Three-dimensional temperature field construction system and method based on infrared sensing temperature measurement data calibration

Abstract

The invention discloses a three-dimensional temperature field construction system and method based on infrared sensing temperature measurement data calibration, belongs to the technical field of temperature field prediction, and solves the problem of overfitting when an existing neural network model processes a complex nonlinear relationship. In order to solve the problems that a single neural network model is limited in generalization ability when facing different scenes and data, the method comprises the following steps: preprocessing input variables; respectively constructing three regression models by using the three gradient boosting decision tree models, pre-constructing a prediction fusion model based on the regression models, executing the trained and optimized prediction fusion model, and outputting a final calibration temperature; according to the method, three regression models are respectively trained through three gradient boosting decision tree models, namely XGBoost, LightGBM and CatBoost, and meanwhile, prediction results of a fusion model through weighting or combining a plurality of base models are predicted, so that a more stable prediction result is provided, and deviation and errors possibly existing in a single model are further reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of temperature field prediction, and particularly relates to a three-dimensional temperature field construction system and method based on infrared sensor temperature measurement data calibration. BACKGROUND

[0002] As a non-contact temperature detection method, infrared sensor temperature measurement technology has been widely used in industrial equipment state monitoring, power system fault diagnosis, building energy-saving evaluation, medical health detection and security monitoring, etc. due to its fast response speed, high spatial resolution, no need to contact the measured object and other advantages. However, the accuracy of infrared temperature measurement is easily affected by multiple factors such as the surface material of the measured object, the measurement distance, the environmental temperature, the air flow (wind speed) and the observation angle, resulting in significant deviation in the original temperature data directly output by the infrared sensor, which is difficult to meet the demand of high-precision temperature field reconstruction. Therefore, how to realize accurate calibration of infrared sensor temperature measurement data through multi-dimensional information fusion and model optimization to construct a reliable three-dimensional temperature field is still a technical problem to be solved in this field.

[0003] At present, when calibrating the infrared sensor temperature measurement data, a neural network is generally used for temperature calibration and prediction. However, the neural network model has the problem of overfitting when dealing with complex nonlinear relationships, has a high requirement for the processing of input features, and the single neural network model has limited generalization ability when facing different scenes and data, which is difficult to adapt to diversified temperature measurement needs. In view of the above problems, we propose a three-dimensional temperature field construction system and method based on infrared sensor temperature measurement data calibration. SUMMARY

[0004] The present application aims to solve the problems of the prior art, such as the overfitting of the existing neural network model when dealing with complex nonlinear relationships, the high requirement for the processing of input features, and the limited generalization ability of the single neural network model when facing different scenes and data, which is difficult to adapt to diversified temperature measurement needs.

[0005] The present application is implemented as follows: a three-dimensional temperature field construction method based on infrared sensor temperature measurement data calibration, the method comprising:

[0006] Based on the infrared sensor, the original measurement temperature and the environmental temperature are collected in real time, and based on the sensor group, the distance parameter, the surface material, the three-dimensional angle and the wind speed are collected, and the original measurement temperature, the environmental temperature, the distance parameter, the surface material, the three-dimensional angle and the wind speed are integrated as input variables;

[0007] The collected input variables are loaded, the input variables are preprocessed, and a preprocessed set is output.

[0008] Three regression models are respectively constructed using three gradient boosting decision tree models, the three regression models are trained on the same base model training set to generate base model prediction results;

[0009] A prediction fusion model based on regression models is pre-constructed, the prediction fusion model is trained and optimized, and the prediction fusion model is super parameter optimized, the base model prediction results are loaded as new feature inputs of the prediction fusion model, the trained and optimized prediction fusion model is executed, and the final calibrated temperature is output;

[0010] A three-dimensional temperature field is constructed based on the calibrated real temperature and three-dimensional perspective information, and the constructed three-dimensional temperature field is output.

[0011] Preferably, the method for preprocessing the input variables comprises:

[0012] The input variables are loaded and cleaned, and missing values and outliers in the input variables are removed;

[0013] The input variables from which the missing values and outliers are removed are obtained, the types of the input variables are identified, and the input variable features are constructed based on the types of the input variables.

[0014] Preferably, the method for constructing the input variable features based on the types of the input variables comprises:

[0015] The types of the input variables are identified, and the types of the input variables are determined;

[0016] If the type of the input variable is a category feature, the input variable is encoded;

[0017] If the type of the input variable is a continuous feature, the input variable is normalized, the original value is preserved, or a derived feature is introduced.

[0018] Preferably, when the three regression models are respectively constructed using the three gradient boosting decision tree models, the three gradient boosting decision tree models are XGBoost regression model, LightGBM regression model and CatBoost regression model.

[0019] Preferably, the method for training the three regression models on the same base model training set comprises:

[0020] Modeling sample sets are extracted from historical input variables, and the modeling sample sets are divided into base model training sets and base model validation sets;

[0021] Pre-built XGBoost regression models, LightGBM regression models, and CatBoost regression models are trained using the base model training set, and hyperparameters are tuned for the XGBoost regression models, LightGBM regression models, and CatBoost regression models to output converged XGBoost regression models, LightGBM regression models, and CatBoost regression models.

[0022] Load the base model validation set, and use the base model validation set to evaluate the prediction accuracy of the XGBoost regression model, LightGBM regression model and CatBoost regression model respectively, and output the prediction values ​​of the XGBoost regression model, LightGBM regression model and CatBoost regression model respectively.

[0023] Obtain the predicted values ​​of the XGBoost regression model, LightGBM regression model, and CatBoost regression model, and concatenate the predicted values ​​of the XGBoost regression model, LightGBM regression model, and CatBoost regression model to form the prediction result of the base model.

[0024] Preferably, when pre-constructing a prediction fusion model based on a regression model, the prediction fusion model is a linear regression model, a small multilayer perceptron (MLP) model, or a ridge regression model.

[0025] Preferably, when performing hyperparameter tuning for XGBoost regression models, LightGBM regression models, and CatBoost regression models, and when performing hyperparameter tuning for prediction fusion models, the hyperparameters adjusted include max_depth, learning_rate, subsampling ratio, or regularization term coefficient. Grid search, Bayesian optimization, or automated hyperparameter tuning tools are used to find the optimal parameter combination during hyperparameter tuning.

[0026] Preferably, the method for constructing a three-dimensional temperature field based on calibrated real temperature and three-dimensional perspective information includes:

[0027] The three-dimensional space is divided into a voxel grid, and the temperature value of each voxel is the average value of all calibrated temperature points within the voxel grid.

[0028] The calibrated real temperature is associated with the corresponding three-dimensional coordinate points to generate point cloud data with temperature attributes;

[0029] Discrete temperature points are fitted to a continuous surface using surface interpolation, and the surface temperature distribution is output.

[0030] On the other hand, the present invention also provides a three-dimensional temperature field construction system based on infrared sensing thermometry data calibration, the system comprising:

[0031] The data acquisition module collects raw measured temperature and ambient temperature in real time based on infrared sensors, and collects distance parameters, surface material, three-dimensional view, and wind speed based on the sensor group. It integrates raw measured temperature, ambient temperature, distance parameters, surface material, three-dimensional view, and wind speed as input variables.

[0032] The preprocessing module is used to load the collected input variables, preprocess the input variables, and output the preprocessed set.

[0033] The base model training module uses three gradient boosting decision tree models to construct three regression models respectively. The three regression models are trained on the same base model training set to generate base model prediction results.

[0034] The fusion model training module pre-builds a prediction fusion model based on a regression model, trains and optimizes the prediction fusion model, performs hyperparameter tuning on the prediction fusion model, loads the prediction results of the base model, uses the prediction results of the base model as new feature inputs for the prediction fusion model, executes the trained and optimized prediction fusion model, and outputs the final calibrated temperature.

[0035] The temperature field construction module constructs a three-dimensional temperature field based on the calibrated real temperature and three-dimensional perspective information, and outputs the constructed three-dimensional temperature field.

[0036] Preferably, the preprocessing module includes:

[0037] The data cleaning unit is used to load input variables, clean the input variables, and remove missing and outlier values ​​from the input variables.

[0038] The type recognition unit is used to acquire input variables after removing missing and outlier values ​​from the input variables, identify the type of the input variables, and construct and process the features of the input variables based on the type of the input variables.

[0039] The feature processing unit identifies the type of the input variable and determines its type. If the type of the input variable is a categorical feature, it encodes the input variable. If the type of the input variable is a continuous feature, it normalizes the input variable, preserves the original value, or introduces derived features.

[0040] Compared with the prior art, the embodiments of this application have the following main advantages:

[0041] This invention employs three gradient boosting decision tree models—XGBoost, LightGBM, and CatBoost—to train three regression models. These models then predict temperature using the same set of input features, allowing the base model predictions to complement the biases of individual models. This effectively reduces the risk of overfitting in individual models and improves overall generalization ability. Simultaneously, the prediction fusion model, by weighting or combining the predictions of multiple base models, smooths out the prediction fluctuations of individual models, providing more stable prediction results. This further reduces the potential biases and errors of individual models, effectively improving the robustness and prediction accuracy of the overall model.

[0042] In this embodiment of the invention, in the calibration of infrared sensing temperature data and the construction of a three-dimensional temperature field, the input variables include distance and surface material factors. By considering the surface material factors, the predictive fusion model can better capture the true relationships in the data, rather than relying solely on the local features of the training data. This helps reduce the risk of overfitting, improves the performance of the predictive fusion model on new data, and the introduction of distance and surface material factors can significantly improve the accuracy of temperature measurement and the robustness of the model, enhance the generalization ability of the predictive fusion model, and adapt to diverse measurement scenarios and objects. Attached Figure Description

[0043] Figure 1 This is a schematic diagram illustrating the implementation process of the three-dimensional temperature field construction method based on infrared sensing temperature data calibration provided by the present invention.

[0044] Figure 2 This is a schematic diagram of the structure of the three-dimensional temperature field construction system based on infrared sensing temperature data calibration provided by the present invention. Detailed Implementation

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0046] Currently, neural network models suffer from overfitting when dealing with complex nonlinear relationships, have high requirements for input feature processing, and single neural network models have limited generalization ability when facing different scenarios and data, making it difficult to adapt to diverse temperature measurement needs. To address these issues, we propose a three-dimensional temperature field construction system and method based on infrared sensing temperature data calibration. In short, the method first integrates the original measured temperature, ambient temperature, distance parameters, surface material, three-dimensional viewpoint, and wind speed as input variables, preprocesses the input variables, and outputs a preprocessed set. Three regression models are constructed using three gradient boosting decision tree models, and the three regression models are trained on the same base model training set. A prediction fusion model based on the regression models is preconstructed, trained and optimized, and its hyperparameters are tuned. The prediction results of the base models are loaded and used as new feature inputs for the prediction fusion model. The trained and optimized prediction fusion model is then executed, and the final calibrated temperature is output. This invention employs three gradient boosting decision tree models—XGBoost, LightGBM, and CatBoost—to train three regression models. These models then predict temperature using the same set of input features, allowing the base model predictions to complement the biases of individual models. This effectively reduces the risk of overfitting in individual models and improves overall generalization ability. Simultaneously, the prediction fusion model, by weighting or combining the predictions of multiple base models, smooths out the prediction fluctuations of individual models, providing more stable prediction results. This further reduces the potential biases and errors of individual models, effectively improving the robustness and prediction accuracy of the overall model.

[0047] This invention provides a method for constructing a three-dimensional temperature field based on infrared sensing thermometric data calibration. Figure 1 A schematic diagram illustrating the implementation process of a three-dimensional temperature field construction method based on infrared sensor temperature data calibration is shown. The method specifically includes:

[0048] S10 collects raw measured temperature and ambient temperature in real time based on infrared sensors, and collects distance parameters, surface material, three-dimensional view and wind speed based on sensor group, and integrates raw measured temperature, ambient temperature, distance parameters, surface material, three-dimensional view and wind speed as input variables;

[0049] S20: Load the collected input variables, preprocess the input variables, and output the preprocessed set;

[0050] S30: Three regression models are constructed using three gradient boosting decision tree models. The three regression models are trained on the same base model training set to generate the base model prediction results.

[0051] S40: Pre-build a prediction fusion model based on a regression model, train and optimize the prediction fusion model, perform hyperparameter tuning on the prediction fusion model, load the prediction results of the base model, use the prediction results of the base model as new feature inputs of the prediction fusion model, execute the trained and optimized prediction fusion model, and output the final calibrated temperature.

[0052] S50 constructs a three-dimensional temperature field based on the calibrated real temperature and three-dimensional perspective information, and outputs the constructed three-dimensional temperature field.

[0053] This invention employs three gradient boosting decision tree models—XGBoost, LightGBM, and CatBoost—to train three regression models. These models then predict temperature using the same set of input features, allowing the base model predictions to complement the biases of individual models. This effectively reduces the risk of overfitting in individual models and improves overall generalization ability. Simultaneously, the prediction fusion model, by weighting or combining the predictions of multiple base models, smooths out the prediction fluctuations of individual models, providing more stable prediction results. This further reduces the potential biases and errors of individual models, effectively improving the robustness and prediction accuracy of the overall model.

[0054] This invention provides a method for preprocessing input variables, which specifically includes:

[0055] S101, Load input variables, clean the input variables, and remove missing and outlier values ​​from the input variables;

[0056] S102, Obtain the input variables after removing missing and outlier values ​​from the input variables, identify the type of the input variables, and construct and process the input variable features based on the input variable type.

[0057] In this embodiment, the method for constructing input variable features based on input variable type includes:

[0058] S1021, Identify the type of input variable and determine the type of input variable;

[0059] S1022, If the type of the input variable is a category feature, the input variable is encoded; wherein, in this embodiment, the input variable of the type of category feature can be a surface material parameter.

[0060] S1023, If the type of the input variable is continuous feature, normalize the input variable, retain the original value, or introduce derived features.

[0061] Among them, the input variables of the continuous feature type can be distance, wind speed and angle parameters. When introducing derived feature processing for the angle parameter, the direction cosine component of the angle, the square term of wind speed, etc. can be used to enhance the expressive power of the model.

[0062] It should be noted that in the calibration of infrared sensor temperature data and the construction of the three-dimensional temperature field, the input variables include distance and surface material factors. By considering surface material factors, the predictive fusion model can better capture the true relationships in the data, rather than relying solely on local features of the training data. This helps reduce the risk of overfitting, improves the performance of the predictive fusion model on new data, and the introduction of distance and surface material factors can significantly improve the accuracy of temperature measurement and the robustness of the model, enhance the generalization ability of the predictive fusion model, and adapt to diverse measurement scenarios and objects.

[0063] In this embodiment of the invention, when constructing three regression models using three gradient boosting decision tree models, the three models are XGBoost regression, LightGBM regression, and CatBoost regression. Compared to neural networks, GBDT models, based on additive modeling with a tree structure, naturally possess anti-overfitting capabilities and have higher tolerance for categorical features and missing values, eliminating the need for complex feature engineering. Furthermore, XGBoost regression excels at handling sparse data, LightGBM regression efficiently handles large-scale data, and CatBoost regression automatically processes categorical features. Based on different optimization strategies and assumptions, the three regression models, after training on the same training data, can complement each other's prediction results to reduce the bias of a single model. This fusion of "multi-view" prediction results effectively reduces the risk of overfitting in a single model and improves the overall generalization ability.

[0064] This invention provides a method for training three regression models on the same base model training set. The method specifically includes:

[0065] S201, extract the modeling sample set from historical input variables, and divide the modeling sample set into the base model training set and the base model validation set;

[0066] S202 uses the base model training set to train the pre-built XGBoost regression model, LightGBM regression model and CatBoost regression model respectively, and performs hyperparameter tuning on the XGBoost regression model, LightGBM regression model and CatBoost regression model to output the converged XGBoost regression model, LightGBM regression model and CatBoost regression model.

[0067] S203, Load the base model validation set, use the base model validation set to evaluate the prediction accuracy of the XGBoost regression model, LightGBM regression model and CatBoost regression model respectively, and output the prediction values ​​of the XGBoost regression model, LightGBM regression model and CatBoost regression model respectively;

[0068] S204: Obtain the predicted values ​​of the XGBoost regression model, LightGBM regression model, and CatBoost regression model, and concatenate the predicted values ​​of the XGBoost regression model, LightGBM regression model, and CatBoost regression model to form the prediction result of the base model.

[0069] In this embodiment, regression models are constructed using XGBoost, LightGBM, and CatBoost, respectively. The three models are trained on the same training data. The prediction accuracy of each model on the validation set is evaluated, and its prediction results on the validation set are output. Each model generates a predicted value for the same set of input features, which serves as an intermediate feature for subsequent fusion. By integrating multiple models, the generalization ability of the models can be improved, adapting to diverse temperature measurement needs.

[0070] In this embodiment, when pre-constructing the prediction fusion model based on the regression model, the prediction fusion model is a simple and robust model such as a linear regression model, a small multilayer perceptron (MLP) model, or a ridge regression model. The prediction results of the three base models XGBoost, LightGBM, and CatBoost on the base model validation set are concatenated into a new feature vector. Using this feature vector as input and the calibration temperature as output, a prediction fusion model is trained. If the prediction fusion model is a linear regression model, the loss function can be the mean squared error (MSE), and the optimizer is the Adam optimizer. If the prediction fusion model is a small MLP model, the activation function is ReLU, the loss function is MSE, and the optimizer is the Adam optimizer. When the prediction fusion model is a ridge regression model, the optimal hyperparameters can be selected through cross-validation.

[0071] It should be noted that when performing hyperparameter tuning for XGBoost regression models, LightGBM regression models, and CatBoost regression models, and when performing hyperparameter tuning for prediction fusion models, the hyperparameters adjusted include max_depth, learning_rate, subsampling ratio, or regularization term coefficient. Grid search, Bayesian optimization, or automated hyperparameter tuning tools are used to find the optimal parameter combination during hyperparameter tuning.

[0072] This invention provides a method for constructing a three-dimensional temperature field based on calibrated real temperature and three-dimensional viewpoint information. The method specifically includes:

[0073] S301 divides the three-dimensional space into a voxel grid, and the temperature value of each voxel is the average of all calibrated temperature points within the voxel grid.

[0074] S302 establishes a correlation between the calibrated real temperature and the corresponding three-dimensional coordinate points to generate point cloud data with temperature attributes;

[0075] S303 uses surface interpolation to fit discrete temperature points into a continuous surface and outputs the surface temperature distribution.

[0076] In this embodiment, by employing refined voxel mesh generation, temperature correlation of point cloud data, and fitting of continuous surfaces, a high-precision, intuitive, and dynamically updated three-dimensional temperature field distribution can be provided. Furthermore, the method combines calibrated temperature data with three-dimensional perspective information, enabling it to adapt to complex three-dimensional environments. In practical applications, environments may contain various obstacles, surfaces of different materials, and complex heat source distributions. Using three-dimensional perspective information, this method can accurately handle these complex situations and generate reliable temperature field distributions.

[0077] On the other hand, embodiments of the present invention also provide a three-dimensional temperature field construction system based on infrared sensing temperature data calibration. Figure 2 A schematic diagram of a three-dimensional temperature field construction system based on infrared sensor temperature data calibration is shown. This system specifically includes:

[0078] The data acquisition module 100 collects the original measured temperature and ambient temperature in real time based on the infrared sensor, and collects distance parameters, surface material, three-dimensional view and wind speed based on the sensor group, and integrates the original measured temperature, ambient temperature, distance parameters, surface material, three-dimensional view and wind speed as input variables;

[0079] The preprocessing module 200 is used to load the collected input variables, preprocess the input variables, and output a preprocessed set.

[0080] The base model training module 300 uses three gradient boosting decision tree models to construct three regression models respectively. The three regression models are trained on the same base model training set to generate base model prediction results.

[0081] The fusion model training module 400 pre-builds a prediction fusion model based on a regression model, trains and optimizes the prediction fusion model, performs hyperparameter tuning on the prediction fusion model, loads the prediction results of the base model, uses the prediction results of the base model as the new feature input of the prediction fusion model, executes the trained and optimized prediction fusion model, and outputs the final calibration temperature.

[0082] The temperature field construction module 500 constructs a three-dimensional temperature field based on the calibrated real temperature and three-dimensional perspective information, and outputs the constructed three-dimensional temperature field.

[0083] In this embodiment, the preprocessing module 200 includes:

[0084] The data cleaning unit 210 is used to load input variables, clean the input variables, and remove missing and outlier values ​​from the input variables.

[0085] Type recognition unit 220 is used to acquire input variables after removing missing and outlier values ​​from the input variables, identify the type of the input variables, and construct and process the features of the input variables based on the type of the input variables.

[0086] The feature processing unit 230 identifies the type of the input variable and determines the type of the input variable. If the type of the input variable is a categorical feature, the input variable is encoded. If the type of the input variable is a continuous feature, the input variable is normalized, the original value is retained, or a derived feature is introduced.

[0087] In summary, this invention provides a three-dimensional temperature field construction system and method based on infrared sensing temperature data calibration. The embodiments of this invention train three regression models using three gradient boosting decision tree models: XGBoost, LightGBM, and CatBoost. These models then predict temperature for the same set of input features, allowing the base model predictions to complement the biases of individual models. This effectively reduces the risk of overfitting in individual models and improves overall generalization ability. Simultaneously, the prediction fusion model, by weighting or combining the predictions of multiple base models, smooths out the prediction fluctuations of individual models, providing more stable prediction results. This further reduces the potential biases and errors of individual models, effectively improving the robustness and prediction accuracy of the overall model.

[0088] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0089] It should be understood that the disclosed apparatus can be implemented in other ways, given the several embodiments provided in this application. For example, the apparatus embodiments described above are merely illustrative; the division of units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or communication connections shown or discussed may be through some interfaces; the indirect coupling or communication connections between devices or units may be telecommunications or other forms.

[0090] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. A method for constructing a three-dimensional temperature field based on infrared sensing thermometry data calibration, characterized in that, The method includes: The system uses infrared sensors to collect raw measured temperature and ambient temperature in real time, and uses sensor arrays to collect distance parameters, surface material, three-dimensional view, and wind speed. The raw measured temperature, ambient temperature, distance parameters, surface material, three-dimensional view, and wind speed are integrated as input variables. Load the collected input variables, preprocess the input variables, and output the preprocessed set; Three regression models were constructed using three gradient boosting decision tree models. The three regression models were trained on the same base model training set to generate the base model prediction results. A pre-built prediction fusion model based on a regression model is trained and optimized, and hyperparameters of the prediction fusion model are tuned. The prediction results of the base model are loaded and used as new feature inputs to the prediction fusion model. The trained and optimized prediction fusion model is then executed, and the final calibrated temperature is output. A three-dimensional temperature field is constructed based on the calibrated real temperature and three-dimensional perspective information, and the constructed three-dimensional temperature field is output.

2. The method for constructing a three-dimensional temperature field based on infrared sensing thermometry data calibration as described in claim 1, characterized in that: Methods for preprocessing input variables include: Load the input variables, clean the input variables, and remove missing and outlier values ​​from the input variables; Obtain the input variables after removing missing and outlier values, identify the type of the input variables, and construct and process the features of the input variables based on the input variable type.

3. The method for constructing a three-dimensional temperature field based on infrared sensing thermometry data calibration as described in claim 2, characterized in that: The method for constructing and processing input variable features based on input variable type includes: Identify and determine the type of input variables; If the type of the input variable is a categorical feature, the input variable is encoded. If the type of input variable is continuous feature, normalize the input variable, retain the original value, or introduce derived features.

4. The method for constructing a three-dimensional temperature field based on infrared sensing thermometry data calibration as described in claim 1, characterized in that: When constructing three regression models using three gradient boosting decision tree models, the three gradient boosting decision tree models are XGBoost regression model, LightGBM regression model, and CatBoost regression model, respectively.

5. The method for constructing a three-dimensional temperature field based on infrared sensing thermometry data calibration as described in claim 4, characterized in that: The method of training three regression models on the same base model training set includes: Extract a modeling sample set from historical input variables and divide the modeling sample set into a base model training set and a base model validation set; Pre-built XGBoost regression models, LightGBM regression models, and CatBoost regression models are trained using the base model training set, and hyperparameters are tuned for the XGBoost regression models, LightGBM regression models, and CatBoost regression models to output converged XGBoost regression models, LightGBM regression models, and CatBoost regression models. Load the base model validation set, and use the base model validation set to evaluate the prediction accuracy of the XGBoost regression model, LightGBM regression model and CatBoost regression model respectively, and output the prediction values ​​of the XGBoost regression model, LightGBM regression model and CatBoost regression model respectively. Obtain the predicted values ​​of the XGBoost regression model, LightGBM regression model, and CatBoost regression model, and concatenate the predicted values ​​of the XGBoost regression model, LightGBM regression model, and CatBoost regression model to form the prediction result of the base model.

6. The method for constructing a three-dimensional temperature field based on infrared sensing thermometry data calibration as described in claim 5, characterized in that: When pre-constructing a prediction fusion model based on a regression model, the prediction fusion model can be a linear regression model, a small multilayer perceptron (MLP) model, or a ridge regression model.

7. The method for constructing a three-dimensional temperature field based on infrared sensing thermometry data calibration as described in claim 5, characterized in that: When performing hyperparameter tuning for XGBoost regression models, LightGBM regression models, and CatBoost regression models, and when performing hyperparameter tuning for prediction fusion models, the hyperparameters to be adjusted include max_depth, learning_rate, subsampling ratio, or regularization term coefficient. Grid search, Bayesian optimization, or automated hyperparameter tuning tools are used to find the optimal parameter combination during hyperparameter tuning.

8. The method for constructing a three-dimensional temperature field based on infrared sensing thermometry data calibration as described in claim 7, characterized in that: The method for constructing a three-dimensional temperature field based on calibrated real temperature and three-dimensional perspective information includes: The three-dimensional space is divided into a voxel grid, and the temperature value of each voxel is the average value of all calibrated temperature points within the voxel grid. The calibrated real temperature is associated with the corresponding three-dimensional coordinate points to generate point cloud data with temperature attributes; Discrete temperature points are fitted to a continuous surface using surface interpolation, and the surface temperature distribution is output.

9. A three-dimensional temperature field construction system based on infrared sensing temperature data calibration, used to implement the three-dimensional temperature field construction method based on infrared sensing temperature data calibration as described in any one of claims 1-8, characterized in that: The system includes: The data acquisition module collects raw measured temperature and ambient temperature in real time based on infrared sensors, and collects distance parameters, surface material, three-dimensional view, and wind speed based on the sensor group. It integrates raw measured temperature, ambient temperature, distance parameters, surface material, three-dimensional view, and wind speed as input variables. The preprocessing module is used to load the collected input variables, preprocess the input variables, and output the preprocessed set. The base model training module uses three gradient boosting decision tree models to construct three regression models respectively. The three regression models are trained on the same base model training set to generate base model prediction results. The fusion model training module pre-builds a prediction fusion model based on a regression model, trains and optimizes the prediction fusion model, performs hyperparameter tuning on the prediction fusion model, loads the prediction results of the base model, uses the prediction results of the base model as new feature inputs for the prediction fusion model, executes the trained and optimized prediction fusion model, and outputs the final calibrated temperature. The temperature field construction module constructs a three-dimensional temperature field based on the calibrated real temperature and three-dimensional perspective information, and outputs the constructed three-dimensional temperature field.

10. The three-dimensional temperature field construction system based on infrared sensing thermometry data calibration as described in claim 9, characterized in that: The preprocessing module includes: The data cleaning unit is used to load input variables, clean the input variables, and remove missing and outlier values ​​from the input variables. The type recognition unit is used to acquire input variables after removing missing and outlier values ​​from the input variables, identify the type of the input variables, and construct and process the features of the input variables based on the type of the input variables. The feature processing unit identifies the type of the input variable and determines its type. If the type of the input variable is a categorical feature, it encodes the input variable. If the type of the input variable is a continuous feature, it normalizes the input variable, preserves the original value, or introduces derived features.

Images