820 results about "Logit" patented technology

The local intrinsic dimensionality (LID) for a diffusion model with respect to a particular data sample is determined by using the diffusion model's diffusion process to apply noise to a data sample and evaluate how the estimated log probability of the data sample changes at different levels of noise. Particularly, the differential of change in noise to change in log probability can be used to determine the local intrinsic dimensionality. This may be determined by evaluating the log probability at several noise levels and determining a slope of the difference. In additional examples, the differential is evaluated directly at a selected noise level. The selected noise level can be optimized by calculating the estimated LID for various data samples at a variety of noise levels and selecting the LID that corresponds to a “knee” where the estimated LID sharply changes.

The invention discloses a ground motion intensity measure optimization method for seismic response prediction. The method includes the following steps: (1) building a building damage database; (2) optimizing ground motion intensity measures; (3) constructing a machine learning model; and (4) predicting damage states. According to the method, a combination of optimal ground motion intensity measures is selected by an elastic-net algorithm, wherein the efficiency and practicability in a current selection criterion are updated based on the goodness of fit and regression coefficient of the elastic-net model, thereby avoiding an assumption that intensity measures and structure response need to obey a logarithmic linear relationship and the limitation of intensity measure dimensions in traditional methods. By using the combination of optimal ground motion intensity measures as input variables, the prediction accuracy can be remarkably improved, and meanwhile the number of samples required is reduced.

Provided are a method and an apparatus for constructing a road congestion prediction model, a device, a medium, and a product. A road traffic network is defined as a directed weighted graph. Historical dynamic traffic features of each road segment in the road traffic network are obtained as sample data, including recent dynamic traffic features and periodic dynamic traffic features. The sample data is input into a mixture of adaptive graph learners (MAGL) model for learning, and a probability prediction vector is output. The sample data is input into a trend expert model, and a trend distribution vector of a predicted probability of future traffic conditions is output. The periodic dynamic traffic features are fused to determine a periodicity prediction vector. An aggregated logit vector is obtained. An objective function is determined based on the aggregated logit vector. Congestion prediction training is performed to obtain a road congestion prediction model.

Systems and methods for unsupervised multi-modal causal structure learning for root cause analysis. System logs of a cloud system can be transformed to time-series data using a log-tailored language model to obtain system log features of the cloud system. A metric causal graph and a log causal graph can be predicted from modality-specific representations and modality-invariant representations of extracted system metric features and system log features, respectively, using the deep neural network. The metric causal graph and log causal graph can be fused to obtain a fused causal graph. Root causes of system failure can be flagged for system maintenance based on ranked entities obtained from the fused causal graph to obtain flagged root causes. System maintenance can be performed autonomously based on the flagged root causes from identified system entities to optimize the cloud system with an updated configuration.

This document discloses methods and systems for cohort identification. The methods and systems include improved calculations to perform cohort identification and practical applications of the improved calculations. Specifically, the systems and methods described herein may utilize key components that include enhancements of existing cohort clustering techniques with regard to selecting a number of cohort input dimensions, normalizing input data using a logarithm kernel-function, treatment of categorical data with mutually exclusive and not-mutually exclusive values, methods and visualization tool to determine appropriate number of cohorts, methods and visualization tool to compare cohorts extracted from different input dimensions, and methods to quantify the difference in cohorts. Beyond improvements to the cohort clustering techniques, also disclosed are ancillary tools to prepare input data by joining CRM and product usage data and facilitate subsequent automated action via an API to retrieve cohort results.

The invention discloses a method for embedding a text watermark during Logs generation based on a large language model. The method comprises the following steps of: analyzing a generated text tl-1 of a large model through dependency syntax analysis, constructing a dependency syntax tree of the generated text tl-1, sorting according to scores of a dependency syntax relationship, and selecting and adding biased Logs; a candidate word list is generated, probability distribution corresponding to each dependency type is calculated, the selection probability of the candidate word list is adjusted, vocabularies are sampled and generated at key positions from the candidate word list according to a set rule, and a watermark embedded text is obtained; the candidate word list comprises a first list and a second list, and the selection probability of the first list is higher than that of the second list; and extracting a to-be-detected fragment in the watermark-embedded text, detecting the watermark of the text, and outputting a detection result. According to the method, the problem that the anti-attack capability of adding watermarks to the generated text is insufficient is solved, and meanwhile, the problems of text quality and large long text calculation amount of a current watermarking method in the Logits generation period based on a large model are solved.

The invention provides a publicly verifiable large language model watermark detection method and system. The method comprises the following steps: S1, embedding watermark information into a large language model by utilizing a watermark embedding model; s2, extracting watermark information from the text generated by the large language model by using a watermark extraction model, and generating a corresponding correctness proof; according to the watermark embedding model, logits and probability distribution of candidate words in a large language model generation process are dynamically modified, so that the large language model is more likely to select candidate words in a green word set as output words, and watermark information is embedded in an output text; the watermark extraction model is used for judging whether the text contains watermark information or not by counting the number of green words and the number of red words contained in the text and utilizing a statistical test method.

A GNSS positioning slowly-varying fault detection method based on residual-SVR regression comprises the steps that an observation information sequence is acquired based on a Kalman filter, and a covariance matrix of the observation information sequence is calculated; accumulating multi-step information through a sliding window, and constructing chi-square statistics; based on the fault-free data, constructing a training set by taking an innovation sequence as input and chi-square statistics as output, and generating an innovation-statistics mapping function; and fitting a normal slope threshold value based on an SVR predicted value, carrying out least square fitting on an observation statistic curve by sliding a window in real time, and judging whether to start a slow change fault alarm or not. According to the method, the residual error sequence is directly used as model input, and the dynamic chi-square statistical magnitude is used for replacing a traditional dichotomy label, so that the detection delay is reduced; an SVR detection model based on grid search and cross validation collaborative optimization is utilized, and an optimal parameter combination of a minimum mean square error (MSE) is screened through logarithm uniform sampling, interval linear sampling and five-fold cross validation, so that the average absolute error of slowly varying fault detection is reduced.

The invention discloses a weld defect detection method and system, and relates to the technical field of defect detection.The method comprises the steps that a visible light image sequence, an infrared image sequence and a pixel point cloud sequence are collected, windows are moved in geometric coding results of pixel point cloud of each frame and a visible light image through window attention, and the windows are extracted and sorted into visual feature maps; extracting a temperature feature map from each frame of infrared image through thermal gradient convolution and cavity convolution, and arranging the temperature feature map into a geometric feature sequence and a defect feature sequence based on cross attention and decoupling head mapping; capturing the time sequence dependence of the defect feature sequence through a Transform encoder, carrying out potential space modeling, and generating a defect probability vector by utilizing defect priori knowledge in combination with a Bayesian network; and a Gaussian regression model and a logarithmic probability equation are adopted to map the geometric feature sequence into a probability threshold of each defect type, and the probability threshold is compared with a defect probability vector to determine the defect, so that high-precision defect detection fusing geometry, temperature, time sequence and priori knowledge is realized.

The invention provides an adjacent tunnel excavation disturbance vulnerability evaluation method based on Bayesian updating, which introduces a concept of system stiffness to represent uncertainty of tunnel damage caused by adjacent excavation disturbance, and brings obtained system stiffness data into an adjacent excavation disturbance model through a fitting equation. Calculating a large number of numerical values to establish a tunnel damage database; establishing an agent model between the parameters corresponding to the disturbance working condition and the tunnel damage index; collecting data for parameter characteristic statistics to obtain prior probability distribution of tunnel damage indexes, and combining engineering actual monitoring data for Bayesian updating to obtain posterior probability distribution of the tunnel damage indexes; and correcting the disturbance probability demand model by using the updated mean value and standard deviation of the tunnel damage indexes, and updating the vulnerability curve in combination with tunnel damage state threshold division and a logarithmic normal probability distribution function. The probability of different damage degrees of the tunnel structure can be made to fit the reality, and the applicability of an existing disturbance damage database is improved.

The invention belongs to the technical field of power system optimization scheduling, and provides a source network load storage intelligent collaborative optimization method, which comprises the following steps of: deploying sensors at four ends of a source network load storage respectively, collecting in real time by utilizing a cloud data center, enabling data of the four ends to be consistent in time sequence through a PTP protocol, constructing a topological graph according to parameters and data, and establishing a source network load storage intelligent collaborative optimization system. Selecting a model in a digital twinning environment for simulation; dividing independent agents at four ends of a source network load storage, setting observation data, an execution space and excitation feedback, forming an excitation item by economy, stability and environmental protection, interactively circulating actual data, a prediction instruction and an excitation value, recording into a sequence, inputting the sequence into a strategy network, and calculating and outputting logarithmic probability gradient to update the parameters of the strategy network; and the intelligent agent completes interactive circulation according to the strategy network, generates a local scheduling instruction, aggregates the instruction to perform weighted calculation, generates a global scheduling scheme, issues the global scheduling scheme to execution equipment, updates parameters by using an average deviation calculated by a deviation vector, resolves the global scheduling scheme and issues the global scheduling scheme to form a closed-loop mechanism.

The invention relates to the technical field of power distribution cabinet detection, and discloses an intelligent power distribution cabinet state dynamic monitoring method and system, and the method comprises the following steps: collecting multi-source data generated in the operation process of a power distribution cabinet; preprocessing the multi-source data to generate a standardized data set; a random field model is constructed and used for describing the spatial relevance between units in the power distribution cabinet and the relation between the state of each unit and multi-source data, and model parameters are estimated through the maximum logarithmic posterior probability. An intelligent power distribution simulation model and a multi-source data acquisition technology are adopted, multi-dimensional parameters are acquired in real time through a high-precision sensor, monitoring efficiency and decision accuracy are remarkably improved, data analysis integrity and model sensitivity are ensured through dynamic time modeling and capturing of space and time relevance of the power distribution cabinet, and the power distribution cabinet can be monitored more accurately. And the state distribution visualization and maintenance priority strategy generation module optimizes the maintenance plan, reduces the operation and maintenance cost, and guarantees the stability of the power distribution system.

The embodiment of the invention provides a multi-modal data fused postoperative plastic surgery infection early warning method and system. The method comprises the following steps: firstly, performing quality evaluation and restoration on postoperative wound images, clinical texts, wearable vital signs and baseline features; performing double-flow coding and priori segmentation on the repaired image to obtain a coarse mask; constructing a super-pixel graph, obtaining a topological consistent fine mask through a graph neural network, and quantifying the erythema / exudation area; the text and the vital signs are coded respectively and then fused with the image features and the masks through cross-attention and self-attention, and a fusion implicit vector and uncertainty are obtained through quality weight gating; an individualized dynamic threshold value is constructed by combining baseline risk, logarithmic probability calibration and a sliding window trend, a confidence band is output by using conformal prediction, and early reliable infection early warning is realized.

The invention discloses a multi-modal visual language model illusion relieving method based on illusion induction optimization, and relates to the field of artificial intelligence and computer vision. According to the method, when a generative large language model generates new word marks, firstly, the probability of each word mark in a vocabulary is generated according to an input mark sequence, the contrast ratio of the current input mark sequence is calculated through a contrast model, the contrast ratio is mapped into a preset vector space, and the corrected logarithmic probability is obtained. And selecting a plurality of illusion tokens with the maximum contrast from the illusion tokens, and adjusting the probability of the illusion tokens according to a preset optimization strategy. If the illusion token belongs to the selected illusion token, amplifying the probability of the illusion token according to preset optimization intensity; otherwise, the current probability is the corrected probability. And selecting a new word mark according to the corrected probability, and circulating the process to generate a final text. According to the method, by introducing the illusion induction optimization strategy, the illusion phenomenon generated when the large visual language model generates the text is remarkably reduced.

Techniques for autocontrastive decoding of a machine learning model are provided. In one aspect, a system for machine learning includes: a multi-layer machine learning model; and an autocontrastive decoding module configured to obtain prediction probabilities from multiple, different layers of the multi-layer machine learning model as data propagates through the multi-layer machine learning model, and aggregate in a contrastive manner the prediction probabilities from the multiple, different layers of the multi-layer machine learning model to provide a final output from the multi-layer machine learning model. The multi-layer machine learning model can be a transformer-based machine learning model. The autocontrastive decoding module can be configured to redistribute a prediction probability distribution of the transformer-based machine learning model by maximizing a difference between log-probabilities of a final layer and one or more intermediate layers of the transformer-based machine learning model. A machine learning method using the present system is also provided.

The invention discloses a multi-dimensional knowledge distillation method based on Logit cross correction, and the method comprises the steps: inputting a to-be-classified image into a primary feature extraction module of a teacher model and a primary feature extraction module of a student model, and obtaining two primary features; secondly, inputting the two primary features into a category result sensing module to obtain a self-adaptive weighted Logit score, and calculating category result loss; and then inputting the adaptively weighted Logit score and the primary feature into a Logit cross correction module and a category generation perception module, and calculating correction loss and category generation loss. And finally, performing weighted fusion on the category result loss, the correction loss and the category generation loss through variable weights to obtain a final loss value, and minimizing the loss value through back propagation to complete multi-dimensional knowledge distillation. According to the method, feature knowledge and Logit knowledge are combined, so that the problems of poor distillation interpretability and insufficient utilization of local semantic information in a knowledge distillation task are solved.

The invention provides an industrial data quality evaluation method based on micro-service, a medium and a system, and belongs to the technical field of electrical digital data process.The industrial data quality evaluation method comprises the steps that firstly, multi-source heterogeneous data is collected through a distributed network, and an evaluation system containing deviation rate, completeness rate and timeliness indexes is established; extracting features by using a deep learning model, and generating a feature probability distribution matrix; then, constructing a state transition matrix based on the characteristic entropy and the fluctuation coefficient, deploying an evaluation unit through the micro-service architecture, and establishing a transverse and longitudinal evaluation network to calculate information gain and association strength; and finally, performing feature fusion through an improved pyramid structure, realizing quality scoring by adopting a three-layer evaluation equation set, ensuring scientificity of a scoring result through logarithm mapping, and realizing dynamic evaluation of industrial data quality. The technical problem that in the prior art, an industrial data quality evaluation method is difficult to adapt to dynamic changes of multi-source heterogeneous data is solved.

The invention provides an unsupervised industrial anomaly detection method based on an improved full-convolution cross-scale flow network. The method comprises the steps of performing data preprocessing on an industrial image data set; industrial image data is used as input, a pre-trained visual backbone network is used for extracting multi-scale features, a resolution-perceived channel attention module is used for enhancing expression of the multi-scale features, and an enhanced multi-scale feature map is obtained; performing multilayer reversible transformation by taking the enhanced multi-scale feature map as input and taking an ICSF-Net model based on hierarchical attention and expansion convolution as a cross-scale normalized flow network, and modeling multi-scale feature distribution; in the training stage, optimizing and updating ICSF-Net model parameters by maximizing the log likelihood of a normal sample under potential Gaussian distribution based on multi-scale feature distribution and combining an LSGR mechanism; in the reasoning stage, probability density estimation is carried out based on multi-scale feature distribution, an abnormal score graph is generated, and defect detection and positioning are achieved. According to the invention, the accuracy, robustness and pixel-level positioning precision of industrial product defect detection are improved.

The invention discloses a distribution line early-stage fault time sequence hidden Markov modeling and identification method and system based on multistage evolution characteristics, and belongs to the field of distribution line early-stage fault identification. Comprising the steps of obtaining a current waveform sample sequence of an early fault in a distribution line; for each current waveform sample in the sequence, extracting a multi-dimensional time-frequency feature, and constructing a feature vector of each sample; performing fault stage identification on each sample by using the first-level hidden Markov model group, and outputting a fault stage tag sequence corresponding to each sample; combining the fault stage label sequences of the current sample and a plurality of previous historical samples to form a stage label sequence window; and respectively inputting the stage label sequence window into a second-level tree line fault hidden Markov model and a second-level non-tree line fault hidden Markov model, calculating a corresponding first average log-likelihood value and a corresponding second average log-likelihood value, comparing the two average log-likelihood values, and judging whether a current sample belongs to a tree line early fault or not.

The invention provides a control method and system for high-temperature protection of a hydraulic cylinder, and relates to the field of hydraulic cylinders, and the control method specifically comprises the steps that data are collected through temperature sensors arranged in a cylinder body, a piston rod, oil and the environment, and a three-dimensional temperature field model is constructed based on the Fourier heat conduction law; an oil temperature threshold value is set, a graded cooling mechanism is started when the threshold value is exceeded, and cooling heat is calculated according to a Newton cooling formula and a logarithmic average temperature difference method. Fuzzy PI D control is combined with a neural network prediction model, the cooling intensity is dynamically adjusted according to the temperature rising rate and the load state, and the high-temperature risk is pre-judged. The failure of the sensor is processed by a Kalman filtering algorithm, and the cooling system is automatically switched to a standby loop and alarms if the cooling system fails. The system comprises a temperature acquisition module, a control module, a cooling execution module, an alarm module and a remote communication module, and can solve the problems that an existing protection method is poor in passive cooling effect and unreasonable in timing cooling.

The invention relates to the technical field of power system operation and control, and particularly discloses a time-sharing electric quantity prediction method based on a logarithmic load density growth curve, which comprises the following steps of: firstly, performing causal detection and dynamic time-delay optimization on historical load and multivariate external data through convergence cross mapping and mutual information technologies, and constructing a causal time-delay feature set; and the problems of multi-element coupling and time-delay effect quantization are solved. Secondly, fitting a load trend by using time-frequency decomposition in cooperation with a segmented logistic model, extracting dynamic parameters representing a growth rate and a saturation capacity, and endowing the model with a sensing ability for a load evolution stage; then, causal features, growth parameters and load components are deeply fused through cross-domain modulation and a gating mechanism, the nonlinear modulation effect of an external environment on a load mode is explicitly modeled, and finally, a probability interval is generated in combination with quantile regression and residual error correction. According to the scheme, accurate and probabilistic prediction of the time-sharing electric quantity in a complex scene is realized, and the scientificity of an agent electricity purchase decision is improved.

The invention provides a civil structure deformation anomaly detection method based on time series data, and relates to the field of civil structure deformation anomaly detection. A disturbance intensity response value, a disturbance curvature and a local disturbance folding feature are constructed, a range adjustment enhancement value is formed by combining a symbol jump mark and a neighborhood disturbance difference value, a latent guide feature is generated based on the range adjustment enhancement value, a disturbance reconstruction feature is constructed through superposition of a diffusion residual error and an asymmetric difference item, normalization mapping is completed to obtain a normalization feature, and the normalization feature is obtained. A disturbance spinor modulation factor is constructed in combination with a nonlinear suppression correlation coefficient, a disturbance spinor tensor is formed through a logarithmic compression channel and a square amplification channel and by applying quadrature phase coding, a tensor potential mapping map is generated under a path coupling and self-coupling mechanism, disturbance energy offset and extreme value deflection are constructed based on the tensor potential mapping map, and disturbance energy offset and extreme value deflection are obtained. And a probability potential index is formed, and civil structure deformation anomaly detection model training is completed based on the probability potential index, so that civil structure deformation anomaly detection is realized.

The invention relates to the technical field of oil and gas field development, in particular to a fracture parameter inversion method based on a Bayesian neural network, which comprises the following steps: establishing a bottom hole net pressure conversion model based on an actual construction curve, and drawing a bottom hole net pressure curve; calculating a bottom hole net pressure index sequence and a corresponding time sequence; establishing a shaft bottom crack extension mode judgment criterion on the basis of a classic double logarithmic curve analysis method; taking the net pressure index sequence and the time sequence obtained in the previous step as input data, combining actual physical parameter constraints, and establishing an inversion fracture parameter model based on a Bayesian neural network; and inputting the pressure index sequence to be inverted into the Bayesian neural network model to obtain a specific fracture parameter inversion result. According to the technical scheme, the confidence interval of the prediction result can be given, the prediction result and uncertainty quantification capability can be synchronously provided, and the reliability and decision support value of the inversion result are greatly improved.

Techniques are provided for calibrating distillation learning from a teacher model to a student model. In particular, the present disclosure presents systems and methods that provide high quality and high speed convergence. That is, the proposed example system not only enables the distillation loss to be minimized at a probability mean in the probability domain of the teacher prediction distribution, but also provides fine (e.g., symmetric and / or strong) convex losses around logi t and / or optimal values in the probability domain (e.g., including away from the minimum) to encourage the rapid convergence (e.g., fast convergence) of gradient-based methods. And regardless of the distance from the minimum value).

The invention discloses an oil spill diffusion prediction method and device for realizing dispersant response, relates to the technical field of sea surface oil spill simulation, and aims to solve the problems that oil spill diffusion simulation of an existing oil spill model is not fine enough and cannot support the dispersant response. The method comprises the following steps: acquiring simulation parameters of an oil spilling area and dispersant action parameters; determining a target entrainment probability of the oil spilling area through an entrainment algorithm according to the dispersant action parameters and the simulation parameters; determining the target particle diameter of each oil particle in the oil spilling region by adopting an oil drop particle size formula and a logarithmic normal distribution probability density function according to the dispersant action parameter and the simulation parameter; determining the target position of each oil particle based on the initial position, the simulation parameter, the target entrainment probability, the target particle diameter and an oil spill model; determining a prediction result of oil spill area diffusion based on the target position; according to the oil spill diffusion prediction method provided by the invention, prediction of oil spill diffusion under response of the dispersant can be realized, and the accuracy of a prediction result is improved.

The invention discloses a deformable electron beam lithography machine beam spot size calculation method and system, and the method comprises the steps: S1, employing a hyperbolic cosine function as a primary function according to a relation curve between the current intensity and the beam spot position, and constructing a composite function model through translation and superposition; s2, assuming that the observation value is a noise version of the composite function model, and establishing a likelihood function; a logarithm likelihood function is maximized and converted into a minimum residual sum of squares, partial derivatives of the minimum residual sum of squares about all parameters are solved, and a gradient equation set is obtained; s3, calculating the measurement signal, and extracting key feature points in the curve; calculating an initial parameter value according to the position of the feature point: taking the initial parameter as input, and iteratively solving a gradient equation set to obtain an exact solution of each parameter; and S4, calculating the actual size, the central position and the sharpness of the left and right edges of the electron beam spot according to the exact solutions of the parameters. The method has the advantages of high calculation precision and the like.

The invention relates to a PAC adding control method, system and device for an intelligent water plant and a medium. The method comprises the steps that multiple anti-interference processing and standardized feature extraction are conducted on historical operation data, and a temperature flow coupling data set is constructed by implementing double barrel separation processing on the basis of the water inlet temperature and flow; in each subset, establishing a steady-state prediction equation of the PAC dosage and the inflow and outflow water turbidity through a dynamic balance equation and logarithmic space mapping; carrying out online parameter calibration by adopting a recursive mean square error minimization algorithm, and forming a self-adaptive control model in combination with offline simulation residual error correction; and deploying the adaptive control model to a PAC dosing control end, dynamically adjusting control parameters through error back propagation of a real-time data stream and a prediction result, and generating an optimal dosing instruction. The technical problems that in a traditional scheme, due to dynamic changes of water inlet conditions, model adaptability is poor, precision is poor, and adjustment depends on artificial experience are solved, and accurate automatic control over coagulant adding under complex working conditions is achieved.

The invention discloses a bridge support displacement online probability prediction method, which comprises the following steps: acquiring historical monitoring data of main beam temperature and support displacement, and introducing a group of false points to construct a probability prediction model based on a flow sparse Gaussian process, according to the approximate posterior distribution and the real posterior distribution, constructing a variational free energy boundary as a lower bound of a logarithmic marginal likelihood function, deriving an optimal variational free energy boundary and optimal approximate posterior distribution, and obtaining a specific expression of a probability prediction model; when new monitoring data is obtained, new monitoring data and old historical monitoring data are summarized in combination with new false points and old posterior distribution, new approximate posterior distribution is derived, new real posterior distribution is derived by using the Bayesian theorem, then model parameters and posterior distribution are updated online, and a new specific expression of the probability prediction model is obtained. According to the method, massive health monitoring flow data of the bridge is processed based on the flow sparse Gaussian process model, online probability prediction of bridge support displacement is achieved, and the method has high engineering application value.

The invention discloses a model reasoning method, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence. In the model reasoning process, various information is integrated, and the logarithm probability of candidate lexical elements to be generated is corrected, so that thinking switching of a model in the later stage of reasoning is avoided, and the reasoning efficiency is improved. In this way, the situation that the reasoning process is too long due to frequent switching of reasoning ideas is avoided, and waste of model reasoning resources is avoided while the accuracy of the model reasoning result is guaranteed.

The invention provides a zero-sample large model text detection method and system based on comparison prompt differential analysis, and the method comprises the steps: S1, coding a to-be-detected text into a lexical element sequence through a word segmentation device, and defining a human style prompt and a large model style prompt; s2, conditional probability calculation: respectively calculating the conditional probability of each lexical element position under two prompt contexts; s3, correction probability difference calculation: calculating a corrected logarithmic probability difference value of the lexical element at each position; s4, summarizing a difference sequence, and recording the difference values of all the lexical elements in the difference sequence in sequence; s5, parameter fitting and score generation: fitting negative exponential function parameters through a difference sequence; and S6, threshold classification judgment: comparing the fitted parameters with a preset threshold, and judging that the text is a text generated by a large language model or written by human. Based on the technical scheme of the invention, the detection can be completed without any training or fine tuning; the interpretability is high, and the word-level visualization capability is achieved; and the robustness is good and the fault tolerance is high.