188 results about "Linear correlation" patented technology

The invention provides a marine meteorological data quality control method based on a cross-parameter correlation network, which belongs to the technical field of marine meteorology, and comprises the following steps: constructing a cross-parameter correlation network model comprising an air temperature and air pressure correlation rule, a humidity and air temperature linear correlation rule and a wind speed and air pressure gradient extraction correlation rule; a sliding window algorithm is adopted to calculate a correlation coefficient in real time and identify correlation abnormity, a multivariate abnormity detection mechanism of four dimensions of parameter threshold overrun, spatio-temporal change rate abnormity, probability density distribution offset and correlation verification failure is established, a fuzzy comprehensive evaluation method is adopted to calculate a comprehensive abnormity index, and data abnormity is judged. And the authenticity of the abnormal data is confirmed in combination with a multi-sensor cross validation mechanism, and finally the abnormal data is restored by adopting a virtual sensor data reconstruction algorithm based on correlation network reverse calculation, so that the technical problem of poor abnormal data detection effect in the prior art is solved.

The invention discloses a model processing method and device, equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence, in the method, singular value decomposition is carried out on a network weight matrix of a hybrid expert model, and in an obtained right singular vector matrix, columns with singular values smaller than a threshold value are screened to serve as a null-space basis matrix. In this way, the null-space basis matrix can represent the activation mode direction which contributes little to the activation of the expert network. Furthermore, a sub-network weight matrix is constructed based on each column of the null-space basis matrix and the network weight matrix. Therefore, according to the maximum singular value and the minimum singular value of the sub-network weight matrix, the closeness degree of linear correlation between each column of the network weight matrix and the null-space base matrix can be determined, then the invalid feature sensitivity of each expert network can be determined, and according to the invalid feature sensitivity, the invalid feature sensitivity of each expert network can be determined. And cutting network parameters of the hybrid expert model. In this way, the model parameter quantity can be reduced.

The invention discloses an enterprise budget dynamic analysis and optimization method based on a multi-modal generative adversarial network and variational Bayesian optimization, and belongs to the technical field of enterprise financial intelligent management. Enterprise subitem budget and actual overhead data are input, and subitems and total difference are calculated to construct a difference matrix; a bidirectional LSTM layer of a variational auto-encoder (VAE) is used to extract time sequence features, a mean value and a variance of hidden variables are generated through a full connection layer, and an evidence lower bound loss function is optimized to realize dimension reduction coding of high-dimensional data; and constructing a generative adversarial network (GAN), and adopting a variational Bayesian optimizer to iteratively search an optimal strategy considering the rate of return on investment and the risk score based on a Gaussian process proxy model and an expectation improvement criterion. The method can effectively capture the non-linear correlation between budget subjects, dynamically generates an optimization strategy conforming to business constraints, improves the model generalization ability in a small sample scene through a self-evolution mechanism, and is suitable for budget management of enterprises in multiple industries.

The invention relates to the technical field of asphalt pavement performance evaluation, in particular to an asphalt pavement technical condition evaluation method based on relative entropy information fusion, and the method comprises the steps: S1, building two parallel probability distribution paths; s2, inputting the average maintenance interval time theta into a Bayesian dynamic linear model, and adjusting a PQI distribution mean value mu in real time through a prediction step and an updating step; s3, non-linear correlation between subitem indexes and the PQI is established, a logarithm-logarithm regression model is adopted for a truncated normal path, and a logarithm regression model of a scale parameter lambda is adopted for a Weibull path; s4, respectively calculating the KL divergence of the distribution functions under the two paths; and S5, solving the minimum value of the target function h by adopting an SLSQP algorithm, and iteratively optimizing a regression coefficient beta-beta through gradient descent by taking a least square analytical solution as an initial value. According to the method, the relative entropy information fusion technology is introduced, and a double-probability model of truncated normal distribution and Weibull distribution is combined, so that a dynamic and accurate evaluation system is constructed, and the reliability and the practical value of an evaluation result are improved.

The invention provides a rocket aircraft flow field correction method based on multi-fidelity data fusion, and the method comprises the following steps: 1, collecting low-fidelity data, and carrying out the construction of a low-fidelity data set: simulating the change process of an unsteady flow field around an object in a finite time step through a Reynolds time-average simulation method or a large vortex simulation method; 2, data correlation analysis is carried out, wherein a linear relation model yH = rho (x) yL + delta (x) of a non-viscous flow field yL and a viscous flow field yH of the rocket aircraft is established; step 3, neural network architecture design: constructing a low-fidelity data approximation network NNL; 4, performing hyper-parameter learning and optimization; defining a loss function; 5, performing data acquisition and preprocessing: performing high-fidelity viscous flow field data acquisition, and preprocessing the acquired high-fidelity and low-fidelity flow field data; and step 6, model training and verification.

The invention relates to the technical field of electric digital data processing, in particular to a real-time monitoring method and system for sludge solidification stirring equipment, and the method comprises the steps: obtaining multi-dimensional parameter data of the sludge solidification stirring equipment at a plurality of historical time nodes; constructing the parameter data of the same dimension into a parameter sequence in a set time window; acquiring the local fluctuation degree of the parameter data of any time node in the parameter sequence to which the parameter data belong, and the linear correlation degree between the parameter data and other parameter sequences; determining the parameter data greater than a linear correlation degree threshold as strong cooperation parameter data; and respectively calculating first abnormal scores of the parameter data and the strong cooperation parameter data of any time node by using an improved local outlier factor algorithm, averaging the first abnormal scores to obtain a second abnormal score, and responding to the parameter data of which the second abnormal score is greater than a set threshold value as abnormal parameter data. The problem that an existing algorithm is not high in detection precision is solved.

The invention provides a multi-state-quantity integrated high-voltage cable distributed online monitoring method and system, and belongs to the field of high-voltage cable monitoring, and the method comprises the steps: S1, deploying monitoring nodes, and constructing a distributed monitoring network; s2, acquiring multi-dimensional state data of the high-voltage cable, and performing preprocessing and feature extraction to form a state vector; s3, calculating linear correlation among the state vectors, calculating nonlinear correlation intensity by adopting a kernel correlation analysis method, and performing weighted cooperative calculation; s4, calculating a connection weight between nodes according to the monitored network topology, and executing a distributed consistency protocol; and S5, establishing a normal state reference vector and a fault state reference vector based on historical data, and calculating a comprehensive health degree evaluation value by adopting a relative distance method. According to the method, collaborative decision-making of multi-dimensional state vectors is realized by using a distributed consistency protocol, the comprehensive assessment capability of the overall health state of the high-voltage cable is formed, and the accuracy and reliability of fault identification are effectively improved.

This invention discloses a downconversion fluorescence thermometry method based on bilinear mode, implemented using a hardware system. The hardware system includes a 280nm light-emitting diode, two convex lenses, and a Tb... 3+ :Cs2Na 0.9 Ag 0.1 The system comprises InCl6 temperature-sensing material, a grating spectrometer, an oscilloscope, a signal generator, a computer, and a thermostat. The specific implementation steps are: signal excitation and acquisition; temperature calibration; actual temperature measurement; and high measurement accuracy: VPR mode linear correlation coefficient R² ≥ 0.998, FL mode linear correlation coefficient R² ≥ 0.992, avoiding measurement errors introduced by the exponential bias or correction term of traditional FIR technology, thus fundamentally improving temperature measurement accuracy.

The invention relates to the technical field of atmospheric boundary layer height prediction, and discloses a multi-meteorological parameter coupling-oriented atmospheric boundary layer height prediction method, which comprises the following steps of: acquiring meteorological data, and preprocessing the meteorological data; analyzing linear correlation between meteorological parameters and pollution parameters in the meteorological data and the height of an atmospheric boundary layer, and screening thermal characteristics, dynamic characteristics and pollution characteristics from the linear correlation as a multivariate characteristic set; constructing a prediction model based on a bidirectional long-short-term memory network and an attention mechanism, and performing automatic optimization on model hyper-parameters by adopting a Bayesian optimization method to obtain an optimized prediction model; and predicting the height of the atmospheric boundary layer by using the optimized prediction model. According to the invention, high-precision and high-reliability atmospheric boundary layer height prediction data support is provided for short-term pollution early warning and environment management.

The invention relates to the technical field of model construction, in particular to a displacement fluid and shale rock imbibition law numerical simulation method and system. According to the method, by quantitatively analyzing the linear correlation degree of granularity and experimental capillary constant distribution and the disturbance degree of local feature points, regularization parameters of a regression algorithm are dynamically adjusted, and on the basis of accurately-calibrated capillary constants, a contact angle is determined by combining permeability characteristic data of probe liquid, so that the permeability of the probe liquid is determined. According to the method, shale surface free energy is obtained through surface energy component theoretical model decoupling, the interface action state of displacement fluid, shale and crude oil is reflected through a three-phase contact angle determined through a three-phase equilibrium relation, and the adhesion work needed by shale oil stripping is obtained. According to the method, the high-precision fitting model of the granularity and the capillary constant is established, and the surface energy component theory is combined, so that more accurate calculation of the shale oil stripping adhesion work is realized, and a reliable numerical simulation basis is provided for improving the shale oil recovery ratio.

The present application relates to a kind of low wave number initial model establishment method, device and medium of depth domain seismic inversion, method includes: extracting the well logging data of target layer section, the logging attribute of the low wave number initial model to be built is calculated;According to well seismic calibration result, extract well seismic data of well logging section, and according to logging depth sampling interval is up-sampled, obtain well seismic data on well logging depth section;The amplitude attribute of well seismic data section is calculated;With the amplitude attribute of well seismic data section as independent variable, the logging attribute of the low wave number initial model to be built is function value, linear fitting is carried out;From the fitting result, the amplitude attribute of optimal linear correlation is selected, and the optimal slope parameter and optimal intercept parameter are obtained;The minimum amplitude attribute of corresponding depth domain seismic data body is calculated;Depth domain seismic data body amplitude attribute is converted into the attribute of low wave number initial model to be built;Low wave number initial model attribute to be built is carried out Gaussian filtering, and the final low wave number initial model is obtained.

The invention discloses a health state monitoring method and system based on big data multi-dimensional evaluation, and relates to the cross technical field of health monitoring, big data processing and quantum machine learning. Through multi-source health data collection and in combination with quantization coding and high-dimensional data dimension reduction technologies, core low-dimensional feature vectors are effectively extracted, further, nonlinear correlation features in the feature vectors are deeply mined through a quantum neural network model, comprehensive and accurate evaluation of the health state of the user is achieved, and the user health state evaluation accuracy is improved. The multi-dimensional and deep evaluation mode can reflect the health state of the user more accurately, and provides a more reliable basis for subsequent early warning and intervention.

The application discloses a design tide level value method based on short-term data error consideration in the field of ocean engineering technology, and aims to solve the problem of difficult engineering design tide level value in the area with short tide level data. The method comprises the following steps: calculating the design high and / or low tide level value of a long-term tide level station according to the annual maximum and / or minimum tide level sequence of the long-term tide level station, and estimating the sampling error; obtaining the same period hourly tide level data of a temporary tide level station and a long-term tide level station in the sea area where the engineering is located, and forming the daily maximum and / or minimum tide level sequence; establishing a simple linear correlation equation, and obtaining the regression line error and the sampling error of the correlation coefficient; and then calculating the design high and / or low tide level value of the temporary tide level station according to the above-mentioned data. The application is suitable for the moisture-proof and corrosion-proof of engineering structures, solves the problem of unreliable results when designing tide level value according to short-term tide level data, shortens the required long-term tide level data to 5 years, and changes the current situation that it is difficult to design tide level value when the long-term tide level data of the engineering sea area is less than 20 years.

The invention relates to a heavy-load locomotive wheel remaining service life prediction method based on an abrasion rule, and relates to the technical field of heavy-load locomotive operation and maintenance, and the method comprises the steps: collecting wheel basic data, turning repair data, operation data and line data; analyzing the linear correlation between the continuous variable and the abrasion loss by adopting a Pearson's correlation coefficient, carrying out hypothesis testing, and outputting a significance degree factor influencing the wheel abrasion; the wheel diameter average ten thousand kilometer abrasion loss and the wheel flange average ten thousand kilometer abrasion loss corresponding to all the working conditions are obtained; in each iteration, according to the current wheel state and the working condition, calling the average ten thousand kilometer abrasion loss of the wheel diameter and the average ten thousand kilometer abrasion loss of the wheel flange corresponding to each working condition; and training the random forest model by taking the running mileage, the abrasion loss and the line condition as input characteristics, and outputting a corrected accumulated running mileage prediction result. Scientific basis is provided for operation and maintenance management of heavy-duty locomotive wheels, and safety and economical efficiency of heavy-duty railway transportation are improved.

The present invention discloses a method for screening a riboswitch that specifically recognizes doxycycline, as well as its rational modification and application, relating to the field of gene expression regulation. The invention provides a doxycycline-specific riboswitch and establishes a high-throughput screening method based on flow cytometry. Computer-assisted prediction of riboswitch binding sites and calculation of mutation sites are used to obtain a doxycycline riboswitch with a further enhanced activation factor. Dose-response curves show that the riboswitch exhibits a linear correlation at 40-100 μg / L. The constructed whole-cell sensor is low-cost, highly stable, and simple to operate, making it suitable for rapid on-site detection.

Apparatus and methods for detecting non-linear data dependencies in machine learning are provided. The methods may include receiving a data set comprising a plurality of data features, the data set for processing by a machine learning model operating on a machine learning system. The methods may also include plotting the data set in multi-dimensional space and linearizing the plotted, multi-dimensional space by feeding the plotted data to an algorithm programmed to execute the Delaunay triangulation method. The methods may further include receiving from the algorithm the data set linearized in a faceted space and detecting linear correlations in the linearized data set. The methods may additional include identifying a first data feature that has a correlation coefficient with a second data feature that is greater than an accuracy metric and removing the first data feature from the data set to create a reduced data set.

This invention discloses a lightweight modeling method, apparatus, system, and storage medium for high-dimensional data meteorological forecasting tasks, belonging to the technical field of computation, extrapolation, or counting. The method constructs prediction vectors and target variables from the original high-dimensional data required for the target meteorological forecasting task. Through an improved variational autoencoder strategy, it reconstructs latent predictor variables that influence or potentially influence the target variables from the prediction vectors. It then selects latent variables highly correlated with the original predictor variables to construct a candidate predictor variable dataset. After removing candidate variables that are significantly uncorrelated with the target variables and those with strong linear correlations from the candidate predictor variable dataset, it performs key predictor variable selection and constructs a lightweight prediction model based on the key predictor variables. This invention significantly reduces the dimensionality of the model input data while maintaining prediction accuracy, and enhances the model's prediction accuracy by constructing predictor variables that influence the target variables and have clear physical meaning.

The invention relates to the technical field of rural carbon source abnormity diagnosis, and discloses a machine learning diagnosis method for rural carbon source abnormity. The method comprises the following steps: step 1, setting inspection points, acquiring management data of all carbon sources of a country, actual measurement data of all the inspection points and management data of fluctuation factors, and classifying to form a data set; 2, the distribution rationality of rural carbon source inspection points is evaluated, a coverage index is generated, different rural carbon source structures are adapted, and the comprehensive adaptation precision is high; 3, evaluating the linear correlation degree of the greenhouse gas total CO2 equivalent of all inspection points and the linear distance between the inspection points, generating a correlation coefficient, and accurately judging the risk of carbon source leakage; 4, a fixed threshold value is set, corresponding management measures are triggered, the problems of insufficient routing inspection coverage, carbon source leakage and the like are rapidly solved, and the prevention and control timeliness is improved; and 5, monitoring and evaluating the abnormal degree of the rural carbon source, generating an abnormal score, and triggering management measures, so that the abnormal response is more timely.

This invention discloses a method and system for joint inversion of structural constraints based on Cauchy's inequality. It proposes a Cauchy inequality constraint applicable to structural constraints, achieving a linear correlation between the model's gradient vector set. Simultaneously, it calculates structural constraints separately according to different directions, constructing a structural constraint inversion. The directions can be freely combined as needed, including four directions: horizontal, vertical, 45° inclined, and 135°. The corresponding gradients and partial derivatives are obtained by discretizing the directions, with different weighting coefficients configured. Different directional combinations can better simulate the distribution characteristics of actual geological subsurface space. This method overcomes the limitation of previous structural constraint inversion techniques that lack consideration of the spatial differences of actual geological bodies. Furthermore, by explicitly expressing the gradients in different directions, it reduces the computational complexity of joint inversion using Cauchy's inequality constraints.

The invention relates to a qualitative and quantitative method for liensinine perchlorate. The qualitative method comprises the following steps: weighing 1mg of liensinine perchlorate and 1mg of liensinine, uniformly grinding and tabletting the liensinine perchlorate and the liensinine with 100mg of potassium bromide respectively, acquiring an infrared spectrum of 4000-400cm <-1 > by adopting an infrared spectrometer, and comparing spectrograms to find that the peak shapes of the liensinine perchlorate and the liensinine are similar, the peak shapes of the liensinine perchlorate and the liensinine perchlorate are common characteristic peaks and the peak shapes of the liensinine perchlorate are unique characteristic peaks of the liensinine perchlorate and the liensinine are common characteristic peaks and the peak shapes of the The quantitative method comprises the following steps: weighing a series of liensinine perchlorate reference substances with different masses, uniformly grinding and tabletting the liensinine perchlorate reference substances and 100mg of potassium bromide respectively, collecting an infrared spectrum, carrying out integration on a 625cm <-1 > characteristic peak to obtain a peak area, and drawing a standard curve by taking the peak area as a vertical coordinate and taking the sampling mass of the reference substances as a horizontal coordinate; the method comprises the following steps: weighing a sample to be detected, collecting an infrared spectrum, integrating a 625cm <-1 > characteristic peak to obtain a peak area, substituting the peak area into the standard curve equation, calculating the mass of the liensinine perchlorate in the sample, and substituting the mass into the content calculation formula to obtain the content of the liensinine perchlorate in the sample. The method is good in precision, repeatability and stability, wide in linear range, high in linear correlation coefficient and high in recovery rate, the problem that liensinine interference cannot be avoided in an existing content determination method is effectively solved, whether liensinine perchlorate is generated or not can be qualitatively judged, and the quantity of liensinine perchlorate generated can be quantitatively detected; and a feasible new method is provided for production process optimization and quality evaluation of liensinine perchlorate.

Disclosed are a monitoring method and a monitoring device for irrigation water use (IWU), a computer device and a non-transitory computer readable medium. The method includes: determining a linear correlation between a first evapotranspiration (ET), a first soil water storage, and a monthly reference ET of a target region according to a first data set in response to a monitoring instruction from a user; determining multiple second soil water storages of the target region based on a second data set, and the linear correlation of the first ET, the first soil water storage, and the monthly reference ET of the target region; determining the IWU of the target region according to the first ET, the second ET, the first soil water storage, the second soil water storage of the target region, and the first preset correlation.

The present invention provides a method and system for fitting interference fringes for a cold atom interferometer gyroscope. The method comprises: obtaining a mathematical model of the interference fringes of the cold atom interferometer gyroscope, linearizing the interference fringes model based on trigonometric functions to obtain a corresponding linear function; solving the linear correlation coefficient in the linear function using the least squares method based on the currently sampled transverse fringe phase and atomic transition probability; and expanding the range of fringe initial phase values. Within the expanded range of fringe initial phase values, the interference fringes are fitted based on the linear correlation coefficient and the real-time sampled transverse fringe phase and atomic transition probability. This solution ensures the accuracy of interference fringe fitting and improves the fitting speed, meeting the real-time requirements of interference fringe fitting.

The application is a modified asphalt mixing equipment monitoring method and system based on MEMS, and relates to the technical field of equipment state monitoring, comprising: a redundant sensor deployment and pairing module; a consistency index real-time calculation module; a judgment and alarm triggering module.In the application, the standardized deployment and pairing of the redundant sensors eliminates the system error of traditional sensors due to parameter and installation differences from selection, installation to software binding; the two-dimensional analysis of the time domain linear correlation coefficient and the frequency domain energy correlation coefficient is adopted, and then the weight factor is fused into a consistency index, which not only avoids the limitations of single-dimensional judgment, but also comprehensively and accurately identifies the potential faults of the sensor itself and the key components of the equipment, effectively reduces false positives and false negatives, and provides a reliable basis for the preventive maintenance of the modified asphalt mixing equipment.

Disclosed are a method for screening a riboswitch capable of specifically recognizing doxycycline as well as rational engineering and application of the riboswitch, and belongs to the field of gene expression regulation. The present disclosure provides a doxycycline specific riboswitch, and establishes a high-throughput screening method based on flow cytometry, and obtains the doxycycline riboswitch with further improved activation fold by way of computer-aided prediction of binding sites of the riboswitch and calculation of mutation sites. According to a dose-response curve, the riboswitch shows linear correlation at 40-100 μg / L, and meanwhile, a whole-cell sensor constructed is low in price, high in stability and simple to operate, and can be used for on-site rapid detection.

The present invention discloses a calculation method and device, electronic device, and storage medium for time downscaling of river material flux. The method comprises: collecting water quality data, hydrological data, and meteorological data of a river; calculating the daily scale flux of river material based on the data and setting it as a response variable; accumulating and calculating the monthly scale flux based on the daily scale flux, and forming a prediction variable of a model with the hydrological data and meteorological data; constructing a random forest based on the response variable and the prediction variable, and calculating the feature importance of the prediction variable; optimizing the prediction variable based on the feature importance result of the prediction variable; retraining the random forest model based on the optimized prediction variable, fitting the daily scale material flux, and evaluating the overall performance based on the linear correlation result with the response variable value; inputting the new monthly scale flux of the river and the hydrological data and meteorological data into the random forest prediction model, estimating the predicted value of the daily scale flux of the river material, and completing the material flux downscaling.

The application discloses a detection method for cathode fluid distribution of a proton exchange membrane fuel cell stack, which is characterized by utilizing the relationship between the characteristic frequency of the flow channel impedance and the stoichiometric ratio of the cathode fluid (f ch The linear correlation between 1 / ln(1-1 / λ) and 1 / ln(1-1 / λ) is used to determine the cathode fluid distribution of each cell in the stack. The detection method is simple, practical and effective. All parameters are extracted from the real-time measured electrochemical impedance spectrum, without the need for prior correction, which is of great significance for high consistency stack design and health state monitoring of each cell in the stack.

The invention discloses a financial risk control method and device based on a large language model, equipment and a medium, and relates to the technical field of finance, and the method comprises the steps: obtaining an initial financial risk control feature set, determining a corresponding amplification template from a preset template library, carrying out the amplification of the initial financial risk control feature set based on each amplification template, and obtaining a final financial risk control feature set; obtaining an amplified financial risk control feature set; grouping the financial risk control features in the amplified financial risk control feature set to obtain a grouped financial risk control feature set; and filtering the grouped financial risk control feature set according to the linear correlation among the financial risk control features to obtain a filtered financial risk control feature set, compressing the filtered financial risk control feature set to obtain a compressed financial risk control feature set, and storing the compressed financial risk control feature set in the financial risk control feature set. And generating a target financial risk control rule set by using the compressed financial risk control feature set and the target large language model. The risk control rule set is automatically generated by using the target large language model, so that the dependence on experts is avoided.

The application relates to the technical field of network catering services, in particular to a catering takeout platform scoring method based on XGBoost-PCA, which comprises the following steps: constructing a modular scoring system; performing normalization processing on the data of each module; calculating the original module scores of each module based on an XGBoost algorithm; performing dimension reduction on the original module scores by adopting a PCA principal component analysis algorithm; and performing weighted summation on the module dimension reduction scores to obtain a final platform score; the application constructs multiple decision trees by XGBoost, calculates the gain of different features at a split node, identifies the index with the largest contribution to the prediction result and eliminates the redundant weak features, and further combines PCA to analyze the linear correlation between the features, eliminates the linear redundancy by maximizing the variance, and obtains the dimension reduction comprehensive score, which removes the repeated potential factors in the original data, avoids repeated calculation of these factors in subsequent evaluation, and thus ensures the objectivity of the evaluation.

The invention belongs to the technical field of drainage equipment fault diagnosis, and particularly relates to a water pump room drainage equipment abnormity diagnosis method and device. The method comprises the following steps of collecting multi-source time sequence data of a water pump in a complete operation period, analyzing linear correlation between water pump flow and pipeline pressure based on a Pearson's correlation coefficient, diagnosing water leakage of a pipeline valve, and performing clustering analysis on the water pump flow data by adopting a K-means clustering algorithm, the water pump packing water leakage time period is recognized, and the actual drainage capacity is calculated; the device comprises an attribute monitoring component, a server, an alarm platform and a visual platform. According to the method, the subjectivity of manual inspection and the dependence of deep learning on samples and computing power are overcome, automatic, accurate and self-adaptive diagnosis of the abnormality of the drainage equipment is realized, and the diagnosis efficiency and reliability are remarkably improved.

The application relates to the technical field of welding process tracing, and discloses a welding process tracing method based on multi-source data fusion; first, welding multi-source data is collected and data cleaning is performed, entity types and relation types are set based on a rule base knowledge extraction method, and attribute types are matched to output welding process triplets; second, process parameter correlation mining is performed, linear correlation coefficients between process parameters are calculated through correlation analysis, welding process control relations are identified, quality relation weights are calculated, and a welding process knowledge graph is established; then, a welding process quality prediction model is established based on a full connection neural network, welding process quality is predicted, and welding process quality prediction results are output; finally, a multi-path fusion and sorting algorithm is used to determine a tracing priority and find a reverse tracing path; the welding multi-source data is fused through the knowledge graph, the purpose of welding process tracing is achieved, and the method is objective and accurate.