232 results about "Model migration" patented technology

The invention relates to the field of equipment state monitoring in industrial Internet of Things, and discloses an equipment state monitoring, analysis and evaluation method and system based on big data, and the method comprises the steps: carrying out the adaptation of a multi-source heterogeneous data protocol, and carrying out the cleaning of a dynamic mask, and generating a standardized data stream; constructing a dynamic hypergraph of an embedded constraint equation based on physical topology; combining incremental tensor decomposition with manifold constraint to update a core tensor; abnormal association is positioned based on singular value distribution and a hyperedge propagation algorithm; cross-equipment model migration is realized through topological optimal transmission and knowledge distillation, and a target equipment evaluation model is generated; the system comprises a data preprocessing module, a hypergraph modeling module, a tensor analysis module, a state evaluation module, a transfer learning module and a dynamic tuning module. According to the method, through multi-source data dynamic cleaning, physical constraint hypergraph modeling, incremental tensor decomposition and manifold constraint, and in combination with an abnormal positioning closed loop and cross-equipment topology migration, equipment state monitoring and rapid model adaptation are realized.

The invention discloses an intelligent glasses audio processing method and system. The method comprises the following steps: data preparation: collecting source domain data and target domain data; pre-training the model; converting the audio signal into sequence data; an RNN model is constructed and trained; model migration: migrating the pre-trained RNN model to target domain data, and processing noise in the target domain data by using parameters and features of the pre-trained model; intelligent analysis and processing, wherein noise in target domain data is analyzed and processed by using the migrated RNN model; real-time processing and feedback are carried out, audio signals captured by a microphone of the intelligent glasses are analyzed in real time, and a noise reduction strategy is dynamically adjusted to eliminate background noise; and optimizing noise reduction parameters and strategies according to the feedback. By collecting the source domain data and performing model pre-training, the RNN model can identify and process various known noise types and features and quickly adapt to different noise environments, so that the noise reduction effect is improved, and a user can obtain the optimal hearing experience in different noise environments.

The invention relates to the field of information automation management, and particularly discloses a water electrolysis efficiency dynamic scheduling method and system based on reinforcement learning. According to the method, a state space is constructed through multi-parameter dynamic perception, an improved depth deterministic strategy gradient (DDPG) algorithm is used for learning, a strategy is optimized by means of a self-adaptive exploration strategy and a multi-time scale learning mechanism, model migration and rapid starting are achieved, real-time feedback and closed-loop control are constructed, the water electrolysis efficiency is improved, and energy consumption is reduced. The method comprises a reward function and an exploration rate adjustment formula, so that the strategy optimization direction is accurate, and the exploration process is reasonable. The system comprises a data acquisition module, a processing module, a reinforcement learning module, a control execution module and a feedback monitoring module, and all the modules work cooperatively to improve the system performance.

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

The invention discloses a spatio-temporal cross-scale dimension reduction characterization method, system and device for a multi-modal physiological signal. The method comprises the steps of multi-modal signal time sequence alignment, sliding variable scale segmentation, single-modal spatio-temporal feature pre-training, cross-modal coupling alignment and model migration application. In the single-mode spatial-temporal feature pre-training, single-mode signal fragments under different time processes or spatial arrangements are reconstructed through a codec, and multi-level spatial-temporal modeling and analysis of a signal source are realized; in cross-modal coupling alignment, spatial-temporal characteristics of physiological signals of other modals are reconstructed through a codec, and a coupling relationship among different modals of a physiological system is learned; the model after cross-modal coupling alignment can be migrated to a specific task for fine tuning, and the comprehensive characterization capability of physiological signals in a few modals, specified space or limited time collected by a backward device on a physiological system is enhanced. According to the method, through bottom-layer dimension reduction and migration, effective characterization of the whole complex physiological state by the easily obtained signals is achieved.

The invention discloses a bearing variable working condition fault diagnosis method fusing model migration and feature migration learning, and the method comprises the steps: processing bearing vibration signals of a source domain and a target domain through wavelet transform, and extracting a time-frequency diagram; expanding the two-dimensional time-frequency graph data set by using DCGAN, and balancing the number of the two-dimensional time-frequency graph data set; then, model parameter migration is adopted, AlexNet network parameters pre-trained in a source domain are migrated, a migrated AlexNet network is constructed, and depth features are extracted; then, a domain adaptation method based on improved migration joint matching is provided, multiple strategies are fused, and a low-dimensional feature space with small distribution difference and good discrimination performance is obtained; and finally, on the basis of a labeled source domain feature data training model after domain adaptation, realizing identification and classification of unlabeled target domain feature data. The method is ideal in diagnosis performance and high in accuracy under variable working conditions and data imbalance, domain data distribution difference can be reduced by improving the migration joint matching method, and feature discrimination performance and fault diagnosis accuracy are improved.

The invention discloses an implicit gradient optimization-based large language model prison break attack resisting method, which is characterized in that continuous gradient optimization on resistance tokens is realized through a Gumbel-Softmax technology, calculation cost is reduced in combination with a two-stage proxy model screening mechanism, and semantic concealment is kept by adopting a dynamic regularization strategy. The system comprises a gradient optimization module, an agent screening module and a migration enhancement module, and can effectively improve the attack success rate and the cross-model migration capability of resistance prompts. The technical problems that a traditional attack resisting method is low in efficiency and poor in concealment are solved, the attack cost is reduced while the large model attack success rate is increased, the API calling frequency is effectively reduced, and the method is suitable for the field of large language model security testing.

The invention discloses a defect detection method and a model migration method. The model migration method comprises the following steps: obtaining a source domain image and a corresponding label, and obtaining a baseline model from a source domain; obtaining a target domain image, wherein the target domain image is completely labeled or partially labeled or completely unlabeled; inputting the annotation of the source domain image into a de-noising device to obtain a de-noised annotation; inputting the initial baseline model into an unlabeled target domain image to obtain a second target domain prediction result, performing data expansion on the unlabeled target domain image to obtain a second target domain expansion image, and taking the second target domain prediction result as a pseudo label of the second target domain expansion image; and training a baseline model by using the source domain image and the corresponding de-noised annotation, the annotated target domain image and the corresponding annotation and / or the second target domain expansion image and the corresponding pseudo annotation to obtain a migration model. According to the method, training can be completed with a small number of labeled target domain images, and the problem that the model migration performance is poor under the condition that the target domain images are insufficient is solved.

The invention discloses a medical multi-mode base model training method, system and device and a storage medium, and relates to the technical field of model training, and the method comprises the steps: constructing a medical multi-mode base model; each data processing branch is trained; and after training of each data processing branch is completed, training the multi-task decoder by using output data of each data processing branch. On one hand, based on the model migration capability, a small amount of multi-modal data can be utilized to better train a multi-task decoder, and the obtained trained medical multi-modal base model can support input of the multi-modal data; and on the other hand, by performing decoupling training on the data processing branch and the multi-task decoder, the interdependence between the data processing branch and the multi-task decoder can be reduced, so that the stability, generalization ability and prediction accuracy of the trained medical multi-modal base model are improved.

The invention relates to the technical field of model transfer learning, and discloses a model adaptive optimization method based on transfer learning. The method comprises the steps that source domain model structure parameters and target domain task initial data distribution are obtained, the feature mapping relation of all levels of a source domain model is extracted, and a cross-domain feature migration reference topological framework is generated; dividing a migratable feature layer and a to-be-reconstructed feature layer according to a target domain data distribution difference, and dynamically adjusting a migration priority in combination with a sample distribution density; freezing and unfreezing the transferable feature layer layer by layer based on the priority, synchronously constructing a local feature reconstructor, and optimizing domain offset through iterative feature alignment; collecting a feature reconstruction error and a migration feature retention degree in each iteration, and calculating a dynamic balance coefficient to adjust a freezing proportion and reconstruction intensity; and fusing the two types of features through a global model integrator, and generating mixed feature representation to drive end-to-end training of a target domain task.

The invention discloses an ultra-short-term output scene generation method and system for a newly-built new energy station, and the method comprises the steps: selecting a target station and a source station according to the geographic position of the target station, and further constructing a target domain data set and a source domain data set; training an improved denoising diffusion probability model on the source domain data set to obtain a source domain model; comprehensively selecting a source domain optimal model from the source domain models according to the model evaluation indexes; in combination with a model migration mechanism and a target domain data set, performing model parameter fine tuning on the source domain optimal model to obtain multiple groups of target domain models; and comprehensively selecting a target domain optimal model from the obtained multiple groups of target domain models in combination with a model evaluation index so as to complete the generation of the ultra-short-term output scene of the new energy station. According to the method, high-quality generation of a newly-built new energy station output scene can be completed under the small sample condition, planning and scheduling decisions of a subsequent novel power system can be effectively supported, and the new energy consumption capability of the system is further improved.

The invention discloses a large model migration optimization system driven by cross-modal feature alignment, particularly relates to the field of simulated confrontation, and is used for solving the problem of high-dimensional data storage and calculation efficiency of multi-source information processing in an adversarial environment. A closed loop is formed with multi-scale time agglomeration, context entropy pruning and combined change trend dynamic control, so that cross-modal high-dimensional information is balanced between compression storage and time sequence integrity, the weight in the mapping stage is adaptively adjusted according to the situation, and the residual error is corrected at any time; the long-term memory is continuously consolidated through incremental drilling and drift monitoring, and disaster forgetting is avoided; reasoning and scheduling real-time write-back performance indexes, and driving self-evolution of the structure; according to the method, the dual burden of video memory and delay caused by dimension explosion is inhibited, the migration alignment precision is kept, the strategy sequence is ensured to keep high-consistency and high-confidence output in the complex situation, meanwhile, the computing power demand is linearly reduced, and the resource utilization rate is remarkably improved.

The invention provides a domain self-adaption-based generative attack and defense confrontation method. Directional interference on a black-box intelligent system is realized by constructing a generative confrontation network framework. The method takes a domain self-adaption theory in transfer learning as a core, builds a dynamic mapping relation between a white box model group and a black box target system, designs a dual-channel generative adversarial network, jointly supervises sample authenticity and cross-model transfer consistency through cross entropy loss and KL divergence constraint, builds a dynamic boundary feedback system, and achieves the dynamic boundary feedback. And a disturbance generation strategy of the generator is adjusted according to the real-time attack effect, and a self-adaptive attack and defense game closed loop is formed. Compared with a traditional black box attack method, the method has the advantages that migration robustness of an adversarial sample in a model heterogeneous scene is remarkably improved, stable attack efficiency can still be maintained when a defense system has protection mechanisms such as adversarial training and input reconstruction, and the method can be applied to the fields such as penetration testing of an intelligent security system and the like. And a technical support is provided for vulnerability evaluation of the intelligent system in a complex environment.

The invention discloses a meta reinforcement learning model migration training method and device based on topological transformation, and relates to the technical field of machine learning. Obtaining power data of a plurality of regions, and constructing a region data set; performing data feature comparison on the region data set and a model data set of a preset training model, and determining an updating strategy of the preset training model according to data features; and migrating data in the regional data set into the model data set, and executing an updating operation according to the updating strategy to obtain a target model. According to the method, the regional data set is constructed by collecting the multi-regional power data, the regional data set is subjected to data feature comparison with the model data set of the preset training model to determine the model updating strategy, and the migration training method is effectively applied to model updating. According to the method, the training process is accelerated, the model performance is improved, and the computing resource demand is reduced, so that an efficient and economical model updating mode is realized, and the adaptability and accuracy of the model in power fault positioning are enhanced.

The invention relates to the technical field of information security identification, and particularly discloses a multi-modal data artificial intelligence security classification identification method in combination with transfer learning, which comprises the following steps of: in a process of constructing a multi-modal behavior data set, acquiring file access frequency and network connection frequency in real time; fluctuation characteristic values are extracted through fast Fourier transform and a K-means clustering algorithm and used for evaluating internal threat and external attack risks faced by the system, the two types of characteristics are fused into a comprehensive characteristic vector, the comprehensive characteristic vector serves as input to construct a transfer learning model based on a support vector machine, intelligent recognition of the safety state of a target object is achieved, and the safety of the target object is improved. In order to improve the recognition accuracy, a proportional-integral-differential control mechanism is introduced into the system, the sampling period and the detection granularity are dynamically adjusted according to recognition errors, a closed-loop feedback optimization structure is formed, effective integration of multi-source heterogeneous behavior data, cross-scene migration of the model and online self-adaptive adjustment are achieved, and the recognition accuracy is improved. And the intelligent level of the security identification system is improved.

The invention relates to the technical field of aviation large model optimization, provides a multi-model migration optimization method and system for an aviation large model, and aims to solve the problem that the stability and consistency of knowledge migration are difficult to maintain in a complex aerodynamic environment due to dependence on a static feature space in the prior art. The method comprises the following steps: constructing a flight mechanics feature library based on flight mechanics parameters and control response data of a source aircraft model and a target aircraft model; executing coupling simulation to generate a compensation parameter set; establishing a model knowledge migration channel, reconstructing the knowledge migration channel according to the compensation parameter set, and generating a target migration channel and a migration weight regulation and control function; and adjusting the target migration channel and monitoring the distribution offset of the aerodynamic characteristic knowledge component, updating the aerodynamic difference characteristics, and iteratively executing the operation from coupling simulation to adjustment of the transmission state register until the distribution offset enters a steady-state operation interval. According to the method, high-precision knowledge migration from a wide-body aircraft to a narrow-body aircraft is realized, and the aviation airworthiness safety standard is met.

The invention relates to a data set distribution difference-oriented electronic medical record data representation learning method and system, and the method comprises the steps: pre-training a source domain teacher model through a source domain data set of an existing disease, and extracting the health state representation information of a source domain disease patient; training a domain invariant feature extractor as a transition model, modeling general features through an adversarial training strategy, and establishing an independent extraction channel for private features to realize feature alignment among different domains; and migrating parameters of the transition model to a target domain emerging disease prediction model, and performing fine tuning in combination with target domain data to realize accurate prediction of emerging diseases. Through domain invariant feature extraction and model migration optimization, the generalization ability and prediction precision of the model on different data sets are significantly improved, clinical results of patients can be accurately predicted, accurate support is provided for medical decision, and a reliable foundation is laid for the cross-domain prediction problem in medical data analysis.

The invention discloses a lottery hypothesis-based structured pruning method and system, and the method comprises the steps: carrying out the random initialization and iteration amplitude pruning of the weight of a target neural network model, generating a structured sparse sub-network, and carrying out the weight playback technology, thereby obtaining a structured sparse sub-network; parameters corresponding to the sparse sub-network structure in the randomly initialized dense neural network are migrated to the sparse sub-network, a small amount of training is carried out to obtain a convergent sparse neural network, the trained model is migrated to a server equipped with an Ampere architecture GPU, reasoning acceleration is carried out, and efficient image recognition task reasoning is realized; the system comprises a model initialization module, a model trimming module, a parameter migration module, a model training module, a model migration module and a reasoning module. According to the method, the architecture characteristics of the Ampere architecture GPU equipment are combined, neural network pruning is combined with lottery hypotheses, efficient pruning is performed on the neural network, and finally efficient reasoning is realized.

The invention belongs to the technical field of inference model migration and optimization, and relates to an AIOT board card-oriented inference model seamless migration and adaptive optimization method. A hardware performance analysis tool is used for comprehensively evaluating the computing power of AIOT equipment to obtain hardware performance indexes, an original reasoning model is layered, the analysis tool is used for evaluating performance data of each layer, a conversion strategy of the model is optimized on the basis of the hardware performance indexes and model performance data, and the conversion strategy of the AIOT equipment is optimized. Hardware limitation and model performance requirements can be considered in the conversion process of the model. Converting the original reasoning model based on the determined conversion strategy, and deploying the original reasoning model to the AIOT board card, thereby achieving the purpose of seamless migration; and secondly, a deep learning model and real-time load monitoring are adopted to dynamically adjust the reasoning precision, the batch size and the execution path, so that the reasoning model can be adaptively optimized according to the actual operation condition, and the efficiency and the performance of the model under limited resources are further improved.

The invention belongs to the technical field of mobile traffic prediction, and particularly relates to a cross-domain migration space-time integrated network algorithm, which comprises the following steps: S1, constructing an integrated learning model, carrying out initial training on a basic learner by utilizing integrated learning, and capturing space-time characteristics in combination with a space-time network, the method is used for overcoming the defects of instability and overfitting in a multi-dimensional space-time cross-domain environment, and an integrated learning model comprises a traffic data set structure, spectral clustering, long-short term memory (LSTM) and a convolutional neural network (CNN). The invention provides a new migration learning algorithm, the algorithm trains a mobile traffic prediction model from source domain data with rich data, then the model is migrated to a mobile edge network environment with sparse data so as to improve the prediction precision of the model, and meanwhile, a new migration mapping framework is provided so as to improve the prediction precision of the model. That is, a reflection function is learned from empirical data and optimized to predict when or how to migrate to achieve optimal predictive performance.

The invention relates to the technical field of sensitive target recognition and adversarial attack in remote sensing images, in particular to an adversarial patch generation method of a remote sensing image detection model based on particle swarm optimization, which comprises the following steps: firstly, generating an initial aggressive patch by adopting a particle swarm optimization algorithm, and introducing a cosine annealing mechanism to dynamically adjust inertia weight and improve search diversity; then, an AdamW optimizer is used for local refined updating, an L2 regular term is added to prevent overfitting, the patch generalization ability is enhanced, the whole process is combined with a multi-stage dynamic weight mechanism and a time sequence sensing strategy, attack loss, non-printability loss and smoothness loss are subjected to self-adaptive weighting, and the patch generalization ability is improved. And the balance of attack intensity, concealment and physical implementability is realized. The adversarial patch generated by the method can significantly reduce the performance of a target detection model in digital simulation, and has good cross-model migration ability and physical deployment potential.

The invention discloses a model mobility measurement method for a remote sensing image cross-domain migration scene, and the method comprises the steps: constructing a feature space of a target domain for a neural network model cross-scene migration task for remote sensing earth surface element classification; extracting the features of each category of the target domain by using the feature space of the target domain, and inputting the features into a pre-established feature calculation model to obtain the feature separability of all dimensions in the feature space of the target domain; applying a weight to the feature separability of the corresponding dimension to obtain the feature separability of the target domain after weighted correction; and a mobility measurement index is calculated based on the feature separability of the target domain after weighted correction, and comparative analysis is performed in combination with the precision of the source domain model after the source domain model is migrated to the target domain to obtain the mobility of the source domain model to the target domain. According to the method, the weight is applied to the separability of each dimension in the feature space, the importance of the features of different dimensions is considered, and the stable correlation between the mobility measurement and the precision after migration is improved.

Briefly, embodiments of a system, method, and article for receiving a user input requesting that an initial application container compatible with a first programming model be modified to generate an adjusted application container compatible with a second programming model. The second programming model is different from the first programming model. A first set of development artifacts may be determined for the initial application container. A first subset of the first set of development artifacts may be automatically adjusted to enable compatibility with the second programming model. automatically generate One or more behavior definitions associated with the automatically adjusted first subset may be automatically generated. The adjusted application container comprising the initial application container and the one or more behavior definitions may be generated.

The invention discloses a comprehensive traffic low-altitude global intelligent sensing method based on autonomous evolution, and belongs to the crossing field of intelligent traffic and computer vision. The method comprises the following steps: collecting multi-source sensing data and extracting a structured causal variable; a dynamic causal graph is constructed, and attribution analysis of perception errors is realized through intervention learning; realizing cross-scene model migration and few-sample self-adaption based on the knowledge graph; carrying out adaptive reasoning by adopting an environment-aware dynamic graph neural network; and realizing a causal-driven autonomous evolution closed loop based on multi-source feedback. According to the method, the crossing of the perception system from'perception-optimization 'to'understanding-evolution' in a low-altitude complex traffic environment is realized, the perception precision, the scene adaptability and the system interpretability are remarkably improved, and the manual operation and maintenance cost is reduced.

The invention provides an automatic operation and maintenance method and system suitable for a closed system. The technical problem that a traditional operation and maintenance scheme is difficult to apply due to network isolation and sample scarcity in closed environments such as finance and energy is solved. The method comprises the following steps: generating a unique and traceable migration identifier for all operation and maintenance data, models and reasoning results through a data and migration management module; quantitatively calculating a transferability score between the source domain and the target domain through a transferability evaluation module; an optimal model migration strategy is dynamically selected according to the mobility score, and efficient self-adaption of the model is achieved; closed-loop operation and maintenance are executed through a multi-agent cooperation system comprising detection, diagnosis and repair agents, and self-learning and self-optimization of the system are realized through small sample active learning and a knowledge base evolution mechanism. According to the method, rapid construction, continuous evolution and whole-process traceability of the artificial intelligence operation and maintenance capability in the closed system are realized, and the method has remarkable innovativeness and industrial application value.

The invention provides an axial plunger pump fault diagnosis method suitable for marking sample scarcity conditions, and belongs to the technical field of fault diagnosis, aiming at performance degradation caused by marking sample scarcity in axial plunger pump fault diagnosis of a deep learning model. According to the method, the dynamic pressure simulation model is constructed under the condition that the marked samples are scarce, and unmarked samples can be supplemented by generating high-quality simulation samples; a model migration framework and a semi-supervised multi-adversarial pre-adaptation strategy are constructed, and the generalization ability of the model in a target domain task is improved. Experimental results show that high-quality simulation samples are generated, and the plunger pump fault sample recognition rate under the condition that marked samples are scarce is increased.

The present application relates to a model migration method in the field of artificial intelligence, including: obtaining sample data of a target task, wherein the sample data includes multiple image samples; evaluating N pre-trained models based on the sample data to obtain N evaluation values, which are used to characterize the degree of fit between the pre-trained model and the target task, where N≥2; determining K pre-trained models from the N pre-trained models based on the N evaluation values, wherein the K pre-trained models are models corresponding to the first K evaluation values after sorting the N evaluation values from large to small, where 1≤K≤N; processing the sample data based on the K pre-trained models to obtain a target model for processing the target task, wherein the target model includes K pre-trained models. In this way, a pre-trained model suitable for the current task can be quickly found from a model library composed of a large number of pre-trained models, and the capabilities of multiple optimal pre-trained models can be integrated to jointly solve the current task, effectively improving the distribution-outside generalization capability of the pre-trained model.

The invention discloses a model migration method and related equipment, and aims to solve the problem that in the prior art, when a deep learning model is quantized to enable a GPU (Graphic Processing Unit) to have resources to execute an inference task, the prediction capability and accuracy of the model are reduced, so that the model is not applicable to an application scene with high accuracy and precision requirements. The method comprises the following steps: determining a calling frequency of calling an algorithm model in a graphic processing unit to execute an inference task; if the calling frequency is smaller than a preset calling frequency threshold value, migrating hierarchical structure information of the idle algorithm model with the calling frequency smaller than the preset calling frequency threshold value and model parameter information related to the reasoning task to a CPU and a memory, and enabling a thread of the idle algorithm model to enter a dormant state; when a calling request of the idle algorithm model is received, the thread in the dormant state is awakened, and hierarchical structure information and model parameter information of the idle algorithm model are migrated to the graphic processing unit from the CPU and the memory.

The invention relates to the technical field of spectral data processing, and discloses a Raman spectrum model migration transfer method based on a standard substance, which comprises the following steps: S1, acquiring Raman spectrums under different devices as training data; s2, carrying out beam alignment; s3, performing point-by-point soft weighting on the spectrum after beam alignment by adopting a multi-scale parallel gating convolution strategy to obtain an enhanced feature map; s4, windowing segmentation is carried out based on the enhanced feature map, and intra-segment point-by-point gain is carried out; s5, superposing the gained segments, and carrying out normalization by using window energy to obtain local equalization output; s6, calculating a loss function according to local equalization, and optimizing model parameters; aiming at the common bottlenecks of baseline noise inconsistency, peak position offset, noise interference and the like in the existing Raman spectrum cross-device analysis process, the method is not limited to traditional piecewise linear migration, and self-adaptive spectrum migration is realized through the convolutional neural network, so that the spectrum consistency and comparability between different devices are improved.

The invention discloses an intelligent customer service cross-project experience migration method in combination with migration learning. The method comprises the following steps: S1, generating a source project feature set; s2, generating a source project optimization model containing a shared feature layer; s3, performing global optimization on shared feature layer parameters in the source project optimization model through Bayesian optimization; s4, generating an optimized target project feature set; s5, constructing a target project model; s6, performing fine tuning training on the target project model; s7, performing semantic consistency evaluation, response accuracy evaluation and business logic matching degree evaluation on the fine-tuned target project model on the target project data set; and S8, deploying the target project model passing the performance verification to a target project intelligent customer service system. According to the method, efficient deployment of the cross-project intelligent customer service system is realized, the migration efficiency and the service quality of the model are remarkably improved, and the method can be widely applied to intelligent customer service scenes of multiple industries.