212 results about "Single sample" patented technology

The invention discloses an interpretability visualization method and system based on chronic disease dynamic prediction, and belongs to the technical field of intelligent medical treatment, and the method comprises the steps: carrying out the distributed processing, standardized storage, distributed verification and undersampling of medical data, and obtaining a balanced data set comprising effective samples; extracting a disease tag of a previous time node from each patient as a feature, and performing feature importance evaluation by using a tree model to screen out an important feature subset; inputting the effective samples into a chronic disease prediction model, calculating the marginal contribution of each important feature, and visualizing the marginal contribution into a force diagram for analyzing single sample prediction logic; and calculating the interaction importance between every two features, screening feature interaction pairs, visualizing the feature interaction pairs into text rules and corresponding influence factors, and constructing the interaction pairs into new features for retraining the chronic disease prediction model. According to the method, the interpretability of the model in a chronic disease prediction technology can be enhanced, and the transparency and traceability of decision logic of a complex model are realized.

The invention relates to a cross-bearing single sample intelligent diagnosis method based on cognitive guidance and Riemannian manifold, and belongs to the technical field of rotating machinery fault diagnosis. Aiming at the problems of insufficient global task distribution learning ability, small sample over-fitting, Euclidean modeling limitation and the like of the existing meta learning method in a cross-domain single-sample scene, a cognitive guidance Riemannian meta learning framework is provided. According to the technical scheme, the method comprises the following steps: 1) constructing cognitive prototype learning global task distribution, and guiding a model to extract high-quality general meta-knowledge from multiple tasks; 2) designing a cognitive adaptive factor to dynamically adjust source domain memory, enhancing target domain adaptation and reducing single sample deviation; and 3) introducing a Riemann metric driving strategy, mapping the data to a Grassmann manifold space, and enhancing the non-linear feature discrimination ability by using geodesic distance. According to the method, the average diagnosis accuracy in a cross-bearing single sample task reaches 93.46% and is improved by 10.04% compared with an existing optimal method, and the accuracy and generalization ability under complex working conditions are improved.

The invention discloses a single-sample time sequence knowledge graph reasoning method based on long sequence modeling, and belongs to the field of time sequence knowledge graph reasoning. According to the method, an attention mechanism and co-occurrence analysis are combined, features are aggregated from adjacent entities, and therefore tracking of the historical evolution trend is achieved, specifically, the attention mechanism is adopted to integrate relation information in entity neighborhoods, and the relevance between the adjacent entities and the importance of the neighborhoods of the adjacent entities are clearly explored in combination with the co-occurrence analysis. And furthermore, an encoder based on a state space model is adopted to encode the time sequence interaction of the entity, so that the method not only can efficiently model a long sequence, but also can capture long-term dependency. And finally, introducing a measurement network, applying the learned representation to the measurement network, evaluating the similarity between the support entity pair and the query entity pair from multiple perspectives, and ensuring that the possibility of reflecting future events is comprehensively evaluated. According to the invention, a comprehensive experiment on a widely used time sequence knowledge graph data set shows that the method is obviously superior to a baseline model.

The invention relates to the field of environmental monitoring, and particularly discloses a non-negative matrix factorization and adaptive peak recognition fluorescence feature extraction and traceability system, which comprises a spectral data preprocessing module, a spectral data non-negative matrix factorization module, a component number automatic selection module, an adaptive peak recognition module, a feature library construction module and a similarity comparison module. An improved non-negative matrix factorization model is adopted to decompose the three-dimensional fluorescence spectrum matrix of a single sample, and an optimal component number K is automatically determined through multiplicative update rule iterative optimization; the self-adaptive peak identification module carries out selective filtering, accurately extracts the position and intensity of a fluorescence peak through multiple mechanisms, and carries out peak position calibration in a neighborhood; the Hungary algorithm is adopted to carry out characteristic peak matching to calculate the comprehensive similarity between the samples, and rapid and accurate identification of the pollution source is realized. The method has the advantages of high resolution, strong anti-interference capability, low requirement on the number of samples, automation and the like, and is suitable for water quality fingerprint feature extraction of a water sample in a complex environment and real-time source tracing of sewage.

The invention relates to the technical field of data analysis, and discloses a time series data anomaly detection method and system based on convergence trajectory feature analysis, and the method comprises the steps: collecting and preprocessing multi-dimensional time series data of at least one industrial device belonging to the same type, and obtaining a standardized single device feature sequence of each industrial device; training a baseline auto-encoder model based on the standardized feature tensor to obtain baseline parameters and baseline features of the baseline auto-encoder model; performing disturbance training on the baseline auto-encoder model by taking the standardized single device feature sequence of each industrial device as a single sample based on the baseline parameters to obtain convergence trajectory features of the baseline auto-encoder model, and obtaining a standard digital portrait library based on all industrial devices; and carrying out feature attribution and root cause analysis on the digital portrait. According to the method, the sensitivity and the detection accuracy of weak signals, small samples and complex behavior anomalies are remarkably improved.

The invention discloses a shield tunneling stratum intelligent identification method and system based on strict single sample prototype network and verification set scale calibration, and the method comprises the steps: dividing samples collected from a shield tunneling machine in real time into a training set, a verification set and a test set according to a set time and an engineering interval; carrying out normalized preprocessing on the multi-dimensional mechanism parameters by adopting a training set, and then carrying out supervised pre-training on the lightweight MLP encoder by adopting a cosine interval classifier to obtain a discriminative embedding space; according to the method, a high-precision, reproducible and low-time-delay online recognition function on the shield tunneling stratum under the condition of extremely few labels is achieved, the overall precision and minority class recall can be considered on the premise that the labeling cost is not increased, so that the output probability is more calibrated, the expandability and rapid deployment capacity are achieved, and the method is suitable for popularization and application. And the real-time performance and the reliability of shield tunneling stratum identification are obviously improved.

The invention discloses a gear breakage identification method and device, equipment and a medium, and relates to the technical field of fault diagnosis, and the method comprises the steps: carrying out the single-sample fault diagnosis of impact single-sample data of each gear measurement point of a current gear, and obtaining a single-sample fault diagnosis result; performing trend analysis of gear impact and meshing spectrum impact based on the impact single sample data to obtain a gear impact trend result and a meshing spectrum impact trend result respectively, and obtaining gear impact continuity statistics of the impact single sample data; and according to the current health level of the current gear, the single sample fault diagnosis result, the gear impact trend result, the meshing spectrum impact trend result and the gear impact continuity statistical magnitude, determining whether the current gear has a gear breakage fault, and judging whether gear breakage early warning is needed. Multi-angle features are extracted through single sample analysis and trend analysis. And comprehensive gear breaking fault identification is carried out in combination with health levels and multi-angle features, so that misjudgment of a single index is avoided.

The invention discloses a jewelry value evaluation knowledge base construction method based on a large language model. The method comprises the following steps: S1, collecting and preprocessing jewelry value evaluation text data; s2, keeping title enhancement of the hierarchical information; s3, standardizing table data; s4, converting the text including the table into a Markdown format; s5, extracting jewelry value evaluation knowledge based on single sample learning by using a large language model; s6, processing and storing a result; according to the method, the title is converted into the Markdown format, the absolute path of each title in the text is reserved, and meanwhile, the cue word for guiding the large language model to extract information in a structured manner is constructed for the extraction task, so that structured extraction and tree structure organization of jewelry value evaluation knowledge are realized; the problem that a hierarchical structure is difficult to construct in a knowledge base due to the fact that a current text information extraction result based on a large language model lacks all levels of title link relations is solved.

The invention discloses an artificial leather quality rapid detection method based on spectral analysis, and particularly relates to the field of quality detection.The method comprises the steps that component composition characteristics, surface and microcosmic characteristics and dynamic response characteristics of artificial leather are synchronously obtained through multi-dimensional original data acquisition, and a time-space correlation data set is formed through timestamp alignment and space calibration; constructing component analysis, surface association and dynamic response models, and eliminating space-time migration through a synchronous optimization model; after multi-dimensional features are extracted, optimized features are generated through cross-direction tensor fusion, attention enhancement and gating cycle unit time sequence fusion; and finally, quantizing the quality index based on a multi-task model, and outputting a quality grade and a traceable report in combination with a dynamic threshold value through hierarchical coding, graph neural network association enhancement and evidence fusion. The single sample collection duration is less than 3 minutes, the response time is less than 1 second, and full-dimension, rapid and accurate detection and quality traceability are realized.

The invention belongs to the technical field of image processing, and discloses a rice bud stage drought resistance prediction method, which comprises the following steps: acquiring sample images of a plurality of individuals of a rice variety to be detected under control and stress conditions, performing foreground segmentation and structural region marking on the images, and identifying seed, bud and root regions. And multi-dimensional phenotypic features are extracted for each region to construct a single-sample high-dimensional feature vector. And carrying out group feature extraction on a plurality of single sample vectors under the same processing condition, and carrying out double correction on a sample confidence coefficient weight and a feature steady-state weight to obtain a group feature vector. And constructing a multi-dimensional drought resistance response feature vector according to the difference between the group feature vectors of the stress group and the control group, and inputting the multi-dimensional drought resistance response feature vector into a pre-trained classification model to obtain a drought resistance level prediction result. The method achieves the automation of drought resistance evaluation, effectively inhibits the interference of abnormal samples, and improves the evaluation accuracy.

The invention relates to a single-sample 3D object normalization method and system based on geometric and semantic consistency and application, and the method comprises the steps: 1, zero-sample object semantic correspondence mapping: building a semantic correspondence relationship between a test model and a prior model; a second stage of normalized hypothesis generation, including: generating a set of normalized attitude hypotheses for each sampling initial attitude of the test model; a joint energy function considering semantic and geometric clues at the same time is provided, so that a test model is aligned with a prior model; and a third stage, normative attitude selection: determining a final normative attitude by evaluating the consistency of relative semantic positions between the prior model and the test model according to the generated normative attitude hypothesis. The invention provides an elaborately designed joint strategy, and the strategy uniquely integrates semantic and geometric clues to realize more accurate alignment.

The invention provides a deep learning adaptive learning rate optimization method based on gradient variance and time sequence attenuation, and the method comprises the steps: collecting single sample gradients of all samples, forming a gradient set in a batch, and calculating a batch gradient global variance signal in the batch based on the single sample gradients; performing random disturbance reasoning on the deep learning model to obtain prediction output; on the basis of the confidence trajectory unit and the historical trajectory thereof, constructing a composite dynamic feature containing historical time step information; jointly inputting the composite dynamic features and the confidence calibration variance, and inputting the combined input to a scheduler to generate a dynamic learning rate; and updating the deep learning model according to the dynamic learning rate, generating new model parameters, completing model parameter updating in combination with a gradient calculated on the current batch of training data, and transmitting training dynamic information to the next round of signal extraction process through a state feedback mechanism to perform closed-loop optimization.

The invention discloses a section bearing steel strip-shaped carbide identification method fusing an attention efficient layer aggregation network, and the method comprises the steps: collecting a section bearing steel image, and constructing a section bearing steel strip-shaped carbide data set; marking section bearing steel strip-shaped carbides in each image in the data set, and constructing a target detection data set; according to the method, a YOLOv8 network is improved, and a Swin Transform module and three ELAN-A modules are added to a neck network; the improved YOLOv8 network is trained, and a section bearing steel strip-shaped carbide recognition model is constructed; and detecting the test set by adopting the trained section bearing steel strip-shaped carbide identification model, and verifying through detection evaluation indexes. The method can effectively meet the requirement of section bearing steel banded carbide identification, and meanwhile solves the problem that existing manual detection of a single sample is time-consuming.

The present invention provides a method and system for constructing and operating a multi-modal large model for reservoir dam safety. The method includes the following steps: constructing a training set: a single sample of the training set includes; multi-modal data as model input: sensor data deployed on the dam, historical hidden danger records, engineering technical documents formed in each stage of the dam's planning, design, construction, and operation, information resources related to dam safety, geometric and material information of the BIM model, timestamp and spatial location information; and the probability value of whether a hidden danger event occurs as the training label; training a multi-modal pre-trained model based on the Transformer architecture using the training set; performing real-time prediction and early warning on dam safety hidden dangers; calculating the actual impact degree, comprehensive risk score, dynamically set early warning threshold, and emergency measure suggestions of the hidden danger event according to the probability value of whether the hidden danger event occurs output by the model. The present invention comprehensively improves the scientificity and effectiveness of dam safety monitoring.

The application discloses a water quality and soil detection and analysis method and system based on multi-standard automatic evaluation and belongs to the technical field of environmental monitoring data processing. The system comprises a data import module, an intelligent identification module, a batch sample processing module, a multi-standard evaluation module, a comprehensive analysis module, a water chemistry verification module and a result output module. The system automatically analyzes the Excel table structure through an intelligent identification algorithm, extracts index names, units, detection results and background values, supports batch sample format identification, can analyze multiple samples at a time, integrates three environmental quality standards of underground water GB / T 14848-2017, surface water GB 3838-2002 and soil GB 36600-2018, automatically performs quality classification evaluation, calculates single pollution index P i , exceeding multiple, Nemerow comprehensive pollution index P N , performs water chemistry anion and cation balance verification, Shukarev classification, TDS and hardness checking, and supports single sample and batch sample export and summary table generation. The application solves the problems of weak Excel data recognition ability, batch sample processing difficulty, lack of multi-standard automatic switching, single comprehensive evaluation method and insufficient data quality checking in the prior art, and greatly improves the efficiency and accuracy of detection data analysis.

The invention provides a cactus plant species identification method based on high-throughput sequencing data. A specific variation region in a cactus chloroplast genome is screened based on a high-throughput sequencing data system, and a standardized cactus plant species rapid identification process is developed by combining Pi value screening and specific variation region marking. Compared with a traditional DNA bar code fragment, the identification precision of the targeted specific variation region is remarkably improved, so that species with highly overlapped forms can be effectively distinguished, and the identification accuracy is improved. Meanwhile, the method greatly reduces the technical threshold and cost, PCR amplification is performed on the DNA of the sample to be detected by designing a specific primer of a targeted specific variation region, whole genome sequencing and assembly are avoided, the detection cost and detection period of a single sample are reduced, and the method is compatible with a conventional PCR platform and has high practicability. And on-site rapid detection of scenes such as a medicinal material market and a customs port becomes possible.

The invention relates to the technical field of surface defect detection, in particular to a synthetic leather surface defect detection method based on space instance dissimilatory self-attention, which comprises the following steps: constructing a space instance dissimilatory self-attention bottleneck structure which comprises a main branch and an auxiliary branch; wherein the output of the main branch and the output of the auxiliary branch are jointly fused with the input of the space instance special dissimilatory self-attention bottleneck structure, and the fusion result is used as the output of the space instance special dissimilatory self-attention bottleneck structure; replacing a bottleneck structure in a YOLOv8 algorithm backbone network with a space instance dissimilatory self-attention bottleneck structure to obtain a surface defect detection model; acquiring a surface defect sample data set to train a surface defect detection model; inputting to-be-detected data into the trained surface defect detection model to obtain a detection result; the space instance dissimilatory self-attention bottleneck structure is introduced, the problems of false detection and missing detection caused by the fact that the specificity of a single sample is not fully excavated can be reduced, and the target detection precision is improved.

The application discloses a graphite ore grade identification method and device, equipment and medium, and relates to the technical field of graphite ore grade identification. The application introduces Laplacian wavelet convolution instead of a traditional ordinary convolution kernel, realizes parallel extraction of multi-scale features, combines a multi-head attention mechanism to capture scale features of the ore under a global view, constructs a sparse sample relation graph based on an enhanced feature matrix, and uses a Chebyshev graph convolution network to perform multi-order neighborhood information propagation and aggregation on a graph structure to realize grade identification. The process reconstructs the grade identification problem from'single sample independent judgment' to'sample relation learning' under a small sample on the basis of global multi-scale feature extraction, so that the final prediction of each sample node depends not only on its own features, but also on similar feature information of all neighbor samples, thereby realizing high-precision identification of the grade of the graphite ore under a small sample.

The present application relates to the technical field of surface defect detection, and particularly relates to a synthetic leather surface defect detection method based on spatial instance-specific self-attention, which comprises constructing a spatial instance-specific self-attention bottleneck structure, which comprises a main branch and an auxiliary branch; wherein the output of the main branch and the output of the auxiliary branch are fused together with the input of the spatial instance-specific self-attention bottleneck structure, and the fusion result is taken as the output of the spatial instance-specific self-attention bottleneck structure; the spatial instance-specific self-attention bottleneck structure is used to replace the bottleneck structure in the backbone network of the YOLOv8 algorithm to obtain a surface defect detection model; a surface defect sample dataset is obtained to train the surface defect detection model; and the detection data to be detected is input into the trained surface defect detection model to obtain a detection result; the spatial instance-specific self-attention bottleneck structure introduced in the present application is helpful to reduce the false detection and missed detection problems caused by insufficient mining of single sample specificity, and improve the accuracy of target detection.

A system for the simultaneous thermal analysis of a plurality of single samples of, in particular biological, material by means of differential scanning calorimetry, with at least one sample carrier having several sample vessels, wherein a single sensor for measuring an amount of heat emitted or absorbed by the single sample during the thermal analysis is assigned to each sample vessel; a heating and / or cooling unit for the simultaneous temperature application of the single samples included in the sample vessels, with a receptacle for the at least one sample carrier; a measuring instrument, which is connected to the single sensors and which is formed to simultaneously capture a measuring value for the emitted or absorbed amount of heat of the single samples during the thermal analysis, a sample carrier, in particular for use in this system as well as a method for the simultaneous analysis of a plurality of single samples or groups of single samples by means of differential scanning calorimetry.

The present invention discloses a genome-wide association analysis algorithm at the gene level based on an EMS population, which relates to the technical field of bioinformatics. The present invention conducts association analysis with genes as the basic unit. The MAF index of SNPs in the EMS population is much lower than that of the normal population, but the MAF value at the gene level is higher than that at the SNP level. Weights are assigned according to the mutation effects of each mutation site, and the total weighted value of all mutations in each gene of a single sample is statistically calculated as the basis for association analysis, greatly reducing the false positive rate of the results and increasing the statistical power of the analysis. At the same time, multiple statistical methods are used for comprehensive evaluation to find the most reliable candidate genes. Compared with GWAS that can only locate a fuzzy interval that may contain multiple genes, the present invention can directly and accurately locate to a single gene, improving the experimental efficiency of functional verification of candidate genes and effectively solving the problems that the prior art cannot afford large-scale analysis of the EMS population and has low analysis power.

The invention provides a rapid pollutant detection method based on spectral feature analysis, and relates to the technical field of environmental pollutant spectral detection, and the method comprises the following steps: obtaining an original absorption spectrum of a sample, and preprocessing and reconstructing an intrinsic spectral signal; extracting three types of features of morphology, frequency domain energy distribution and phase disturbance, and splicing to obtain a multi-dimensional joint feature tensor; pollutant classification and confidence judgment are completed through an improved CNN, and concentration is predicted through a linear regression model synchronously; in combination with a Mahalanobis distance verification result, deep residual network secondary feature mining is triggered for a low-confidence sample, a final type is determined according to a fusion result, and the concentration is corrected; and generating a detection report and dynamically updating the spectral feature intrinsic library. The method is adaptive to a portable ultraviolet-visible spectrophotometer, the single sample detection time consumption is less than 10min, the classification accuracy is greater than or equal to 90%, the unknown pollutant judgment accuracy is greater than or equal to 89%, and the method is suitable for on-site rapid detection.

The invention discloses a method suitable for on-site analysis of the valence state of thallium in water. The method comprises the following operation steps: step 1, collecting and separating a water sample; step 2, acidizing treatment; 3, Tl (I) and Tl (III) are separated; 4, total thallium is measured; step 5, Tl (I) determination; the invention relates to the technical field of atomic spectrum metal element analysis, and the method has the beneficial effects that by combining an electrochemical separation technology with a Chelex-100 solid-phase extraction column and cooperating with a portable liquid cathode glow discharge thallium measurement instrument, high-sensitivity field detection of Tl (I) and Tl (III) is realized, the detection limit is lower than 1.0 g / L, and the repeatability is RSDlt; the analysis period of a single sample is less than or equal to 30 minutes, the weight of equipment is only 5kg, battery power supply is supported, and acidification-separation-detection integrated operation can be completed outdoors without professional laboratory conditions.

This invention, entitled "A Method and Application for Target Gene Quantification in qPCR Using a Quantitative Internal Standard and a Comprehensive Correction Coefficient," belongs to the field of nucleic acid detection technology. The technical problem it addresses is that existing qPCR techniques require the establishment of a standard curve using a reference sample of known concentration, increasing reagent costs and experimental complexity, and resulting in poor reproducibility. The key technical solution is to provide a method for quantitative viral load qPCR analysis of a single sample. Specifically, it provides an internal standard with a known copy number, i.e., a quantitative internal standard, and establishes a correction method. Through a systematic error compensation mechanism, it achieves accurate quantification of the target gene. This method ensures both accuracy and reproducibility while simplifying detection design in clinical applications.

The invention discloses a deep coal rock pressure maintaining gas content determination method, device and equipment and a storage medium, and belongs to the technical field of well logging. According to the method, a rock core in a pressure maintaining barrel is segmented according to lithology on the basis of the correlation between the occurrence state of coal rock analytic gas and the pore structure of a coal rock reservoir, then the integral cut weighting coefficient of the pressure maintaining gas is determined, and the content of the pressure maintaining gas in a single sample is further obtained. The testing method can provide more accurate and reliable pressure-maintaining gas content data of the deep single coal rock, is applied to evaluation work of the gas content of the deep coal rock, and provides a basis for accurately calculating the total gas content of the single coal rock, delineating a favorable exploration area and improving the success rate of natural gas exploration and development.

This invention relates to a lightweight adaptive method for testing non-stationary data streams. Its core lies in constructing a two-branch adaptive architecture that decouples observation and execution statistics. A shadow observation branch, independent of the main inference branch, maintains a statistical reference baseline in real time, unaffected by adaptive calibration actions, thus suppressing the closed-loop coupling problem between drift perception and statistical updates at the mechanism level. Based on this decoupled architecture, the shadow branch calculates a composite drift surrogate metric to map and generate dynamic momentum parameters, and introduces relaxed cumulative variables to perform three-level state determination and spatiotemporal misalignment mask generation. Finally, during inference forward propagation, dynamic online calibration of statistics is performed only for the key levels selected by the mask. This invention alleviates the adaptation lag and statistical oscillation risks caused by fixed momentum, suppresses the statistical degradation risk caused by single-sample noise, and is suitable for real-time model optimization in single-sample, non-stationary streaming scenarios.

The invention discloses a neutron activation detection station system for rapidly detecting the element content of a material. The system can be used for rapidly detecting the element content of solid bulk materials in industrial production processes of building materials, coal, metallurgy, mines and the like. The system is composed of an element analyzer, a mechanical arm, a weighing scale, a coding sample barrel, a coding card reader, an electric roller conveyor and a PLC control system. The measurement time of a single sample is 30-300 seconds, the weight of the single sample is 1-10 kg, and a plurality of samples can be continuously and automatically measured. The method expands the application process points of the neutron activation technology, breaks through the traditional material detection mode, and has the advantages of wide application range, convenience in operation, high measurement precision, effective improvement of the detection efficiency and the like.

This invention discloses a fault diagnosis method for the polyester esterification stage. This method combines global supervised learning and contextual metric meta-learning. Using the attribute information of a single sample and similarity information from a sample group, it first uses variational modal decomposition to obtain multi-scale data through global supervised training. Multi-scale components with fault characteristics are extracted, and multi-scale feature fusion learning is performed. Triplet loss is used to learn finer, more subtle features. A fixed multi-scale feature fusion module is then used for task meta-learning training to learn a single feature, converting the raw data of the meta-task into a basic feature space. Finally, a dimensional variational prototype module is used to adaptively measure the feature similarity of sample pairs. The statistical method of variational inference automatically learns metric scaling parameters to transform the embedding space. The method is simple and solves the fault diagnosis problem in scenarios with limited data and a fully open set.

The invention belongs to the technical field of nuclear reactor physical simulation and safety analysis, and discloses a nuclear reactor passive start-up simulation method based on a Monte Carlo method, which comprises the following steps: calculating a single sample based on a Monte Carlo direct simulation method, modifying a random number seed in an input card to obtain a plurality of calculation results of mutually independent samples, and calculating the single sample according to the calculation results. And finally, obtaining a probability distribution result of the weak neutron field through statistics. The method can accurately describe the physical process of the weak neutron field stage, and is suitable for a three-dimensional complex reactor.

The invention belongs to the technical field of seismic oscillation, and particularly relates to a local terrain seismic oscillation simulation method considering site soil parameter randomness. The method comprises the following steps: S1, based on a random field theory and site soil exploration data, obtaining site soil parameters with soil layer variability; s2, based on the random field soil parameters, solving local terrain seismic oscillation by using an IBEM to obtain seismic response data of a single sample; s3, establishing a CNN model for solving a local terrain earthquake response; random field soil parameters are set as CNN input parameters, an earth surface point acceleration peak value of each group of independent samples is solved by using an IBEM, the earth surface point acceleration peak value is set as CNN output parameters, a data set is constructed based on the input parameters and the output parameters of each independent sample, and model training is carried out; and S4, based on the trained CNN model, obtaining local terrain ground motion considering site soil parameter randomness by using MCS. The method provided by the invention has good accuracy and applicability.