Cement-water glass gel strength evaluation method and system based on big data

Abstract

The invention discloses a cement-water glass gel strength evaluation method and system based on big data. The method comprises the steps of data acquisition, feature mapping, hidden variable estimation, hybrid model training, interpretability output and strength evaluation. The invention relates to the technical field of intelligent evaluation of materials, in particular to a cement-water glass gel strength evaluation method and system.The cement-water glass gel strength evaluation method based on big data comprises the steps that persistent coherent information is converted into continuous persistent images with physical weights and discretized into vectors, and differentiable microcosmic features are constructed in combination with spectral peak features, porosity and particle size; by taking generalized Afframi dynamics as a skeleton, mapping microscopic characteristics into kinetic parameters, introducing temperature and humidity coupling, and realizing online estimation and uncertainty quantification of gel reactivity by combining extended Kalman filtering; a hybrid model of physical prior and neural network correction is constructed, and physical constraint and data-driven training are combined, so that the accuracy, physical consistency and generalization ability of evaluation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material intelligent evaluation, and particularly relates to a cement-sodium silicate gel strength evaluation method and system based on big data. BACKGROUND

[0002] With the rapid development of big data, the Internet of Things and artificial intelligence, material science and engineering are evolving towards high-throughput characterization, online sensing and digital twinning. The improvement of a large number of characterization means and computing power enables the acquisition of multi-modal large sample data of the formulation, environment, microstructure and mechanical properties of the cement-sodium silicate gel, a multi-scale dynamic system, which promotes the transformation from empirical experiments to data-driven and physically coupled real-time prediction, monitoring and closed-loop control. At the same time, the industry has increasingly high requirements for interpretability, uncertainty quantification and engineering deployability.

[0003] Traditional methods mainly rely on offline experimental design, artificial feature engineering or single physical models, which have many shortcomings: feature extraction often ignores the continuous differentiable representation of topology and spectrum, making it difficult to be used for end-to-end learning; pure physical or empirical models are difficult to absorb complex micro-characterization and sample heterogeneity, and are prone to underfitting or loss of generalization ability; pure data-driven methods are flexible, but lack physical constraints, are prone to non-physical solutions, and usually cannot provide reliable uncertainty estimates; in addition, traditional processes are not suitable for real-time estimation of unobservable states and maintaining stable prediction ability under sensor data drift or formulation changes, thus being difficult to meet the needs of real-time evaluation, interpretable decision-making and reliability assurance in modern production. SUMMARY

[0004] In view of the above problems, in order to overcome the defects of the prior art, the cement-sodium silicate gel strength evaluation method and system based on big data are provided. In view of the problems that the traditional feature extraction method has discontinuous, non-differentiable features, and it is difficult to associate the micro-topological information with the physical formulation, the present application converts the persistent homology information into a continuous persistent image with physical weights and discretizes it into a vector, and at the same time extracts the physical quantitative indicators of peak area and half-peak width from the diffraction spectrum, to construct a micro-feature vector coupled with the formulation and pore size; in view of the problems that the traditional dynamic modeling and empirical fitting method is difficult to estimate the gel reaction rate in real time, difficult to couple micro-features with environmental disturbances, and lack of uncertainty quantification, the present application uses a generalized physics-informed neural network to construct a multi-scale dynamic model, which can simultaneously estimate the gel reaction rate, micro-structure and mechanical properties, and provide uncertainty estimates. The dynamics is a physical skeleton, the micro feature vector is mapped to the dynamics parameter, the temperature and humidity coupling is introduced, the extended Kalman filter is combined to carry out prior promotion and observation update on the hidden variable gel degree; in view of the problems that the traditional pure physical model is prone to underfitting and the pure data-driven model is prone to losing physical constraints, a hybrid predictor composed of a physical prior strength curve and a differentiable neural network output correction term is constructed, data fitting error, physical residual consistency penalty and monotonicity constraint are jointly used for training, so that the prediction not only follows the dynamics physical consistency but also has the flexible correction ability of data driving.

[0005] The technical scheme adopted by the present application is as follows: a cement-silicate gel strength evaluation method based on big data, which comprises the following steps: Step S1: data acquisition, collecting experimental formula records, time series feature data, micro characterization data and strength labels of historical samples at corresponding moments; Step S2: feature mapping, extracting differentiable topological and frequency domain features from the micro characterization data, obtaining a continuous persistent image with physical weight and discretizing it, and constructing a micro feature vector; Step S3: hidden variable estimation, establishing a low-dimensional state space model with the gel reaction degree as a hidden variable, mapping the dynamics parameters by using the micro features and time series feature data, and predicting and updating by using the extended Kalman filter to obtain a virtual measurement; Step S4: hybrid model training, constructing a hybrid predictor composed of a physical prior and a data-driven correction term, the physical prior is given by the micro feature vector and the virtual measurement, and fitting error, physical residual and monotonicity penalty are added during joint training; Step S5: explainability output, providing confidence intervals and local feature contributions for the prediction results, using deep integration and heterogeneous variance regression to model the mean and data noise at the same time, and using local linear approximation to make explicit feature contribution; Step S6: strength evaluation, performing feature mapping and hidden variable estimation on the sample to be evaluated, inputting the micro feature vector, time series feature data and virtual measurement into the hybrid model, and outputting the strength prediction value, confidence interval and local feature contribution in real time.

[0006] Further, in step S1, the data acquisition collects historical cement-silicate gel experiment and formula data, time series feature data, micro characterization data and strength label data; the experiment and formula data includes: unique identifier, water-binder ratio, silica mass fraction, sodium oxide mass fraction, type and mass fraction of admixture, initial mixing temperature, mixing time; the time series feature data includes: temperature, humidity, solution pH, conductivity; the micro characterization data includes: electron micrograph, X-ray diffraction spectrum, Fourier infrared spectrum, porosity, particle median diameter; the strength label data refers to the compressive strength of the cement-silicate gel.

[0007] Further, in step S2, the feature mapping constructs a differentiable topological and frequency domain feature mapping for micro-characterization data to obtain continuous persistent image vectors while extracting physical quantification indicators of peak position and peak intensity, specifically including the following steps: Step S21: Constructing a morphological subset, extracting binary sub-layers from electron micrographs and constructing an upstream filtering sequence to generate a morphological subset; Step S22: Obtain persistent images, calculate persistent homology information generated by morphological filtering, and only take 0 and 1 dimensions: obtain persistent images; Step S23: Persistent image transformation, perform a physical weight transformation on the persistent image; Step S24: Constructing micro-feature vectors, discretizing the persistent image after the physical weight transformation on the fixed grid to obtain vectorized features; then extracting spectral energy features from the Fourier infrared spectrum; constructing the final micro-feature vector.

[0008] Further, in step S3, the latent variable estimation establishes a low-dimensional physical driving state space model, regarding the gel degree as an unobservable latent variable, and uses the time series features and micro-features to perform posterior estimation on the latent variable through extended Kalman filtering to obtain the estimated virtual measurement, specifically including the following steps: Step S31: Constructing a dynamic model, using a generalized model with temperature and humidity coupling and mapping the micro-features to the dynamic constants through parameters; Step S32: Discretization, forward Euler discretization of the latent variable, and construction of the observation vector; Step S33: Extended Kalman filtering, performing posterior estimation on the latent variable through extended Kalman filtering. First, combine the time step and the dynamic rate information to obtain the posterior estimation of the latent variable at the current time; then calculate the covariance matrix; finally, perform overall update.

[0009] Further, in step S4, the hybrid model training constructs a physically regularized hybrid predictor, taking the micro-feature vector, time series features, and virtual measurement as input to predict the target compression strength, specifically including the following steps: Step S41: Constructing an intensity prior physical model, taking the parameters determined by the micro-feature vector and the time-varying virtual measurement as input, and outputting a physically empirical form of the intensity prediction curve over time; Step S42: Define a data-driven correction term, which takes the micro-features, time series features, and virtual measurements as input through a differentiable neural network, and outputs a correction to the physical prior; Step S43: final prediction, add the physical prior prediction and the data-driven correction term to obtain the final hybrid model strength prediction value as the model output during training and inference; Step S44: training loss, the model loss includes data fitting error, physical residual penalty, monotonicity penalty.

[0010] Further, in step S5, the explainability output provides confidence interval and local feature contribution for the prediction result, adopts deep integration and heterogeneous variance regression to simultaneously model the mean and data noise, and then uses local linear approximation to make explicit feature contribution, and specifically includes the following steps: Step S51: model integration, integrate the mean and variance outputs of Step S44: training loss, the model loss includes data fitting error, physical residual penalty, monotonicity penalty. Step S44: training loss, the model loss includes data fitting error, physical residual penalty, monotonicity penalty. Step S52: generate confidence interval, give the confidence interval based on the approximate Gaussian assumption, and the confidence interval width is determined by the total standard deviation. Step S53: generate contribution, estimate the feature contribution for the input vector. Step S54: uncertainty splitting, split the total variance into two parts, the first part is the process uncertainty of the dynamic hidden variable propagated to the intensity, and the second part is the observation uncertainty introduced by the data-driven part and the observation noise. Step S55: chain propagation, map the extended Kalman filter posterior covariance to the intensity prediction uncertainty through the physical prior model based on the chain rule.

[0011] Further, in step S6, the strength evaluation, acquires the cement-silicate gel data to be evaluated, inputs the hybrid model, and uses the hybrid model to predict the strength of the cement-silicate gel in real time as the evaluation strength, and generates the explainability output.

[0012] The cement-silicate gel strength evaluation system based on big data provided by the application comprises a data acquisition module, a feature mapping module, a hidden variable estimation module, a hybrid model training module, an explainability output module and a strength evaluation module. The data acquisition module acquires historical cement-silicate gel experiment and formula data, time series feature data, microscopic characterization data and strength label data, and sends the data to the feature mapping module, the hidden variable estimation module, the hybrid model training module, the explainability output module and the strength evaluation module. The feature mapping module receives data sent by the data acquisition module, constructs a differentiable topological and frequency domain feature mapping for microcosmic characterization data, obtains a continuous persistent image vector, simultaneously extracts physical quantitative indicators of peak position and peak strength, and sends data to the latent variable estimation module; The latent variable estimation module receives data sent by the data acquisition module and the feature mapping module, establishes a low-dimensional physical driving state space model, regards the gel degree as an unobservable latent variable, uses time series features and microcosmic features to perform posterior estimation on the latent variable through extended Kalman filtering to obtain an estimated virtual measurement, and sends data to the hybrid model training module; The hybrid model training module receives data sent by the data acquisition module and the latent variable estimation module, constructs a physically regularized hybrid predictor, takes the microcosmic feature vector, time series features and virtual measurement as input, predicts the target compression strength, and sends data to the explainability output module; The explainability output module receives data sent by the data acquisition module and the hybrid model training module, provides a confidence interval and local feature contribution for the prediction result, uses deep integration and heterogeneous variance regression to simultaneously model the mean and data noise, and uses local linear approximation to explicitly express the feature contribution, and sends data to the strength evaluation module; The strength evaluation module receives data sent by the data acquisition module and the explainability output module, obtains cement-silicate gel data to be evaluated, inputs the hybrid model, uses the hybrid model to predict the strength of the cement-silicate gel in real time as the evaluation strength, and generates explainability output.

[0013] The above scheme has the following beneficial effects: (1) In view of the problems of traditional feature extraction methods, such as discontinuous features, non-differentiable features, and difficulty in associating micro-topological information with physical formulations, the scheme converts persistent homology information into a continuous persistent image with physical weights and discretizes it into a vector, simultaneously extracts physical quantitative indicators of peak area and half-peak width from the diffraction spectrum, constructs a microcosmic feature vector coupled with formulations and pore size, and thus obtains a differentiable, physically perceptible feature table that can be used for end-to-end training.

[0014] (2) In view of the problems of traditional dynamic modeling and empirical fitting methods, such as difficulty in real-time estimation of gel reaction rate, difficulty in coupling microcosmic features and environmental disturbances, and lack of uncertainty quantification, the scheme uses generalized dynamics as a physical skeleton, maps the microcosmic feature vector into dynamic parameters and introduces temperature and humidity coupling, and combines extended Kalman filtering to perform prior propagation and observation update on the latent variable gel degree, which realizes real-time virtual measurement estimation and provides posterior covariance for uncertainty propagation. ​

[0015] (3) In view of the problems that the traditional pure physical model is prone to underfitting and the pure data-driven model is prone to losing physical constraints, the scheme is a hybrid predictor composed of a physical prior strength curve and a differentiable neural network outputting a correction term, and jointly uses data fitting error, physical residual consistency penalty and monotonicity constraint for training, so that the prediction not only follows the dynamic physical consistency but also has the flexible correction ability of data driving, thereby improving the generalization and suppressing non-physical solutions. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A schematic diagram of the cement-silicate gel strength evaluation method based on big data provided by the present application is shown in the figure; Figure 2 A schematic diagram of the cement-silicate gel strength evaluation system based on big data provided by the present application is shown in the figure; Figure 3 A schematic diagram of step S2 is shown in the figure; Figure 4 A schematic diagram of step S3 is shown in the figure; Figure 5 A schematic diagram of step S4 is shown in the figure; Figure 6 A schematic diagram of step S5 is shown in the figure.

[0017] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, and are used to explain the present application together with embodiments of the present application, and do not constitute a limitation on the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0019] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0020] Embodiment one, refer to Figure 1 The cement-silicate gel strength evaluation method based on big data provided by the present application comprises the following steps: Step S1: data collection, collecting experimental formula records, time series feature data, microscopic characterization data and intensity labels of historical samples; Step S2: feature mapping, extracting differentiable topological and frequency domain features from microscopic images and diffraction spectra, obtaining continuous persistent images with physical weights and discretizing them, and constructing microscopic feature vectors; Step S3: hidden variable estimation, establishing a low-dimensional state space model with the degree of gel reaction as the hidden variable, mapping the dynamics parameters with microscopic features and time series features, and predicting and updating through extended Kalman filtering to obtain virtual measurements; Step S4: hybrid model training, constructing a hybrid predictor composed of physical priors and data-driven correction terms, the physical priors are given by the microscopic feature vectors and virtual measurements, and the fitting error, physical residual and monotonicity penalty are added during joint training; Step S5: explainable output, providing confidence intervals and local feature contributions for the prediction results, using deep integration and heterogeneous variance regression to model the mean and data noise, and then using local linear approximation to do explicit feature contribution; Step S6: strength evaluation, performing feature mapping and hidden variable estimation on the sample to be evaluated, inputting the microscopic feature vectors, time series feature data and virtual measurements into the hybrid model, and outputting the strength prediction value, confidence interval and local feature contribution in real time.

[0021] Embodiment two, see Figure 1 This embodiment is based on the above embodiment, in step S1, the data collection collects experimental and formula data, time series feature data, microscopic characterization data and strength label data of historical cement-silicate gel; the experimental and formula data includes: unique identifier , water-binder ratio , mass fraction of silicon dioxide , mass fraction of sodium oxide , type and mass fraction of admixture , initial mixing temperature , mixing time ; the time series feature data includes: temperature , humidity , solution pH , conductivity ; the microscopic characterization data includes: electron micrograph , X-ray diffraction spectrum , Fourier infrared spectrum , porosity , median particle diameter ; the strength label data refers to the compressive strength of cement-silicate gel .

[0022] Embodiment three, see Figure 1 and Figure 3 , based on the above embodiment, in step S2, the feature mapping, the micro-characterization data constructs a differentiable topological and frequency domain feature mapping, obtains a continuous persistent image vector, and at the same time extracts the peak position and peak intensity physical quantitative index from the diffraction spectrum, including the following steps: Step S21: Constructing a morphological subset, extracting binary sub-layers from electron micrographs and constructing an upstream filtering sequence, the threshold set is , generating a morphological subset , as follows: ; Wherein, represents the gray value of the electron micrograph at the pixel point p, represents the threshold sequence obtained by method, represents the binary morphological subset composed of threshold values; Step S22: Obtain persistent image, calculate persistent homology information generated by morphological filtering, only take 0 dimension and 1 dimension: get persistent image, as follows: ; Wherein, represents the persistent homology point set obtained by morphological filtering, represents the threshold pair of the kth topological feature, represents the persistent threshold, represents the abort threshold; Step S23: Persistent image transformation, the persistent image is transformed with physical weight, represented as: ; wherein, represents the persistent image after the physical weight transformation, represents a two-dimensional Gaussian kernel function, represents a Gaussian kernel covariance matrix, represents a transpose symbol, represents an exponential function with a natural constant as the base, represents a physical weighting function, defined as: ; wherein, represents linear normalization, normalized to , , and represent physical weighting hyperparameters; Step S24: Construct micro-feature vector, discretize on the fixed grid G to get the vectorized feature ; then extract the spectral energy features from the Fourier infrared spectrum, including: peak area with full width at half maximum ; constructing a spectrum vector ; obtaining a final micro-feature vector ; wherein, denotes vector concatenation, denotes a discretized persistent image vector with length p; and denote the peak area and the full width at half maximum of the i-th peak, respectively, denotes the peak area with full width at half maximum denotes the total number of peaks in the Fourier infrared spectrum, and denote the peak area and the full width at half maximum of the i-th peak, respectively, denotes the peak area with full width at half maximum denotes the final micro-feature vector.

[0023] By performing the above operations, in view of the problems of feature discontinuity, non-differentiability, and difficulty in associating micro-topological information with physical formulations existing in traditional feature extraction methods, the present scheme converts persistent homology information into a continuous persistent image with physical weights and discretizes it into a vector, simultaneously extracts physical quantitative indicators of peak area and half peak width from the diffraction spectrum, constructs a micro-feature vector coupled with formulations and pore size, and thus obtains a differentiable, physically perceptible feature table that can be used for end-to-end training.

[0024] In an embodiment, referring to Figure 1 and Figure 4 , this embodiment is based on the above-mentioned embodiments, in step S3, the latent variable estimation establishes a low-dimensional physically driven state space model, taking the gel degree as an unobservable latent variable , uses the time series features and the micro features to perform posterior estimation of the latent variable through extended Kalman filtering to obtain the estimated pseudo-measurement , specifically comprising the following steps: Step S31: constructing a dynamic model, adopting a generalized model with temperature and humidity coupling and mapping the micro features into dynamic constants through parameters, represented as follows: ; wherein, denotes the gel reaction degree, which is a latent variable, and the initial value ; and denote the reaction rate constant and the reaction order, respectively, which are mapped from the micro features and the environment; and denote the learnable dynamic weight parameters, and These represent the linear sensitivity coefficients of temperature and relative humidity to the rate, respectively. Indicates reference temperature. Indicates reference humidity. Indicates the initial reaction constant; Step S32: Discretization. Perform forward Euler discretization on the latent variables and construct the observation vector, as follows: ; in, Indicates the discrete time step. Indicates at time Discretized latent variables, Gaussian white noise represents the random perturbation in the dynamic model. Represents the observation vector. This represents the reaction rate constant at time t. Indicates the reaction order at time t; Step S33: Extended Kalman Filtering. The latent variables are estimated posteriorly using extended Kalman filtering to obtain the virtual measurement. ; First, combine the time step In addition to the dynamic rate information, we can advance to obtain the posterior estimated latent variables at the current time: ; in, Let represent the latent variables of the prior estimate at time t. Indicates at time The latent variable estimated posteriorly is the filter output from the previous time step. ; Indicates time The reaction rate constant, Indicates time The reaction stage; Then calculate the covariance matrix: ; in, Let represent the prior covariance matrix, and let represent the variance of the prior covariance matrix. Uncertainty, This represents the posterior covariance of the previous time step. Represents the process noise covariance. Let Jacobian matrix represent the state transition matrix, specifically the derivative of the state update mapping with respect to the state at point [0, 1]. The evaluation at this point is expressed as: ; Finally, perform a complete update: ; in, This represents the Kalman gain, used to map observation information into a weight matrix for state updates. The Jacobian matrix representing the observation mapping with respect to the state is taken at the prediction point. Location: ; Represents the observation noise covariance matrix; Let represent the posterior estimate at time t, where is the time t. The real-time virtual measurement estimate integrates observation information; Indicates time The actual observed vector, This represents an observational mapping based on predicted states and microscopic features; Indicates the posterior covariance. Represents the identity matrix.

[0025] By performing the above operations, this scheme addresses the problems of traditional kinetic modeling and empirical fitting methods, such as difficulty in real-time estimation of gel reactivity, difficulty in coupling microscopic features with environmental disturbances, and lack of uncertainty quantification. It utilizes a generalized... Using dynamics as the physical framework, microscopic feature vectors are mapped to dynamic parameters and temperature-humidity coupling is introduced. Combined with extended Kalman filtering, the latent variable gelation degree is advanced a priori and updated by observation. This not only realizes real-time virtual measurement estimation, but also provides posterior covariance for uncertainty propagation.

[0026] Example 5, see Figure 1 and Figure 5 This embodiment is based on the above embodiment. In step S4, the hybrid model training constructs a physically regularized hybrid predictor, which incorporates the micro-feature vectors. Time series characteristics and virtual measurement As input, predict the target compressive strength. Specifically, it includes the following steps: Step S41: Construct a priori physical model of intensity, using microscopic feature vectors Determined parameters and time-varying virtual measurements As input, the output is a predicted intensity curve over time based on physical experience, as shown below: ; in, This represents the compressive strength predicted a priori by physics. This represents the set of parameters of the physical prior model. Represents the microscopic eigenvectors The final intensity upper bound function obtained by mapping, This represents a small network with a three-layer perceptron structure. The scale parameter, which represents the microscopic characteristics, controls how quickly the intensity increases with reactivity. The power exponent, determined by microscopic characteristics, controls... The nonlinear delay effect; Step S42: Define a data-driven correction term, which is a differentiable neural network that accepts microscopic features, temporal features, and dummy measurements as inputs, and outputs a correction amount for the physical prior. ; in, This represents the intensity correction term of the neural network output. This represents a parameterized neural network mapper, with the parameter set as follows: , Represents a time-series feature data vector; Step S43: Final prediction. The physical prior prediction is added to the data-driven correction term to obtain the final mixed model strength prediction value, which is used as the model output during training and inference. ;in, This represents the intensity prediction by the model at time t; Step S44: Training loss, model loss includes data fitting error, physical residual penalty, and monotonicity penalty: ; in, Indicates the overall training loss. Represents the number of training samples. Indicates the training sample index. Denotes the set of observation times for sample i. This represents the weighting coefficient at sample time. The model represents the samples At any moment The prediction Indicates sample At any moment The measured compressive strength, Indicates the weight of the physical residual term. Indicates the monotonicity penalty weight. Indicates short-run differences, Set to 0.5 hours. This means that a penalty is only applied when the forecast shows a decline; the loss is a one-sided constraint, and the physical residual is... Defined as the consistency residual between the intensity model and the dynamic variables: .

[0027] By performing the above operation, in view of the problems that the traditional pure physical model is prone to underfitting and the pure data-driven model is prone to losing physical constraints, the scheme constructs a hybrid predictor composed of a physical prior strength curve and a differentiable neural network outputting a correction term, and jointly uses data fitting error, physical residual consistency penalty and monotonicity constraint for training, so that the prediction not only follows the dynamic physical consistency but also has the flexibility of data-driven correction, thereby improving the generalization and suppressing non-physical solutions.

[0028] Embodiment six, see Figure 1 and Figure 6 , this embodiment is based on the above embodiment, in step S5, the explainability output provides a confidence interval and a local feature contribution for the prediction result, adopts deep integration and heterogeneous variance regression to simultaneously model the mean and data noise, and then uses local linear approximation to make explicit feature contribution, which specifically includes the following steps: Step S51: model integration, integrating the mean and variance outputs of different initialization models of the same structure, the integration is obtained by independently training each model outputs: , ; the total variance is divided into two parts of observable noise and parameter uncertainty, and is represented as follows: ; wherein, denotes the number of integrated models, denotes the prediction mean of the mth model at time t, denotes the variance output by the mth model, and denotes the observable noise estimation; denotes the integrated mean prediction, denotes the total variance after integration; Step S52: generate confidence interval, give the confidence interval based on the approximate Gaussian assumption, and the confidence interval width is determined by the total standard deviation, and is represented as follows: ; wherein, denotes the closed interval of the confidence level , denotes the standard quantile point, and the total standard deviation; Step S53: generate contribution, for the input vector , the feature contribution is estimated as: ; wherein, denotes the contribution of the input component to the prediction at time , ​represents the combined input vector, including micro-features, time-series features, dummy measurements; represents the vector represents the first component of the vector represents the baseline value of the oth feature, which is the mean value of the training set; represents the coefficient of the local linear regression, which is the coefficient of the linear model fitted by weighting at the prediction point, and the kernel function uses the Gaussian kernel function; Step S54: uncertainty splitting, splitting the total variance into two parts, the first part is the process uncertainty caused by the uncertainty of the dynamic hidden variable propagating to the intensity, and the second part is the observation uncertainty introduced by the data-driven part and the observation noise, which is represented as follows: wherein, represents the variance component caused by the process model, which comes from the propagation of the hidden variable uncertainty of the extended Kalman filter, represents the observation uncertainty of the second part; Step S55: chain propagation, the extended Kalman filter posterior covariance is mapped to the intensity prediction uncertainty by the physical prior model, which is represented as follows: wherein, represents the partial derivative of the physical prior with respect to at the current time, using the current and .

[0029] Embodiment seven, see Figure 1 , which is based on the above-mentioned embodiment, in step S6, the intensity evaluation, the cement-silicate gel data to be evaluated is obtained, which is input into the mixed model, and the mixed model is used to predict the strength of the cement-silicate gel in real time as the evaluation strength, and an interpretable output is generated.

[0030] Embodiment eight, see Figure 1 and Figure 2 , which is based on the above-mentioned embodiment, the cement-silicate gel strength evaluation system based on big data provided by the application, comprising a data acquisition module, a feature mapping module, a hidden variable estimation module, a mixed model training module, an interpretable output module and an intensity evaluation module; The data acquisition module acquires historical cement-silicate gel experiment and formula data, time-series feature data, micro-characterization data and strength label data, and sends the data to the feature mapping module, the hidden variable estimation module, the mixed model training module, the interpretable output module and the intensity evaluation module.​​ The feature mapping module receives the data sent by the data acquisition module, constructs a differentiable topological and frequency domain feature mapping for the microcosmic characterization data, obtains a continuous persistent image vector, extracts physical quantitative indicators of peak position and peak intensity from the diffraction spectrum, and sends the data to the latent variable estimation module; The latent variable estimation module receives the data sent by the data acquisition module and the feature mapping module, establishes a low-dimensional physical driving state space model, regards the gel degree as an unobservable latent variable, uses the time series features and microcosmic features to perform posterior estimation on the latent variable through an extended Kalman filter, obtains an estimated virtual measurement, and sends the data to the hybrid model training module; The hybrid model training module receives the data sent by the data acquisition module and the latent variable estimation module, constructs a physically regularized hybrid predictor, takes the microcosmic feature vector, the time series features and the virtual measurement as inputs, predicts the target compressive strength, and sends the data to the explainability output module; The explainability output module receives the data sent by the data acquisition module and the hybrid model training module, provides a confidence interval and a local feature contribution for the prediction result, uses deep integration and heterogeneous variance regression to simultaneously model the mean and data noise, uses local linear approximation to make an explicit feature contribution, and sends the data to the strength evaluation module; The strength evaluation module receives the data sent by the data acquisition module and the explainability output module, obtains the cement-silicate gel data to be evaluated, inputs the hybrid model, uses the hybrid model to predict the strength of the cement-silicate gel in real time as the evaluation strength, and generates an explainability output.

[0031] It should be noted that in this document, the terms "comprise", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to such a process, method, article or device.

[0032] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application.

[0033] The above describes the present application and its embodiments, which are not limited, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired thereby, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution are not creative, and should belong to the protection scope of the present application.

Claims

1. A method for evaluating the strength of a cement-silicate hydrate gel based on big data, characterized by, The method comprises the following steps: Step S1: data collection, collecting experimental formula records, time series feature data, microcosmic characterization data and intensity labels of historical samples; Step S2: feature mapping, extracting differentiable topological and frequency domain features from the microcosmic characterization data, obtaining continuous persistent images with physical weights and discretizing them, and constructing a microcosmic feature vector; Step S3: hidden variable estimation, establishing a low-dimensional state space model with the degree of gel reaction as a hidden variable, mapping the dynamics parameters with the microcosmic features and time series feature data, and predicting and updating through extended Kalman filtering to obtain a virtual measurement; Step S4: hybrid model training, constructing a hybrid predictor composed of a physical prior and a data-driven correction term, the physical prior being given by the microcosmic feature vector and the virtual measurement, and the fitting error, physical residual and monotonicity penalty being added during joint training; Step S5: explainable output, providing confidence intervals and local feature contributions for the prediction results, using deep integration and heterogeneous variance regression to model the mean and data noise at the same time, and then using local linear approximation to make explicit feature contributions; Step S6: strength evaluation, performing feature mapping and hidden variable estimation on the samples to be evaluated, inputting the microcosmic feature vector, time series feature data and virtual measurement into the hybrid model, and outputting the strength prediction value, confidence interval and local feature contribution in real time.

2. The big data based cement-silicate gel strength evaluation method according to claim 1, characterized in that: In step S1, the data collection collects experimental and formula data, time series feature data, microcosmic characterization data and strength label data of historical cement-silicate gel; the experimental and formula data includes: unique identifier, water glass ratio, silica mass fraction, sodium oxide mass fraction, additive type and mass fraction, initial mixing temperature, mixing time; the time series feature data includes: temperature, humidity, solution pH, conductivity; the microcosmic characterization data includes: electron micrograph, X-ray diffraction spectrum, Fourier infrared spectrum, porosity, particle median diameter; the strength label data refers to the compressive strength of the cement-silicate gel.

3. The big data based cement-silicate gel strength evaluation method according to claim 1, characterized in that: In step S2, the feature mapping constructs differentiable topological and frequency domain feature mapping for the microcosmic characterization data, obtains a continuous persistent image vector, and extracts physical quantitative indicators of peak position and peak strength, comprising the following steps: Step S21: building a morphology subset, extracting binary sub-layers from the electron micrograph and constructing an upstream filtering sequence to generate a morphology subset; Step S22: obtaining a persistent image, calculating persistent homology information generated by morphology filtering, and only taking 0D and 1D: obtaining a persistent image; Step S23: persistent image transformation, transforming the persistent image with physical weights; Step S24: constructing a microcosmic feature vector, discretizing the persistent image after the physical weight transformation on a fixed grid to obtain a vectorized feature; then extracting spectral energy features from the Fourier infrared spectrum; and constructing a final microcosmic feature vector.

4. The big data based cement-silicate gel strength evaluation method of claim 1, wherein: In step S3, the latent variable estimation, a low-dimensional physically driven state space model is established, the gel degree is regarded as an unobservable latent variable, the posterior estimation of the latent variable is performed through extended Kalman filtering using the time series features and micro features, and an estimated pseudo-measurement is obtained. Specifically, the following steps are included: Step S31: constructing a kinetic model, using a generalized model with temperature and humidity coupling and mapping the micro-features into kinetic constants through parameters; Step S32: discretization, forward Euler discretization is performed on the latent variable, and an observation vector is constructed; Step S33: extended Kalman filtering, the posterior estimation of the latent variable is performed through extended Kalman filtering. First, the posterior estimation of the latent variable at the current time is obtained by forward propagation combined with the time step and the kinetic rate information. Then, the covariance matrix is calculated. Finally, the overall update is performed.

5. The big data based cement-silicate gel strength evaluation method as claimed in claim 1, wherein: In step S4, the mixed model training, a physically regularized mixed predictor is constructed, the micro feature vector, the time series feature and the pseudo-measurement are taken as inputs, and the target compression strength is predicted. Specifically, the following steps are included: Step S41: constructing a strength prior physical model, taking the parameters determined by the micro feature vector and the time-varying pseudo-measurement as inputs, and outputting a predicted curve of the strength over time based on a physical empirical form; Step S42: defining a data-driven correction term, taking the micro feature, the time series feature and the pseudo-measurement as inputs, and outputting a correction amount for the physical prior; Step S43: final prediction, adding the physical prior prediction and the data-driven correction term to obtain the final mixed model strength prediction value as the model output during training and inference; Step S44: training loss, the model loss includes data fitting error, physical residual penalty and monotonicity penalty.

6. The big data based cement-silicate gel strength evaluation method of claim 1, wherein: In step S5, the explainability output, a confidence interval and a local feature contribution are provided for the prediction result. Deep integration and heterogeneous variance regression are used to simultaneously model the mean and data noise, and local linear approximation is used to explicitly express the feature contribution. Specifically, the following steps are included: Step S51: Model integration, integration The mean and variance outputs of each model are ensembled through independent training. The total variance is obtained from different initialization models with the same structure; the total variance is decomposed into two parts: observable noise and parameter uncertainty. Step S52: generating a confidence interval, a confidence interval based on the approximate Gaussian assumption is given, and the confidence interval width is determined by the total standard deviation; Step S53: generating a contribution degree, estimating the feature contribution of the input vector; Step S54: uncertainty splitting, splitting the total variance into two parts, the first part is the process uncertainty of the uncertainty of the dynamic latent variable propagated to the strength, and the second part is the observation uncertainty introduced by the data-driven part and the observation noise; Step S55: chain propagation, the posterior covariance of the extended Kalman filtering is mapped to the uncertainty of the strength prediction through the physical prior model based on the chain rule.

7. The big data based cement-silicate gel strength evaluation method of claim 1, wherein: In step S6, the strength evaluation, the cement-silicate gel data to be evaluated is obtained, the mixed model is input, the strength of the cement-silicate gel is predicted in real time by using the mixed model, and the explainability output is generated.

8. A cement-silicate gel strength evaluation system based on big data for implementing the cement-silicate gel strength evaluation method based on big data according to any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a feature mapping module, a latent variable estimation module, a mixed model training module, an explainability output module and a strength evaluation module.

9. The cement-silicate gel strength evaluation system based on big data according to claim 8, characterized in that: The data collection module collects historical experimental and formula data, time series characteristic data, microscopic characterization data and strength label data of cement-silicate gel, and sends the data to the feature mapping module, the latent variable estimation module, the mixed model training module, the explainability output module and the strength evaluation module. The feature mapping module receives the data sent by the data collection module, constructs a differentiable topological and frequency domain feature mapping for the microscopic characterization data, obtains a continuous persistent image vector, extracts physical quantitative indicators of peak position and peak strength, and sends the data to the latent variable estimation module. The latent variable estimation module receives the data sent by the data collection module and the feature mapping module, establishes a low-dimensional physical driving state space model, regards the gel degree as an unobservable latent variable, uses the time series characteristics and microscopic characteristics to perform posterior estimation on the latent state through extended Kalman filtering, obtains an estimated virtual measurement, and sends the data to the mixed model training module. The mixed model training module receives the data sent by the data collection module and the latent variable estimation module, constructs a physically regularized mixed predictor, takes the microscopic characteristic vector, the time series characteristics and the virtual measurement as input, predicts the target compressive strength, and sends the data to the explainability output module. The explainability output module receives the data sent by the data collection module and the mixed model training module, provides a confidence interval and a local feature contribution for the prediction result, uses deep integration and heterogeneous variance regression to model the mean and data noise at the same time, uses local linear approximation to make explicit feature contribution, and sends the data to the strength evaluation module. The strength evaluation module receives the data sent by the data collection module and the explainability output module, obtains the cement-silicate gel data to be evaluated, inputs the mixed model, uses the mixed model to predict the strength of the cement-silicate gel in real time as the evaluation strength, and generates an explainability output.

Images