segments is achieved; then, executing physical layer deletion according to the optimized interval parameters, and reconnecting the network structures before and after pruning; and finally, implementing an endpoint tuning strategy, only carrying out all-parameter fine tuning on key layers on two sides of the sheared interval, and recovering the model performance at the lowest calculation overhead. According to the method, through combination of differential interval search and end point directional optimization, accurate compression and high-performance maintenance of the depth dimension of the large model are realized, model storage occupation and reasoning delay are remarkably reduced, the model output reliability in a key task scene is guaranteed, and the method is suitable for large-scale popularization and application. The method is particularly suitable for efficient deployment of the large language model in a resource-constrained environment.

This invention discloses an incremental federated causal structure learning method in dynamic scenarios, involving the interdisciplinary field of computer causal inference and federated learning. The method includes the following steps: S1: Historical federated causal structure learning; S2: Adaptive selection of new clients; S3: Weighted federated causal structure learning. A weighted federated causal structure learning mechanism is designed to achieve efficient collaborative incremental learning between the server and high-quality, non-redundant new clients. This invention uses an improved clustering method and a multi-dimensional quality assessment strategy to select high-quality nodes from new clients. Among these high-quality clients, redundant new clients with historical causal structures similar to those on the server are identified to reduce redundant computation. Weighted federated aggregation learning is then performed on the selected new clients based on the historical causal structure, eliminating the need for full learning of both historical and new clients, effectively reducing communication overhead while maintaining learning accuracy.

The invention provides a hyperspectral and multispectral image fusion method and system. The method comprises the following steps: acquiring hyperspectral and multispectral data of the same spatial region; performing up-sampling on the low-resolution hyperspectral image by adopting bilinear interpolation; one-dimensional wavelet transform is utilized to extract reflection peak and spectrum change characteristics in a spectrum dimension, two-dimensional wavelet transform is utilized to extract texture and edge information in a space dimension, and pixel-by-pixel wavelet basis adaptive selection is realized through Softmax gating; multi-granularity feature modeling is realized for the spectral branches by adopting grouped multi-scale convolution, and a self-adaptive weighting mechanism is formed for the spatial branches through global average pooling, a convolution layer and Sigmoid activation so as to enhance spatial details and compensate spatial feature loss by utilizing MSI; a cross-scale two-way attention mechanism is introduced, and long-range dependence modeling and information interaction of spectrum and spatial characteristics are achieved; and a high-resolution hyperspectral image is generated through a feature aggregation and reconstruction module.

The embodiment of the invention provides a photovoltaic power prediction method based on dynamic weight allocation and optimization. The method comprises the following steps: acquiring photovoltaic resource data; performing data preprocessing operation on the photovoltaic resource data; selecting a prediction model according to the photovoltaic resource data subjected to the data preprocessing operation; through a particle swarm optimization and reinforcement learning framework, fusing an expert strategy and a multi-objective optimization function, and training the prediction model; and predicting the photovoltaic power by using the trained prediction model, and recording a prediction result. According to the technical scheme of the invention, adaptive selection and optimal fusion of the prediction model can be realized, and the calculation efficiency and generalization ability are improved while the prediction precision is ensured. The embodiment of the invention also provides a system using the photovoltaic power prediction method based on dynamic weight distribution and optimization, and computer equipment implementing the photovoltaic power prediction method based on dynamic weight distribution and optimization.

This invention relates to the field of passive underwater target identification technology, and in particular to a propeller noise cyclic coherent demodulation method based on an improved real-number genetic algorithm. Based on cyclic stationarity analysis theory, an optimal coherent weighted quadratic envelope spectrum is constructed by replacing the fixed integration frequency band with a weighting function, and an adaptive detection threshold is constructed based on the false alarm probability. The observed spectrum is modeled as the sum of harmonics and noise, and a sparse non-negative optimization problem is solved to extract the harmonics. Using the total harmonic intensity of the envelope spectrum as the cost function, an improved real-number genetic algorithm is used to optimize and search for the optimal weighting coefficient vector. The demodulated spectrum is obtained by coherently weighting and integrating the normalized cyclic modulation spectrum. This invention achieves autonomous allocation of weighting coefficients and adaptive selection of demodulation frequency bands, effectively improving the signal-to-noise ratio gain of cyclic demodulation, and is applicable to passive target identification in data backtracking and review systems.

The application discloses a kind of processing and early warning method of monitoring data, it is related to offshore platform jacket monitoring technical field, comprising: obtaining jacket monitoring data, setting noise reduction method obtains analysis data;Adaptive selection method is constructed to carry out analysis, judge whether to replace noise reduction method to monitoring data is reduced, and constitute noise reduction dataset;According to the different noise reduction data, construct early warning method selection rule, select early warning method to obtain early warning result;According to noise reduction dataset and IDW interpolation algorithm, construct key data prediction model to obtain key data prediction result;According to early warning result and prediction result, determine jacket structure health condition;The application can select different noise reduction method for different data, adopt the most suitable method to carry out early warning, improve early warning speed and accuracy, based on interpolation algorithm according to the sea condition data that has occurred, predict the sea condition data that has not occurred, according to the early warning condition and key data condition obtained, assess the health condition of jacket.

The present application relates to a kind of weed classification detection method and system based on YOLOv8 improved algorithm.The method comprises: the image to be detected containing weed is input to YOLOv8 improved algorithm model, the image data set after processing is input to the backbone network of YOLOv8 improved algorithm model, different scale image feature information is extracted by C2f-FADC module and input to Neck network, DASI module carries out low resolution high semantic information and high resolution low semantic information fusion, produces multi-scale semantic feature information and is transmitted to task alignment detection head, extracts different task interaction features, calculates classification features, obtains weed classification result.FADC module is used to replace Bottleneck module in C2f module, C2f-FADC module is proposed, and the module is integrated into backbone network, which can improve the receptive field of convolution;DASI module can well enhance the accuracy of small target detection by adaptive selection and fine fusion of high and low dimensional features, improve the performance of weed detection model.

The invention discloses a multi-modal database-oriented adaptive index structure selection method, which comprises the following steps of: analyzing statistical characteristics such as data dimensions, variances, sparseness and distance distribution of each modal, calculating hidden dimensions of the modals, and describing effective data complexity of the modals; based on a preset index adaptation rule, automatically mapping each mode to an optimal index type in the candidate index structure set, and completing automatic construction of a local index and binding of the local index and a global routing structure according to the optimal index type; on the basis, the multi-modal query can be automatically routed to the corresponding index to execute retrieval according to the index mapping relation. According to the method, an index structure self-adaptive selection mechanism without manual configuration is realized, the index maintenance cost of the multi-modal database can be remarkably reduced, and the query efficiency and expandability in a complex retrieval scene are improved.

The invention discloses an unmanned aerial vehicle remote sensing small target detection method and system based on space-frequency domain aggregation and multi-scale feature enhancement, and belongs to the field of computer vision. According to the method, RT-DETR is adopted as a basic network, a backbone network for extracting spatial domain and frequency domain information is designed, and a self-adaptive selection mechanism and a multi-scale feature fusion module are added, so that the small target detection capability is greatly improved. The method comprises the following steps: step 1, inputting an image into a cross-stage space-frequency domain aggregation feature extraction network, and extracting feature maps of four levels of P2, P3, P4 and P5 from the input image;

The present application relates to the technical field of machine learning, and particularly relates to a preclinical drug experiment quality evaluation method based on machine learning, comprising: obtaining a plurality of experiment samples, each experiment sample comprising a dose level and an observation value of at least one biological index; determining a trend residual of each biological index of each experiment sample; determining a biological index evaluation weight sequence composed of quality evaluation weights of all biological indexes; determining a composite distance between any two experiment samples; determining a quality evaluation index of each experiment sample; and evaluating the quality of preclinical drug experiments based on the quality evaluation index of each experiment sample. By constructing a dose effect trend function, combining the trend residual and the weight, establishing a composite distance model fusing pharmacological trends, realizing adaptive selection of neighborhood parameters, improving the accuracy and stability of anomaly detection, overcoming the limitations of traditional machine learning LOF algorithm, and enhancing the evaluation reliability.

The invention provides a deep collaborative multi-task feature learning method based on mutual information regularization. Aiming at the technical problems of low feature sharing efficiency, serious inter-task interference, poor generalization ability and the like in the existing multi-task learning, the invention constructs a deep learning framework based on mutual information constraint, and designs a mutual information calculation module of variational inference and a KL divergence regularization optimization mechanism. And cross-task feature efficient sharing is realized through a multi-layer encoder structure and a feature adaptive selection mechanism. According to the scheme, mutual information calculation is optimized by adopting a batch estimation strategy and a parallel calculation mechanism, and a dynamic task weight distribution and cross-task knowledge migration mechanism is introduced to enhance the generalization ability of the model. Experiments show that compared with the prior art, the classification accuracy, convergence speed, calculation efficiency and the like of the method are improved by 31.2%, 65.3% and 44.8% respectively, and the method has higher cross-domain migration ability and can be widely applied to multi-task learning scenes such as computer vision and natural language processing.

The invention relates to the technical field of ship transportation, and discloses a self-adaptive unloading control system of a tank cleaning robot for a bulk cargo ship. The invention aims to solve the technical problems that the existing bulk cargo ship cabin cleaning operation seriously depends on manual operation, the automation degree is low, the operation efficiency is low, and the cabin cleaning cleanliness is difficult to guarantee. According to the invention, a set of closed-loop control architecture is constructed, multi-angle vision and mechanical sensing information are deeply fused, and adaptive selection of operation strategies of goods with different physical characteristics and dynamic cooperation among multiple units of the robot are realized. The core of the method is that a coarse-fine separation parallel collaborative operation mode is adopted, and a dynamic task assignment mechanism based on comprehensive cost benefit evaluation is introduced, so that the maximization of the operation efficiency of the whole ship is realized. By constructing a set of complete and full-automatic cabin cleaning operation and a good process, the automation level and the operation efficiency of the cabin cleaning operation of the bulk cargo ship are remarkably improved.

This invention discloses a method and system for assessing tumor content in liquid biopsies. The method performs quality control on high-throughput sequencing data from body fluid samples, filtering germline variations, clonal hematopoietic-related variations, and sequencing errors to obtain copy number variation (CNV) information and somatic mutation characteristic information. A weighted mean squared error loss function is constructed based on the CNV information, and the tumor content is solved using a constrained gradient descent method. When the CNV signal does not meet preset judgment conditions, the system switches to the mutation characteristic module, constructing a negative log-likelihood function based on a binomial distribution based on the principal clonal cluster, and solving for the tumor content through iterative optimization. In extreme cases, a backoff mechanism is triggered. This invention combines the stability of CNV at high tumor content with the sensitivity of mutational VAF at low content, achieving adaptive assessment from low to high tumor content.

The invention relates to artificial intelligence, and discloses an agent collaborative navigation method based on communication suggestions, and the method comprises the following steps: each agent obtains current local observation, and receives communication input at the last moment-1; generating a communication graph of the communication connection state between the intelligent agents based on the neighborhood relation at the previous moment and the current local observation and opinion set; adding with motion to generate a joint observation feature; inputting a strategy network, and outputting a current action and an action opinion; and the intelligent agent executes actions to push system state evolution, receives rewards fed back by the environment to obtain new local observation and communication input, and enters a next decision cycle. According to the method, the communication suggestions and the action suggestions are introduced, so that the self-adaptive selection of the communication relationship between the intelligent agents and the dynamic coordination of the behavior decision are realized, and the overall coordination and safety of a group are improved while the information transmission efficiency is ensured.

The invention relates to the technical field of financial transaction data visualization, in particular to a transaction data analysis method and system based on a regular polygon block thermodynamic diagram. By constructing a unified time-price two-dimensional data matrix, adaptive selection and rendering of various regular polygons such as rectangles, hexagons, parallelograms, triangles, linear views and the like are supported, and on this basis, information of untransacted buyers and sellers is superposed. According to the transaction data analysis method and system based on the regular polygon block thermodynamic diagram provided by the invention, multi-shape rendering is supported through the unified matrix, repeated calculation of data for different shapes is avoided, and the calculation complexity is reduced; the optimal shape is automatically selected according to the data density, the visual definition is improved, the problems of grid shape solidification, insufficient transaction granularity distinguishing, low multi-view linkage efficiency, missing of non-transaction data and the like in the prior art are solved, and the adaptability, precision and comprehensiveness of transaction data visualization are remarkably improved.

The invention discloses a student comprehensive physical fitness interesting exercise test method based on motion quality grading and self-adaptive combination, which comprises the following steps: acquiring basic physical fitness data of a student through face recognition, and calculating initial difficulty scores and increment scores of five test items of squatting, boxing, jumping, high leg raising and sideways bending by adopting a threshold grading algorithm; the first 5N groups are trained in a fixed sequence, the coordinates of the key points of the actions of the students are obtained through the visual capture technology, and the quality score and the comprehensive score of each action are calculated; executing adaptive selection based on the quality accumulated score, and dynamically selecting a training item by adopting a minimum score selection strategy and a minimum value optimization principle of a ratio of an item selection score to a residual increment score; according to the method, the test content can be adaptively adjusted according to the individual difference of the students, the scientificity and interestingness of the test are improved, and the method is suitable for school physical fitness test and physical education.

The application provides a large model-based intelligent question analysis method, device and equipment and a storage medium. The application performs step-by-step semantic analysis on a question through a thinking chain reasoning framework, simulates a complete thinking process of a human being from understanding a problem to disassembling a structure, breaks through the limitation of only making sequence translation in the prior art, and significantly improves the accuracy of complex SQL generation. A multi-layer progressive retrieval mechanism is adopted, thinking chain information is used to drive focusing from a business scene to field association layer by layer, the retrieval result is ensured to be highly matched with the query requirement, a dynamically generated personalized SQL sub-template is used, and the adaptation capability for an unseen question is greatly enhanced. A multi-dimensional feature vector integrating semantic similarity, structural complexity and business matching degree is constructed, adaptive selection of a decoding strategy is realized, an optimal generation path can be used in different complexity scenes, and efficiency and quality are taken into account.

The invention provides a product residual life prediction method considering multi-parameter time sequence correlation modeling, and the method comprises the steps: firstly, carrying out the standardization processing of multi-performance parameter degradation data of a mechanical electronic product, and carrying out the function type principal component analysis based on the adaptive selection of a mixed basis function, screening out key performance parameters representing the degradation state of the product; secondly, aiming at the screened key performance parameters, a parameter estimation method based on a nonlinear Wiener process of Markov Monte Carlo is adopted, and a degradation model and a life prediction model of each performance parameter are established; and finally, on the basis of edge distribution of each parameter, modeling multi-parameter dynamic time sequence correlation through a Copula function, and forming a joint residual life prediction model. The method has an adaptive degradation feature extraction capability, and can effectively fuse multi-parameter time sequence information so as to predict the residual life of the product more comprehensively and more accurately.