Concrete chloride ion concentration prediction method based on deep learning

Abstract

The invention relates to the technical field of chloride ion concentration prediction, and discloses a concrete chloride ion concentration prediction method based on deep learning, the concrete chloride ion concentration is predicted through a deep learning model, the prediction accuracy and reliability are improved, and the prediction accuracy and reliability are improved by calculating the prediction uncertainty and dynamically adjusting the learning rate. And the training process of the model is optimized. Through evaluation of prediction uncertainty, prediction errors of the model under specific conditions can be found in time, the learning rate is adjusted according to the prediction errors, and it is ensured that the model can keep the optimal performance and stability in different training stages. And a correction factor is generated by calculating the mean value and the standard deviation of the environmental parameters, and the predicted value is adjusted. The correction method based on the environmental parameters not only considers the influence of the external environment on the concrete chloride ion concentration, but also improves the robustness and adaptability of the prediction result. And a more accurate and reliable chloride ion concentration prediction result is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of chloride ion concentration prediction, and in particular to a method for predicting concrete chloride ion concentration based on deep learning. Background Art

[0002] Concrete is one of the most commonly used building materials in modern construction projects, and its durability is directly related to the safety and service life of building structures. Chloride ion corrosion is one of the main causes of early damage to concrete structures. Chloride ions mainly come from environmental factors such as sea salt and deicing salt. They can penetrate into the interior of concrete and cause steel corrosion after contact with steel bars, thereby weakening the overall performance of concrete structures. Therefore, accurately predicting the chloride ion concentration and its diffusion rate in concrete is crucial for evaluating and preventing early damage to concrete structures.

[0003] The process of chloride ion corrosion involves multiple physical and chemical phenomena: chloride ions migrate to the interior through pores and cracks in concrete, and the diffusion follows Fick's second law. The diffusion rate is affected by factors such as the density and porosity of concrete. When the chloride ion concentration exceeds the critical value, it will destroy the protective layer on the surface of the steel bar and cause steel corrosion. Deep learning is a machine learning method based on artificial neural networks that can automatically learn complex nonlinear relationships from a large amount of data. It has significant advantages in processing high-dimensional and nonlinear data and is particularly suitable for solving complex problems such as concrete chloride ion concentration prediction. However, concrete structures include multiple scenarios for rapid prediction, such as bridges, tunnels, coastal areas, saline-alkali areas, and concrete curing processes. Deep learning models usually contain multiple hidden layers and a large number of neurons, which leads to high computational complexity of the model. Even on high-performance computing devices, the model's inference time may be long. In scenarios such as bridges, tunnels, coastal areas, and saline-alkali areas, real-time monitoring of chloride ion concentration is required, and traditional deep learning models are difficult to meet this demand. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for predicting chloride ion concentration of concrete based on deep learning, comprising: Step 1, collect concrete structure samples, and obtain chloride ion concentration, environmental parameters, curing conditions, and component information of the concrete structure samples; Step 2, pre-processing the chloride ion concentration, environmental parameters, curing conditions, and component information; Step 3, initialize the parameters of the deep learning model, the deep learning model traverses the pre-treated chloride ion concentration, environmental parameters, maintenance conditions, and component information, performs forward propagation to calculate the predicted value, calculates the loss value of the loss function, performs back propagation to update the parameters, and completes the iterative process; repeat the iterative process until the maximum number of iterations is reached or the loss value of the loss function reaches the minimum, and the iteration is stopped to obtain the trained deep learning model; after traversing the pre-treated chloride ion concentration, environmental parameters, maintenance conditions, and component information, the deep learning model is verified using the verification set to obtain the verification prediction value and the verification loss value; Step 4, obtain the environmental parameters, curing conditions, and component information of the concrete structure, and input them into the trained deep learning model to predict the chloride ion concentration value of the concrete structure, and obtain the prediction uncertainty based on the verified prediction value; Step 5, preset a prediction threshold, compare the prediction uncertainty with the prediction threshold, if the prediction uncertainty is greater than or equal to the prediction threshold, make a first adjustment to the learning rate of the deep learning model, retrain and continue to execute step 4 until the prediction uncertainty is less than the prediction threshold, then stop re-adjusting the learning rate and retraining; if the prediction uncertainty is less than the prediction threshold, execute step 6; Step 6, calculate the correction factor according to the environmental parameters, obtain the chloride ion concentration value of the concrete structure according to the correction factor and the prediction, and obtain the final chloride ion concentration.

[0005] Furthermore, chloride ion concentration was used as a prediction label for the deep learning model.

[0006] Furthermore, the correction factor is calculated according to the environmental parameters, including: Step 61, obtaining parameter values ​​of concrete structure environmental parameters; Step 62, calculating the mean of the parameter value and the standard deviation of the parameter value; Step 63, calculating a correction factor according to the parameter value of the environmental parameter, the mean value of the parameter value and the standard deviation of the parameter value.

[0007] Furthermore, the standard deviation of the verified prediction value is taken as the prediction uncertainty.

[0008] Furthermore, the adjustment of learning rate includes: Step 51, preset the maximum number of training times and the maximum value of verification loss; Step 52, obtaining the number of training times; Step 53, calculating the prediction uncertainty after the current round of traversal and the prediction uncertainty after the previous round of traversal to obtain the uncertainty change; Step 54, calculate the learning rate according to the prediction uncertainty, the uncertainty change, the verification loss value, the number of training times, the maximum number of training times, and the maximum verification loss value.

[0009] Furthermore, the learning rate is: ; ; In the formula, is the learning rate, is the initial learning rate, U is the prediction uncertainty, T is the prediction threshold, L is the validation loss value, is the maximum value of the verification loss, t is the number of training times, is the maximum number of training times, , , are weight coefficients respectively, max is the maximum value operation, is the lower limit of the learning rate.

[0010] Furthermore, the correction factor is: ; In the formula, C is the correction factor, m is the adjustment coefficient, is the parameter value of the i-th environmental parameter, is the mean value of the parameter value of the i-th environmental parameter, is the standard deviation of the parameter value of the i-th environmental parameter.

[0011] Further, the final chloride ion concentration is: ; In the formula, is the final chloride ion concentration, In order to predict the chloride ion concentration of concrete structure, is the sigmoid function, is the exponential decay coefficient, f is the current traversal round, F is the total traversal round, is the dynamic weight coefficient.

[0012] The embodiments of the present invention have the following technical effects: The present invention predicts the chloride ion concentration of concrete through a deep learning model, which not only improves the accuracy and reliability of the prediction, but also dynamically adjusts the learning rate by calculating the prediction uncertainty, and further optimizes the training process of the model. Specifically, by evaluating the prediction uncertainty, the present invention can timely find the prediction error of the model under specific conditions, and adjust the learning rate accordingly to ensure that the model can maintain the best performance and stability at different training stages. In addition, the present invention introduces the correction factor of the environmental parameter, generates the correction factor by calculating the mean and standard deviation of the environmental parameter, and further adjusts the predicted value. This correction method based on environmental parameters not only takes into account the impact of the external environment on the chloride ion concentration of concrete, but also improves the robustness and adaptability of the prediction result. Finally, by combining the correction factor and the model prediction value, the present invention can provide a more accurate and reliable chloride ion concentration prediction result, which provides powerful technical support for the maintenance and management of concrete structures. This method has significant advantages in real-time monitoring, early warning system and maintenance decision support, and can effectively prevent and delay the corrosion of concrete structures, extend their service life, reduce maintenance costs, and improve the safety and reliability of infrastructure. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 This is a flow chart of a method for predicting chloride ion concentration in concrete based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0016] Figure 1 Schematic diagram of a method for predicting chloride ion concentration in concrete based on deep learning provided by an embodiment of the present invention. Figure 1 , including: Step 1: Collect concrete structure samples and obtain chloride ion concentration, environmental parameters, curing conditions, and component information of the concrete structure samples.

[0017] Environmental parameters include but are not limited to: temperature, humidity, rainfall, wind speed, salt content, pH, ultraviolet intensity, and air pollution index. Curing conditions include but are not limited to: curing time, curing temperature, curing humidity, curing method, curing medium, curing frequency, curing pressure, and curing ventilation. Component information includes but is not limited to: cement information, aggregate information, admixture information, and admixture information.

[0018] Step 2: pre-process the chloride ion concentration, environmental parameters, maintenance conditions, and composition information.

[0019] Step 3, initialize the parameters of the deep learning model, the deep learning model traverses the pretreated chloride ion concentration, environmental parameters, maintenance conditions, and component information, performs forward propagation to calculate the predicted value, calculates the loss value of the loss function, and performs back propagation to update the parameters to complete the iterative process; repeat the iterative process until the maximum number of iterations is reached or the loss value of the loss function is minimized, and the iteration is stopped to obtain a trained deep learning model; after traversing the pretreated chloride ion concentration, environmental parameters, maintenance conditions, and component information, the deep learning model is verified using the verification set to obtain the verified predicted value and the verified loss value.

[0020] Select a suitable deep learning model, such as multi-layer perceptron (MLP), convolutional neural network (CNN), recurrent neural network (RNN), etc., as the model architecture of deep learning. Determine the number of layers of the model and the number of neurons in each layer. For example, you can choose an MLP model with 3 hidden layers, each with 128 neurons. Model parameters include: weights, biases, activation functions, loss functions, optimizers, etc. Divide the preprocessed data into a training set and a validation set. The training set accounts for 80% and the validation set accounts for 20%. Input the data in the training set into the model, perform forward propagation to calculate the predicted value, and use the selected loss function to calculate the loss value between the predicted value and the true value. Calculate the gradient of the loss function with respect to the model parameters through the back-propagation algorithm, and use the optimizer to update the model parameters according to the calculated gradient. Set the maximum number of iterations (epochs), for example 100 times. Repeat the process of forward propagation, loss calculation, back-propagation, and parameter update until the maximum number of iterations is reached or the loss value of the loss function reaches the minimum value.

[0021] Use the pre-divided validation set data, input the trained model, and perform forward propagation to calculate the validation prediction value. Use the selected loss function to calculate the validation loss value between the validation prediction value and the true value. Calculate the performance indicators on the validation set, such as mean square error (MSE), mean square logarithmic error (MSLE), R² score, etc., and record the validation prediction value and validation loss value for subsequent model adjustment and evaluation.

[0022] The chloride ion concentration is used as the prediction label of the deep learning model, and the predicted value obtained by the model is the chloride ion concentration.

[0023] Step 4: Obtain the environmental parameters, curing conditions, and component information of the concrete structure, and input them into the trained deep learning model to predict the chloride ion concentration value of the concrete structure. Based on the verified predicted value, the prediction uncertainty is obtained.

[0024] The standard deviation of the verified predicted value is used as the prediction uncertainty. The standard deviation reflects the degree of dispersion of the predicted value. The larger the standard deviation, the higher the prediction uncertainty of the model.

[0025] Step 5, preset a prediction threshold, compare the prediction uncertainty with the prediction threshold, if the prediction uncertainty is greater than or equal to the prediction threshold, make a first adjustment to the learning rate of the deep learning model, retrain and continue to execute step 4 until the prediction uncertainty is less than the prediction threshold, then stop the learning rate readjustment and retraining; if the prediction uncertainty is less than the prediction threshold, execute step 6.

[0026] In this embodiment, the average value and standard deviation of all verification prediction values ​​in the training process are calculated, and the sum of the average value and the standard deviation is used as the prediction threshold.

[0027] Step 51, preset the maximum number of training times and the maximum value of verification loss.

[0028] In this embodiment, the maximum number of training times is set to 100 to prevent the model from being overtrained. A reasonable maximum validation loss value is set to prevent the model from having too low performance on the validation set. For example, 0.05 can be set.

[0029] Step 52, obtaining the number of training times.

[0030] Specifically, it refers to the current number of training sessions.

[0031] Step 53, calculating the prediction uncertainty after the current round of traversal and the prediction uncertainty after the previous round of traversal to obtain the uncertainty change.

[0032] An epoch refers to the process in which the model performs a complete forward propagation and backward propagation on the entire training dataset. A round traversal refers to a forward propagation and backward propagation for all samples in the training set, that is, performing one epoch of training. The updated model is used to perform forward propagation on each sample in the validation set to obtain the validation prediction value, and the prediction uncertainty after the current round traversal is obtained based on the validation prediction value. At the end of the previous epoch, the prediction uncertainty has been calculated and recorded.

[0033] ; is the forecast uncertainty of the rth round, is the prediction uncertainty of the r-1th round.

[0034] Step 54, calculate the learning rate according to the prediction uncertainty, the uncertainty change, the verification loss value, the number of training times, the maximum number of training times, and the maximum verification loss value.

[0035] ; ; In the formula, is the learning rate, is the initial learning rate, U is the prediction uncertainty, T is the prediction threshold, L is the validation loss value, is the maximum value of the verification loss, t is the number of training times, is the maximum number of training times, , , are weight coefficients respectively, max is the maximum value operation, is the lower limit of the learning rate.

[0036] This embodiment sets a maximum value operation and sets a learning rate lower limit for the learning rate. When the calculated learning rate is less than or equal to the learning rate lower limit, , when the calculated learning rate is greater than the lower limit of the learning rate, the learning rate is calculated .

[0037] By dynamically adjusting the learning rate by considering prediction uncertainty, uncertainty variation, validation loss, number of trainings, maximum number of trainings, and maximum validation loss, this method can significantly improve the performance of the model in predicting chloride ion concentration in concrete. Specifically, this method can not only flexibly adjust the learning rate according to the current uncertainty of the model and the performance on the validation set, but also dynamically balance the exploration and development of the model during the training process to avoid overfitting and underfitting. The introduction of prediction uncertainty enables the model to automatically reduce the learning rate when the uncertainty is high, thereby optimizing the parameters more carefully and improving the stability of the prediction. The consideration of the uncertainty variation helps the model identify fluctuations during the training process and adjust the learning strategy in time. The combination of validation loss and maximum validation loss ensures that the model's performance on unseen data is always in the optimal state, improving the generalization ability of the model. In addition, by limiting the number of trainings and the maximum number of trainings, the model can achieve the best performance within a limited training time and save computing resources. In summary, this adaptive learning rate adjustment method not only improves the prediction accuracy and stability of the model, but also enhances the adaptability of the model under different environments and conditions, providing more reliable technical support for the subsequent chloride ion concentration prediction.

[0038] In another embodiment, when the prediction uncertainty is less than the prediction threshold, in this embodiment, the learning rate is preferably also appropriately adjusted: ; In this formula, the minimum validation loss is taken into account Even if the uncertainty is low, the model may not have fully converged. Increasing the learning rate appropriately can speed up convergence, especially in the early stages of training. In the early stages of training, the uncertainty of the model is usually high, and a larger learning rate can be used to quickly approach the optimal solution. In the later stages of training, the uncertainty of the model is low, and a smaller learning rate can be used to fine-tune the model parameters and improve the accuracy of the model.

[0039] Step 6, calculate the correction factor according to the environmental parameters, obtain the chloride ion concentration value of the concrete structure according to the correction factor and the prediction, and obtain the final chloride ion concentration.

[0040] Step 61, obtaining parameter values ​​of concrete structure environmental parameters.

[0041] Step 62, calculating the mean of the parameter value and the standard deviation of the parameter value.

[0042] Step 63, calculating a correction factor according to the parameter value of the environmental parameter, the mean value of the parameter value and the standard deviation of the parameter value.

[0043] By using the parameter values, mean values ​​and standard deviations of environmental parameters to calculate the correction factors, the accuracy and reliability of the model in predicting chloride ion concentration in concrete structures can be significantly improved. This method can not only capture the changing trends of environmental parameters, but also quantify the impact of these changes on chloride ion concentration. Specifically, the mean and standard deviation of the parameter values ​​provide statistical characteristics of environmental parameters, which help to identify the distribution characteristics and fluctuations of the parameters. The introduction of the correction factor enables the model to dynamically adjust its prediction results under different environmental conditions, so as to better adapt to changes in actual application scenarios. For example, in an environment with large fluctuations in temperature and humidity, the correction factor can be adjusted according to the actual values ​​and statistical characteristics of these parameters to ensure that the prediction results are closer to the actual situation. In addition, this method can also reduce the risk of overfitting of the model, improve its generalization ability, and make the prediction performance of the model under different environmental conditions more stable and accurate. In summary, the correction factor calculation method based on the statistical characteristics of environmental parameters provides a more scientific and effective means for the prediction of chloride ion concentration.

[0044] ; In the formula, C is the correction factor, m is the adjustment coefficient, is the parameter value of the i-th environmental parameter, is the mean value of the parameter value of the i-th environmental parameter, is the standard deviation of the parameter value of the i-th environmental parameter.

[0045] ; In the formula, is the final chloride ion concentration, In order to predict the chloride ion concentration of concrete structure, is the sigmoid function, is the exponential decay coefficient, f is the current traversal round, F is the total traversal round, is the dynamic weight coefficient.

[0046] Calculating the final prediction value through the above formula has the following outstanding advantages: First, the formula cleverly integrates multiple important factors, including the initial prediction value, the sigmoid function, the exponential decay coefficient, the current iteration round, the total iteration round and the dynamic weight coefficient. This makes the final prediction result not only rely on the original prediction value, but also fully considers the dynamic changes in the model training process and the influence of environmental factors, thereby greatly improving the accuracy of the prediction.

[0047] Secondly, by introducing the sigmoid function, the changes in the predicted values ​​can be smoothed to a certain extent, avoiding the large overall prediction deviation caused by individual extreme values. At the same time, the addition of the exponential decay coefficient and the dynamic weight coefficient enables the model to converge quickly in the early stage of training, and slowly adjust when it is close to convergence, effectively avoiding the problem of premature stopping or overfitting.

[0048] Finally, the flexibility of this formula lies in that it can adapt to different application scenarios and data characteristics by adjusting the values ​​of each parameter, thereby achieving personalized prediction results. In general, such a design not only ensures the accuracy of the prediction results, but also takes into account the robustness and scope of application of the model. It is an efficient and practical prediction method.

[0049] It should be noted that the terms used in the present invention are only for describing specific embodiments, rather than limiting the scope of the present application. As shown in the present specification, unless the context clearly indicates an exception, the words "one", "a", "a kind of" and / or "the" do not specifically refer to the singular, but may also include the plural. The terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of more restrictions, the elements defined by the sentence "include one..." do not exclude the presence of other identical elements in the process, method or device including the elements.

[0050] It should also be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. Unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", etc. should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting chloride ion concentration in concrete based on deep learning, characterized in that: include: Step 1, collect concrete structure samples, and obtain chloride ion concentration, environmental parameters, curing conditions, and component information of the concrete structure samples; Step 2, pre-processing the chloride ion concentration, environmental parameters, curing conditions, and component information; Step 3, initialize the parameters of the deep learning model, the deep learning model traverses the pre-treated chloride ion concentration, environmental parameters, maintenance conditions, and component information, performs forward propagation to calculate the predicted value, calculates the loss value of the loss function, performs back propagation to update the parameters, and completes the iterative process; repeat the iterative process until the maximum number of iterations is reached or the loss value of the loss function reaches the minimum, and the iteration is stopped to obtain the trained deep learning model; after traversing the pre-treated chloride ion concentration, environmental parameters, maintenance conditions, and component information, the deep learning model is verified using the verification set to obtain the verification prediction value and the verification loss value; Step 4, obtain the environmental parameters, curing conditions, and component information of the concrete structure, and input them into the trained deep learning model to predict the chloride ion concentration value of the concrete structure, and obtain the prediction uncertainty based on the verified prediction value; Step 5, preset a prediction threshold, compare the prediction uncertainty with the prediction threshold, if the prediction uncertainty is greater than or equal to the prediction threshold, make a first adjustment to the learning rate of the deep learning model, retrain and continue to execute step 4 until the prediction uncertainty is less than the prediction threshold, then stop re-adjusting the learning rate and retraining; if the prediction uncertainty is less than the prediction threshold, execute step 6; Step 6, calculating the correction factor according to the environmental parameters, obtaining the chloride ion concentration value of the concrete structure according to the correction factor and the prediction, and obtaining the final chloride ion concentration.

2. A method for predicting chloride ion concentration of concrete based on deep learning according to claim 1, characterized in that: Chloride ion concentration was used as the predicted label for the deep learning model.

3. A method for predicting chloride ion concentration of concrete based on deep learning according to claim 2, characterized in that: Correction factors are calculated based on environmental parameters, including: Step 61, obtaining parameter values ​​of concrete structure environmental parameters; Step 62, calculating the mean of the parameter value and the standard deviation of the parameter value; Step 63, calculating a correction factor according to the parameter value of the environmental parameter, the mean value of the parameter value and the standard deviation of the parameter value.

4. A method for predicting chloride ion concentration of concrete based on deep learning according to claim 3, characterized in that: The standard deviation of the verified prediction value is taken as the prediction uncertainty.

5. A method for predicting chloride ion concentration of concrete based on deep learning according to claim 4, characterized in that: Adjustment of learning rate, including: Step 51, preset the maximum number of training times and the maximum value of verification loss; Step 52, obtaining the number of training times; Step 53, calculating the prediction uncertainty after the current round of traversal and the prediction uncertainty after the previous round of traversal to obtain the uncertainty change; Step 54, calculate the learning rate according to the prediction uncertainty, the uncertainty change, the verification loss value, the number of training times, the maximum number of training times, and the maximum verification loss value.

6. A method for predicting chloride ion concentration of concrete based on deep learning according to claim 5, characterized in that: The learning rate is: ； ； In the formula, is the learning rate, is the initial learning rate, U is the prediction uncertainty, T is the prediction threshold, L is the verification loss value, is the maximum value of the verification loss, t is the number of training times, is the maximum number of training times, , , are weight coefficients respectively, max is the maximum value operation, is the lower limit of the learning rate.

7. A method for predicting chloride ion concentration of concrete based on deep learning according to claim 3, characterized in that: The correction factor is: ； In the formula, C is the correction factor, m is the adjustment coefficient, is the parameter value of the i-th environmental parameter, is the mean value of the parameter value of the i-th environmental parameter, is the standard deviation of the parameter value of the i-th environmental parameter.

8. The method for predicting chloride ion concentration of concrete based on deep learning according to claim 1, characterized in that: The final chloride ion concentration is: ； In the formula, is the final chloride ion concentration, In order to predict the chloride ion concentration of concrete structure, is the sigmoid function, is the exponential decay coefficient, f is the current traversal round, F is the total traversal round, is the dynamic weight coefficient.

Images