Graphene floor heating module temperature drift real-time compensation control method

Abstract

The application discloses a kind of graphene floor heating module temperature drift real-time compensation control methods, it is related to industrial control system technical field, comprising: based on historical several graphene floor heating module's full life cycle test data, construct graphene floor heating module temperature time evolution trend model, output graphene floor heating module's full life cycle temperature distribution vector;Graphene floor heating module abnormal type causal network is constructed, for full life cycle temperature distribution vector is divided, obtains temperature drift local anomaly attribution path;Temperature anomaly early warning model is constructed, verify real-time graphene floor heating module's operating data relative temperature drift local anomaly support variable, predict future graphene floor heating module's temperature drift local anomaly critical time;Establish drift-inhibition prediction control network, according to temperature drift dynamic compensation function, generate graphene floor heating module local anomaly optimal real-time temperature drift compensation scheme.The application reduces redundant energy consumption and operation cost.

Description

Technical Field

[0001] This invention relates to the field of industrial control system technology, specifically to a real-time temperature drift compensation control method for graphene floor heating modules. Background Technology

[0002] With the rapid development of electric heating film floor heating technology, graphene heating elements have been widely used in mid-to-high-end building heating fields due to their excellent thermal response characteristics, electrothermal conversion efficiency, and far-infrared radiation performance. However, under long-term power-on operation and intermittent temperature rise cycles, local temperature drift inevitably occurs between the graphene heating layer and the electrode interface, as well as between the encapsulation substrate and the electrode, due to the mismatch of thermal expansion coefficients and the accumulation of electrothermal stress. This manifests as uneven heating in the heating area, attenuation of edge hot spots, or resistance jumps and oscillations. Existing floor heating control systems generally adopt fixed PID regulation or simple over-temperature protection strategies, lacking real-time temperature field monitoring. The dynamic modeling capability of spatial evolution trends cannot identify drift precursor features in advance, nor can it establish a quantitative relationship between anomaly types and physical attribution paths. A few temperature monitoring methods based on threshold alarms have defects such as strong lag, high false alarm rate, inability to distinguish between reversible fluctuations and irreversible aging drift, and difficulty in generating optimal compensation action sequences in real time. There is an urgent need for a real-time temperature drift compensation control method for graphene underfloor heating modules that can integrate full life cycle test data, causal inference and predictive control technology to solve the problems of delayed anomaly warning, ambiguous attribution paths and lack of global optimization capability of compensation actions in existing technologies. Summary of the Invention

[0003] To address the aforementioned technical problems, a real-time temperature drift compensation control method for graphene underfloor heating modules is provided. This technical solution resolves the problems mentioned above.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for real-time temperature drift compensation and control of a graphene underfloor heating module includes: S1. Based on the full life cycle test data of several graphene underfloor heating modules in history, analyze the time series distribution of temperature field of several graphene underfloor heating modules, construct a time series evolution trend model of temperature of graphene underfloor heating modules, and output the full life cycle temperature distribution vector of graphene underfloor heating modules. S2. Based on the known abnormality data of temperature drift of graphene floor heating module, construct the causal network of abnormality types of graphene floor heating module, divide the temperature distribution vector of the whole life cycle of graphene floor heating module, and obtain the local anomaly attribution path of temperature drift of graphene floor heating module. S3. Using the local anomaly attribution path of temperature drift of graphene floor heating module, construct a temperature anomaly early warning model for graphene floor heating module, verify the supporting variables of local anomaly of temperature drift relative to the real-time operation data of graphene floor heating module, and predict the critical moment of local anomaly of temperature drift of graphene floor heating module in the future. S4. Based on the critical moment of local anomalies in temperature drift of the future graphene floor heating module and the selectable drift suppression control action, establish a drift-suppression predictive control network, and generate the optimal real-time temperature drift compensation scheme for local anomalies of the graphene floor heating module according to the temperature drift dynamic compensation function.

[0005] Preferably, step S1 specifically includes: Based on the full life cycle test data of several historical graphene underfloor heating modules, a two-dimensional matrix of the full life cycle test data of a single graphene underfloor heating module is established with the collection time point as the row and the spatial temperature measurement points of the historical graphene underfloor heating module as the column, and data preprocessing is performed. For several graphene underfloor heating modules, a two-dimensional matrix of the full life cycle test data of each graphene underfloor heating module is laid out sequentially according to time sequence to generate several graphene underfloor heating module dataset matrices and calculate the spatial covariance matrix. The spatial covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalues ​​and corresponding unit eigenvectors of the spatial covariance matrix. The eigenvalues ​​of the spatial covariance matrix are sorted in descending order according to the eigenvalues ​​of the spatial covariance matrix, and the unit eigenvectors corresponding to the spatial covariance matrix are taken as the principal component directions. The ratio of the eigenvalue corresponding to each principal component direction to the sum of all eigenvalues ​​is calculated as the variance contribution rate of the principal component direction. The cumulative variance contribution rate is continuously calculated from front to back. A preset cumulative variance contribution rate threshold is set. The top few principal component directions that meet the threshold are extracted, and the number of extracted principal component directions is used as the number of principal components to be retained. The row vector of the temperature field corresponding to any moment in the dataset matrix of several graphene underfloor heating modules is projected onto the top few principal component directions and arranged in chronological order to generate a dimension-reduced representation matrix of the temperature distribution vector of several graphene underfloor heating modules throughout their entire life cycle, thus obtaining the temporal distribution of the temperature field of several graphene underfloor heating modules throughout their entire life cycle. The preset cumulative variance contribution rate threshold specifically refers to a percentage value pre-set based on the maximum number of principal components allowed by computing resource limitations, which determines the number of principal component directions that need to be retained during the dimensionality reduction process. Based on the time-series temperature field distribution of several graphene underfloor heating modules throughout their entire life cycle, and using the data acquisition time point as input, a combined kernel function for the graphene underfloor heating modules is designed according to the physical characteristics of temperature drift during the operation of the modules. The hyperparameters of the combined kernel function are optimized by maximizing the logarithmic marginal likelihood, and a time-series temperature evolution trend model of the graphene underfloor heating modules is constructed, outputting the temperature distribution vector of the graphene underfloor heating modules throughout their entire life cycle.

[0006] Preferably, step S2 specifically includes: Based on the full life cycle test data of several historical graphene floor heating modules and the temperature drift anomaly data of graphene floor heating modules with known anomaly type labels obtained from accelerated aging experiments in the laboratory, a graphene floor heating module anomaly type performance dataset is generated by combining the known anomaly type labels. The known anomaly type labels include: edge cold zone diffusion type, center hot spot attenuation type, resistance jump oscillation type, and adhesive layer peeling slow change type. Based on the temperature distribution vector of the graphene underfloor heating module throughout its entire life cycle, a sliding time window is set up. The spatiotemporal statistical characteristics and temperature control behavior characteristics of the temperature field of the graphene underfloor heating module within the time window are calculated and denoted as candidate statistics. The dataset of abnormal types of graphene underfloor heating modules is used as positive samples, and the time window samples in the temperature distribution vector of the graphene underfloor heating module throughout its entire life cycle without any abnormal types are used as negative samples. The independent samples t-test is used to calculate the mean difference and significance of the difference between the positive samples and the negative samples on each candidate statistics. Candidate statistics with a significance of difference less than the preset significance threshold are selected as observation node variables. Based on a two-dimensional matrix of full life-cycle test data for a single graphene underfloor heating module, we obtain historical temperature data of the graphene underfloor heating module at known times. Substituting this data into the time-series evolution trend model of the graphene underfloor heating module temperature, we predict the nominal temperature field of the graphene underfloor heating module at the same time in the future. We calculate the weighted root mean square value of the residual between the nominal temperature field and the measured temperature of the graphene underfloor heating module at the same time in the future, and obtain the index of the severity of abnormal temperature drift of the graphene underfloor heating module at that time, which is recorded as the effect variable. We record the sub-parameters of the spatiotemporal statistical characteristics and temperature control behavior characteristics of the graphene underfloor heating module temperature field within the time window as candidate causal variables, and construct a time-slice aligned causal network variable set for the temperature drift of the graphene underfloor heating module. Based on the time-slice aligned causal network variable set and observation node variables of the graphene floor heating module temperature drift, a graphene floor heating module temperature drift observation data matrix is ​​constructed. The observation node variables are used as nodes and any two different node pairs are connected by undirected edges to establish a completely undirected graph of graphene floor heating module temperature drift. Based on the completely undirected graph of temperature drift of graphene floor heating module, the partial correlation coefficient of each node pair is calculated under the condition of empty set and the conditional independence test is performed. If the node pair is conditionally independent under the condition of empty set, the undirected edge corresponding to the node pair is deleted and the empty set is recorded as the separation set of the node pair. All node pairs are traversed to generate the completely undirected skeleton graph of temperature drift of graphene floor heating module. Using the V structure, we traverse the undirected edge triplet corresponding to the complete undirected skeleton graph of temperature drift of graphene floor heating module, set the test conditions, identify the undirected edge triplet that meets the collision structure and determine the direction of the undirected edge, and construct the partial directed acyclic graph of temperature drift equivalence class of graphene floor heating module. Find all directed edges pointing to the effect variable in the partial directed acyclic graph of the graphene underfloor heating module temperature drift equivalence class. Mark the starting node of the directed edge as the direct cause of the effect variable. For any direct cause of the effect variable and an upstream cause node exists, mark the upstream node as the indirect cause of the effect variable. When there is an undirected edge between the effect variable and any candidate cause variable in the partial directed acyclic graph of the graphene underfloor heating module temperature drift equivalence class, determine the candidate cause variable as a potential confounding factor of the graphene underfloor heating module temperature drift. Based on the directed acyclic graph of the temperature drift equivalence class of the graphene underfloor heating module, all variables with direct directed edges that are direct causes of the temperature drift anomaly severity index of the graphene underfloor heating module are extracted and denoted as the treatment variable set. All remaining variables outside the treatment variable set are denoted as the control variable set. The attribution path weight coefficients of the variables in the treatment variable set to the effect variables are calculated to obtain the causal effect strength estimate of each direct cause variable on the temperature drift anomaly severity index. The causal effect strength estimate is used as the weight of the corresponding directed edge and the directed acyclic graph of the temperature drift equivalence class of the graphene underfloor heating module is labeled to construct the causal network of the graphene underfloor heating module anomaly type.

[0007] Preferably, step S2 further includes: Based on the dimensionality reduction representation of the temperature distribution vector throughout the entire life cycle of the graphene underfloor heating module, the principal component score vector of the temperature throughout the entire life cycle of the graphene underfloor heating module is obtained. The Euclidean norm of the difference vector of the principal component score vector of the temperature of the graphene underfloor heating module at adjacent time steps is calculated, and the time series vector of the temperature field change rate of the graphene underfloor heating module throughout the entire life cycle is obtained. Substitute the principal component score vector of the graphene underfloor heating module throughout its entire life cycle into the temperature time series evolution trend model of the graphene underfloor heating module, output the principal component score prediction variance corresponding to each time step of the graphene underfloor heating module, and obtain the time series vector sequence of principal component score prediction variance corresponding to the entire life cycle of the graphene underfloor heating module. By splicing the time series vector of the temperature field change rate throughout the entire life cycle of the graphene underfloor heating module with the time series vector sequence of the principal component score prediction variance corresponding to the entire life cycle of the graphene underfloor heating module, a joint feature matrix of temperature field change rate and prediction uncertainty throughout the entire life cycle of the graphene underfloor heating module is established. The composite index of the temperature field change rate-prediction uncertainty joint feature matrix of the graphene underfloor heating module throughout its entire life cycle is calculated for each time step, and Bayesian change point detection is performed. The number of change points is set to 2. The entire life cycle of the graphene underfloor heating module is divided into three continuous intervals, which are defined in chronological order as the stable baseline segment, the trend initiation segment, and the significant deviation segment. The state segment label of each time step of the graphene underfloor heating module is obtained. The historical temperature data corresponding to the time steps of the trend initiation segment and the significant deviation segment of the graphene underfloor heating module are selected as the input data matrix of the causal network of the graphene underfloor heating module's anomaly type, and the local anomaly attribution path of temperature drift of the graphene underfloor heating module is obtained.

[0008] Preferably, step S3 specifically includes: Based on the graphene floor heating module sensor, the real-time temperature field vector, real-time input power, cumulative power-on time and real-time ambient temperature of the graphene floor heating module are obtained according to a fixed sampling period. These are combined as the real-time operation data of the graphene floor heating module. The real-time temperature field vector of the graphene floor heating module is reduced in dimension by combining the dimension reduction representation matrix of the temperature distribution vector of several graphene floor heating modules throughout their entire life cycle, and the dimension reduction representation of the real-time temperature distribution vector of the graphene floor heating module is obtained. The principal component score vector of the real-time temperature of the graphene floor heating module is then calculated. Based on the local anomaly attribution path of temperature drift in the graphene underfloor heating module, the real-time observation node variables of the graphene underfloor heating module in the current time window are calculated, and the real-time support variable vector of the local anomaly attribution path of temperature drift in the graphene underfloor heating module is constructed. The historical temperature data corresponding to the time step of the stable benchmark segment to which the graphene floor heating module belongs is traversed. The real-time support variable vector sample set of the local anomaly attribution path of temperature drift of the graphene floor heating module is extracted. The mean and covariance matrix of the real-time support variable vector sample set are calculated as the health benchmark of the graphene floor heating module. The Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path for temperature drift in the graphene underfloor heating module and the health benchmark of the graphene underfloor heating module is calculated. A health benchmark threshold is set. If the Mahalanobis distance is less than or equal to the health benchmark threshold, it is determined that the real-time operating data of the graphene underfloor heating module does not support any local anomaly attribution path for temperature drift in the graphene underfloor heating module, and the graphene underfloor heating module is in a healthy operating state. If the Mahalanobis distance is greater than the health benchmark threshold, it is determined that the real-time operating data of the graphene underfloor heating module supports any local anomaly attribution path for temperature drift in the graphene underfloor heating module, and the graphene underfloor heating module triggers a temperature anomaly warning signal. Based on the temperature anomaly warning signal triggered by the graphene floor heating module, the real-time operation data of the graphene floor heating module in the current time window is extracted. According to the temperature time series evolution trend model of the graphene floor heating module, the temperature drift anomaly severity index of the graphene floor heating module in the current time window is calculated and denoted as the real-time temperature drift level of the graphene floor heating module. The real-time temperature drift rate of the graphene floor heating module is calculated and the real-time temperature drift level and rate of the graphene floor heating module are used as the real-time state vector of the graphene floor heating module. Based on the physical characteristics of temperature drift during the operation of the graphene floor heating module, and combined with the Wiener process with random acceleration, the graphene floor heating module's temperature real-time drift level, rate, and time interval product are used as the drift level of the graphene floor heating module at the previous moment, and the graphene floor heating module's temperature real-time drift rate is used as the drift rate of the graphene floor heating module at the previous moment, thus establishing the graphene floor heating state transition matrix. The Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path of temperature drift in the graphene floor heating module and the health benchmark of the graphene floor heating module is used as the observation. Based on the temperature anomaly warning signal triggered by the graphene floor heating module, the weighted sum of the estimated causal effect intensity of each direct cause variable on the local anomaly attribution path of temperature drift is used as the initial temperature drift level of the graphene floor heating module. The average temperature drift rate of the graphene floor heating module in the current time window is used as the initial temperature drift rate of the graphene floor heating module. The random number seed of the graphene floor heating module is set, and the initial particle set is generated by sampling from the preset prior distribution and the particles are assigned equal initial weights. Based on the graphene floor heating state transition matrix, the particle state of the graphene floor heating module at the next moment is predicted. According to the Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path of the temperature drift of the graphene floor heating module and the health benchmark of the graphene floor heating module, the likelihood weight of each particle is calculated, and normalization and resampling are performed to generate a new particle set. The posterior expectation of the state of the real-time support variable vector of the local anomaly attribution path of the temperature drift of the graphene floor heating module is calculated as the state estimate at the current moment. Taking the new particle set as the starting point, it is substituted into the graphene floor heating state transition matrix to generate the set of temperature state evolution paths of the graphene floor heating module within a specified time period in the future. The minimum value of the lower boundary effect variable of the significantly deviated segment of the local anomaly attribution path of temperature drift in the graphene underfloor heating module is used as the critical threshold for significant anomaly. The time point when the subset of the temperature state evolution path set of the graphene underfloor heating module is first greater than or equal to the critical threshold for significant anomaly within a specified time period in the future is found. The statistical distribution of the time when all particles cross the critical threshold for significant anomaly is calculated. The median of the time when all particles cross the critical threshold for significant anomaly is used as the critical moment of local anomaly of temperature drift in the future graphene underfloor heating module.

[0009] Preferably, step S4 specifically includes: Using the critical moment of local anomaly in temperature drift of the graphene floor heating module, the posterior expectation of the state of the real-time support variable vector of the local anomaly attribution path of temperature drift of the graphene floor heating module, and the dataset of anomaly type performance of the graphene floor heating module as input, and based on the physical controllable input of the graphene floor heating module, the power regulation amount, the duty cycle regulation amount of the on / off cycle, and the preheating / slow cooling timing offset of the graphene floor heating module are combined and encoded into a discretized set of candidate drift suppression control actions. Using the model predictive control algorithm, the control time domain and prediction time domain are set, the discrete sampling time interval is determined, the temperature drift state vector of the graphene floor heating module in the future prediction step is used as the state vector, and the drift suppression control action applied in the future control step is used as the control action. A state space model of the drift-suppression predictive control network is constructed. The weighted sum of the drift suppression error term, the control energy penalty term and the control smoothness penalty term is used as the temperature drift dynamic compensation function. The state constraints and control input constraints are used as constraints to generate several candidate suppression action sequences for the temperature drift of the graphene floor heating module. Based on generating several candidate suppression action sequences for temperature drift of the graphene underfloor heating module, the dynamic compensation function value of temperature drift corresponding to each candidate sequence is calculated, and the candidate sequence with the smallest compensation function value is selected as the optimal real-time temperature drift compensation scheme for local anomalies of the graphene underfloor heating module.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a real-time temperature drift compensation control method for graphene underfloor heating modules. By integrating full-lifecycle temperature field time-series evolution modeling, causal network anomaly attribution, particle filter critical moment prediction, and model predictive control, it achieves early warning, precise location, and dynamic suppression of temperature drift in graphene underfloor heating modules. The temperature evolution model constructed based on principal component analysis and combined kernel functions can effectively extract dimensionality reduction features of the temperature field, reducing computational burden. Causal networks and statistical tests are used to automatically identify the anomaly types and attribution paths of temperature drift, solving the problems of high false alarm rate and poor interpretability of traditional threshold methods. Combining stochastic accelerated Wiener processes and particle filter algorithms, it can evaluate the temperature drift level and rate in real time and accurately predict local anomaly critical moments, providing a time window for early intervention. By constructing a drift-suppression predictive control network, under the condition of satisfying state and control constraints, it dynamically optimizes compensation actions such as power adjustment, duty cycle, and time-series offset, significantly extending the effective life of the underfloor heating module, improving thermal field uniformity and operational safety, while reducing redundant energy consumption and maintenance costs. Attached Figure Description

[0011] Figure 1 This is a flowchart of a real-time temperature drift compensation control method for a graphene underfloor heating module. Detailed Implementation

[0012] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0013] Reference Figure 1 As shown, a real-time temperature drift compensation control method for a graphene underfloor heating module includes: S1. Based on the full life cycle test data of several graphene underfloor heating modules in history, analyze the time series distribution of temperature field of several graphene underfloor heating modules, construct a time series evolution trend model of temperature of graphene underfloor heating modules, and output the full life cycle temperature distribution vector of graphene underfloor heating modules. Step S1 specifically includes: Based on the full life cycle test data of several historical graphene underfloor heating modules, a two-dimensional matrix of the full life cycle test data of a single graphene underfloor heating module is established with the collection time point as the row and the spatial temperature measurement points of the historical graphene underfloor heating module as the column, and data preprocessing is performed. To further explain, the data collection time points are arranged sequentially along the entire life cycle of the graphene floor heating module, and the historical graphene floor heating module spatial temperature measurement points include multiple temperature sensing locations distributed on the surface, inside, and at the interface of the graphene floor heating module; the "several graphene floor heating modules" refers to multiple graphene floor heating modules of the same specification. For several graphene underfloor heating modules, a two-dimensional matrix of the full life cycle test data of each graphene underfloor heating module is laid out sequentially according to time sequence to generate several graphene underfloor heating module dataset matrices and calculate the spatial covariance matrix. The spatial covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalues ​​and corresponding unit eigenvectors of the spatial covariance matrix. The eigenvalues ​​of the spatial covariance matrix are sorted in descending order according to the eigenvalues ​​of the spatial covariance matrix, and the unit eigenvectors corresponding to the spatial covariance matrix are taken as the principal component directions. To further explain, the graphene floor heating module dataset matrix uses the sum of all graphene floor heating module acquisition time points as rows and the number of historical graphene floor heating module spatial temperature measurement points as columns; the principal component direction is specifically: the unit eigenvector corresponding to the largest eigenvalue of the spatial covariance matrix is ​​taken as the first principal component direction, the unit eigenvector corresponding to the second largest eigenvalue of the spatial covariance matrix is ​​taken as the second principal component direction, and so on until all principal component directions are obtained; The ratio of the eigenvalue corresponding to each principal component direction to the sum of all eigenvalues ​​is calculated as the variance contribution rate of the principal component direction. The cumulative variance contribution rate is continuously calculated from front to back. A preset cumulative variance contribution rate threshold is set. The top few principal component directions that meet the threshold are extracted, and the number of extracted principal component directions is used as the number of principal components to be retained. The row vector of the temperature field corresponding to any moment in the dataset matrix of several graphene underfloor heating modules is projected onto the top few principal component directions and arranged in chronological order to generate a dimension-reduced representation matrix of the temperature distribution vector of several graphene underfloor heating modules throughout their entire life cycle, thus obtaining the temporal distribution of the temperature field of several graphene underfloor heating modules throughout their entire life cycle. The preset cumulative variance contribution rate threshold specifically refers to a percentage value pre-set based on the maximum number of principal components allowed by computing resource limitations, which determines the number of principal component directions that need to be retained during the dimensionality reduction process. Based on the time-series distribution of temperature fields throughout the entire life cycle of several graphene underfloor heating modules, and taking the acquisition time point as input, a combined kernel function of graphene underfloor heating modules is designed according to the physical characteristics of temperature drift during the operation of graphene underfloor heating modules. The hyperparameters of the combined kernel function of graphene underfloor heating modules are optimized by maximizing the logarithmic marginal likelihood, and a time-series evolution trend model of temperature of graphene underfloor heating modules is constructed, outputting the temperature distribution vector of the entire life cycle of graphene underfloor heating modules. To further explain, the combined kernel function is used to characterize the similarity of temperature distribution vectors between different acquisition time points. Specifically, it includes a linear combination of radial basis function kernel and white noise kernel, or a combination of periodic kernel and trend kernel designed based on the physical characteristics of temperature drift of graphene floor heating module.

[0014] S2. Based on the known abnormality data of temperature drift of graphene floor heating module, construct the causal network of abnormality types of graphene floor heating module, divide the temperature distribution vector of the whole life cycle of graphene floor heating module, and obtain the local anomaly attribution path of temperature drift of graphene floor heating module. Step S2 specifically includes: Based on the full life cycle test data of several historical graphene floor heating modules and the temperature drift anomaly data of graphene floor heating modules with known anomaly type labels obtained from accelerated aging experiments in the laboratory, a graphene floor heating module anomaly type performance dataset is generated by combining the known anomaly type labels. The known anomaly type labels include: edge cold zone diffusion type, center hot spot attenuation type, resistance jump oscillation type, and adhesive layer peeling slow change type. Based on the temperature distribution vector of the graphene underfloor heating module throughout its entire life cycle, a sliding time window is set up. The spatiotemporal statistical characteristics and temperature control behavior characteristics of the temperature field of the graphene underfloor heating module within the time window are calculated and denoted as candidate statistics. The dataset of abnormal types of graphene underfloor heating modules is used as positive samples, and the time window samples in the temperature distribution vector of the graphene underfloor heating module throughout its entire life cycle without any abnormal types are used as negative samples. The independent samples t-test is used to calculate the mean difference and significance of the difference between the positive samples and the negative samples on each candidate statistics. Candidate statistics with a significance of difference less than the preset significance threshold are selected as observation node variables. Further explanation: the spatiotemporal statistical characteristics are: the mean and maximum values ​​of the temperature field mean, variance, skewness, kurtosis, and spatial gradient modulus of the graphene underfloor heating module within the time window, as well as the temperature change rate, autocorrelation coefficient, and power spectral density energy in a specific frequency band at adjacent time points; the temperature control behavior characteristics are: the temperature rise slope, steady-state fluctuation amplitude, overshoot, and response delay of the temperature field of the graphene underfloor heating module at adjacent time points within the time window; the preset significance threshold is: the difference in mean and significance of the positive and negative samples on each candidate statistic calculated by independent samples t-test, and the preset significance threshold is set based on engineering experience in the field of graphene underfloor heating module anomaly detection; Based on a two-dimensional matrix of full life-cycle test data for a single graphene underfloor heating module, we obtain historical temperature data of the graphene underfloor heating module at known times. Substituting this data into the time-series evolution trend model of the graphene underfloor heating module temperature, we predict the nominal temperature field of the graphene underfloor heating module at the same time in the future. We calculate the weighted root mean square value of the residual between the nominal temperature field and the measured temperature of the graphene underfloor heating module at the same time in the future, and obtain the index of the severity of abnormal temperature drift of the graphene underfloor heating module at that time, which is recorded as the effect variable. We record the sub-parameters of the spatiotemporal statistical characteristics and temperature control behavior characteristics of the graphene underfloor heating module temperature field within the time window as candidate causal variables, and construct a time-slice aligned causal network variable set for the temperature drift of the graphene underfloor heating module. Based on the time-slice aligned causal network variable set and observation node variables of the graphene floor heating module temperature drift, a graphene floor heating module temperature drift observation data matrix is ​​constructed. The observation node variables are used as nodes and any two different node pairs are connected by undirected edges to establish a completely undirected graph of graphene floor heating module temperature drift. Based on the completely undirected graph of temperature drift of graphene floor heating module, the partial correlation coefficient of each node pair is calculated under the condition of empty set and the conditional independence test is performed. If the node pair is conditionally independent under the condition of empty set, the undirected edge corresponding to the node pair is deleted and the empty set is recorded as the separation set of the node pair. All node pairs are traversed to generate the completely undirected skeleton graph of temperature drift of graphene floor heating module. Using the V structure, we traverse the undirected edge triplet corresponding to the complete undirected skeleton graph of temperature drift of graphene floor heating module, set the test conditions, identify the undirected edge triplet that meets the collision structure and determine the direction of the undirected edge, and construct the partial directed acyclic graph of temperature drift equivalence class of graphene floor heating module. To further explain, the test conditions specifically refer to the following: if the undirected edge triple is ABC and the edge between AB is deleted in the conditional independence test, then the corresponding separation set is denoted as Sep(A, B). If C is not included in Sep(A, B), then the triple is determined to be a V structure and C is a collision node, with the orientation being A→C←B. If C is included in Sep(A, B), then the triple is determined to be a non-V structure and cannot be oriented, and the undirected edge between AC and CB is retained. It should be noted that this determination is only valid when the edge between AB is in the graphene floor heating module temperature drift completely undirected skeleton graph and is an undirected edge. Find all directed edges pointing to the effect variable in the partial directed acyclic graph of the graphene underfloor heating module temperature drift equivalence class. Mark the starting node of the directed edge as the direct cause of the effect variable. For any direct cause of the effect variable and an upstream cause node exists, mark the upstream node as the indirect cause of the effect variable. When there is an undirected edge between the effect variable and any candidate cause variable in the partial directed acyclic graph of the graphene underfloor heating module temperature drift equivalence class, determine the candidate cause variable as a potential confounding factor of the graphene underfloor heating module temperature drift. Based on the directed acyclic graph of the temperature drift equivalence class of the graphene underfloor heating module, all variables with direct directed edges that are direct causes of the temperature drift anomaly severity index of the graphene underfloor heating module are extracted and denoted as the treatment variable set. All remaining variables outside the treatment variable set are denoted as the control variable set. The attribution path weight coefficients of the variables in the treatment variable set to the effect variables are calculated to obtain the causal effect strength estimate of each direct cause variable on the temperature drift anomaly severity index. The causal effect strength estimate is used as the weight of the corresponding directed edge and the directed acyclic graph of the temperature drift equivalence class of the graphene underfloor heating module is labeled to construct the causal network of the graphene underfloor heating module anomaly type.

[0015] Step S2 also includes: Based on the dimensionality reduction representation of the temperature distribution vector throughout the entire life cycle of the graphene underfloor heating module, the principal component score vector of the temperature throughout the entire life cycle of the graphene underfloor heating module is obtained. The Euclidean norm of the difference vector of the principal component score vector of the temperature of the graphene underfloor heating module at adjacent time steps is calculated, and the time series vector of the temperature field change rate of the graphene underfloor heating module throughout the entire life cycle is obtained. Substitute the principal component score vector of the graphene underfloor heating module throughout its entire life cycle into the temperature time series evolution trend model of the graphene underfloor heating module, output the principal component score prediction variance corresponding to each time step of the graphene underfloor heating module, and obtain the time series vector sequence of principal component score prediction variance corresponding to the entire life cycle of the graphene underfloor heating module. By splicing the time series vector of the temperature field change rate throughout the entire life cycle of the graphene underfloor heating module with the time series vector sequence of the principal component score prediction variance corresponding to the entire life cycle of the graphene underfloor heating module, a joint feature matrix of temperature field change rate and prediction uncertainty throughout the entire life cycle of the graphene underfloor heating module is established. The composite index of the temperature field change rate-prediction uncertainty joint feature matrix of the graphene underfloor heating module throughout its entire life cycle is calculated for each time step, and Bayesian change point detection is performed. The number of change points is set to 2. The entire life cycle of the graphene underfloor heating module is divided into three continuous intervals, which are defined in chronological order as the stable baseline segment, the trend initiation segment, and the significant deviation segment. The state segment label of each time step of the graphene underfloor heating module is obtained. The historical temperature data corresponding to the time steps of the trend initiation segment and the significant deviation segment of the graphene underfloor heating module are selected as the input data matrix of the causal network of the graphene underfloor heating module's anomaly type, and the local anomaly attribution path of temperature drift of the graphene underfloor heating module is obtained. To further explain, the specific expression for the composite index at each time step in the joint characteristic matrix of the temperature field change rate and prediction uncertainty throughout the entire life cycle of the graphene underfloor heating module is as follows: , In the formula, The joint characteristic matrix of temperature field change rate and prediction uncertainty throughout the entire life cycle of the graphene underfloor heating module at time step Composite indicators, This represents the current time step in the entire lifecycle of the graphene underfloor heating module. This represents the total number of time steps throughout the entire lifecycle of the graphene underfloor heating module. The Euclidean norm of the difference vector between the principal component score vectors of the graphene underfloor heating module at adjacent time steps represents the rate of change of the temperature field throughout the entire life cycle of the graphene underfloor heating module. This is a time index variable, representing any time step in the entire lifecycle of the graphene underfloor heating module, distinct from the current time step. , To ensure the entire lifecycle of graphene underfloor heating modules Within, the maximum value among the rates of change of the temperature field of the graphene underfloor heating module corresponding to all time steps. To avoid numerical calculation errors caused by a zero denominator, the default value is set to an extremely small positive number. , The time-series evolution trend model of temperature for graphene underfloor heating modules at time step The trace of the predicted variance of the principal component scores in the output represents the degree of accumulated uncertainty in the model's prediction of the temperature field at that moment. The mean of the principal component score prediction variance trace of the graphene underfloor heating module within the stationary baseline period is used as the baseline uncertainty. This is the nonlinear enhancement coefficient, set to 10 to amplify the contribution of the rate of change exceeding the threshold to the composite index. The abnormal threshold for the rate of change of temperature field in graphene underfloor heating modules is defined as the rate of change within a stable reference range. The sum of the mean and twice the standard deviation, This is an indicator function. It takes a value of 1 when the rate of change of the temperature field in the current time step of the graphene floor heating module is greater than the abnormal threshold, and a value of 0 otherwise. The stable baseline segment, the trend emergence segment, and the significant deviation segment are specifically divided as follows: Stable baseline segment: The continuous interval from the start of the graphene underfloor heating module's entire life cycle to the first change point. The composite index of the joint characteristic matrix of the graphene underfloor heating module's temperature field change rate and prediction uncertainty throughout its entire life cycle is less than or equal to the 95th quantile of the composite index within this stable baseline segment. Trend emergence segment: The continuous interval from the first change point to the second change point. The composite index of the joint characteristic matrix of the graphene underfloor heating module's temperature field change rate and prediction uncertainty throughout its entire life cycle is greater than the 95th quantile of the stable baseline segment and less than or equal to the 99th quantile of the composite index of the joint characteristic matrix of the graphene underfloor heating module's temperature field change rate and prediction uncertainty throughout its entire life cycle. Significant deviation segment: The continuous interval from the second change point to the end of the life cycle. The composite index of the joint characteristic matrix of the graphene underfloor heating module's temperature field change rate and prediction uncertainty throughout its entire life cycle is greater than the 99th quantile of the composite index of the joint characteristic matrix of the graphene underfloor heating module's temperature field change rate and prediction uncertainty throughout its entire life cycle.

[0016] S3. Using the local anomaly attribution path of temperature drift of graphene floor heating module, construct a temperature anomaly early warning model for graphene floor heating module, verify the supporting variables of local anomaly of temperature drift relative to the real-time operation data of graphene floor heating module, and predict the critical moment of local anomaly of temperature drift of graphene floor heating module in the future. Step S3 specifically includes: Based on the graphene floor heating module sensor, the real-time temperature field vector, real-time input power, cumulative power-on time and real-time ambient temperature of the graphene floor heating module are obtained according to a fixed sampling period. These are combined as the real-time operation data of the graphene floor heating module. The real-time temperature field vector of the graphene floor heating module is reduced in dimension by combining the dimension reduction representation matrix of the temperature distribution vector of several graphene floor heating modules throughout their entire life cycle, and the dimension reduction representation of the real-time temperature distribution vector of the graphene floor heating module is obtained. The principal component score vector of the real-time temperature of the graphene floor heating module is then calculated. Based on the local anomaly attribution path of temperature drift in the graphene underfloor heating module, the real-time observation node variables of the graphene underfloor heating module in the current time window are calculated, and the real-time support variable vector of the local anomaly attribution path of temperature drift in the graphene underfloor heating module is constructed. The historical temperature data corresponding to the time step of the stable benchmark segment to which the graphene floor heating module belongs is traversed. The real-time support variable vector sample set of the local anomaly attribution path of temperature drift of the graphene floor heating module is extracted. The mean and covariance matrix of the real-time support variable vector sample set are calculated as the health benchmark of the graphene floor heating module. The Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path for temperature drift in the graphene underfloor heating module and the health benchmark of the graphene underfloor heating module is calculated. A health benchmark threshold is set. If the Mahalanobis distance is less than or equal to the health benchmark threshold, it is determined that the real-time operating data of the graphene underfloor heating module does not support any local anomaly attribution path for temperature drift in the graphene underfloor heating module, and the graphene underfloor heating module is in a healthy operating state. If the Mahalanobis distance is greater than the health benchmark threshold, it is determined that the real-time operating data of the graphene underfloor heating module supports any local anomaly attribution path for temperature drift in the graphene underfloor heating module, and the graphene underfloor heating module triggers a temperature anomaly warning signal. To further explain, the health benchmark threshold refers to the mean of the Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path of temperature drift of the graphene underfloor heating module and the health benchmark of the graphene underfloor heating module, plus three standard deviations, within the stable benchmark time window. Based on the temperature anomaly warning signal triggered by the graphene floor heating module, the real-time operation data of the graphene floor heating module in the current time window is extracted. According to the temperature time series evolution trend model of the graphene floor heating module, the temperature drift anomaly severity index of the graphene floor heating module in the current time window is calculated and denoted as the real-time temperature drift level of the graphene floor heating module. The real-time temperature drift rate of the graphene floor heating module is calculated and the real-time temperature drift level and rate of the graphene floor heating module are used as the real-time state vector of the graphene floor heating module. Based on the physical characteristics of temperature drift during the operation of the graphene floor heating module, and combined with the Wiener process with random acceleration, the graphene floor heating module's temperature real-time drift level, rate, and time interval product are used as the drift level of the graphene floor heating module at the previous moment, and the graphene floor heating module's temperature real-time drift rate is used as the drift rate of the graphene floor heating module at the previous moment, thus establishing the graphene floor heating state transition matrix. The Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path of temperature drift in the graphene floor heating module and the health benchmark of the graphene floor heating module is used as the observation. Based on the temperature anomaly warning signal triggered by the graphene floor heating module, the weighted sum of the estimated causal effect intensity of each direct cause variable on the local anomaly attribution path of temperature drift is used as the initial temperature drift level of the graphene floor heating module. The average temperature drift rate of the graphene floor heating module in the current time window is used as the initial temperature drift rate of the graphene floor heating module. The random number seed of the graphene floor heating module is set, and the initial particle set is generated by sampling from the preset prior distribution and the particles are assigned equal initial weights. To further explain, the preset prior distribution is as follows: based on the historical temperature data statistics of the stable reference segment to which the graphene floor heating module belongs, the temperature drift level and temperature drift rate of all time steps in the stable reference segment are extracted, the mean and variance are calculated, and the drift level and drift rate are set to follow a normal distribution as the prior distribution for the initial state sampling of particles. Each particle contains the temperature drift level and temperature drift rate. Based on the graphene floor heating state transition matrix, the particle state of the graphene floor heating module at the next moment is predicted. According to the Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path of the temperature drift of the graphene floor heating module and the health benchmark of the graphene floor heating module, the likelihood weight of each particle is calculated, and normalization and resampling are performed to generate a new particle set. The posterior expectation of the state of the real-time support variable vector of the local anomaly attribution path of the temperature drift of the graphene floor heating module is calculated as the state estimate at the current moment. Taking the new particle set as the starting point, it is substituted into the graphene floor heating state transition matrix to generate the set of temperature state evolution paths of the graphene floor heating module within a specified time period in the future. The minimum value of the lower boundary effect variable of the significantly deviated segment of the local anomaly attribution path of temperature drift in the graphene underfloor heating module is used as the critical threshold for significant anomaly. The time point when the subset of the temperature state evolution path set of the graphene underfloor heating module is first greater than or equal to the critical threshold for significant anomaly within a specified time period in the future is found. The statistical distribution of the time when all particles cross the critical threshold for significant anomaly is calculated. The median of the time when all particles cross the critical threshold for significant anomaly is used as the critical moment of local anomaly of temperature drift in the future graphene underfloor heating module.

[0017] S4. Based on the critical moment of local anomaly of temperature drift in the future graphene floor heating module and the selectable drift suppression control action, establish a drift-suppression predictive control network, and generate the optimal real-time temperature drift compensation scheme for local anomalies of the graphene floor heating module according to the temperature drift dynamic compensation function. Step S4 specifically includes: Using the critical moment of local anomaly in temperature drift of the graphene floor heating module, the posterior expectation of the state of the real-time support variable vector of the local anomaly attribution path of temperature drift of the graphene floor heating module, and the dataset of anomaly type performance of the graphene floor heating module as input, and based on the physical controllable input of the graphene floor heating module, the power regulation amount, the duty cycle regulation amount of the on / off cycle, and the preheating / slow cooling timing offset of the graphene floor heating module are combined and encoded into a discretized set of candidate drift suppression control actions. To further explain, the optional drift suppression control action set includes: power adjustment of the graphene floor heating module, with an adjustment range of [-20%, +20%] of the rated power; duty cycle adjustment of the on / off cycle, which adjusts the duty cycle of the pulse width modulation control signal, with an adjustment range of [-15%, +15%]; and preheating / slow cooling timing offset, which shifts the time phase of the temperature control curve forward or backward, with an adjustment range of [-30, +30] minutes. Using the model predictive control algorithm, the control time domain and prediction time domain are set, the discrete sampling time interval is determined, the temperature drift state vector of the graphene floor heating module in the future prediction step is used as the state vector, and the drift suppression control action applied in the future control step is used as the control action. A state space model of the drift-suppression predictive control network is constructed. The weighted sum of the drift suppression error term, the control energy penalty term and the control smoothness penalty term is used as the temperature drift dynamic compensation function. The state constraints and control input constraints are used as constraints to generate several candidate suppression action sequences for the temperature drift of the graphene floor heating module. Further explanation: The discrete sampling time interval is determined to be consistent with the data acquisition time interval for the entire lifecycle test. Control time domain: The future time steps for executing optional drift suppression control are set as the control time domain. Based on the physical characteristics of the thermal inertia of the graphene underfloor heating module, the control time domain value is set to 3 to 5 control cycles, with each control cycle being 15 to 30 minutes. Prediction time domain: The future time steps for predicting the evolution of the temperature drift state are set as the prediction time domain. The product of the prediction time domain and the discrete sampling time interval is greater than or equal to the difference between the critical moment of the future local anomaly of the temperature drift of the graphene underfloor heating module and the current moment, and the prediction time domain is greater than or equal to the control time domain. The state vector includes: the cumulative temperature drift of the graphene underfloor heating module. The drift level represents the degree of temperature deviation relative to the healthy baseline at the current moment and the temperature drift rate of the graphene floor heating module, representing the change in temperature drift level per unit time; the drift suppression error term is the weighted sum of the squares of the differences between the drift level of each prediction step and the healthy baseline of the graphene floor heating module within the prediction time domain; the control energy penalty term is the weighted sum of the squares of the control action amplitudes of each control step within the control time domain; the control smoothness penalty term is the weighted sum of the squares of the differences in control actions between adjacent control steps within the control time domain; the specific constraints are: state constraint: the temperature drift level of the graphene floor heating module does not exceed 80% of the lower boundary threshold of the significant deviation segment; control input constraint: optional drift suppression control action set parameter constraints; Based on generating several candidate suppression action sequences for temperature drift of the graphene floor heating module, calculate the temperature drift dynamic compensation function value corresponding to each candidate sequence, and select the candidate sequence with the smallest compensation function value as the optimal real-time temperature drift compensation scheme for local anomalies of the graphene floor heating module. To further explain, for each candidate temperature drift suppression action sequence of the graphene underfloor heating module, the drift state evolution trajectory after applying the control action sequence is predicted according to the drift-suppression predictive control network state space model. The drift suppression error term corresponding to the trajectory is calculated, as well as the control energy penalty term and control smoothness penalty term corresponding to the sequence. The three terms are weighted and summed to obtain the comprehensive score of the sequence. The candidate sequence with the smallest compensation function value is selected as the optimal real-time temperature drift compensation scheme for local anomalies of the graphene underfloor heating module. The specific optimal real-time temperature drift compensation scheme for local anomalies of the graphene underfloor heating module is as follows: For each sampling time, starting from the current temperature drift state of the graphene underfloor heating module, the constraints are calculated in the control time domain. If the temperature drift state compensation function has a convex quadratic form with respect to the control action and the constraints are linear, then the optimal real-time temperature drift compensation scheme is obtained through quadratic programming analysis. The optimal real-time temperature drift compensation scheme for local anomalies in the graphene underfloor heating module is obtained by solving the problem. If the temperature drift dynamic compensation function or constraint conditions are nonlinear, a differential evolution algorithm is used to obtain several candidate suppression action sequences for the temperature drift of the graphene underfloor heating module. Specifically, a population containing 50 candidate suppression action sequences for the temperature drift of the graphene underfloor heating module is initialized. Each candidate sequence corresponds to a set of continuous control actions in the control time domain. The mutation factor is set to 0.8 and the crossover probability is set to 0.9. The temperature drift dynamic compensation function value is used as the fitness function. After 100 generations of iterative evolution, the candidate sequence with the smallest fitness is selected as the approximate optimal solution. In each sampling period, only the first control action at the current moment is executed, and rolling optimization is performed again at the next sampling moment to obtain the optimal real-time temperature drift compensation scheme for local anomalies in the graphene underfloor heating module.

[0018] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for real-time temperature drift compensation and control of a graphene underfloor heating module, characterized in that, include: S1. Based on the full life cycle test data of several graphene underfloor heating modules in history, analyze the time series distribution of temperature field of several graphene underfloor heating modules, construct a time series evolution trend model of temperature of graphene underfloor heating modules, and output the full life cycle temperature distribution vector of graphene underfloor heating modules. S2. Based on the known abnormality data of temperature drift of graphene floor heating module, construct the causal network of abnormality types of graphene floor heating module, divide the temperature distribution vector of the whole life cycle of graphene floor heating module, and obtain the local anomaly attribution path of temperature drift of graphene floor heating module. S3. Using the local anomaly attribution path of temperature drift of graphene floor heating module, construct a temperature anomaly early warning model for graphene floor heating module, verify the supporting variables of local anomaly of temperature drift relative to the real-time operation data of graphene floor heating module, and predict the critical moment of local anomaly of temperature drift of graphene floor heating module in the future. S4. Based on the critical moment of local anomalies in temperature drift of the future graphene floor heating module and the selectable drift suppression control action, establish a drift-suppression predictive control network, and generate the optimal real-time temperature drift compensation scheme for local anomalies of the graphene floor heating module according to the temperature drift dynamic compensation function.

2. The method for real-time temperature drift compensation and control of a graphene underfloor heating module according to claim 1, characterized in that, Step S1 specifically includes: Based on the full life cycle test data of several historical graphene underfloor heating modules, a two-dimensional matrix of the full life cycle test data of a single graphene underfloor heating module is established with the collection time point as the row and the spatial temperature measurement points of the historical graphene underfloor heating module as the column, and data preprocessing is performed. For several graphene underfloor heating modules, a two-dimensional matrix of the full life cycle test data of each graphene underfloor heating module is laid out sequentially according to time sequence to generate several graphene underfloor heating module dataset matrices and calculate the spatial covariance matrix. The spatial covariance matrix is ​​decomposed into eigenvalues ​​to obtain the eigenvalues ​​and corresponding unit eigenvectors of the spatial covariance matrix. The eigenvalues ​​of the spatial covariance matrix are sorted in descending order according to the eigenvalues ​​of the spatial covariance matrix, and the unit eigenvectors corresponding to the spatial covariance matrix are taken as the principal component directions. The ratio of the eigenvalue corresponding to each principal component direction to the sum of all eigenvalues ​​is calculated as the variance contribution rate of the principal component direction. The cumulative variance contribution rate is continuously calculated from front to back. A preset cumulative variance contribution rate threshold is set. The top few principal component directions that meet the threshold are extracted, and the number of extracted principal component directions is used as the number of principal components to be retained. The row vector of the temperature field corresponding to any moment in the dataset matrix of several graphene underfloor heating modules is projected onto the top few principal component directions and arranged in chronological order to generate a dimension-reduced representation matrix of the temperature distribution vector of several graphene underfloor heating modules throughout their entire life cycle, thus obtaining the temporal distribution of the temperature field of several graphene underfloor heating modules throughout their entire life cycle. The preset cumulative variance contribution rate threshold specifically refers to a percentage value pre-set based on the maximum number of principal components allowed by computing resource limitations, which determines the number of principal component directions that need to be retained during the dimensionality reduction process. Based on the time-series temperature field distribution of several graphene underfloor heating modules throughout their entire life cycle, and using the data acquisition time point as input, a combined kernel function for the graphene underfloor heating modules is designed according to the physical characteristics of temperature drift during the operation of the modules. The hyperparameters of the combined kernel function are optimized by maximizing the logarithmic marginal likelihood, and a time-series temperature evolution trend model of the graphene underfloor heating modules is constructed, outputting the temperature distribution vector of the graphene underfloor heating modules throughout their entire life cycle.

3. The method for real-time temperature drift compensation and control of a graphene underfloor heating module according to claim 2, characterized in that, Step S2 specifically includes: Based on the full life cycle test data of several historical graphene floor heating modules and the temperature drift anomaly data of graphene floor heating modules with known anomaly type labels obtained from accelerated aging experiments in the laboratory, a graphene floor heating module anomaly type performance dataset is generated by combining the known anomaly type labels. The known anomaly type labels include: edge cold zone diffusion type, center hot spot attenuation type, resistance jump oscillation type, and adhesive layer peeling slow change type. Based on the temperature distribution vector of the graphene underfloor heating module throughout its entire life cycle, a sliding time window is set up. The spatiotemporal statistical characteristics and temperature control behavior characteristics of the temperature field of the graphene underfloor heating module within the time window are calculated and denoted as candidate statistics. The dataset of abnormal types of graphene underfloor heating modules is used as positive samples, and the time window samples in the temperature distribution vector of the graphene underfloor heating module throughout its entire life cycle without any abnormal types are used as negative samples. The independent samples t-test is used to calculate the mean difference and significance of the difference between the positive samples and the negative samples on each candidate statistics. Candidate statistics with a significance of difference less than the preset significance threshold are selected as observation node variables. Based on a two-dimensional matrix of full life-cycle test data for a single graphene underfloor heating module, we obtain historical temperature data of the graphene underfloor heating module at known times. Substituting this data into the time-series evolution trend model of the graphene underfloor heating module temperature, we predict the nominal temperature field of the graphene underfloor heating module at the same time in the future. We calculate the weighted root mean square value of the residual between the nominal temperature field and the measured temperature of the graphene underfloor heating module at the same time in the future, and obtain the index of the severity of abnormal temperature drift of the graphene underfloor heating module at that time, which is recorded as the effect variable. We record the sub-parameters of the spatiotemporal statistical characteristics and temperature control behavior characteristics of the graphene underfloor heating module temperature field within the time window as candidate causal variables, and construct a time-slice aligned causal network variable set for the temperature drift of the graphene underfloor heating module. Based on the time-slice aligned causal network variable set and observation node variables of the graphene floor heating module temperature drift, a graphene floor heating module temperature drift observation data matrix is ​​constructed. The observation node variables are used as nodes and any two different node pairs are connected by undirected edges to establish a completely undirected graph of graphene floor heating module temperature drift. Based on the completely undirected graph of temperature drift of graphene floor heating module, the partial correlation coefficient of each node pair is calculated under the condition of empty set and the conditional independence test is performed. If the node pair is conditionally independent under the condition of empty set, the undirected edge corresponding to the node pair is deleted and the empty set is recorded as the separation set of the node pair. All node pairs are traversed to generate the completely undirected skeleton graph of temperature drift of graphene floor heating module. Using the V structure, we traverse the undirected edge triplet corresponding to the complete undirected skeleton graph of temperature drift of graphene floor heating module, set the test conditions, identify the undirected edge triplet that meets the collision structure and determine the direction of the undirected edge, and construct the partial directed acyclic graph of temperature drift equivalence class of graphene floor heating module. Find all directed edges pointing to the effect variable in the partial directed acyclic graph of the graphene underfloor heating module temperature drift equivalence class. Mark the starting node of the directed edge as the direct cause of the effect variable. For any direct cause of the effect variable and an upstream cause node exists, mark the upstream node as the indirect cause of the effect variable. When there is an undirected edge between the effect variable and any candidate cause variable in the partial directed acyclic graph of the graphene underfloor heating module temperature drift equivalence class, determine the candidate cause variable as a potential confounding factor of the graphene underfloor heating module temperature drift. Based on the directed acyclic graph of the temperature drift equivalence class of the graphene underfloor heating module, all variables with direct directed edges that are direct causes of the temperature drift anomaly severity index of the graphene underfloor heating module are extracted and denoted as the treatment variable set. All remaining variables outside the treatment variable set are denoted as the control variable set. The attribution path weight coefficients of the variables in the treatment variable set to the effect variables are calculated to obtain the causal effect strength estimate of each direct cause variable on the temperature drift anomaly severity index. The causal effect strength estimate is used as the weight of the corresponding directed edge and the directed acyclic graph of the temperature drift equivalence class of the graphene underfloor heating module is labeled to construct the causal network of the graphene underfloor heating module anomaly type.

4. The method for real-time temperature drift compensation and control of a graphene underfloor heating module according to claim 3, characterized in that, Step S2 also includes: Based on the dimensionality reduction representation of the temperature distribution vector throughout the entire life cycle of the graphene underfloor heating module, the principal component score vector of the temperature throughout the entire life cycle of the graphene underfloor heating module is obtained. The Euclidean norm of the difference vector of the principal component score vector of the temperature of the graphene underfloor heating module at adjacent time steps is calculated, and the time series vector of the temperature field change rate of the graphene underfloor heating module throughout the entire life cycle is obtained. Substitute the principal component score vector of the graphene underfloor heating module throughout its entire life cycle into the temperature time series evolution trend model of the graphene underfloor heating module, output the principal component score prediction variance corresponding to each time step of the graphene underfloor heating module, and obtain the time series vector sequence of principal component score prediction variance corresponding to the entire life cycle of the graphene underfloor heating module. By splicing the time series vector of the temperature field change rate throughout the entire life cycle of the graphene underfloor heating module with the time series vector sequence of the principal component score prediction variance corresponding to the entire life cycle of the graphene underfloor heating module, a joint feature matrix of temperature field change rate and prediction uncertainty throughout the entire life cycle of the graphene underfloor heating module is established. The composite index of the temperature field change rate-prediction uncertainty joint feature matrix of the graphene underfloor heating module throughout its entire life cycle is calculated for each time step, and Bayesian change point detection is performed. The number of change points is set to 2. The entire life cycle of the graphene underfloor heating module is divided into three continuous intervals, which are defined in chronological order as the stable baseline segment, the trend initiation segment, and the significant deviation segment. The state segment label of each time step of the graphene underfloor heating module is obtained. The historical temperature data corresponding to the time steps of the trend initiation segment and the significant deviation segment of the graphene underfloor heating module are selected as the input data matrix of the causal network of the graphene underfloor heating module's anomaly type, and the local anomaly attribution path of temperature drift of the graphene underfloor heating module is obtained.

5. The method for real-time temperature drift compensation and control of a graphene underfloor heating module according to claim 4, characterized in that, Step S3 specifically includes: Based on the graphene floor heating module sensor, the real-time temperature field vector, real-time input power, cumulative power-on time and real-time ambient temperature of the graphene floor heating module are obtained according to a fixed sampling period. These are combined as the real-time operation data of the graphene floor heating module. The real-time temperature field vector of the graphene floor heating module is reduced in dimension by combining the dimension reduction representation matrix of the temperature distribution vector of several graphene floor heating modules throughout their entire life cycle, and the dimension reduction representation of the real-time temperature distribution vector of the graphene floor heating module is obtained. The principal component score vector of the real-time temperature of the graphene floor heating module is then calculated. Based on the local anomaly attribution path of temperature drift in the graphene underfloor heating module, the real-time observation node variables of the graphene underfloor heating module in the current time window are calculated, and the real-time support variable vector of the local anomaly attribution path of temperature drift in the graphene underfloor heating module is constructed. The historical temperature data corresponding to the time step of the stable benchmark segment to which the graphene floor heating module belongs is traversed. The real-time support variable vector sample set of the local anomaly attribution path of temperature drift of the graphene floor heating module is extracted. The mean and covariance matrix of the real-time support variable vector sample set are calculated as the health benchmark of the graphene floor heating module. The Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path for temperature drift in the graphene underfloor heating module and the health benchmark of the graphene underfloor heating module is calculated. A health benchmark threshold is set. If the Mahalanobis distance is less than or equal to the health benchmark threshold, it is determined that the real-time operating data of the graphene underfloor heating module does not support any local anomaly attribution path for temperature drift in the graphene underfloor heating module, and the graphene underfloor heating module is in a healthy operating state. If the Mahalanobis distance is greater than the health benchmark threshold, it is determined that the real-time operating data of the graphene underfloor heating module supports any local anomaly attribution path for temperature drift in the graphene underfloor heating module, and the graphene underfloor heating module triggers a temperature anomaly warning signal. Based on the temperature anomaly warning signal triggered by the graphene floor heating module, the real-time operation data of the graphene floor heating module in the current time window is extracted. According to the temperature time series evolution trend model of the graphene floor heating module, the temperature drift anomaly severity index of the graphene floor heating module in the current time window is calculated and denoted as the real-time temperature drift level of the graphene floor heating module. The real-time temperature drift rate of the graphene floor heating module is calculated and the real-time temperature drift level and rate of the graphene floor heating module are used as the real-time state vector of the graphene floor heating module. Based on the physical characteristics of temperature drift during the operation of the graphene floor heating module, and combined with the Wiener process with random acceleration, the graphene floor heating module's temperature real-time drift level, rate, and time interval product are used as the drift level of the graphene floor heating module at the previous moment, and the graphene floor heating module's temperature real-time drift rate is used as the drift rate of the graphene floor heating module at the previous moment, thus establishing the graphene floor heating state transition matrix. The Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path of temperature drift in the graphene floor heating module and the health benchmark of the graphene floor heating module is used as the observation. Based on the temperature anomaly warning signal triggered by the graphene floor heating module, the weighted sum of the estimated causal effect intensity of each direct cause variable on the local anomaly attribution path of temperature drift is used as the initial temperature drift level of the graphene floor heating module. The average temperature drift rate of the graphene floor heating module in the current time window is used as the initial temperature drift rate of the graphene floor heating module. The random number seed of the graphene floor heating module is set, and the initial particle set is generated by sampling from the preset prior distribution and the particles are assigned equal initial weights. Based on the graphene floor heating state transition matrix, the particle state of the graphene floor heating module at the next moment is predicted. According to the Mahalanobis distance between the real-time support variable vector of the local anomaly attribution path of the temperature drift of the graphene floor heating module and the health benchmark of the graphene floor heating module, the likelihood weight of each particle is calculated, and normalization and resampling are performed to generate a new particle set. The posterior expectation of the state of the real-time support variable vector of the local anomaly attribution path of the temperature drift of the graphene floor heating module is calculated as the state estimate at the current moment. Taking the new particle set as the starting point, it is substituted into the graphene floor heating state transition matrix to generate the set of temperature state evolution paths of the graphene floor heating module within a specified time period in the future. The minimum value of the lower boundary effect variable of the significantly deviated segment of the local anomaly attribution path of temperature drift in the graphene underfloor heating module is used as the critical threshold for significant anomaly. The time point when the subset of the temperature state evolution path set of the graphene underfloor heating module is first greater than or equal to the critical threshold for significant anomaly within a specified time period in the future is found. The statistical distribution of the time when all particles cross the critical threshold for significant anomaly is calculated. The median of the time when all particles cross the critical threshold for significant anomaly is used as the critical moment of local anomaly of temperature drift in the future graphene underfloor heating module.

6. The method for real-time temperature drift compensation and control of a graphene underfloor heating module according to claim 5, characterized in that, Step S4 specifically includes: Using the critical moment of local anomaly in temperature drift of the graphene floor heating module, the posterior expectation of the state of the real-time support variable vector of the local anomaly attribution path of temperature drift of the graphene floor heating module, and the dataset of anomaly type performance of the graphene floor heating module as input, and based on the physical controllable input of the graphene floor heating module, the power regulation amount, the duty cycle regulation amount of the on / off cycle, and the preheating / slow cooling timing offset of the graphene floor heating module are combined and encoded into a discretized set of candidate drift suppression control actions. Using the model predictive control algorithm, the control time domain and prediction time domain are set, the discrete sampling time interval is determined, the temperature drift state vector of the graphene floor heating module in the future prediction step is used as the state vector, and the drift suppression control action applied in the future control step is used as the control action. A state space model of the drift-suppression predictive control network is constructed. The weighted sum of the drift suppression error term, the control energy penalty term and the control smoothness penalty term is used as the temperature drift dynamic compensation function. The state constraints and control input constraints are used as constraints to generate several candidate suppression action sequences for the temperature drift of the graphene floor heating module. Based on generating several candidate suppression action sequences for temperature drift of the graphene underfloor heating module, the dynamic compensation function value of temperature drift corresponding to each candidate sequence is calculated, and the candidate sequence with the smallest compensation function value is selected as the optimal real-time temperature drift compensation scheme for local anomalies of the graphene underfloor heating module.

Images