Mass concrete temperature prediction method based on multi-source mixed data input
Through a deep neural network method based on multi-source data, combined with temperature field theoretical analysis and actual measured data, a CNN-LSTM-Attention model was established, which solved the data dependence and reliability problems in the temperature prediction of large-volume concrete, and achieved the construction period Accurate prediction of temperature changes and adaptability analysis in complex environments.
Patent Information
- Application Number
- CN202411021499.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-19
AI Technical Summary
The prior art has insufficient data dependence in the temperature prediction of large volume concrete, poor model reliability, and unpredictable temperature changes at undeployed sensor points, and traditional methods lack flexibility in complex construction environments.
A deep neural network method based on multi-source mixed data input is adopted, combined with temperature field theory analysis model and measured data, a CNN-LSTM-Attention model is established, and a variety of influencing factors are considered to predict the temperature changes of large-volume concrete.
It realizes accurate prediction of temperature changes in large-volume concrete construction period, without the need for actual measured data at points to be predicted, enhances the reliability and adaptability of the model, and can conduct holistic analysis in complex environments.
Smart Images

Figure CN118983019A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of concrete construction temperature prediction, and in particular to a mass concrete temperature prediction method based on multi-source mixed data input. Background Art
[0002] During the hydration process, large-volume concrete releases a lot of heat, causing the internal temperature to rise. If the cooling is uneven, a temperature gradient will be generated, resulting in a large temperature difference between the inside and the surface, causing thermal stress, which will lead to concrete cracking.
[0003] Temperature control is crucial to the quality and safety of large-volume concrete. Its temperature changes need to be accurately monitored throughout the construction and maintenance period. In this process, it is important to predict the development of structural temperature changes. On the one hand, temperature prediction helps to optimize the construction plan, including pouring time, layer thickness control, pouring sequence, etc., thereby improving construction efficiency and reducing construction risks; on the other hand, temperature prediction can help identify and control possible structural quality risks in advance and implement dynamic and differentiated maintenance measures.
[0004] Traditional methods for temperature prediction mainly include numerical simulation and empirical formula. Numerical simulation requires a large amount of initial data and computing resources, and the accuracy of the model depends on the accuracy of the input parameters. If the input data is inaccurate, the prediction results may deviate from the actual situation. The empirical formula method is based on a large amount of experimental data and engineering experience to establish an empirical formula to predict the temperature change of concrete. It has low accuracy and is only suitable for quick estimation under similar conditions. It cannot adapt to complex and changeable construction environments and lacks flexibility.
[0005] With the development of artificial intelligence technology, methods for predicting the temperature of large-volume concrete using deep learning have gradually attracted attention, including convolutional neural networks (CNNs) and long short-term memory networks (LSTMs). Although these methods can obtain relatively accurate temperature prediction values, they generally have limitations. The model is trained using the existing sampling data of the target point to be predicted. However, training only relies on the data of the target point, the data volume is small, and the model reliability is poor. At the same time, in actual projects, predictions require the support of measured data of the target point, and it is impossible to predict the temperature changes of points where sensors are not deployed. In addition, the change of concrete temperature is affected by many factors, including ambient temperature, ambient humidity, curing shed temperature, and temperature changes of adjacent measuring points. Summary of the invention
[0006] Purpose of the invention: To propose a mass concrete temperature prediction method based on multi-source mixed data input to solve the above problems existing in the prior art.
[0007] In a first aspect of the present invention, a method for predicting the temperature of mass concrete based on multi-source mixed data input is proposed, the method comprising the following steps:
[0008] S1. Construct a theoretical analysis model of temperature field;
[0009] S2 establishes a mass concrete temperature influencing factor index S according to the temperature field theoretical analysis model, and integrates theoretical calculation data with measured data to form a mass concrete hydration heat data sample library;
[0010] S3 establishes a deep neural network model, the input of the model is the sample data in the mass concrete hydration heat data sample library, and the output is the predicted value of the mass concrete temperature change;
[0011] S4. Divide the sample data in the large-volume concrete hydration heat data sample library into a training data set and a test data set; use the training data set to train the deep neural network model, use the test data set to test the training result, and use the trained deep neural network model to predict the temperature change during the construction period of the large-volume concrete.
[0012] In a further embodiment, the factors affecting the temperature field during the mass concrete construction period include pouring temperature, adiabatic temperature rise caused by cement hydration, and thermal conductivity.
[0013] In a further embodiment, the expression of the temperature field model in step S1 is as follows:
[0014]
[0015] Where T is the pouring temperature, θ is the adiabatic temperature rise caused by cement hydration, τ is the pouring time, x is the length direction, y is the width direction, and z is the thickness direction.
[0016] In a further embodiment, calculating the temperature field change theoretical data according to the temperature field model in step S2 specifically includes:
[0017] The concrete flat plate structure is divided into n-1 layers along the thickness direction. The three adjacent layers are numbered n-1, n, and n+1. The thickness of each layer is H represents the total thickness of the concrete slab structure;
[0018] The pouring time is divided into multiple finite segments Δt. In the τth time, the temperatures corresponding to the three adjacent points are T n-1,τ , T n,τ , T n+1,τ ;
[0019] Based on the finite difference method of one-dimensional temperature field of mass concrete, the temperature T of the nth layer at the τ+1th time is predicted.n,τ+1 :
[0020]
[0021] In the formula, represents the thermal conductivity, ρ is density; c is specific heat; λ is thermal conductivity; θ τ It represents the adiabatic temperature rise caused by cement hydration at the τth time.
[0022] In a further embodiment, the adiabatic temperature rise generated by cement hydration at the τth time satisfies the following expression:
[0023]
[0024] In the formula, m c For each m 3 The mass of cement required for concrete; Q0 is the heat released per kg of cement; ρ is the apparent density of concrete; c is the specific heat of concrete; m f For each m 3 The mass of fly ash required for concrete; m is the hydration coefficient, ordinary cement m = 0.43 + 0.0018 (m c +m f ), slag cement m = 0.55 + 0.001 (m c +m f ).
[0025] Through the above theory, it can be seen that at a certain point T n Temperature variation with T n-1 , T n+1 、Concrete insulation temperature riseθ τ At the same time, the temperature change of concrete hydration heat is also affected by the temperature of the curing shed, ambient temperature and ambient humidity. Therefore, an index S of influencing factors affecting the temperature change of large-volume concrete can be established.
[0026] In a further embodiment, the mass concrete temperature influencing factor index includes ambient temperature, ambient humidity, adjacent layer temperature, and pouring time, and the expression is as follows:
[0027] S=(T c ,T n-1 ,T n+1 ,T env ,τ,RH,θ τ )
[0028] Where, T n-1 , T n+1 Respectively represent the temperature of the n-1th layer and the n+1th layer; T env represents the ground temperature; RH represents the ambient humidity; T crepresents the temperature inside the concrete curing shed (external ambient temperature); τ is the pouring time; θ τ It is the adiabatic temperature rise of concrete hydration heat;
[0029] The temperature field change theoretical data are combined with the actual monitored data to form a large volume concrete hydration heat data sample library M:
[0030]
[0031] In the formula, data RH ,dataθ τ They represent the temperature inside the concrete curing shed (external ambient temperature), the temperature of the n-1th layer, the temperature of the n+1th layer, the foundation temperature, the ambient humidity, θ τ This is the sample data corresponding to the adiabatic temperature rise of concrete hydration heat.
[0032] In a further embodiment, the deep neural network model includes a CNN layer and an LSTM layer; the CNN layer is used to extract high-dimensional features from the input data; the LSTM layer is used to capture the temporal correlation in the historical sequence, and an attention mechanism is introduced to optimize the weight of the LSTM output;
[0033] The LSTM layer consists of a forget gate, an input gate, and an output gate; the forget gate uses a sigmoid function to filter memory information and newly input information at the previous moment. When the value of the forget gate is 1, all information will be retained; when the value is 0, all information will be discarded;
[0034] The output h at time step t t Calculated as:
[0035] i t =σ(W i x t +U i h t-1 +b i )
[0036] f t =σ(W f x t +U f h t-1 +b f )
[0037] o t =σ(W o x t +U o h t-1 +b o )
[0038]
[0039] H t =Vh t
[0040] In the formula, σ represents the sigmoid activation function, tanh represents the hyperbolic tangent activation function, and f t Forget the door, i t represents the input gate, o t represents the output gate, Represents the product of vector elements; W * , U * , b * , V represents the network parameters to be learned, *∈{i,f,o,c} and represents the candidate state obtained by the nonlinear function waiting to be stored in the long-term memory; H t is the output of the LSTM layer at time t.
[0041] The attention mechanism is introduced to optimize the weight of the LSTM layer output, which is calculated as follows:
[0042] e t =utanh(wh t +b)
[0043]
[0044] In the formula, the attention score function e t Responsible for calculating the correlation between the temperature value and the output vector of the LSTM network layer at time t; the weight coefficients are u and w; b is the bias coefficient; the output of the Attention layer at time t is represented by s t .
[0045] In a further embodiment, step 4 specifically includes:
[0046] The sample data in the large volume concrete hydration heat data sample library is divided into a training data set and a test data set according to a ratio of 4:1; the training data set and the test data set are normalized:
[0047]
[0048] In the formula, x n is the normalized data, x o is the original data, x mean is the mean value of the input data, x max is the maximum value of the input data, x min is the minimum value of the input data;
[0049] The deep neural network model is trained using the training data set, and the training results are tested using the test data set; mean absolute error MAE, mean absolute percentage error MAPE, and root mean square error RMSE are used as evaluation indicators:
[0050]
[0051] In the formula, y i represents the true value; represents the predicted value; n represents the number of data.
[0052] In a further embodiment, the input of the deep neural network model is the sample data in the mass concrete hydration heat data sample library, and the output of the model is the predicted value of the mass concrete temperature change, which is expressed as follows:
[0053] ω(τ,n)=F(T c ,T n-1 ,T n+1 ,T env ,τ,RH)
[0054] Where, T n-1 , T n+1 Respectively represent the theoretical temperatures of the n-1th layer and the n+1th layer; T c Indicates the temperature of the concrete surface; T env represents the ground temperature; RH represents the ambient humidity; F(*) represents the deep neural network model; ω(τ,n) is the predicted value of the temperature change at the τth time in the nth layer output after prediction by the deep neural network model.
[0055] Beneficial effects: The present invention establishes a temperature prediction influencing index for mass concrete construction, comprehensively considers the factors affecting temperature development, and uses mixed data of multiple types of influencing factors as input to establish a temperature prediction model based on a deep neural network CNN-LSTM-Attention. The training data of the model comes from the detailed finite element analysis data of the target structure in the early stage. The temperature change trend can be accurately predicted without the measured data of the predicted point, realizing the overall analysis of the temperature change of the construction structure. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 Schematic diagram of the deep neural network model in the embodiment.
[0057] Figure 2 LSTM network structure diagram in the embodiment.
[0058] Figure 3 This is a diagram of the attention mechanism (Attention) network structure in the embodiment.
[0059] Figure 4 This is a framework diagram of the CNN-LSTM-Attention model in the embodiment.
[0060] Figure 5 This is a diagram of the deep neural network model training results in the embodiment.
[0061] Figure 6 This is a statistical diagram of mixed data input into the temperature prediction model in the embodiment.
[0062] Figure 7 It is a comparison chart of the predicted value and the measured value in the embodiment.
[0063] Figure 8 Schematic diagram of the temperature measurement point arrangement in the embodiment.
[0064] Fig. 9 Flow chart of the method for predicting temperature of mass concrete in the embodiment. DETAILED DESCRIPTION
[0065] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features known in the art are not described.
[0066] This embodiment discloses a method for predicting the temperature of mass concrete based on multi-source mixed data input, and the process is as follows:
[0067] S1: Determine the factors affecting the temperature change during the construction period of mass concrete based on theoretical analysis;
[0068] The main factors affecting the maximum temperature rise inside large-volume concrete are: (1) pouring temperature; (2) adiabatic temperature rise caused by cement hydration; (3) thermal conductivity. As concrete hardens, the heat exchange between the core of the concrete and the environment produces an uneven temperature field, and its internal hydration heat is constantly changing, which is an unstable temperature field.
[0069] S2: Obtain a sample library of mass concrete hydration heat data;
[0070] In the absence of analytical data, finite element analysis is used to construct a temperature field theoretical analysis model, calculate the theoretical data of temperature field changes based on the temperature field theoretical analysis model, and combine it with the actual monitored data to form a sample library of large-volume concrete hydration heat data.
[0071] The temperature field theoretical analysis model is established through partial differential equations:
[0072]
[0073] Mass concrete is often extra long and extra wide, but thin, and can be assumed to be an infinitely large flat plate structure. Since its thickness is much smaller than its length and width, in engineering calculations, the horizontal temperature distribution of mass concrete is generally always very close, and the very small heat conduction effect in the horizontal direction can be ignored. The heat conduction in the vertical direction is much greater than the heat conduction in the horizontal direction. Therefore, mass concrete can be simplified to one-dimensional heat conduction in the vertical direction, and the concrete heat conduction differential equation can be solved by the finite difference method.
[0074] Assuming that the temperature field of mass concrete can be solved as the temperature field of an infinitely large plate, let the thickness of the mass concrete plate be H, divide the mass concrete into n-1 layers along the thickness direction, and number the three adjacent points as n-1, n, and n+1. Let the thickness of each small layer be h, that is, At the same time, the pouring time is divided into multiple finite segments Δt. In the τth time, the temperatures corresponding to the three adjacent points are T n-1,τ , T n,τ , T n+1,τ ; where T n,τ Represents the temperature of the nth point at time τ. According to the difference principle, ignoring the truncation error, the partial derivatives of each order at each point inside the concrete can be calculated.
[0075] Based on the finite difference method of one-dimensional temperature field of mass concrete, the temperature T of the nth layer at the τ+1th time is predicted. n,τ+1 :
[0076]
[0077] The adiabatic temperature rise generated by cement hydration at the τth time satisfies the following expression:
[0078]
[0079] m c For each m 3 The mass of cement required for concrete, kg / m 3 ; Q0 is the heat released per kg of cement, kJ / kg; ρ is the apparent density of concrete, generally 2400kg / m3; c is the specific heat of concrete, generally (0.92~1.0)×103J / (kg·℃); m f For each m 3 The mass of fly ash required for concrete, kg / m 3; m is the hydration coefficient, ordinary cement: m = 0.43 + 0.0018 (m c +m f ), slag cement: m = 0.55 + 0.001 (m c +m f ).
[0080] Through the above theory, it can be seen that at a certain point T n Temperature variation with T n-1 , T n+1 、Concrete insulation temperature riseθ τ At the same time, the temperature change of concrete hydration heat is also affected by the temperature of the curing shed, the ambient temperature, and the ambient humidity. Therefore, we can establish the influencing factor index S that affects the temperature change of large-volume concrete.
[0081] The influencing factors of mass concrete temperature include ambient temperature, ambient humidity, adjacent layer temperature, and pouring time. The expressions are as follows:
[0082] S=(T c ,T n-1 ,T n+1 ,T env ,τ,RH,θ τ )
[0083] Where, T n-1 , T n+1 Respectively represent the temperature of the n-1th layer and the n+1th layer; T env represents the ground temperature; RH represents the ambient humidity; T c represents the temperature inside the concrete curing shed (external ambient temperature); τ is the pouring time; θ τ It is the adiabatic temperature rise of concrete hydration heat;
[0084] The temperature field change theoretical data are combined with the actual monitored data to form a large volume concrete hydration heat data sample library M:
[0085]
[0086] In the formula, data RH ,dataθ τ They represent the temperature inside the concrete curing shed (external ambient temperature), the temperature of the n-1th layer, the temperature of the n+1th layer, the foundation temperature, the ambient humidity, θ τ This is the sample data corresponding to the adiabatic temperature rise of concrete hydration heat.
[0087] In order to determine the required temperature field, the specific temperature field at a certain time period must be calculated based on known initial conditions and boundary conditions. Boundary conditions refer to the law of temperature interaction between the concrete surface and the surrounding medium (such as air or water); initial conditions and boundary conditions are collectively called boundary conditions; the influence of many factors such as the initial temperature field (placement temperature of large-volume concrete structures); the geometric characteristics of the concrete structure (the size and connection topology of each component), material properties (thermal conductivity, density, specific heat and diffusion coefficient of concrete); and boundary conditions that change with time (ambient temperature, radiant heat, wind speed, structural position). After determining the above factors, the temperature field of the large-volume concrete structure at a given time can be solved. In the process of calculating the temperature field of large-volume concrete, the thermal parameter a, which is the thermal conductivity, also needs to be considered. ρ is density; c is specific heat; λ is thermal conductivity; θ is the value and distribution of adiabatic temperature rise.
[0088] From the above analysis, we can see that the small heat conduction effect in the horizontal direction can be ignored for large-volume concrete, while the heat conduction in the vertical direction often has a greater impact. In addition, the upper surface boundary temperature and the foundation temperature also need to be considered in the calculation of the differential method.
[0089] S3: Establish a deep neural network model CNN-LSTM-Attention with mixed data input;
[0090] In order to achieve accurate temperature prediction of the hydration heat process of large-volume concrete, a deep neural network model CNN-LSTM-Attention with mixed data input is established. The one-dimensional CNN layer is used to extract high-dimensional features from the input data, and the LSTM layer is used to capture the temporal correlation in the historical sequence. At the same time, the attention mechanism is introduced to optimize the weight of the LSTM output, enhance the influence of key information, and optimize the overall prediction model.
[0091] The network structure diagram of Convolutional Neural Networks (CNN) is as follows Figure 1 As shown in Figure 1, its main goal is to extract significant features from the input data, which is usually composed of convolutional layers, pooling layers, discard layers, and fully connected layers. The convolutional layer plays a vital role in feature extraction, and the convolution kernel captures relevant features from the input data. The maximum pooling technique and the RELU activation function are used to discard irrelevant features of piecewise linearity, and the fully connected layer expands the pooled neurons into a one-dimensional vector form.
[0092] The LSTM network structure is as follows Figure 2As shown in the figure, it consists of three gates: forget, input, and output. The forget gate uses the sigmoid function to filter memory information and newly input information from the previous moment. When the value of the gate is 1, all information is retained, and when the value is 0, all information is discarded. This function is essential for LSTM's ability to retain long-term dependencies in time series data. However, LSTM still has the problem of losing important information when the input sequence is too long. Therefore, CNN is combined with the long short-term memory (LSTM) network to enhance performance. CNN is used to preprocess the data and filter out irrelevant information, thereby improving the accuracy of the prediction. H is the output of LSTM, and the output h at time step t t It can be calculated as:
[0093] i t =σ(W i x t +U i h t-1 +b i )
[0094] f t =σ(W f x t +U f h t-1 +b f )
[0095] o t =σ(W o x t +U0h t-1 +b0)
[0096]
[0097] H t =Vh t
[0098] Here, σ represents the sigmoid activation function, tanh represents the hyperbolic tangent activation function, and f t Forget the door, i t represents the input gate, o t represents the output gate, represents the product of vector elements, W * , U * , b * , V represents the network parameters to be learned, *∈{i,f,o,c} and represents the candidate states obtained by the nonlinear function waiting to be stored in long-term memory.
[0099] Attention mechanism represents a resource allocation mechanism that mimics human attention by focusing on relevant areas while reducing attention to irrelevant areas. By discarding unimportant information and amplifying key information, this mechanism enhances the importance of key information and improves the accuracy of the model. The attention mechanism achieves this by adopting a probability distribution to pay enough attention to key messages, thereby compensating for the information loss caused by LSTM due to long sequences.
[0100] See the network structure diagram of the attention mechanism (Attention) Figure 3 , x t represents the input of the LSTM model, h t represents the output of the hidden layer obtained by LSTM. The attention probability distribution of the attention mechanism output to the hidden layer of LSTM is represented by α t , the output of the attention-optimized LSTM is denoted as y. The input of the attention layer consists of the output of the LSTM layer after extracting the time series features. The weights of the attention layer are calculated as follows:
[0101] e t =utanh(wh t +b)
[0102]
[0103] Attention scoring function t Responsible for calculating the correlation between the temperature value and the output vector of the LSTM network layer at time t. The weight coefficients are u and w, b is the bias coefficient; the output of the Attention layer at time t is represented by s t .
[0104] The CNN-LSTM-Attention model framework is as follows Figure 4 As shown in the figure, it includes input layer, CNN layer, LSTM, Attention layer and output layer. Multi-dimensional time series data is used as the input of CNN layer. CNN layer performs convolution operation, increases depth, compresses the number of parameters and reduces feature dimension through pooling. The fully connected layer converts the features into a one-dimensional structure and completes feature vector extraction. LSTM and attention layers learn the internal variation law of the load from the extracted features, thereby realizing the prediction function.
[0105] The model training sample data includes input samples and output samples. According to the theoretical analysis in S2, the input samples of the prediction model are parameters related to the temperature change of a certain point of mass concrete, including time, temperature values of adjacent points, ambient temperature values, temperature values in the curing shed, humidity values and other information. The output samples are historical mass concrete temperature change values. Before being used as training samples, the data samples are preprocessed by removing outliers, filling in missing values and normalizing the data. The data samples are provided by the simulation analysis in S2 to ensure that the deep learning model is fully trained.
[0106] S4: Use the data in the sample library to train the model and predict the temperature changes during the construction period of mass concrete;
[0107] The large volume concrete simulation data in the sample library in S2 is used as the model training data, and 28 days of hydration thermal analysis data under 5 different environmental conditions are selected, and the data collection cycle is 15 minutes. The ratio of the training data set to the test data set is 4:1. First, the input data is normalized:
[0108]
[0109] In the formula, x n is the normalized data, x o is the original data, x mean is the mean value of the input data, x max is the maximum value of the input data, x min is the minimum value of the input data.
[0110] In order to test the effectiveness of the model and evaluate the prediction effect, mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) were selected as evaluation indicators. The calculation formulas of each evaluation indicator are as follows:
[0111]
[0112] The proposed CNN-LSTM-Attention based model was optimized and implemented in the TensorFlow framework. The experimental hardware includes Intel(R) Core(TM) i7-10510 U CPU and 16GB RAM. The Adam optimization algorithm updates the network parameters with an initial learning rate of 0.01. The neural network training parameters are set to a maximum of 50 iterations and a minimum batch size of 1024. Table 1 shows the specific structural parameters of the model.
[0113] Table 1: Model structure parameters
[0114]
[0115]
[0116] The input layer of the model receives raw time series data with 6 dimensions (48×6) in 48 time steps to predict the target data. The convolution layer is used for data feature extraction, followed by a dropout layer that discards 20% of neurons to avoid overfitting. Finally, the feature sequence is converted to a 48×64 vector and input to the LSTM layer. The output of the LSTM layer is integrated into the attention layer through a fully connected layer to calculate the attention distribution and optimize the weight distribution. The model generates a set of output prediction results every 48 time steps. The above model hyperparameters are designed to adapt to the given data and ensure the accuracy of the prediction model. The model training results are shown in Figure 2. Figure 5 As shown in Figure 2, the CNN-LSTM-Attention model converges quickly in the early stages of training.
[0117] The trained deep learning model is used to predict the temperature of large-volume concrete in actual projects, and the temperature values of points without sensors are predicted by the measured values of points with temperature sensors. Figure 6 As shown in the figure, the mixed data of the input temperature prediction model includes the upper and lower temperature values of the target point, the ambient temperature value, the temperature value in the maintenance shed, and the humidity value. The time step of the model prediction is set to 48, that is, the next set of data is predicted by the first 48 sets of data.
[0118] The input of the deep neural network model is the sample data in the mass concrete hydration heat data sample library, and the output of the model is the predicted value of the mass concrete temperature change, which is expressed as follows:
[0119] ω(τ,n)=F(T c ,T n-1 ,T n+1 ,T env ,τ,RH)
[0120] Where, T n-1 , T n+1 Respectively represent the theoretical temperatures of the n-1th layer and the n+1th layer; T c Indicates the temperature of the concrete surface; T env represents the ground temperature; RH represents the ambient humidity; F(*) represents the deep neural network model; ω(τ,n) is the predicted value of the temperature change at the τth time in the nth layer output after prediction by the deep neural network model.
[0121] The comparison between the predicted value and the measured value is as follows Figure 7 As shown, the concrete temperature value predicted by the present invention is in good agreement with the measured value.
[0122] Corresponding to the above embodiment, the present embodiment also discloses a mass concrete temperature prediction system based on multi-source mixed data input, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a series of steps corresponding to the method in the above embodiment are implemented.
[0123] The scheme proposed in the above embodiment was tested at a nuclear power project raft foundation construction site. The main body of the raft foundation is a cylindrical structure with a radius of 27m and a thickness of 3.70m. The raft foundation concrete is C45P8. There is a concrete prestressed tensioning corridor at the bottom of the raft foundation. The corridor is a circular ring structure with the same center as the raft foundation. The radius R = 20.68m, the wall thickness of the two sides of the corridor is 0.8m, the corridor width is 2.5m, and the top is a concrete cover plate.
[0124] The hydration heat during the raft foundation pouring process is monitored by embedding temperature sensors. The monitoring frequency is 15 minutes / time. The overall design of the temperature measurement point layout is as follows: Figure 8 shown.
[0125] (1) According to the pouring direction of raft foundation concrete, the temperature sensors are mainly arranged in two radial directions: 0° (BB) and 90° (AA):
[0126] ①, Select three points in the 0° direction, namely 6.5m point (1#), 13.5m point (2#), and 22m point (3#);
[0127] ② Select three points along the radius direction at 90°, namely 6.5m point (4#), 22m point (5#) and radius point (6#);
[0128] (2) Select the 45° (DD) radial direction as the secondary layout direction: select two points along the 45° radial direction, namely the 13.5m point (7#) and the radius point (8#);
[0129] (3) Select the center point to arrange the sensor (10#);
[0130] (4) One atmospheric temperature measuring point;
[0131] (5) Two insulation layer temperature measurement points;
[0132] (6) Two temperature measurement points in the insulation shed;
[0133] In the process of temperature monitoring, the temperature of multiple different points on the raft foundation is predicted by the proposed scheme, and the input is the measured values of the first 48 steps of the parameters in the established temperature index, which are compared with the measured values.
[0134] It should be noted that the terms such as "upper", "lower", "left", "right", "front", "back", etc. cited in the invention are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments in their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.
[0135] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. A method for predicting temperature of mass concrete based on multi-source mixed data input, characterized in that: The steps include: S1. Construct a theoretical analysis model of temperature field; S2 establishes a mass concrete temperature influencing factor index S according to the temperature field theoretical analysis model, and integrates theoretical calculation data with measured data to form a mass concrete hydration heat data sample library; S3 establishes a deep neural network model, the input of the model is the sample data in the mass concrete hydration heat data sample library, and the output is the predicted value of the mass concrete temperature change; S4, dividing the sample data in the large volume concrete hydration heat data sample library into a training data set and a test data set; The deep neural network model is trained using the training data set, the training result is tested using the test data set, and the temperature change during the construction period of large-volume concrete is predicted using the trained deep neural network model.
2. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 1 is characterized in that: The factors affecting the temperature field during the mass concrete construction period include pouring temperature, adiabatic temperature rise caused by cement hydration, and thermal conductivity.
3. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 2 is characterized in that: The expression of the temperature field model in step S1 is as follows: Where T is the pouring temperature, θ is the adiabatic temperature rise caused by cement hydration, τ is the pouring time, x is the length direction, y is the width direction, and z is the thickness direction.
4. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 3 is characterized in that: Step S2 calculates the temperature field change theoretical data according to the temperature field model, specifically including: The concrete flat plate structure is divided into n-1 layers along the thickness direction. The three adjacent layers are numbered n-1, n, and n+1. The thickness of each layer is H represents the total thickness of the concrete slab structure; The pouring time is divided into multiple finite segments Δt. In the τth time, the temperatures corresponding to the three adjacent points are T n-1,τ , T n,τ , T n+1,τ ; Based on the finite difference method of one-dimensional temperature field of mass concrete, the temperature T of the nth layer at the τ+1th time is predicted. n,τ+1 : In the formula, represents the thermal conductivity, ρ is density; c is specific heat; λ is thermal conductivity; θ τ It represents the adiabatic temperature rise caused by cement hydration at the τth time.
5. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 4 is characterized in that: The adiabatic temperature rise generated by cement hydration at the τth time satisfies the following expression: In the formula, m c For each m 3 The mass of cement required for concrete; Q0 is the heat released per kg of cement; ρ is the apparent density of concrete; c is the specific heat of concrete; m f For each m 3 The mass of fly ash required for concrete; m is the hydration coefficient, ordinary cement m = 0.43 + 0.0018 (m c +m f ), slag cement m = 0.55 + 0.001 (m c +m f ).
6. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 5 is characterized in that: The influencing factors of mass concrete temperature include ambient temperature, ambient humidity, adjacent layer temperature, and pouring time. The expressions are as follows: S=(T c ,T n-1 ,T n+1 ,T env ,τ,RH,θ τ ) Where, T n-1 、T n+1 Respectively represent the temperature of the n-1th layer and the n+1th layer; T env represents the ground temperature; RH represents the ambient humidity; T c represents the temperature inside the concrete curing shed (external ambient temperature); τ is the pouring time; θ τ It is the adiabatic temperature rise of concrete hydration heat; The temperature field change theoretical data are combined with the actual monitored data to form a large volume concrete hydration heat data sample library M: In the formula, data RG ,dataθ τ They represent the temperature in the concrete curing shed, the temperature of the n-1th layer, the temperature of the n+1th layer, the foundation temperature, the ambient humidity, θ τ This is the sample data corresponding to the adiabatic temperature rise of concrete hydration heat.
7. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 4 is characterized in that: The deep neural network model includes a CNN layer and an LSTM layer; the CNN layer is used to extract high-dimensional features from input data; The LSTM layer is used to capture the temporal correlation in the historical sequence, and the attention mechanism is introduced to optimize the weight of the LSTM output; The LSTM layer consists of a forget gate, an input gate, and an output gate; the forget gate uses a sigmoid function to filter memory information and newly input information at the previous moment. When the value of the forget gate is 1, all information will be retained; when the value is 0, all information will be discarded; The output h at time step t t Calculated as: i t =σ(W i x t +U i h t-1 +b i ) f t =σ(W f x t +U f h t-1 +b f ) the t =σ(W o x t +U o h t-1 +b o ) H t =Vh t In the formula, σ represents the sigmoid activation function, tanh represents the hyperbolic tangent activation function, and f t Forget the door, i t represents the input gate, o t represents the output gate, Represents the product of vector elements; W * , U * , b * , V represents the network parameters to be learned, *∈{i,f,o,c} and represents the candidate state obtained by the nonlinear function waiting to be stored in the long-term memory; H t is the output of the LSTM layer at time t.
8. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 7 is characterized in that: It also includes the introduction of an attention mechanism to optimize the weights of the LSTM layer output, which is calculated as follows: e t =utanh(wh t +b) In the formula, the attention score function e t Responsible for calculating the correlation between the temperature value and the output vector of the LSTM network layer at time t; The weight coefficients represent u and w; b is the bias coefficient; The output of the Attention layer at time t is represented as s t .
9. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 8 is characterized in that: Step 4 specifically includes: The sample data in the large volume concrete hydration heat data sample library is divided into a training data set and a test data set according to a ratio of 4:1; the training data set and the test data set are normalized: In the formula, x n is the normalized data, x o is the original data, x mean is the mean value of the input data, x max is the maximum value of the input data, x min is the minimum value of the input data; The deep neural network model is trained using the training data set, and the training results are tested using the test data set; mean absolute error MAE, mean absolute percentage error MAPE, and root mean square error RMSE are used as evaluation indicators: In the formula, y i represents the true value; represents the predicted value; n represents the number of data.
10. The method for predicting temperature of mass concrete based on multi-source mixed data input according to claim 6, characterized in that: The input of the deep neural network model is the sample data in the mass concrete hydration heat data sample library, and the output of the model is the predicted value of the mass concrete temperature change, which is expressed as follows: ω(τ,n)=F(T c ,T n-1 ,T n+1 ,T env (( Where, T n-1 、T n+1 Respectively represent the theoretical temperatures of the n-1th layer and the n+1th layer; T c Indicates the temperature of the concrete surface; T env represents the ground temperature; RH represents the ambient humidity; F(*) represents the deep neural network model; ω(τ,n) is the predicted value of the temperature change at the τth time in the nth layer output after prediction by the deep neural network model.
Citation Information
Patent Citations
Mass concrete temperature monitoring data prediction method and system and medium
CN114742260A
Cable temperature prediction method and device and terminal equipment
CN116911107A
Dam concrete bin surface spraying optimization method based on Attention-LSTM model
CN118228582A
Cracking risk assessment and early warning method for mass concrete structure
WO2024140233A1
Cited By
Urban utility tunnel operation environment temperature determination method and system
CN119783324A
Nuclear power safety radiation dose rate predictor based on attention decomposition mechanism
CN120069167A
Concrete hydration heat self-adaptive control method and system based on digital twinning
CN120124266A
A concrete hydration heat self-adaptive control method and system based on digital twinning
CN120124266B
Internet of Things temperature control method and system for precooled concrete production
CN120653033A
