435 results about "Overfitting" patented technology

The invention discloses an attention mechanism and LSTM fused industrial bearing vibration time sequence signal fault prediction method and system. The method comprises the following steps: collecting a bearing vibration signal and carrying out filtering, noise reduction and normalization preprocessing; constructing a deep learning model combining the bidirectional BiLSTM and a coordinate attention mechanism to extract bidirectional time sequence features and enhance key fault features; carrying out model training by adopting a multi-target composite loss function and an Adam optimizer, and introducing an early stop mechanism to prevent overfitting; performing fault type identification and degree evaluation on the real-time vibration signal by using the trained model, and performing quantitative analysis by fusing multi-scale spectrum kurtosis features and nonlinear kinetic parameters; and finally, outputting a fault diagnosis report, and triggering multi-stage early warning based on an adaptive threshold. The method can realize high-precision and high-reliability bearing fault prediction and health state evaluation, and is suitable for intelligent operation and maintenance of industrial equipment.

The invention discloses a data processing system based on an artificial intelligence algorithm, and relates to the technical field of artificial intelligence and data processing, and the system comprises a data input interface which is used for receiving a multi-source heterogeneous data stream; and the preprocessing engine is connected with the data input interface and comprises a dynamic metadata sensing unit, an incremental quality evaluation unit, a bidirectional verification self-repairing unit and a closed-loop feedback optimization unit. According to the data processing system based on the artificial intelligence algorithm, through dynamic metadata perception and an incremental quality evaluation mechanism, the manual intervention requirement of a multi-source heterogeneous data preprocessing stage is reduced, and the problem of adaptation stiffness caused by the fact that a traditional method depends on a static rule is solved; by combining the bidirectional verification design of the business rule and the model feedback, the dual reliability of the data recovery strategy in logic rationality and algorithm compatibility is ensured, the risk of overfitting or scene mismatching caused by a single verification mechanism is avoided, and the input data quality and decision accuracy of a downstream artificial intelligence model are improved.

The invention relates to the technical field of deep learning and small sample learning, in particular to a small sample capacity training method based on deep learning, which comprises the steps of 1, cross-domain data adaptation and feature alignment, 2, meta-knowledge distillation and prototype enhancement, 3, attention-guided small sample fine adjustment, and 4, model uncertainty quantification and iterative optimization. According to the small sample capacity training method based on deep learning, through cross-domain feature alignment, meta-knowledge distillation, prototype enhancement and dynamic iterative optimization, the problems of model overfitting and weak generalization ability in a small sample scene are solved, high-precision model training when the sample size is less than or equal to 50 is realized, and the training efficiency is improved. The method is suitable for data scarce scenes such as medical images and minority language processing.

The invention discloses an integrated wind power prediction method and system based on a multi-source data set. The method comprises the following steps: acquiring historical meteorological factors and fan operation data; preprocessing the historical meteorological factors and the fan operation data to obtain a data set; based on the data set, key features are obtained through a Boruta algorithm, the key features are processed through a sliding window mechanism and a VMD algorithm, and an enhanced feature matrix is obtained; inputting the enhanced feature matrix into a deep learning model for prediction, and obtaining a preliminary prediction value; and carrying out residual error correction and fusion on the preliminary prediction value to obtain a final wind power prediction result. The method effectively improves the capability of processing wind energy intermittency, volatility and randomness, avoids the defects that a physical model is complex in calculation and a statistical model is difficult to process nonlinear and non-stationary features, reduces the over-fitting risk of a single deep learning model, can improve the prediction accuracy and stability, and improves the prediction efficiency. And the method has better generalization ability in practical application.

The invention provides a preference alignment optimization method based on reward-driven selective punishment, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining intelligent question and answer training data, and constructing an intelligent question and answer training sample set which comprises a plurality of intelligent question and answer training samples; taking a to-be-optimized large language model as a strategy model and a reference model; and training the strategy model by using the intelligent question and answer training sample set, and measuring the offset amplitude of the strategy model before and after training by using the reference model to obtain an optimized large language model. According to the method, implicit reward signals in the model are introduced, preference data are divided into multiple categories according to implicit reward distribution generated by the model, a dynamic weight function is designed, differential weighted optimization is carried out on different categories of samples, reinforcement learning of high-quality samples and suppression of low-quality samples are achieved, and the method has the advantages of being high in robustness and high in robustness. Weight of low-quality or conflict samples is reduced while high-quality sample learning is enhanced, noise interference is suppressed, and overfitting is prevented.

The invention relates to the technical field of deep learning, provides an optimization method of a defect detection model and defect detection equipment, and can be used for industrial quality inspection. According to the method, at least one to-be-detected product data set in production is used for carrying out multi-round iterative optimization on a defect detection model after pre-training, the defect that traditional model training is isolated from a production line environment is overcome, and closed-loop iterative detection and optimization of the model are achieved. In each optimization process, confidence coefficient learning and active learning are combined, the uncertainty of a model is quantified through confidence coefficient learning, a global confidence coefficient threshold value is determined according to the confidence coefficient corresponding to the defect type of each piece of to-be-detected product data, and the threshold value is dynamically adjusted by using the recall rate and the number of optimization times. And during active learning, sample data screening is carried out by using the adjusted global confidence threshold, so that blind labeling or over-fitting labeling is avoided, the quality of the screened samples is ensured, and the accuracy and generalization of a retraining defect detection model are improved.

The invention relates to the technical field of hydropower station equipment monitoring, in particular to a deep learning modeling and analysis method for hydropower station equipment operation trend early warning. Comprising the following steps: collecting multi-source parameters and dividing dynamic working conditions; mechanism-data driven fusion feature construction is carried out; training a physical informed deep learning model; carrying out meta-learning migration optimization; performing dynamic threshold early warning judgment; and performing mechanism closed-loop verification. The model is built based on a physical informed neural network framework, a differentiable mechanism constraint loss function is introduced, and dual verification is carried out through an equipment simplified simulation model and a historical fault case, so that model output can be ensured to accord with an equipment operation physical rule, and the situation that a pure data driven model possibly deviates from physical common knowledge is avoided; the reliability of the early warning model is improved; according to the method, the basic model is trained by adopting the meta-learning algorithm guided by the fault type label, so that the problems of model over-fitting and high adaptation cost in a small sample scene in the traditional technology are solved.

The invention discloses a visibility regression prediction method based on multi-modal transfer learning and time coding, and relates to the technical field of artificial intelligence. The method comprises the following steps: S1, dividing a data set in different periods according to illumination characteristics, and splitting each period into a training set, a verification set and a test set; s2, preprocessing the data set; s3, constructing an initial model containing a pre-training deep learning network, a time coding module and a multi-layer perceptron regression head; s4, extracting image visual features and time feature vectors; s5, fusing the features and inputting the features into a regression head for prediction; s6, carrying out scheduling training by using layered parameter freezing, an AdamW optimizer and a dual learning rate, and combining with a mixed early stop strategy until convergence; and S7, evaluating the test set to determine a final model. According to the method, complementarity of image and time information is mined, high-precision prediction is realized, generalization is good under different illumination conditions, a layering strategy and an optimization mechanism guarantee stable and efficient training, a multilayer perceptron combination technology enhances expression, and overfitting is effectively prevented.

The invention relates to the technical field of test and measurement, and discloses a series fault arc detection method, which comprises the following steps: collecting current data of fault arc waveforms and normal arc waveforms of a plurality of different loads; performing wavelet transformation on the preprocessed current data, dynamically extracting detail coefficients of a target layer number, calculating a standard deviation, a normalized energy ratio and an energy entropy based on the wavelet coefficients, splicing peak-to-peak values to form a four-dimensional feature vector, adding a normal or fault label, and converting the feature vector into a three-dimensional tensor; the residual shrinkage module is used for constructing a deep residual shrinkage network model based on a one-dimensional convolutional neural network and introducing an attention threshold generation and soft thresholding mechanism; training the model and monitoring the performance by adopting an early stop method; and inputting label-free sample data to the model, and outputting a detection result. The problems that in the prior art, deep features cannot be reflected, the training cost is high, and overfitting is prone to occurring are solved, and the purposes of improving the accuracy, being high in stability and efficient in detection are achieved.

The invention discloses a data deep learning and intelligent analysis method based on an AI artificial intelligence technology, and relates to the technical field of basic AI models, and the method comprises the steps: employing a multi-modal data preprocessing module to carry out the expansion of small sample data through a generative model, and combining with meta-learning to extract prototype features, meanwhile, an epsilon-differential privacy budget is dynamically allocated based on the data sensitivity level so as to inject dynamic noise; establishing a layered federated learning architecture, training a model by local training nodes through a loss function containing a self-adaptive regularization item, and performing sparse processing and gradient disturbance before uploading parameters; the global aggregation node adopts a weighted federated average algorithm to aggregate parameters, and dynamically adjusts the communication frequency according to the loss convergence speed; and a target model is obtained through iterative training, and a decision interpretation report containing the attention thermodynamic diagram and the desensitization identifier is generated when a result is output. According to the method, the problems of small sample overfitting, data islands and privacy disclosure are effectively solved, and the accuracy and practicability of the model are improved.

The invention discloses a corroded oil and gas pipeline residual strength prediction method considering a physical constraint loss function, and the method comprises the steps: collecting multi-source feature data of a corroded oil and gas pipeline, obtaining a residual strength measured value as a label, and constructing a training data set; an XGBoost regression model is combined with an SHAP interpretability analysis technology, and the influence degree and the influence direction of each feature on the residual intensity are quantified; constructing a neural network model, and determining an optimal architecture of a neural network by adopting a hyper-parameter optimization method; constructing a physical constraint term based on the influence degree and the influence direction of each feature, introducing the physical constraint term into a loss function of a neural network model, and forming a comprehensive loss function together with a data-driven loss term; and training the optimized neural network model by using a comprehensive loss function to obtain a final residual intensity prediction model. The method has the advantages that the prediction precision is improved, the model interpretability is enhanced, overfitting is prevented, and multi-source feature data are effectively integrated.

The invention relates to the technical field of program repair, in particular to an executable invariant and differential signal combined automatic program repair method, which comprises the following steps of: receiving a to-be-repaired program and a test set, and calling a large language model to generate a plurality of candidate patches; forming an entry behavior specification list according to the repair intention description; generating a plurality of executable invariant assertions and injecting the executable invariant assertions into the target function or the calling point, and generating an invariant assertion injection record table; generating a test case covering the boundary condition and the abnormal scene based on the large language model; collecting an execution signal when the candidate patch is operated; converting the patch difference into semantic editing features, calculating an editing stability index, and calculating an overall semantic consistency score of the candidate patches according to the support degree of the standard bar; and calculating an overall comprehensive score of the candidate patches, and outputting an optimal patch according to the overall comprehensive score of the candidate patches. According to the method, the proportion of overfitting patches can be effectively reduced, the patch repairing accuracy is improved, the interpretability is high, and the universality is good.

The invention discloses a three-dimensional magnetotelluric deep learning inversion method, and relates to the technical field of three-dimensional magnetotelluric inversion in electromagnetic exploration, and the method comprises the steps: constructing a three-dimensional layered underground resistivity theoretical model, and forming a sample pair through the structure data of the underground resistivity theoretical model and the corresponding visual parameter data; the method comprises the following steps: constructing a three-dimensional neural network based on a Swin Transform module and jump connection; a forward modeling sub-network is trained for the multiple visual parameters, and the weight of the forward modeling sub-network is frozen to serve as a fixed forward modeling operator, so that rapid forward modeling is achieved; after each fixed forward operator is migrated and spliced to the inversion sub-network, each fixed forward operator is used as an additional loss constraint term, and physical driving of neural network simulation is realized; and end-to-end mapping from each apparent parameter to the underground resistivity is established by fitting the inversion sub-network, and quasi-physics and data dual-drive three-dimensional magnetotelluric deep learning inversion is realized. The three-dimensional magnetotelluric inversion method has high practical value and popularization value in the technical field of three-dimensional magnetotelluric inversion with crossing of deep learning and electromagnetic exploration.

The invention relates to the technical field of data processing, in particular to a BN-based UHPC mix proportion data analysis method and a strength prediction system.The intelligent level of ultra-high performance concrete mix proportion design is improved by introducing a Bayesian network model and a multi-objective optimization algorithm, and the method comprises the steps that firstly, an enhanced feature set is generated through feature engineering; a nonlinear relationship and a cross-level interaction effect among material parameters are fully excavated, and the characterization capability of the model on a complex material system is enhanced; secondly, the Bayesian network model is combined with causal reasoning and a hierarchical regularization strategy, so that the overfitting risk is reduced while the compressive strength prediction precision is ensured; in the multi-objective optimization link, through dynamic weight adjustment and Pareto frontier search, carbon emission and strength requirements are effectively balanced, and candidate schemes with low carbon and excellent mechanical properties are output; and finally, dynamic evaluation and risk analysis are carried out to further screen out a mix proportion with high stability and strong feasibility, and a reliable decision basis is provided for engineering practice.

The invention relates to the technical field of protein prediction, in particular to a protein binding site prediction method based on geometric deep learning. According to the method, protein geometric description and a corresponding geometric diagram learning scheme are introduced, and a feasible way is provided for modeling three-dimensional structural features; in addition, the invention also provides an RSA-guided two-stage hybrid transfer learning strategy, firstly, a model is trained on a larger data set to learn general representation, and then fine tuning is carried out for binding site prediction, so that the overfitting problem in limited data training is relieved; according to the invention, a prediction post-processing module and an integrated learning module based on density clustering are designed, and the high variance of the model in multiple times of training operation is significantly reduced.

The invention belongs to the technical field of communication, and particularly relates to a multipath channel DOA estimation method based on a heterogeneous attention double-branch neural network, and the method comprises the steps: constructing a training set, a verification set and a test set through stratified sampling; introducing a multi-head attention mechanism to construct a heterogeneous attention double-branch neural network for realizing intelligent DOA estimation; designing a frequency weighted loss function as a loss function of neural network model training; a dynamic attention mechanism and an early stop mechanism are designed in a model training process to prevent an overfitting phenomenon. According to the method, the problem of serious DOA estimation model overfitting in a small sample and data non-uniform real acquisition data scene is effectively relieved, and the DOA estimation accuracy is improved.

The invention relates to the field of fault diagnosis methods, in particular to a fault diagnosis method based on multi-modal deep learning, and the method comprises the specific steps: S1, carrying out the processing of an original vibration signal of a bearing through continuous wavelet transform, and converting the original vibration signal into a time-frequency image; s2, constructing a multi-scale Mamba network model, and capturing a long-term dependency relationship in the time sequence data; s3, constructing a high-efficiency network model for extracting time-frequency features; s4, introducing a cross attention mechanism, and dynamically splicing the two modal features; and S5, verifying by using an MEF-Net model, and comparing the performance of the MEF-Net model with other reference models, thereby solving the problems that the integrity, accuracy and reliability of data are influenced due to the introduction of noise in the data acquisition process at the present stage, and the accuracy, accuracy and reliability of the data are influenced due to the inaccuracy, gradient disappearance, overfitting and the like of the data. Therefore, the problems of poor model training effect and low accuracy are solved.

The invention discloses a sand permeability coefficient prediction method based on Bayes and a physical information neural network, which comprises the following steps: constructing a physical-data hybrid driven enhanced data set which comprises actually measured anchor point data and physical enhanced data generated based on a seepage physical mechanism; constructing a dual-channel feature decoupling fusion neural network, extracting gradation morphological features and soil compaction features through a particle size channel and a structure channel, and performing fusion in a deep layer; introducing a multi-physical constraint embedding mechanism and a physical constraint activation layer, constructing a total loss function, and performing posterior inference on network parameters through an HMC algorithm to obtain a parameter sample set; and performing permeability coefficient prediction based on the parameter sample set. According to the method, the problem of overfitting caused by small samples in geotechnical engineering is effectively solved through a two-channel architecture and a data enhancement strategy, generation of non-physical prediction values is avoided through two aspects of network architecture and physical loss embedding, cooperation of high-precision prediction and uncertainty quantification is realized, and engineering adaptability is greatly improved.

The invention provides a hydrological flow long sequence prediction method and system of an improved state space model. The method comprises the steps of collecting multi-source data of a drainage basin to be predicted; constructing a time sequence sample pair by the processed multi-source data through a sliding window method, wherein the time sequence sample pair comprises an input sequence and a target sequence; a HydroMama model is constructed according to the time sequence sample pair, a HydroMama prediction model is trained, and an optimal hyper-parameter combination is searched for; based on the trained HydroMama model, traffic prediction and result restoration are realized by adopting an autoregression mechanism; the calculation efficiency is high, and the continuous state equation discretization of the state space model is utilized, so that the long-distance meteorological-hydrological hysteresis effect which is difficult to capture by a traditional cycle model can be captured; according to the method, skewed distribution of hydrological data is processed through logarithmic transformation, and an anti-overfitting objective function is combined, so that the prediction performance of the model on unseen data is remarkably improved, and the practical application value is high.

The invention discloses a discrete element simulation method for slope unstable seepage, effectively solves the problem of nonlinear flow simulation distortion caused by excessive simplification of a traditional seepage model, and improves the calculation precision of transient processes such as rainstorm infiltration. The dynamic porosity feedback mechanism can reflect the influence of the internal structure change of the slope on the seepage field in real time, and more reliable pore water pressure prediction data is provided for landslide early warning. The establishment of a multi-scale coupling framework provides a new technical approach for unstable seepage analysis under complex geological conditions, a physical information neural network effectively avoids the overfitting risk of a pure data driven model, and an improved multi-objective optimization algorithm can quickly position an optimal parameter combination under complex constraint conditions. A dynamic updating mechanism can automatically adjust model parameters according to slope state changes, and the timeliness and accuracy of landslide early warning under the unstable seepage condition are remarkably improved.

The invention discloses a spare part demand prediction method based on a ridge regression improved algorithm, and relates to the technical field of spare part demand prediction, and the method comprises the following steps: 1, data collection and preprocessing; 2, performing feature extraction on the time series data; step 3, constructing an improved ridge regression algorithm: combining Huber loss and a kernel function; 4, dividing a training set and a test set according to a time sequence, selecting an optimal parameter by adopting a K-fold cross validation method, performing robust kernel ridge regression model training on the training set, predicting a model generated by training on the test set, comparing with a true value of the training set, and calculating MAE, MSE and a value; and step 5, comparing with a traditional ridge regression prediction result. According to the method, the model overfitting problem caused by multiple collinearity among influence factors in spare part demand prediction can be solved, compared with traditional ridge regression, the nonlinear relation among variables can be captured, robustness is higher, and therefore the spare part demand quantity can be predicted more accurately.

The invention discloses a transcritical working medium thermophysical property intelligent prediction method suitable for physical property mutation characteristics, and the method comprises the steps: obtaining the physical property-temperature data of a target working medium under a specified working pressure through a standard physical property database, and employing a piecewise linear regression change point detection algorithm based on residual sum of squares and dynamic monitoring; recognition of inflection points of a working medium physical property-temperature curve and self-adaptive division of temperature intervals are achieved, polynomial order optimization is carried out on each segmented interval based on a Bayesian information criterion to balance fitting precision and complexity, continuity constraints at segmented connection points are applied, and a globally continuous segmented polynomial function is constructed. According to the method, the full-process automatic transcritical working medium physical property-temperature function modeling of data acquisition-inflection point identification-order optimization-function generation can be realized, the subjectivity and limitation of artificial experience segmentation are effectively overcome, and the over-fitting risk is avoided while the model precision is ensured; and reliable support is provided for engineering simulation and thermodynamic system design under the transcritical working condition.

The invention belongs to the technical field of financial science and technology, and discloses a multi-agent collaborative document-driven quantitative research and transaction self-evolution system and method. The system adopts a layered architecture, comprises a data and factor service layer, an execution engine layer, an agent layer, an interface service layer and a front-end control layer, and aims to solve the core problems that research and engineering implementation are separated in quantitative strategy research and development, code generation lacks interface constraints and strategies lack self-evolution ability. According to the system, through cooperative work of a research agent, a factor agent and a coding agent, a strategy idea described by a natural language is automatically converted into a standardized document, configuration and executable code; through a closed loop formed by self-monitoring, self-diagnosis and self-improvement intelligent agents, automatic diagnosis and optimization of problems of strategy overfitting, factor attenuation and the like are realized, continuous self-evolution of the strategy is driven, and full-life-cycle automatic management from strategy conception, factor design, automatic coding and real disk back-testing to feedback autonomous optimization is realized.

The invention discloses a GNN-MAPPO task unloading method based on dominant value noise addition, which is characterized in that an MLP is changed into a GNN on the basis of the existing MAPPO framework, a multi-agent system can be directly modeled into a graph structure, an interaction relationship among multiple agents can be better established, Gaussian noise is added on the dominant value, the exploration capability of a model is enhanced, and overfitting is reduced. According to the method, the powerful graph structure learning ability of GNN is combined with an innovative dominant value noise adding mechanism, and the mixed reward function is elaborately designed, so that the MAPPO algorithm can more effectively learn a cooperation strategy between agents and optimize time delay and energy consumption in the aspect of edge computing task unloading, and the efficiency of the MAPPO algorithm is improved. And the exploration capability of the strategy and the avoidance capability of the communication risk can be obviously enhanced, so that a more robust and efficient intelligent task unloading scheme adapting to a dynamic environment can be obtained.

The invention discloses a rolling bearing fault diagnosis method, device and equipment, and relates to the technical field of fault diagnosis. According to the method, domain classification is introduced, the Timer model, the fault classifier and the domain classifier are trained by minimizing the fault classification loss and maximizing the domain classification loss, an adversarial training mechanism is achieved, and therefore the feature extraction process of the Timer model is forced to focus on common fault features in source domain data and target domain data; therefore, feature distribution alignment between the source domain and the target domain is realized. And an adapter plug-in fine tuning strategy is adopted to realize rapid adaptation of the Timer model to a target domain task, so that the over-fitting risk is effectively reduced. The fault classifier learns a group of parameters with good generalization properties through model-independent meta-learning, and rapid adaptation and effective training can be carried out by only using a small number of samples under the condition of small samples in a target domain. According to the method, the fault diagnosis accuracy under different working condition environments and few samples in a target domain is improved.

The invention relates to the technical field of numerical calculation and data analysis, in particular to a data interpolation and fitting analysis system which comprises a data processing module, a depth generation interpolation module, an uncertainty quantification module, an interpretability optimization module, a multi-source heterogeneous data processing module and a hybrid calculation acceleration module. In the prior art, a traditional interpolation method is easy to generate over-fitting or under-fitting in a complex data distribution and high noise scene, and a single deep learning model is insufficient in generalization ability in a small sample or data sparse region; through the dynamic fusion architecture of the depth generation interpolation module, the advantages of a traditional numerical method and deep learning are combined, the weight is automatically adjusted based on data characteristics, the prediction error under noise data is remarkably reduced, the adaptability to complex distribution is improved, and the stability of a data sparse region is enhanced; the problems of insufficient precision and weak generalization ability of a single model of a traditional method are effectively solved.

The invention discloses a cross-domain small sample class incremental audio classification method based on domain generalization representation. The method comprises the following steps: extracting a logarithmic Mel spectrum from an input audio sample; initializing a residual convolutional neural network, wherein the residual convolutional neural network comprises a domain generalization representation extractor and a classifier; in the basic link, a supervised training method and a saliency weighted adversarial training method are successively adopted to train the residual convolutional neural network; extracting characterization from the source domain audio category training sample of the basic link, and calculating and storing a mean vector of the same category characterization for updating the classifier in the increment link; in an increment link, a classifier is updated by adopting an increment audio category training sample of a target domain and a mean vector represented by an old audio category; and inputting a to-be-tested audio sample into the residual convolutional neural network to obtain a category to which the to-be-tested audio sample belongs. According to the method, the domain generalization representation extractor and the confrontation training method of significance weighting are adopted, so that the cross-domain classification performance of the model is effectively improved while overfitting of the model to a new class and forgetting of the model to an old class are relieved.

The invention discloses a training method and application of a long-tail data semi-supervised segmentation model based on double teachers. The training method comprises the following steps: enabling labeled data input to calculate supervised loss; inputting label-free data to obtain general identification student prediction features, professional student prediction features, general identification false labels and professional false labels; calculating a first cross entropy loss between the general recognition student prediction feature and the professional pseudo tag, and a second cross entropy loss between the professional student prediction feature and the general recognition pseudo tag; calculating the comparative learning loss between the general recognition teacher false label and the general recognition student prediction features; and iteratively updating the parameters. The method is composed of different learning modules and has different pseudo label screening strategies, so that model coupling can be avoided; only some additional parameters are introduced, so that the data occupation is small; and furthermore, a more discriminative feature space is formed through comparative learning, so that the utilization rate of the pseudo labels is remarkably improved while overfitting is avoided, and finally, the model training efficiency and accuracy are improved.

The invention discloses an expressway differential settlement prediction and regulation system based on machine learning, and belongs to the technical field of intelligent traffic infrastructure construction and maintenance. According to the system, a closed-loop system including multi-source data fusion acquisition, spatial-temporal characteristic engineering, physical constraint spatial-temporal diagram network prediction, reinforcement learning reverse regulation and control and visual decision early warning is constructed for solving the problem of differential settlement control in alluvial plain deep and thick soft soil area highway reconstruction and extension projects. According to the system, Biot consolidation theory constraints are embedded in a geological weighted graph network, so that the problems of over-fitting and physical consistency deficiency of a pure data driven model under complex geological conditions are effectively solved; meanwhile, the reinforcement learning strategy network is used for replacing artificial experience, and global optimization regulation and control of construction parameters are achieved on the premise that the differential settlement control requirement is met. The method is mainly used for precise settlement prediction and intelligent construction management of highway reconstruction and extension projects.

The invention discloses a target activity prediction method of sgRNA, which comprises the following steps: step 1, preparing a data set, and obtaining an sgRNA activity sequence data set; 2, performing sequence feature extraction on the sgRNA active sequence data set to obtain multiple pieces of feature information; 3, fusing the multiple pieces of feature information to obtain a feature set; 4, constructing an unbalanced data set processing algorithm based on the graph weighted adversarial network; 5, introducing ensemble learning, and constructing a deep learning prediction model based on a voting algorithm; and 6, carrying out conservative motif analysis on the sgRNA high-activity sequence. According to the method, the prediction effect and robustness of the model can be ensured, different characteristics of data can be captured through diversified base learners, the over-fitting risk is reduced, sequence characteristics can be efficiently extracted, and the prediction precision and efficiency are remarkably improved; in addition, the biological significance of the model is explored, and an interpretable analysis attempt is carried out on the model.