A method for controlling hydration heat of ultra-high performance concrete

Abstract

The application discloses a kind of super high performance concrete hydration heat control methods, comprising: collection historical several super high performance concrete structure in the temperature data and size data of water heat process, material proportioning data, hydration heat environment data, constructs training data set;Random forest network model is constructed, and training data set is used to train prediction model, and the convergent random forest network model of output;The material proportioning data of super high performance concrete structure to be cast soon, design size data and environmental data are input into the convergent random forest network model, predict the temperature data of each time in water heat process, and construct cooling water flow rate control equation, based on the cooling water flow rate of predicted temperature data control hydration heat temperature control process.It is used to the temperature regulation of super high performance concrete structure water heat process, and the temperature regulation of super high performance concrete structure water heat process of complex, changeable material proportioning, environment and size can be effectively coped with.

Description

Technical Field

[0001] This invention relates to the field of concrete hydration heat control, and more specifically to a method for controlling the hydration heat of ultra-high performance concrete. Background Technology

[0002] The heat of hydration in concrete refers to the heat released during the reaction of cement and water in the concrete during the setting and hardening process. This heat of hydration causes the internal temperature of the concrete to rise, potentially reaching 50-60°C. If this heat is not dissipated in time, it can lead to excessive temperature differences between the inside and outside of the concrete, causing temperature stress cracks and affecting the stability of the concrete structure. Ultra-high performance concrete (UHPC), on the other hand, is a material made by mixing cement, mineral admixtures, additives, fine aggregates, and fibers using a specific process. It features high compressive strength and excellent durability. With the widespread application of UHPC, more and more buildings are using UHPC-cast structures. However, due to the special properties of UHPC materials, the temperature control of the hydration process is crucial.

[0003] Existing methods for controlling the heat of hydration temperature all employ real-time temperature feedback control, which has the disadvantages of untimely temperature feedback and low accuracy. After the concrete is poured, the layout of the concrete circulation cooling system cannot be changed, resulting in certain limitations in temperature control and making it unable to adapt to varying material ratios and ambient temperatures. Summary of the Invention

[0004] To address the aforementioned shortcomings of existing technologies, this invention provides a method for controlling the heat of hydration in ultra-high performance concrete.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for controlling the heat of hydration of ultra-high performance concrete is provided, which includes the following steps: S1: Collect temperature data of several historical ultra-high performance concrete structures during the hydrothermal process to obtain the output dataset; collect dimensional data, material ratio data and hydration thermal environment data of several historical ultra-high performance concrete structures, and perform normalization processing to construct the input dataset to obtain the training dataset. S2: Construct a random forest network model. In the decision tree of the random forest network model, construct the prediction model of the decision tree and the prediction model of the leaf node in the leaf node of the decision tree. Train the prediction model using the training dataset. The prediction model of the leaf node outputs the predicted temperature data. The prediction model of the decision tree uses the predicted temperature data output by the prediction model of the leaf node to predict the temperature data of the hydrothermal process. Output a converged random forest network model. S3: The material mix proportions, design dimensions, and environmental data of the ultra-high performance concrete structure to be poured are normalized and then input into a convergent random forest network model to predict the temperature data at each moment during the hydration process. A cooling water flow rate regulation equation is constructed, and the cooling water flow rate is regulated based on the predicted temperature data to control the hydration heat temperature control process.

[0006] Further, step S1 includes: S11: Collect temperature data from several historical ultra-high performance concrete structures during the hydrothermal process. The temperature data includes the core temperature, surface temperature, and intermediate layer temperature of the ultra-high performance concrete structures, resulting in a dataset of core temperature, surface temperature, and intermediate layer temperature data. , , ;in, N The amount of temperature data collected during the hydrothermal process. The time when temperature data was collected. The output dataset is obtained by taking the center temperature data, the intermediate layer temperature data, and the surface temperature data, respectively. S12: Obtain dimensional data of several historical ultra-high performance concrete structures, and normalize the dimensional data to obtain normalized values ​​for the section length and width. l , d ; Material mix proportion data of several ultra-high performance concrete structures from the pouring history were obtained, and the material mix proportion data were normalized to obtain normalized values ​​of the material mix proportion data. , M For the type of material, For the first M Normalized values ​​of the proportions of the materials; Historical hydration thermal environment data for several ultra-high performance concrete structures were acquired, including real-time temperature and humidity data. The temperature and humidity data were then normalized to obtain normalized values. , ; S13: Construct an input dataset using normalized values ​​of dimensional data, material ratio data, and environmental data. Combine the output dataset and input dataset to form a training set. Divide the data in the training set into a training set and a test set in an 8:2 ratio.

[0007] Further, step S2 includes: S21: Construct a random forest network model, which includes three decision trees. The three decision trees are used to predict the core temperature data, intermediate layer temperature data, and surface temperature data of the ultra-high performance concrete structure, respectively. Each decision tree includes three leaf nodes, which use size data, material ratio data, and environmental data to predict the temperature data of the hydration heat process, respectively. S22: Initialize the weight coefficients in the decision tree and leaf nodes, and randomly select them with replacement from the training set. M The input and output data corresponding to each collection time are input into three leaf nodes respectively, and the prediction model in the leaf nodes is trained. S23: Set the accuracy threshold for the prediction model corresponding to the three leaf nodes trained. And based on the weighting coefficients of the prediction accuracy of the three leaf nodes. The prediction accuracy of the three leaf node prediction models trained is evaluated to determine whether it meets the requirements. like ,satisfy If the prediction model of the three leaf nodes has converged, then the converged leaf node prediction model is output. Otherwise, if the prediction model for the three leaf nodes is determined to have not converged, return to step S22 and randomly select nodes with replacement from the training set again. M The input and output data corresponding to each collection time are used to continue training the prediction model for the leaf nodes until a converged prediction model for the leaf nodes is output. S24: Randomly select with replacement from the training dataset V The input and output data corresponding to each acquisition time are used to input the input data into the converged leaf node prediction model, and the output data is used to generate the output data. V One predicted temperature data; S25: Will V Predicted temperature data The actual temperature data corresponding to the output data respectively The difference is calculated to filter out abnormal predicted temperature data; like Then determine the predicted temperature data. If the data is unqualified, delete the unqualified predicted temperature data. Then determine the predicted temperature data. The qualified predicted temperature data is retained and used as the training output data for the decision tree prediction model. S26: Number of training output data collected v Calculate the prediction pass rate of the converged leaf node prediction model. And set a pass rate threshold. ,like If the prediction accuracy of the converged leaf node prediction model is low, proceed to step S27; otherwise, return to step S22 and randomly select new leaf nodes with replacement from the training set. M The input and output data corresponding to each collection time are used to retrain the prediction model in the leaf nodes; S27: Combine the training output data and the corresponding time of the collected temperature data to form the training dataset of the decision tree prediction model, and input it into the decision tree prediction model to train the decision tree prediction model and optimize the weight coefficients in the decision tree prediction model. S28: Set the accuracy threshold for the prediction model in the trained decision tree. To evaluate whether the prediction accuracy of the trained decision tree prediction model meets the requirements; like If the prediction model in the decision tree has converged, then the converged prediction model in the decision tree is output. like If the prediction model in the decision tree has not converged, return to step S24 and continue to randomly select with replacement from the training dataset. V The input and output data corresponding to each collection time are used to continue training the prediction model in the decision tree until the output of the converged prediction model in the decision tree is obtained. S29: Output a convergent random forest network model based on the prediction model in the convergent decision tree and the prediction model in the convergent leaf node.

[0008] Furthermore, the prediction models for the three decision trees to predict the center temperature data, intermediate layer temperature data, and surface temperature data of ultra-high performance concrete structures are as follows: ; The prediction model for the center temperature data, intermediate layer temperature data, and surface temperature data output by the decision tree is as follows: ; in, u Location labels for temperature data. These are the weight coefficients for the three leaf nodes. These are the time-series weighting coefficients of the leaf nodes with respect to the collected temperature data. The input function for the decision tree, For the input function of the leaf nodes, The time linking operation is performed for the leaf nodes. For time linking operations in decision trees, These are the weighting coefficients for length, width, and acquisition time in the first leaf node, respectively. These are the weighting coefficients for the data acquisition time and material ratio in the second leaf node, respectively. These are the weighting coefficients for temperature and humidity data in the third leaf node, respectively. The temperature data output from the three leaf nodes are respectively. This includes center temperature data, intermediate layer temperature data, or surface temperature data.

[0009] Furthermore, the objective loss function for the training process of the prediction model in the leaf node is: ; in, x The leaf node number is used. This represents the actual output data of the leaf nodes. The predicted output data for the leaf nodes. This represents the mean squared error of leaf node prediction during training. During training, the weight coefficients of the leaf nodes are updated through backpropagation; ; in, For the first x The weight coefficients corresponding to the leaf node prediction model The number of iterations for the weight coefficients. The leaf node prediction model Weight coefficients for the next iteration The weight coefficients for updating the leaf node prediction model. The learning rate is used to update the weight coefficients of the leaf node prediction model during iteration. j The number of the weight coefficients corresponding to the leaf node prediction model.

[0010] Furthermore, the objective loss function for the training process of the prediction model in the decision tree is: ; in, For the actual output data of the decision tree, The prediction output data for the decision tree, The mean squared error of the decision tree predictions during training; During training, the weight coefficients of the decision tree are updated through backpropagation; ; in, These are the weight coefficients corresponding to the decision tree prediction model. During the training of the decision tree prediction model, the first Weight coefficients for the next iteration These are the weight coefficients updated during the training of the decision tree prediction model. The learning rate is used for updating the weight coefficients during the training process of the decision tree prediction model. iNumber the weight coefficients for the decision tree prediction model.

[0011] Further, step S3 includes: S31: The material mix design data, design dimensions, and environmental data of the ultra-high performance concrete structure to be poured are normalized and then input into a convergent random forest network model to predict the center temperature data of the ultra-high performance concrete structure at various moments during the hydrothermal process. Intermediate layer temperature data and surface temperature data ; S32: Obtain continuous core temperature data during the hydrothermal process. Continuous intermediate layer temperature data and continuous surface temperature data ; S33: Establish the cooling water flow rate regulation equation for the hydrothermal cooling system embedded in ultra-high performance concrete structures. ; in, The initial cooling water flow rate, For maximum cooling water flow rate, For adjustment coefficients, Maximum permissible temperature Minimum permissible temperature; S34: Input continuous center temperature data, continuous intermediate layer temperature data, and continuous surface temperature data into the cooling water flow rate control equation, and output the cooling water flow rate that continuously controls the center temperature, intermediate layer temperature, and surface temperature during the hydrothermal process of the cast ultra-high performance concrete structure, so as to achieve control of the cooling water flow rate at different times.

[0012] The beneficial effects of this invention are as follows: This invention is used for temperature control of the hydrothermal process of ultra-high performance concrete structures, and can effectively cope with the temperature control of the hydrothermal process of ultra-high performance concrete structures with complex and variable material ratios, environments, and dimensions. Furthermore, by predicting the temperature of the hydrothermal process in the center, intermediate layer, and surface layer of the structure, continuous dynamic temperature data can be obtained, providing data support for the development of targeted concrete cooling systems. It can also be used to determine the cooling water flow rate at different locations within the structure, dynamically adjust the cooling water flow rate, and achieve temperature control of the hydrothermal process, making it suitable for various ultra-high performance concrete projects. Attached Figure Description

[0013] Figure 1 This is a flowchart of a method for controlling the heat of hydration in ultra-high performance concrete. Detailed Implementation

[0014] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0015] like Figure 1 As shown, a method for controlling the heat of hydration of ultra-high performance concrete includes the following steps: S1: Collect temperature data of several historical ultra-high performance concrete structures during the hydrothermal process to obtain the output dataset; collect dimensional data, material proportion data and hydration thermal environment data of several historical ultra-high performance concrete structures, and perform normalization processing to construct the input dataset to obtain the training dataset.

[0016] Step S1 specifically includes the following steps: S11: Collect temperature data from several historical ultra-high performance concrete structures during the hydrothermal process. The temperature data includes the core temperature, surface temperature, and intermediate layer temperature of the ultra-high performance concrete structures, resulting in a dataset of core temperature, surface temperature, and intermediate layer temperature data. , , ;in, N The amount of temperature data collected during the hydrothermal process. The time when temperature data was collected. The output dataset is obtained by taking the center temperature data, the intermediate layer temperature data, and the surface temperature data, respectively. S12: Obtain dimensional data of several historical ultra-high performance concrete structures, and normalize the dimensional data to obtain normalized values ​​for the section length and width. l , d ; Material mix proportion data of several ultra-high performance concrete structures from the pouring history were obtained, and the material mix proportion data were normalized to obtain normalized values ​​of the material mix proportion data. , M For the type of material, For the first M Normalized values ​​of the proportions of the materials; Historical hydration thermal environment data for several ultra-high performance concrete structures were acquired, including real-time temperature and humidity data. The temperature and humidity data were then normalized to obtain normalized values. , ; S13: Construct an input dataset using normalized values ​​of dimensional data, material ratio data, and environmental data. Combine the output dataset and input dataset to form a training set. Divide the data in the training set into a training set and a test set in an 8:2 ratio.

[0017] S2: Construct a random forest network model. In the decision tree of the random forest network model, construct the prediction model of the decision tree and the prediction model of the leaf node in the leaf node of the decision tree. Train the prediction model using the training dataset. The prediction model of the leaf node outputs the predicted temperature data. The prediction model of the decision tree uses the predicted temperature data output by the prediction model of the leaf node to predict the temperature data of the hydrothermal process.

[0018] Step S2 specifically includes the following steps: S21: Construct a random forest network model, which includes three decision trees. The three decision trees are used to predict the core temperature data, intermediate layer temperature data, and surface temperature data of the ultra-high performance concrete structure, respectively. Each decision tree includes three leaf nodes, which use size data, material ratio data, and environmental data to predict the temperature data of the hydration heat process, respectively. The prediction model for the three leaf nodes is as follows: ; The decision tree uses the temperature data predicted by the three leaf nodes to further output the predicted center temperature data, intermediate layer temperature data, and surface temperature data; the prediction model of the decision tree is as follows: ; in, u Location labels for temperature data. These are the weight coefficients for the three leaf nodes. These are the time-series weighting coefficients of the leaf nodes with respect to the collected temperature data. The input function for the decision tree, For the input function of the leaf nodes, The time linking operation is performed for the leaf nodes. For time linking operations in decision trees, These are the weighting coefficients for length, width, and acquisition time in the first leaf node, respectively. These are the weighting coefficients for the data acquisition time and material ratio in the second leaf node, respectively. These are the weighting coefficients for temperature and humidity data in the third leaf node, respectively. The temperature data output from the three leaf nodes are respectively. This includes center temperature data, intermediate layer temperature data, or surface temperature data. S22: Initialize the weight coefficients in the decision tree and leaf nodes, and randomly select them with replacement from the training set. M The input and output data corresponding to each collection time are input into three leaf nodes respectively, and the prediction model in the leaf nodes is trained. The objective loss function for training the prediction model in the leaf nodes is: ; in, x The leaf node number is used. This represents the actual output data of the leaf nodes. The predicted output data for the leaf nodes. This represents the mean squared error of leaf node prediction during training. During training, the weight coefficients of the leaf nodes are updated through backpropagation; ; in, For the first x The weight coefficients corresponding to the leaf node prediction model The number of iterations for the weight coefficients. The leaf node prediction model Weight coefficients for the next iteration The weight coefficients for updating the leaf node prediction model. The learning rate is used to update the weight coefficients of the leaf node prediction model during iteration. j The weight coefficient number corresponding to the leaf node prediction model; The weighting coefficients corresponding to the prediction model with three leaf nodes are used as weighting coefficients. Unified representation, weighting coefficient Including weighting coefficients Weighting coefficients and weighting coefficients .

[0019] S23: Set the accuracy threshold for the prediction model corresponding to the three leaf nodes trained. And based on the weighting coefficients of the prediction accuracy of the three leaf nodes. The prediction accuracy of the three leaf node prediction models trained is evaluated to determine whether it meets the requirements. like ,satisfy If the prediction model of the three leaf nodes has converged, then the converged leaf node prediction model is output. Otherwise, if the prediction model for the three leaf nodes is determined to have not converged, return to step S22 and randomly select nodes with replacement from the training set again. MThe input and output data corresponding to each collection time are used to continue training the prediction model for the leaf nodes until a converged prediction model for the leaf nodes is output. S24: Randomly select with replacement from the training dataset V The input and output data corresponding to each acquisition time are used to input the input data into the converged leaf node prediction model, and the output data is used to generate the output data. V One predicted temperature data; S25: Will V Predicted temperature data The actual temperature data corresponding to the output data respectively The difference is calculated to filter out abnormal predicted temperature data; like Then determine the predicted temperature data. If the data is unqualified, delete the unqualified predicted temperature data. Then determine the predicted temperature data. The qualified predicted temperature data is retained and used as the training output data for the decision tree prediction model. S26: Number of training output data collected v Calculate the prediction pass rate of the converged leaf node prediction model. And set a pass rate threshold. ,like If the prediction accuracy of the converged leaf node prediction model is low, proceed to step S27; otherwise, return to step S22 and randomly select new leaf nodes with replacement from the training set. M The input and output data corresponding to each collection time are used to retrain the prediction model in the leaf nodes; S27: Combine the training output data and the corresponding time of the collected temperature data to form the training dataset of the decision tree prediction model, and input it into the decision tree prediction model to train the decision tree prediction model and optimize the weight coefficients in the decision tree prediction model. The objective loss function for training the prediction model in a decision tree is: ; in, For the actual output data of the decision tree, The prediction output data for the decision tree, The mean squared error of the decision tree predictions during training; During training, the weight coefficients of the decision tree are updated through backpropagation; ; in, These are the weight coefficients corresponding to the decision tree prediction model. During the training of the decision tree prediction model, the first Weight coefficients for the next iteration These are the weight coefficients updated during the training of the decision tree prediction model. The learning rate is used for updating the weight coefficients during the training process of the decision tree prediction model. i Number the weight coefficients for the decision tree prediction model; The weighting coefficients corresponding to the prediction model in the decision tree are used as weighting coefficients. Unified representation, weighting coefficient Including the weight coefficients of the three leaf nodes The time series weighting coefficients of leaf nodes with respect to the collected temperature data .

[0020] S28: Set the accuracy threshold for the prediction model in the trained decision tree. To evaluate whether the prediction accuracy of the trained decision tree prediction model meets the requirements; like If the prediction model in the decision tree has converged, then the converged prediction model in the decision tree is output. like If the prediction model in the decision tree has not converged, return to step S24 and continue to randomly select with replacement from the training dataset. V The input and output data corresponding to each collection time are used to continue training the prediction model in the decision tree until the output of the converged prediction model in the decision tree is obtained. S29: Output a convergent random forest network model based on the prediction model in the convergent decision tree and the prediction model in the convergent leaf node.

[0021] The random forest network model constructed in this invention uses leaf nodes and decision trees to achieve multi-level comprehensive prediction. Leaf nodes predict continuous dynamic hydration temperature from different hydration heat influencing factors, and decision trees integrate the dynamic hydration temperature predicted by different leaf nodes to achieve high-precision prediction of dynamic hydration temperature, reduce the mutual influence of different hydration heat influencing factors, and improve the model's data output capability.

[0022] S3: The material mix proportions, design dimensions, and environmental data of the ultra-high performance concrete structure to be poured are normalized and then input into a convergent random forest network model to predict the temperature data at each moment during the hydration process. A cooling water flow rate regulation equation is constructed, and the cooling water flow rate is regulated based on the predicted temperature data to control the hydration heat temperature control process.

[0023] Step S3 specifically includes the following steps: S31: The material mix design data, design dimensions, and environmental data of the ultra-high performance concrete structure to be poured are normalized and then input into a convergent random forest network model to predict the center temperature data of the ultra-high performance concrete structure at various moments during the hydrothermal process. Intermediate layer temperature data and surface temperature data ; S32: Obtain continuous core temperature data during the hydrothermal process. Continuous intermediate layer temperature data and continuous surface temperature data ; S33: Establish the cooling water flow rate regulation equation for the hydrothermal cooling system embedded in ultra-high performance concrete structures. ; in, The initial cooling water flow rate, For maximum cooling water flow rate, For adjustment coefficients, Maximum permissible temperature Minimum permissible temperature; S34: Input continuous center temperature data, continuous intermediate layer temperature data, and continuous surface temperature data into the cooling water flow rate control equation, and output the cooling water flow rate that continuously controls the center temperature, intermediate layer temperature, and surface temperature during the hydrothermal process of the cast ultra-high performance concrete structure, so as to achieve control of the cooling water flow rate at different times.

[0024] This invention relates to temperature control of the hydrothermal process in ultra-high performance concrete structures. It can effectively address the complex and variable material ratios, environmental conditions, and dimensions of ultra-high performance concrete structures during hydrothermal processing. Furthermore, by predicting the temperatures of the hydrothermal process in the center, intermediate layers, and surface of the structure, continuous dynamic temperature data can be obtained, providing data support for developing targeted concrete cooling systems. It can also be used to determine the cooling water flow rate at different locations within the structure, dynamically adjusting the flow rate to achieve temperature control of the hydrothermal process. This invention is applicable to various ultra-high performance concrete projects.

Claims

1. A method for controlling the heat of hydration in ultra-high performance concrete, characterized in that, Includes the following steps: S1: Collect temperature data of several historical ultra-high performance concrete structures during the hydrothermal process to obtain the output dataset; collect dimensional data, material ratio data and hydration thermal environment data of several historical ultra-high performance concrete structures, and perform normalization processing to construct the input dataset to obtain the training dataset. S2: Construct a random forest network model. In the decision tree of the random forest network model, construct the prediction model of the decision tree and the prediction model of the leaf node in the leaf node of the decision tree. Train the prediction model using the training dataset. The prediction model of the leaf node outputs the predicted temperature data. The prediction model of the decision tree uses the predicted temperature data output by the prediction model of the leaf node to predict the temperature data of the hydrothermal process. Output a converged random forest network model. S3: After normalizing the material mix data, design dimension data, and environmental data of the ultra-high performance concrete structure to be poured, input them into a convergent random forest network model to predict the temperature data at each moment during the hydration process, and construct a cooling water flow rate regulation equation. Based on the predicted temperature data, regulate the cooling water flow rate of the hydration heat temperature control process. Step S2 includes: S21: Construct a random forest network model, which includes three decision trees. The three decision trees are used to predict the core temperature data, intermediate layer temperature data, and surface temperature data of the ultra-high performance concrete structure, respectively. Each decision tree includes three leaf nodes, which use size data, material ratio data, and environmental data to predict the temperature data of the hydration heat process, respectively. S22: Initialize the weight coefficients in the decision tree and leaf nodes, and randomly select them with replacement from the training set. M The input and output data corresponding to each collection time are input into three leaf nodes respectively, and the prediction model in the leaf nodes is trained. S23: Set the accuracy threshold for the prediction model corresponding to the three leaf nodes trained. And based on the weighting coefficients of the prediction accuracy of the three leaf nodes. The prediction accuracy of the three leaf node prediction models trained is evaluated to determine whether it meets the requirements. like ,satisfy If the prediction model for the three leaf nodes has converged, then the converged leaf node prediction model is output. This represents the mean squared error of leaf node predictions during training. x The leaf node number; Otherwise, if the prediction model for the three leaf nodes is determined to have not converged, return to step S22 and randomly select nodes with replacement from the training set again. M The input and output data corresponding to each collection time are used to continue training the prediction model for the leaf nodes until a converged prediction model for the leaf nodes is output. S24: Randomly select with replacement from the training dataset V The input and output data corresponding to each acquisition time are used to input the input data into the converged leaf node prediction model, and the output data is used to generate the output data. V One predicted temperature data; S25: Will V Predicted temperature data The actual temperature data corresponding to the output data respectively The difference is calculated to filter out abnormal predicted temperature data; like Then determine the predicted temperature data. If the data is unqualified, delete the unqualified predicted temperature data. Then determine the predicted temperature data. The qualified predicted temperature data is retained and used as training output data for the decision tree prediction model. u Location labels for temperature data. The time when temperature data was collected; S26: Number of training output data collected v Calculate the prediction pass rate of the converged leaf node prediction model. And set a pass rate threshold. ,like If the prediction accuracy of the converged leaf node prediction model is low, proceed to step S27; otherwise, return to step S22 and randomly select new leaf nodes with replacement from the training set. M The input and output data corresponding to each collection time are used to retrain the prediction model in the leaf nodes; S27: Combine the training output data and the corresponding time of the collected temperature data to form the training dataset of the decision tree prediction model, and input it into the decision tree prediction model to train the decision tree prediction model and optimize the weight coefficients in the decision tree prediction model. S28: Set the accuracy threshold for the prediction model in the trained decision tree. To evaluate whether the prediction accuracy of the trained decision tree prediction model meets the requirements; like If the prediction model in the decision tree has converged, then the converged prediction model in the decision tree is output. The mean squared error of the decision tree predictions during training; like If the prediction model in the decision tree has not converged, return to step S24 and continue to randomly select with replacement from the training dataset. V The input and output data corresponding to each collection time are used to continue training the prediction model in the decision tree until the output of the converged prediction model in the decision tree is obtained. S29: Output a convergent random forest network model based on the prediction model in the convergent decision tree and the prediction model in the convergent leaf node.

2. The method for controlling the heat of hydration of ultra-high performance concrete according to claim 1, characterized in that, Step S1 includes: S11: Collect temperature data from several historical ultra-high performance concrete structures during the hydrothermal process. The temperature data includes the core temperature, surface temperature, and intermediate layer temperature of the ultra-high performance concrete structures, resulting in a dataset of core temperature, surface temperature, and intermediate layer temperature data. , , ;in, N The amount of temperature data collected during the hydrothermal process. The output dataset is obtained by taking the center temperature data, the intermediate layer temperature data, and the surface temperature data, respectively. S12: Obtain dimensional data of several historical ultra-high performance concrete structures, and normalize the dimensional data to obtain normalized values ​​for the section length and width. l , d ; Material mix proportion data of several ultra-high performance concrete structures from the pouring history were obtained, and the material mix proportion data were normalized to obtain normalized values ​​of the material mix proportion data. , For the type of material, For the first Normalized values ​​of the proportions of the materials; Historical hydration thermal environment data for several ultra-high performance concrete structures were acquired, including real-time temperature and humidity data. The temperature and humidity data were then normalized to obtain normalized values. , ; S13: Construct an input dataset using normalized values ​​of dimensional data, material ratio data, and environmental data. Combine the output dataset and input dataset to form a training set. Divide the data in the training set into a training set and a test set in an 8:2 ratio.

3. The method for controlling the heat of hydration of ultra-high performance concrete according to claim 2, characterized in that, The prediction models for the three decision trees used to predict the core temperature data, intermediate layer temperature data, and surface temperature data of ultra-high performance concrete structures are as follows: ； The prediction model for the center temperature data, intermediate layer temperature data, and surface temperature data output by the decision tree is as follows: ； in, These are the weight coefficients for the three leaf nodes. These are the time-series weighting coefficients of the leaf nodes with respect to the collected temperature data. The input function for the decision tree, For the input function of the leaf nodes, The time linking operation is performed for the leaf nodes. For time linking operations in decision trees, These are the weighting coefficients for length, width, and acquisition time in the first leaf node, respectively. These are the weighting coefficients for the data acquisition time and material ratio in the second leaf node, respectively. These are the weighting coefficients for temperature and humidity data in the third leaf node, respectively. The temperature data output from the three leaf nodes are respectively. This includes center temperature data, intermediate layer temperature data, or surface temperature data.

4. The method for controlling the heat of hydration of ultra-high performance concrete according to claim 3, characterized in that, The objective loss function for the training process of the prediction model in the leaf node is: ； in, This represents the actual output data of the leaf nodes. The predicted output data for the leaf nodes; During training, the weight coefficients of the leaf nodes are updated through backpropagation; ； in, For the first x The weight coefficients corresponding to the leaf node prediction model The number of iterations for the weight coefficients. The leaf node prediction model Weight coefficients for the next iteration The weight coefficients for updating the leaf node prediction model. The learning rate is used to update the weight coefficients of the leaf node prediction model during iteration. j The number of the weight coefficients corresponding to the leaf node prediction model.

5. The method for controlling the heat of hydration of ultra-high performance concrete according to claim 3, characterized in that, The objective loss function for the training process of the prediction model in the decision tree is: ； in, For the actual output data of the decision tree, The output data for the prediction of the decision tree; During training, the weight coefficients of the decision tree are updated through backpropagation; ； in, These are the weight coefficients corresponding to the decision tree prediction model. During the training of the decision tree prediction model, the first Weight coefficients for the next iteration These are the weight coefficients updated during the training of the decision tree prediction model. The learning rate is used for updating the weight coefficients during the training process of the decision tree prediction model. i Number the weight coefficients for the decision tree prediction model.

6. The method for controlling the heat of hydration of ultra-high performance concrete according to claim 3, characterized in that, Step S3 includes: S31: Normalize the material mix design data, design dimensions, and environmental data of the ultra-high performance concrete structure to be poured, then input them into a convergent random forest network model to predict the center temperature data of the ultra-high performance concrete structure at various moments during the hydrothermal process. Intermediate layer temperature data and surface temperature data ; S32: Obtain continuous core temperature data during the hydrothermal process. Continuous intermediate layer temperature data and continuous surface temperature data ; S33: Establish the cooling water flow rate regulation equation for the hydrothermal cooling system embedded in ultra-high performance concrete structures. ； in, The initial cooling water flow rate, For maximum cooling water flow rate, For adjustment coefficients, Maximum permissible temperature Minimum permissible temperature; S34: Input continuous center temperature data, continuous intermediate layer temperature data, and continuous surface temperature data into the cooling water flow rate control equation, and output the cooling water flow rate that continuously controls the center temperature, intermediate layer temperature, and surface temperature during the hydrothermal process of the cast ultra-high performance concrete structure, so as to achieve control of the cooling water flow rate at different times.

Images