80 results about "Sequence database" patented technology

The invention provides a circuit board production yield root cause tracing method, which comprises the following steps of: acquiring process parameters, equipment states, environment variables and quality detection results of a whole production process, and constructing a multi-dimensional time sequence database; extracting a typical manufacturing process modeling unit through a sliding time window and dynamic time warping; establishing a cross-process dynamic causal relationship graph in combination with nonlinear Granger causal test, a structural equation model and a dynamic Bayesian network; an intervention and anti-factual reasoning method is applied, the causal effect and path stability under parameter disturbance of each process are evaluated, and the influence of a key causal path is quantified; according to the method, the accuracy of defect rate root cause positioning can be improved, and powerful support is provided for circuit board production process optimization and quality improvement.

The invention relates to the technical field of pile foundation construction, in particular to a pile foundation pore-forming quality intelligent management and control method and system based on drilling machine working parameters, and the method comprises the following steps: constructing a data acquisition array, and obtaining key working parameters of a drilling machine and pile foundation pore-forming quality index data based on the data acquisition array; establishing a standard time sequence database; building a dynamic correlation model based on the standard time sequence database, and building a quality judgment standard of the pile foundation pore-forming based on the dynamic correlation model; dynamically predicting the pile foundation formed hole according to the dynamic correlation model to obtain the quality state of the pile foundation formed hole, and obtaining a quality deviation result of the pile foundation formed hole in combination with the quality judgment standard; and a regulation and control strategy of the drilling machine is constructed based on the quality deviation result, the key working parameters are adjusted according to the regulation and control strategy, and intelligent control over the quality of pile foundation hole forming is achieved. Intelligent control over the pile foundation hole forming quality is achieved through driving of the working parameters of the drilling machine.

The invention discloses a real-time data subscription method and device based on a time sequence database and a medium, belongs to the technical field of databases, and aims to solve the technical problems that the real-time performance is poor, the resource consumption is large and the time sequence data required by the current time sequence database in real time acquisition in mass data storage and processing is poor. According to the technical scheme, the method comprises the steps that a data subscription scheduling task is constructed through a predefined SQL statement, in combination with an efficient data change capture module based on a Raft log mechanism, data subscribed by a user is obtained in real time, and the data subscribed by the user is filtered according to a customization rule; then the screened data is automatically transmitted to a preset target sink component in a specified format, so that the whole data processing flow has the characteristics of real-time performance, accuracy and high availability; the method specifically comprises the following steps: constructing a data subscription scheduling task; executing the data subscription scheduling task; and capturing data change based on the raft log.

The invention provides a guide rail dynamic error real-time compensation system based on multi-sensor fusion, and relates to the technical field of guide rail precision compensation, and the system comprises a calling module which is used for calling and traversing a temperature sequence database, and obtaining a target temperature sequence matched with a target working condition; the prediction module is configured to construct a finite element model of the guide rail, and is combined with the target temperature sequence to obtain a predicted straightness sequence of the guide rail so as to obtain a straightness error sequence; the judgment module is configured to traverse the straightness error sequence so as to take a point location corresponding to the predicted straightness corresponding to the straightness error greater than the preset precision as a to-be-compensated point location; the compensation module is configured to obtain a corresponding thermal deformation compensation amount according to the straightness error corresponding to each to-be-compensated point position, and compensate the straightness error of each to-be-compensated point position by the thermal deformation compensation amount in the operation process of the guide rail under the target working condition through the distributed piezoelectric ceramic array, so as to compensate the straightness error of each to-be-compensated point position. And a more reliable precision guarantee is provided for high-precision machining equipment.

The invention relates to the technical field of machine vision and industrial quality inspection, in particular to a polishing scheme autonomous optimization method and system for surface defect detection and a medium, and the method comprises the following steps: S1, based on a preset polishing parameter grading rule, collecting detection data to construct a defect-polishing sequence database; s2, segmenting and evaluating the image, and constructing a mapping relation model of each type of defect and a lighting parameter sequence based on a decision tree; s3, optimizing a lighting scheme by adopting an intelligent optimization algorithm based on the mapping relation model of each type of defects and the lighting parameter sequence; and S4, guiding an actual industrial defect detection task based on a lighting scheme corresponding to the light source parameter sequence obtained in the step S3. The problem that an existing lighting scheme depends on manual operation and is low in efficiency is solved, the optimal lighting scheme is automatically determined, the manpower burden is relieved, and the accuracy and stability of defect detection are improved.

The invention relates to the technical field of intelligent dimming, in particular to a lighting scheme autonomous optimization method and system based on a grey wolf optimization algorithm and a medium, and the method comprises the steps: S1, based on a preset lighting parameter grading rule, collecting detection data to construct a defect-lighting sequence database, and S2, constructing a preset standard mask pattern; s3, segmenting and evaluating the image, and constructing a mapping relation model of each type of defect and the lighting parameter sequence based on the optimal quality, S4, optimizing a lighting scheme by adopting an improved grey wolf optimization algorithm based on the mapping relation model of each type of defect and the lighting parameter sequence, and S5, carrying out the optimization of the lighting scheme. And guiding an actual industrial defect detection task based on a lighting scheme corresponding to the light source parameter sequence obtained in the step S4. The light source debugging efficiency and precision are greatly improved, the optimization speed and quality are improved, the optimization efficiency of light source parameters can be greatly improved, and intelligent production development can be further promoted.

The invention relates to a tumor neoantigen prediction method and a tumor neoantigen prediction system. The method comprises the following steps: acquiring a plurality of amino acid sequences in a gene of a person to be tested as a plurality of test peptides to be compared with a plurality of human protein sequences in a protein sequence database so as to find out a plurality of similar peptides matched with the human protein sequences; filtering similar peptides from the test peptides, and respectively inputting the filtered test peptides into a plurality of trained human leukocyte antigen models to obtain a plurality of sequencing results of the test peptides; calculating the weighted sum of the sequencing of each test peptide in the sequencing result as the score of the test peptide; and selecting at least one of the test peptides as a neoantigen suitable for the subject according to the fraction.

A dynamic cadastral data updating and management method based on a space-time sequence specifically comprises the following steps: S1, collecting dynamic cadastral data, preprocessing the dynamic cadastral data, and outputting a standardized space-time sequence database; s2, performing dynamic modeling on the standardized space-time sequence database, adjusting granularity, and outputting an optimized space-time model; s3, performing space-time sequence comparison by utilizing the optimized space-time model, automatically identifying data change, and outputting a cadastre change incremental data set; s4, constructing a relation chain according to the change incremental data set, writing change data into a current potential library, and archiving historical data into a difference file library; s5, performing space-time consistency verification and efficiency optimization on the current potential library and the difference file library, and outputting a quality inspection passing library; and S6, performing service packaging on the quality inspection passing library, and deploying the quality inspection passing library to a cloud service platform. The method dynamically adapts to change features of different regions, reduces storage cost and update time consumption, improves verification efficiency, and provides dynamically updated spatial basic data support for smart city construction.

The present disclosure relates to the technical field of big data processing, and discloses an account evaluation method and device, computer equipment, a storage medium and a computer program product. The method comprises: obtaining account resource interaction data of a to-be-evaluated account and a preset sequence mode, wherein the preset sequence mode is obtained based on a non-frequent feature sequence database, the non-frequent feature sequence database is determined according to a non-frequent item set obtained by mining a preset non-frequent mode tree, and the preset non-frequent mode tree and an initial feature database have an association relationship; and evaluating the account resource interaction data according to the preset sequence mode to obtain an evaluation result of the to-be-evaluated account. The present method can improve the accuracy of account evaluation and the security of resource interaction.

The application provides an adaptive resource dynamic scheduling method based on multi-modal behavior feature processing, and relates to the technical field of digital data processing.The application collects multi-source data such as network connection behavior, protocol interaction features, time sequence features and system resource states, constructs a visitor behavior time sequence database, and establishes a visitor normal behavior benchmark trajectory model.Based on a behavior evolution deviation index, an intention risk coupling index and a dynamic scheduling drift index, multi-level quantitative evaluation of the access behavior deviation degree, potential risk intention and scheduling safety state is realized.Through threshold value determination, corresponding strategies are triggered, abnormal access behavior is marked, analyzed and dynamically scheduled and controlled, and model parameters are adaptively optimized to realize continuous learning and dynamic evolution.The application can improve the accuracy of access behavior recognition and the refinement level of scheduling control, thereby improving the overall security protection capability and resource scheduling efficiency of the system.

A method for treating a mental disorder in a subject, comprising the step of normalizing the expression of both DRD1 and DRD5 belonging to the DRD1-like family of the subject to a preset standard level. The DRD1 / 5 level and expression can be increased or decreased bidirectionally by targeting the upstream promoter sequence of DRD1 / 5. The promoter activity of DRD1 / 5 can be attenuated by using antipsychotic drugs that directly target the DRD1 / 5 promoter. The antipsychotic drugs used are typical and / or atypical antipsychotic drugs such as haloperidol, risperidone, clozapine, and cariprazine. This treatment is also useful for subjects in whom the constitutive expression of the DRD1 / 5 gene is impaired. This method can be used to treat schizophrenia, autism, depression, mania, and bipolar disorder by normalizing the DRD1 / 5 expression level. The promoter gene sequencing of DRD1 and / or DRD5 can also be used for quantitative mental health screening by comparing with a standard sequence database.

A method for predicting DNA binding residues based on a three-segment sliding window is disclosed. Firstly, according to the input protein sequence information P for DNA binding residue prediction, the number of residues is L, the protein sequence database UniRef90 is searched using the PSI-BLAST program to generate a position-specific scoring matrix of size L×20, a solvent accessibility feature matrix of size L×3 is obtained using the SANN program, and a feature matrix of size L×20 is obtained by one-hot encoding. Then, the three feature matrices are concatenated into a super feature matrix of size L×43, that is, each protein residue corresponds to a 1×43 feature vector, and the similarity of the feature vectors between two residues is obtained by cosine similarity, that is, the residue similarity. Secondly, the feature view processing is performed on each residue in the super matrix by using the three-segment sliding window technology, and these feature views are expanded into feature vectors. Thirdly, the corresponding feature vectors and labels are generated by using the protein sequence of known DNA binding residues, and input into the support vector machine for training. Finally, the feature vector generated by the protein sequence to be predicted is input into the trained model to predict whether the residues in the protein sequence are DNA binding residues. The present invention has low calculation cost and high prediction accuracy.

The application relates to a method and system for mining and analyzing the correlation of clinical comorbidity of discharged patients. The method comprises the following steps: constructing an individual diagnosis and treatment narrative graph for each patient according to the discharge medical record text, generating a time sequence transaction sequence corresponding to each patient, and constructing a sequence database; based on the database, the original time sequence frequent pattern set is constructed by analyzing through an improved generalized sequence pattern algorithm; each pattern in the set is classified to obtain multiple classification clusters, and the original time sequence frequent pattern in each classification cluster is processed through multi-sequence alignment to construct a generalized clinical path graph, and the information in the generalized clinical path graph is extracted to generate a natural language abstract. The method improves the time sequence logic, knowledge abstraction degree and clinical interpretability of the clinical comorbidity correlation mining by constructing a diagnosis and treatment narrative graph, mining a time sequence frequent pattern, and constructing a generalized clinical path graph through clustering and multi-sequence alignment.

This invention relates to a method for predicting steel plate temperature during quenching heat treatment based on multi-task learning, belonging to the technical field of steel rolling heat treatment methods in the metallurgical industry. The technical solution of this invention is as follows: Key historical data of the quenching furnace heating process are collected through a standardized steel plate embedded coupler test system to construct a spatiotemporal sequence database for medium-thick plate quenching heat treatment; a spatiotemporal prediction model is constructed using a Transformer multi-task learning architecture, and a high-precision prediction model is finally trained; key historical data of the steel plate quenching furnace heating process are used as model input, and the predicted temperature of the core and tail sections of the medium-thick plate is output in real time. The beneficial effects of this invention are: it can effectively cope with the complex thermal field distribution and dynamic operating condition changes within the quenching furnace, achieving real-time and accurate prediction of the core temperature of medium-thick plates during quenching heat treatment, providing highly reliable and timely decision support and technical guidance for optimizing the quenching heat treatment process in hot-rolled steel plants.

The invention provides a construction method of a reference sequence for metagenome data compression, which comprises the following steps: constructing a basic reference sequence database according to a sample source of metagenome data; based on the basic reference sequence database, constructing an index of the basic reference sequence database; according to the index of the basic reference sequence database, a first read length sequence is compared with the basic reference sequence database, a comparison result is obtained, and the first read length sequence is a read length sequence of a part of samples randomly selected in metagenome data to be compressed; and according to the comparison result, determining the sequence abundance distribution of the first read length sequence, and constructing the reference sequence for metagenome data compression.

The application discloses a method for collecting active proteins of water leeches, a detection method, an analysis method and application thereof, and relates to the technical field of biological medicines. The application provides the water leech seedlings treated by hunger and the living host treated by emptying, so that the water leech seedlings adsorb and suck the host, and then the active proteins are injected into the body of the host; the sample of the host after being sucked is collected, and then is made into a freeze-dried powder after being frozen and dried, so that the active proteins of the water leeches are obtained; the proteins are extracted from the freeze-dried powder, and then are subjected to enzymolysis treatment to generate peptide segments; the liquid chromatography-mass spectrometry technology is used to detect the peptide segments; and the mass spectrometry data are subjected to database searching analysis based on the protein sequence database of the water leeches and the host, so that the active proteins from the water leeches are identified and the relative content is determined. The application utilizes the characteristic that the active proteins are injected into the body of the host in the sucking process of the water leeches, and realizes efficient enrichment of the target proteins. In combination with the proteomics technology, high-throughput and high-sensitivity protein identification is realized.

The invention discloses a method, a system and a device for identifying the age of a green-head diving duck based on multi-feature fusion. The method comprises the following steps: shooting a high-definition image of a target individual; obtaining one or more of feather state information of an ear feather region, a wing region, a hypochondrium region, a shoulder and back region, a tail region, a head and neck region, a chest and abdomen region, eye iris and beak color change information and chest and tail distance and beak length data information of the target individual in the high-definition image, and performing quantitative processing to obtain feature parameters corresponding to each feature; performing multi-dimensional fusion analysis on the characteristic parameters through an integrated machine learning model to realize accurate matching with corresponding week age / month age data in a feather moulting standard time sequence database of the green-head diving ducklings / young birds; and outputting a week age / month age identification result of the target individual. According to the method, a moulting standard time sequence database covering a young bird stage of the green-head diving ducks for the first time is constructed, high-precision and high-efficiency non-injury age identification is realized through combination of the database and a machine learning model, and the dependence on professionals is reduced.

This invention relates to the fields of molecular biology and forensic identification, specifically to a method for bird species identification based on multi-gene co-amplification and nanopore sequencing. The method involves extracting genomic DNA from avian biological samples; using the extracted DNA as a template, amplification is performed using four independent PCR primer pools that specifically target the avian mitochondrial cytochrome C oxidase subunit I gene, cytochrome B gene, 12S rRNA gene, and 16S rRNA gene; after purification of the amplification products, a nanopore sequencing library is constructed and sequenced; finally, bioinformatics analysis, including read length clustering, draft consensus sequence generation, and polishing, is performed on the raw sequencing data to obtain consensus sequences for each target gene. Species identification is then completed by comparing the sequence with a avian sequence database. This method effectively solves the problem that existing technologies cannot simultaneously meet the practical needs of large-scale, rapid, and high-success-rate identification of avian biological samples.

The application discloses a kind of based on TCR sequencing lung cancer early screening model construction method, first based on the TCR sequencing data of lung cancer patient and healthy person sample is constructed TCR sequence database enrichment, then the TCR sequencing of each sample is carried out and obtains the TCR characteristics of each sample, including V gene and J gene The proportion characteristics of gene, the statistical characteristics of TCR immune group, convergence sequence characteristics, TCR sequence amino acid proportion characteristics, different length TCR sequence proportion characteristics, different frequency TCR sequence proportion characteristics and TCR enrichment sequence characteristics;The TCR characteristics of each sample are screened using feature selection method, and a machine learning model is constructed by combining other information of the sample to obtain a lung cancer early screening model.Based on the lung cancer early screening model, a lung cancer early screening detection device is constructed, which can realize lung cancer early screening based on TCR sequencing information of the sample.

The invention discloses a multi-objective decision-oriented utility sequence pattern mining method and system and application, and the method comprises the following steps: SF1, scanning a DNA sequence database, obtaining a DNA sequence with only a single item as a T1 sequence, and constructing a quantity utility array as a QU array by the T1 sequence; sF2, performing an SQUD pruning strategy on each T1 sequence; calculating a sequence weighted utility value of each T1 sequence based on the QU array, marking the sequence weighted utility value as SWU, constructing a maximum utility number array as MUQA, and deleting the T1 sequence of which the SWU is smaller than the corresponding MUQA value; sF3, an array SPFA of the sequence Pareto front is initialized, and a T1 sequence of the MUQA after non-dominated sorting is recorded; generating a projection database QUPro for the T1 sequence in the SPFA based on a QU array; traversing each T1 sequence in the MUQA and generating an extension mode; and SF4, calculating a utility upper bound by using the PEU, and recursively expanding the current sequence by controlling any elements in the SPFA by the PEU value and quantity. According to the method, by introducing the utility upper bound and the pruning strategy, unnecessary calculation is reduced, and the efficiency of the algorithm is improved.

The invention provides a distributed base station energy storage resource virtual power plant abnormity monitoring method and system, and the method comprises the steps: firstly obtaining the power grid frequency modulation demands, real-time electricity prices, energy storage states, mechanical stresses and radio frequency signal data of all distributed base stations in a virtual power plant; then, aggregating the frequency modulation demand, the electricity price and the energy storage state data to construct an electricity price load response time sequence database, and meanwhile, fusing mechanical stress and a radio frequency signal to generate a stress electromagnetic coupling fingerprint; based on the time sequence database and the physical fingerprints, analyzing and generating charging and discharging competition behavior characteristics of each base station through a Nash equilibrium model; and finally, accurately positioning the abnormal energy storage node by comparing the deviation degree between the actual behavior characteristics and the expected equilibrium state of the model, and realizing credible evidence storage and tracing of the abnormal behavior by using a block chain smart contract. According to the invention, the accuracy and tracing reliability of virtual power plant energy storage abnormity monitoring are improved.

The invention provides a power distribution unit and a management method thereof, and relates to the technical field of power management. Static service priorities and reference electrical parameters of all branches are configured, and real-time electrical parameters are fused to calculate service dynamic priorities; collecting and preprocessing total output current output by a total power supply and current data of each branch according to a fixed sampling window, and constructing a historical load time sequence database; on the basis of historical data, a mixed prediction model combining a seasonal decomposition algorithm and a long-short term memory network is adopted to predict each branch and total output peak current in the future; performing system-level and branch-level pre-overrun judgment and arbitration based on the predicted peak current, and calculating a total current reduction quota and each branch over-standard current; and dynamically generating a regulation and control instruction sequence according to the service dynamic priority, the reduction quota and the overproof current. Accurate prediction and intelligent regulation and control of the power supply load can be realized, and the safety and efficiency of power supply distribution of the production line are improved.

The invention discloses a diffusion model-based protein binding RNA (Ribonucleic Acid) sequence generation method, which comprises the following steps of: in an off-line training stage, carrying out two-stage training on a constructed mixed generation type framework by utilizing a protein binding RNA sequence database, and coupling a variational auto-encoder unit and a conditional potential diffusion model unit by the mixed generation type framework; an online reasoning stage: inputting a to-be-generated protein sequence into the protein language model to extract condition embedding; the conditional potential diffusion model generates a new potential vector through a reverse denoising process taking protein embedding as a guide condition from Gaussian noise in a potential space of the variational auto-encoder, and finally, the potential vector is mapped into the generated protein binding RNA sequence by a decoder of the variational auto-encoder. The invention also discloses a system for realizing the method. According to the invention, a binding sequence can be generated for a protein target except for training data, and excellent biophysical characteristics and functionality are achieved.

Provided are an artificial intelligence-based early cancer diagnosis method and device using a method of inputting information on a cell-free DNA distribution of a tissue-specific regulatory region to an artificial intelligence model learned to early diagnose cancer and analyzing the information, and an information providing device and a storage medium.SOLUTION: A method for providing information for early cancer diagnosis based on artificial intelligence includes extracting a nucleic acid from a biological sample to obtain sequence information, arranging the obtained sequence information in a reference chromosome sequence database, selecting a nucleic acid fragment of a regulatory region based on the arranged sequence information, generating the selected nucleic acid fragment as image data, and inputting the generated image data to an artificial intelligence model learned to distinguish a normal image and a cancer image, analyzing the image data, and comparing the image data with a reference value to determine the presence or absence of cancer.SELECTED DRAWING: Figure 1

Embodiments of the present disclosure relate to a massively parallel processing database for sequence and graph data structures. The systems and methods disclosed herein may include a parallel processing graph database solution for protein sequence analysis to determine viable therapeutic agents for a given condition, and may include determining a protein sequence for the given condition; using the sequence database to compare a query sequence of a sequence of a given disorder to sequences of other known proteins in the sequence database; based on the comparison, using the sequence database to determine a similarity of the query sequence to sequences of other known proteins in the sequence database; and based on the similarity of the sequences, querying the graph database to identify potential therapeutic agents that can have an inhibitory effect on a given condition.

The invention relates to a computer-implemented method for tagging user action (UAC) data in a user environment (ENV), in particular wherein the user environment (ENV) is a computer (CMP)-implemented user interaction (UIA) environment (ENV), the method is characterized by comprising the following steps: a) providing a pre-trained user action (UAC) pattern recognition model (PRM), the pattern recognition model (PRM) being designed in such a way that the pattern recognition model (PRM) recognizes a user action (UAC) pattern (APT) known to the pattern recognition model (PRM) from a sequence of user action (UAC) parts comprising the user action (UAC) pattern (APT) known to the pattern recognition model (PRM); and outputting a user action (UAC) mode (APT) or a flag associated with the user action (UAC) mode (APT), the invention relates to a method for identifying user actions (UAC) in a user action (UAC) sequence (SQC), b) recording of user actions (UAC), c) analyzing the recorded user actions (UAC) by a user action (UAC) pattern recognition model (PRM) for a user action (UAC) pattern (APT) known by the pattern recognition model (PRM), d) if a known user action (UAC) pattern (APT) is not detected in the user action (UAC) sequence (SQC), generating a user interaction (UIA) as a user interaction (UIA) requesting information about a previous user action (UAC) sequence (SQC), the present invention relates to a method for training a pattern recognition model (PRM) using a user action (UAC) sequence, e) tagging past user action (UAC) sequences (SQC) using information received from an information request, f) storing the tagged user action (UAC) sequences (SQC) in a user action (UAC) sequence (SQC) database (DBS), g) training the pattern recognition model (PRM) using the user action (UAC) sequences (SQC) and corresponding tags (LBL) of the user action (UAC) sequence (SQC) database (DBS).

An artificial intelligence-based system comprises an input preparation module that accesses a sequence database and generates an input base sequence. The input base sequence comprises a target base sequence with target bases, wherein the target base sequence is flanked by a right base sequence with downstream context bases, and a left base sequence with upstream context bases. A sequence-to-sequence model processes the input base sequence and generates an alternative representation of the input base sequence. An output module processes the alternative representation of the input base sequence and produces at least one per-base output for each of the target bases in the target base sequence. The per-base output specifies, for a corresponding target base, signal levels of a plurality of epigenetic tracks.

The invention discloses a course recommendation method based on a large language model. The method comprises the following steps: 1, constructing a learner-course sequence database; 2, constructing a course sequence denoising module for filtering noise courses from the course sequence; 3, constructing a course dependency-enhanced learning interest module, which is used for constructing a course dependency graph and capturing dynamic learning interests; 4, constructing an interest-target comparison learning module for narrowing the potential course selection range of the learner; 5, constructing a prediction module used for obtaining the prediction probability of the course selected by the learner; and 6, training the course recommendation network to obtain an optimal course recommendation model which is used for predicting a course recommendation scheme for the learner to select the next course. According to the method, more accurate modeling of learning interests and targets of learners is facilitated, and the accuracy of course recommendation can be remarkably improved.

A tumor neoantigen prediction method and a tumor neoantigen prediction system are provided. In the method, multiple amino acid sequences in genes of a person to be tested are extracted as multiple test peptides to be compared with multiple human protein sequences in a protein sequence database to find multiple similar peptides that match the human protein sequences. The similar peptides are filtered out from the test peptides and the filtered test peptides are input to multiple trained human leukocyte antigen (HLA) models to obtain multiple ranking results of the test peptides. A weighted sum of rankings of each test peptide in the ranking results is calculated as a score of the test peptide. At least one of the test peptides is selected as a neoantigen adapted for the person to be tested according to the score.

The polypeptide design method and sequence of the present application combine mask language modeling and yeast surface display, comprising the following steps: cleaning a protein sequence database, selecting protein sequences meeting the requirements as a training set of a language model, and performing mask language modeling on the protein sequences contained in the training set; designing a set of downstream tasks on the basis of a pre-trained model, and fine-tuning the downstream tasks; randomly masking residues of a selected reference sequence, and predicting the masked residues; performing virtual screening on polypeptide candidates generated by the model; and determining the protein expression level and affinity of the screened polypeptides through yeast display technology. The present application can generate a large number of polypeptide candidates that may have specific properties or functions by transferring natural language processing technology to the polypeptide generation field, and combines artificial intelligence generation, virtual screening and wet experimental characterization to build a relatively complete "dry-wet combination" design process.