LSTM-based panel concrete temperature prediction method

Abstract

The panel concrete temperature prediction method based on the LSTM comprises the following steps: collecting historical monitoring data of panel concrete; constructing a panel concrete monitoring data acquisition table, and processing data in the data acquisition table; the LSTM neural network model is adopted to predict the panel concrete temperature, and a panel concrete temperature prediction value is obtained; and training the LSTM neural network model by using historical monitoring data samples and storing optimal model parameters. Compared with a traditional panel concrete temperature prediction model, the method has the advantages that the progressive relation of external environment influence factors along with time is considered, and meanwhile, the influence of different pouring time and pouring temperature difference on a panel concrete temperature duration curve is considered; and the temperature prediction result of the panel concrete is more accurate.

Description

Technical Field

[0001] The present invention belongs to the technical field of the construction of concrete face rockfill dams, and particularly relates to a method for predicting the temperature of concrete face based on LSTM. Background Technique

[0002] The concrete face rockfill dam mainly relies on the upstream concrete face to block water. Therefore, it is particularly important to ensure the construction quality of the concrete face. If reasonable temperature control and crack prevention measures are not taken during the construction period, temperature cracks may occur in the concrete face, which is not conducive to the anti-seepage of the dam body. The temperature process of the concrete face is closely related to temperature control indicators such as the temperature difference inside and outside the concrete and the temperature drop rate. Therefore, engineering units pay great attention to the temperature process of the concrete face. However, due to the complexity of the on-site conditions, it is quite common for the maximum temperature of the concrete face to exceed the design allowable maximum temperature. Real-time tracking and forecasting of the pouring process of the concrete face is an effective way to control the maximum temperature of the concrete face.

[0003] In order to predict the temperature of the concrete face, there are currently two approaches: one is to conduct a simulation analysis of the temperature field of the entire process of the concrete face, give the current temperature field of the concrete face, and forecast the future temperature field. If problems are found, countermeasures can be taken in a timely manner; the other is to establish a mathematical equation that quantitatively describes the change law of the temperature monitoring values of the concrete face through intelligent methods such as mathematical statistics analysis or neural networks, and predict future temperature information by mining historical temperature information. The temperature change process of the concrete face is affected by multiple factors jointly, and conventional prediction models are difficult to take into account both time series and various external factors at the same time. The LSTM model is suitable for processing sequence data, including time series data, natural language text, etc. This makes it perform excellently in various prediction tasks. Summary of the Invention

[0004] The present invention provides a method for predicting the temperature of concrete face based on LSTM, which collects temperature data and related pouring information, and predicts the temperature of the concrete face based on the LSTM model, which can reduce the influence of various factors on temperature prediction and improve the accuracy of temperature prediction. Furthermore, it provides an important reference basis for formulating scientific and reasonable dynamic temperature control measures.

[0005] The technical solution adopted by the present invention is as follows:

[0006] The method for predicting the temperature of concrete face based on LSTM includes the following steps:

[0007] Step 1: Collect historical monitoring data of the concrete face.

[0008] Step 2: Construct a data collection form for monitoring the concrete face and process the data in the data collection form.

[0009] Step 3: Use the LSTM neural network model to predict the temperature of the face slab concrete and obtain the predicted value of the face slab concrete temperature;

[0010] Step 4: Use the historical monitoring data samples to train the LSTM neural network model and save the optimal model parameters.

[0011] In the above Step 1, the historical monitoring data of the face slab concrete during the pouring period and the curing stage are obtained as follows:

[0012] The temperature of the face slab concrete is obtained by using the distributed optical fiber temperature monitoring method, and the temperature data T at different spatial positions on the face slab concrete can be obtained at the same temperature monitoring time t; t ;

[0013] The environmental monitoring data are monitored by a weather station. The environmental monitoring data include the environmental temperature T e_t , the environmental humidity RH t , the instantaneous wind speed WD t , the instantaneous solar radiation SR t ; At the same time, record the pouring temperature T b of each measuring point and the concrete pouring time t e .

[0014] In the above Step 2, taking the temperature monitoring time t of the face slab concrete as the reference, match the environmental monitoring data according to the concrete temperature time for all the collected historical monitoring data, construct a face slab concrete monitoring data collection table, and preprocess all the data in the data collection table. Specifically, it includes the following steps:

[0015] S2.1: When matching the temperature data information of the temperature measuring points with the environmental data information, use the temperature monitoring time t as the primary key, select the environmental data information measured at the time t' closest to the temperature monitoring time t, and match it with the temperature data information to form a distributed optical fiber temperature and environmental information table.

[0016] S2.2: Finally, match according to the measured pouring temperature of the temperature measuring points, with the measuring point number as the primary key, and construct a face slab concrete monitoring data collection table.

[0017] S2.3: Average and correct the obvious abnormal data in the face slab concrete monitoring data collection table according to the adjacent monitoring time, and uniformly supplement the missing data according to the adjacent time.

[0018] S2.4: Standardize the data in the face slab concrete monitoring data collection table:

[0019] The data standardization process uses Min - Max normalization to map the data between 0 and 1. Obtain the independent variable time series X t= [T t , T e_t , RH t , WD t , SR t , T b ;

[0020]

[0021] In formula (1), x' j is the result after standardizing a single-column attribute in the data collection table, and x j is the value of the single-column attribute in the data collection table; min(x j ) represents the minimum value in the single-column attribute; max(x j ) represents the maximum value in the single-column attribute.

[0022] In step 3, an LSTM neural network model is used to predict the temperature of the panel concrete. The temperature of the panel concrete within a certain period and the measured environmental data information within the corresponding period are used as the input data of the LSTM neural network model. The output panel concrete temperature of the LSTM neural network model is a standardized predicted value; then, the inverse operation is performed on the output panel concrete temperature data to obtain the predicted value of the panel concrete temperature. Specifically, it includes the following steps:

[0023] S3.1: Sort the data in the standardized data collection table according to the measuring point number and monitoring time sequence. Divide the sorted data into a training set and a test set according to a ratio of 3:1, and construct the time series Y t = [y' j , where y' j is the predicted data.

[0024] S3.2: Use the independent variable time series X t as the input of the LSTM neural network model, and use the output of the LSTM neural network model as the standardized predicted value. Construct the time series X' t from the standardized predicted value;

[0025] S3.3: Select the predicted panel concrete temperature Y t in the time series Y tT of the prediction. According to the inverse operation of standardization, restore the panel concrete temperature prediction vector Y' t of the output time series Y t = [T' t .

[0026] The inverse operation is specifically as follows:

[0027] T' t = Y tT (YtT-max -Y tT-min ) + Y tT-min (2);

[0028] In formula (2), Y tT represents the predicted sequence value of the face slab concrete temperature of the time series Y t ; Y tT-max represents the maximum value in the predicted sequence of the face slab concrete temperature of Y tT ; X' T-min represents the minimum value in the predicted sequence of the face slab concrete temperature of Y tT ; T' t represents the predicted face slab concrete temperature value obtained after the inverse operation.

[0029] In step 4, the root mean square error RMSE is used as the loss function to calculate the RMSE value of the predicted vector Y' of the face slab concrete temperature t and the measured face slab concrete temperature vector X t , specifically as follows:

[0030]

[0031] Taking the minimum loss function value as the optimization goal, the LSTM neural network model is trained using historical monitoring data samples, and the optimal model parameters are saved.

[0032] The technical effects of a face slab concrete temperature prediction method based on LSTM according to the present invention are as follows:

[0033] 1) According to the existing distributed optical fiber temperature measurement data, combining construction information and environmental data, while using the LSTM neural network model, the influence of the time process during pouring and the environmental temperature on the pouring temperature is considered. The distributed optical fiber temperature measurement data (spatial continuous temperature field), construction time series information (pouring time, pouring temperature) and environmental parameters (wind speed, solar radiation intensity, day-night temperature difference) are coupled in multiple physical fields, solving the limitation of traditional single-point temperature measurement for temperature prediction, and combining various environmental influencing factors of concrete to improve the accuracy of face slab concrete temperature prediction.

[0034] 2) Compared with the traditional face slab concrete temperature prediction model, the present invention takes into account the progressive relationship of external environmental influencing factors over time, and at the same time considers the influence of different pouring times and different pouring temperatures on the temperature history curve of the face slab concrete; making the temperature prediction result of the face slab concrete more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described below with reference to the drawings and examples;

[0036] Figure 1It is the flow chart of the panel concrete temperature prediction framework of the present invention.

[0037] Figure 2 It is the process diagram of the matching between the panel concrete temperature and environmental data of the present invention.

[0038] Figure 3 It is the effect diagram of the panel concrete temperature data processing of the present invention.

[0039] Figure 4 It is the comparison diagram between the predicted panel concrete temperature value and the environmental value of the example of the present invention. Detailed implementation mode

[0040] The panel concrete temperature prediction method based on LSTM includes the following steps:

[0041] Step 1: Collect the historical monitoring data of the panel concrete;

[0042] Step 2: Construct a collection table of the panel concrete monitoring data and process the data in the collection table;

[0043] Step 3: Use the LSTM neural network model to predict the panel concrete temperature and obtain the predicted panel concrete temperature value;

[0044] Step 4: Use the historical monitoring data samples to train the LSTM neural network model and save the optimal model parameters.

[0045] In the said Step 1, the historical monitoring data of the panel concrete during the pouring period and the curing stage are obtained as follows:

[0046] The panel concrete is obtained by using the distributed optical fiber temperature monitoring method, and the temperature data T at different spatial positions on the panel concrete can be obtained at the same temperature monitoring time t t ;

[0047] The environmental monitoring data during this period are monitored by a weather station, and the environmental monitoring data include the environmental temperature T e_t , the environmental humidity RH t , the instantaneous wind speed WD t , the instantaneous solar radiation SR t ; At the same time, record the pouring temperature T of each measuring point b , the concrete pouring time t e .

[0048] In the said Step 2, all the collected historical monitoring data are used as the benchmark with the panel concrete temperature monitoring time t, and the environmental monitoring data are matched according to the concrete temperature time to construct a collection table of the panel concrete monitoring data, and all the data in the collection table are preprocessed.

[0049] Specifically, it includes the following steps:

[0050] S2.1: The monitoring time of the temperature measurement points of the concrete of each panel and the monitoring time of the environmental monitoring data are not exactly the same. Therefore, when matching the temperature data information of the temperature measurement points with the environmental data information, taking the temperature monitoring time t as the primary key, the environmental data information measured at the time t' closest to the temperature monitoring time t is selected and matched with the temperature data information to form a distributed optical fiber temperature and environmental information table. The matching of temperature data and environmental data is as Figure 2 shown.

[0051] S2.2: Finally, according to the actually measured pouring temperature of the temperature measurement points, matching is carried out with the measurement point numbers as the primary key to construct a monitoring data acquisition table for the panel concrete. The monitoring data acquisition table for the panel concrete is shown in Table 1:

[0052] Table 1 Monitoring Data Acquisition Table for Panel Concrete

[0053]

[0054] S2.3: The obvious abnormal data in the monitoring data acquisition table of the panel concrete is averaged and corrected according to the adjacent monitoring time, and the missing data is evenly filled according to the adjacent time. Specifically as follows:

[0055] Add Figure 3 As the average correction comparison chart of the panel concrete temperature, the original data of the 30#-02 panel is missing from the age of 6 days to 14 days, and the missing data cannot provide data support for subsequent temperature prediction. Therefore, the data of the adjacent 20#-02 panel from 6 days to 14 days is selected for supplementation.

[0056] S2.4: Standardize the data in the monitoring data acquisition table of the panel concrete:

[0057] The data standardization process uses Min-Max normalization. In order to accelerate the training speed of the neural network, improve the convergence speed and prediction accuracy, all data are processed using this method, and the data is mapped between 0 and 1. The independent variable time series X t =[T t , T e_t , RH t , WD t , SR t , T b ;

[0058]

[0059] In formula (1), x' j is the result after standardization of a single-column attribute in the table data acquisition table, and x jFor the single-column attribute value in the data collection form; min(x j ) represents the minimum value in the single-column attribute; max(x j ) represents the maximum value in the single-column attribute.

[0060] In step 3, an LSTM neural network model is used to predict the temperature of the panel concrete. The temperature of the panel concrete within a certain period and the measured environmental data information within the corresponding period are used as the input data of the LSTM neural network model. The output panel concrete temperature of the LSTM neural network model is a standardized predicted value; then, the inverse operation is performed on the output panel concrete temperature data to obtain the predicted value of the panel concrete temperature. Specifically, it includes the following steps:

[0061] S3.1: Sort the data in the standardized data collection form according to the measuring point number and monitoring time sequence, and divide the sorted data into a training set and a test set according to a ratio of 3:1 to construct the time series Y t =[y′ j , where y′ j is the predicted data.

[0062] S3.2: Use the independent variable time series X t as the input of the LSTM neural network model, and use the output of the LSTM neural network model as the standardized predicted value to construct the time series X′ t ;

[0063] S3.3: Select the predicted panel concrete temperature Y t in the predicted time series Y tT , and restore it according to the inverse operation of the standardization to output the panel concrete temperature prediction vector Y′ t =[T′ t of the time series Y t .

[0064] The inverse operation is specifically as follows:

[0065] T′ t =Y tT (Y tT-max -Y tT-min )+Y tT-min (2);

[0066] In formula (2), Y tT represents the predicted sequence value of the panel concrete temperature of the time series Y t ; Y tT-max represents the maximum value in the predicted sequence of the panel concrete temperature of Y tT ; X′ T-min represents Y tTThe minimum value in the predicted panel concrete temperature sequence; T′ t Indicates the predicted panel concrete temperature value obtained after the inverse operation.

[0067] In step 4, the root mean square error RMSE is used as the loss function to calculate the predicted panel concrete temperature vector Y′ t And the measured panel concrete temperature vector X t The RMSE value is as follows:

[0068]

[0069] Taking the minimum loss function value as the optimization goal, the LSTM neural network model is trained using historical monitoring data samples, and the optimal model parameters are saved. Specifically as follows:

[0070] By adjusting parameters such as epoch and batch_size in the model, gradually adjust and perform operations, compare the RMSE values under different parameters, and select the parameters with the smallest RMSE value as the optimal model parameters.

[0071] Table 2 Statistical table of RMSE values for parameter adjustment of LSTM panel concrete model

[0072] epoch batch_size RMSE 50 72 0.779 100 72 0.504 100 128 0.494 150 128 0.441

[0073] Figure 4 This is a comparison chart of the predicted panel concrete temperature value and the environmental value in the example of the present invention. Figure 4 The curves of the true value and the predicted value in it coincide highly, indicating that in this model, there is a high prediction accuracy in both the temperature rise stage and the temperature drop stage of the panel concrete.

Claims

1. A method for predicting the temperature of panel concrete based on LSTM, characterized in that It includes the following steps: Step 1: Collect historical monitoring data of the panel concrete; Step 2: Construct a collection table for the panel concrete monitoring data and process the data in the collection table; Step 3: Use the LSTM neural network model to predict the temperature of the panel concrete and obtain the predicted value of the panel concrete temperature; Step 4: Use the historical monitoring data samples to train the LSTM neural network model and save the optimal model parameters.

2. The method for predicting the temperature of panel concrete based on LSTM according to claim 1, characterized in that: In the said Step 1, obtain the historical monitoring data of the panel concrete during the pouring period and the curing stage, specifically as follows: The temperature data of the face slab concrete is obtained by means of distributed optical fiber temperature monitoring, and the temperature data T at different spatial positions on the face slab concrete can be obtained at the same temperature monitoring time t. t ; Environmental monitoring data is monitored through a weather station. The environmental monitoring data includes environmental temperature T e_t , environmental humidity RH t , instantaneous wind speed WD t , instantaneous solar radiation SR t ; At the same time, record the pouring temperature T b of each measuring point and the concrete pouring time t e .

3. The method for predicting the temperature of panel concrete based on LSTM according to claim 1, which is characterized in that: In the said Step 2, take all the collected historical monitoring data, with the panel concrete temperature monitoring time t as the benchmark, match the environmental monitoring data according to the concrete temperature time, construct a collection table for the panel concrete monitoring data, and preprocess all the data in the collection table.

4. The method for predicting the temperature of panel concrete based on LSTM according to claim 3, wherein: The said Step 2 includes the following steps: S2.1: When matching the temperature data information of the temperature measurement points with the environmental data information, use the temperature monitoring time t as the primary key, select the environmental data information measured at the time t' closest to the temperature monitoring time t, and match it with the temperature data information to form a distributed optical fiber temperature and environmental information table; S2.2: Finally, match according to the measured pouring temperature of the temperature measurement points, with the measurement point number as the primary key, to construct a collection table for the panel concrete monitoring data; S2.3: Average and correct the obvious abnormal data in the collection table of the panel concrete monitoring data according to the adjacent monitoring time, and uniformly supplement the missing data according to the adjacent time; S2.4: Standardize the data in the collection table of the panel concrete monitoring data.

5. The method for predicting the temperature of panel concrete based on LSTM according to claim 4, characterized in that: Data standardization is performed using Min-Max normalization to map the data between 0 and 1; the independent variable time series X is obtained t = t , e_t , t , t , t , b ; In formula (1), x' j is the result after standardizing a certain single-column attribute in the table data collection table, and x j is the single-column attribute value in the data collection table; min(x j ) represents the minimum value in a single-column attribute; max(x j ) represents the maximum value in a single-column attribute.

6. The method for predicting the temperature of panel concrete based on LSTM according to claim 1, characterized in that: In the said Step 3, use the LSTM neural network model to predict the temperature of the panel concrete. Take the panel concrete temperature within a certain period and the measured environmental data information within the corresponding period as the input data of the LSTM neural network model. The panel concrete temperature output by the LSTM neural network model is the standardized predicted value; then perform an inverse operation on the output panel concrete temperature data to obtain the predicted value of the panel concrete temperature.

7. The method for predicting the temperature of panel concrete based on LSTM according to claim 6, wherein: The said Step 3 includes the following steps: S3.1: Sort the data in the standardized data collection form according to the measurement point number and the monitoring time sequence, divide the sorted data into a training set and a test set according to a ratio of 3:1, and construct the time series Y of the dependent variable t = [y′ j , where y′ j is the predicted data; S3.2: Using the independent variable time series X t as the input of the LSTM neural network model, using the output of the LSTM neural network model as the standardized predicted value, and constructing the time series X′ from the standardized predicted values t ; S3.3: Select the predicted time series Y t The predicted temperature Y of the panel concrete in tT , and restore the output time series Y according to the inverse operation of standardization t The panel concrete temperature prediction vector Y' of t =[T' t .

8. The method for predicting the temperature of panel concrete based on LSTM according to claim 7, wherein: The specific inverse operation is as follows: T′ t = Y tT (Y tT-max - Y tT-nin ) + Y tT-min (2); In formula (2), Y tT represents the predicted sequence values of the face slab concrete temperature of the time series Y t ; Y tT-max represents the maximum value in the predicted sequence of the face slab concrete temperature of Y tT ; X′ T-min represents the minimum value in the predicted sequence of the face slab concrete temperature of Y tT ; T′ t represents the predicted face slab concrete temperature value obtained after the inverse operation.

9. The method for predicting the temperature of panel concrete based on LSTM according to claim 1, wherein: In step 4, the root mean square error (RMSE) is used as the loss function to calculate the predicted panel concrete temperature vector Y'. t and the measured panel concrete temperature vector X t The RMSE value is as follows: Take the minimum loss function value as the optimization goal, use the historical monitoring data samples to train the LSTM neural network model, and save the optimal model parameters.

Images