Cement strength prediction method based on adaptive fusion of data driving and mechanism model

Abstract

The application provides a cement strength prediction method based on adaptive fusion of data driving and mechanism model, comprising: acquiring data, the data comprising process variables, the process variables comprising linearly related variables and nonlinearly related variables; inputting the linearly related variables into a trained mechanism model to obtain a first prediction value; a trained data driving model obtaining a second prediction value according to the nonlinearly related variables; and adding the first prediction value and the second prediction value to obtain a final prediction value. By fusing the mechanism model and the data driving model, the mechanism model guarantees the reliability of the model, and the data driving model uses a convolutional neural network to extract highly complex nonlinear relationships, thereby improving the prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of deep learning algorithm technology, and more particularly to a method for predicting cement strength based on the adaptive fusion of data-driven and mechanism model. Background Technology

[0002] Cement is one of the basic raw materials for construction projects and occupies an important position in national economic construction. How to effectively control cement quality and improve cement production efficiency is the focus of the entire cement production industry.

[0003] Cement strength is a key indicator for measuring cement quality. Currently, industry professionals generally consider 28 days as the age at which cement strength basically stabilizes, and use the 28-day strength as the representative strength of general-purpose cement. Predicting the 28-day strength of cement clinker is a complex problem involving multiple variables, nonlinearity, and a long lag time. Various mineral components in clinker undergo a series of complex chemical reactions during cement hydration and hardening. Therefore, the composition and relative content of various minerals in clinker have a very complex impact on clinker strength. Traditional simple linear functions cannot accurately describe the complex process of cement production. The 28-day strength of cement clinker needs to be obtained manually on the 28th day, which is time-consuming and has a long lag time, making it impossible to pre-adjust the amount of admixtures. Therefore, it is necessary to identify the main factors affecting clinker strength, explore and further improve methods for predicting clinker 28-day strength that are suitable for actual production engineering, and establish a reasonable and accurate prediction model. This will enable the early detection of quality problems in cement clinker, timely adjustment of raw meal ratio and the amount of admixtures in cement mix proportion, and reasonable determination of cement grinding fineness and grinding time. This will ensure that the cement production process is under scientific control, thereby achieving higher profits while ensuring the grade of the cement leaving the factory.

[0004] Traditional mechanistic modeling methods predict cement strength using linear functional relationships. This simplification of highly nonlinear relationships into linear ones leads to insufficient prediction accuracy. Currently, data-driven methods, such as neural network algorithms like Back Propagation (BP) and Deep Neural Networks (DNN), are more commonly used to establish nonlinear relationships. However, purely data-driven methods are black-box modeling processes; the quality of the model depends excessively on the data, thus compromising its reliability. Furthermore, well-trained models are not highly adaptable to different data sources.

[0005] To address the above problems, this invention proposes a cement strength prediction method based on the adaptive fusion of data-driven and mechanistic models. Summary of the Invention

[0006] To address the aforementioned shortcomings, this invention provides a cement strength prediction method based on the adaptive fusion of data-driven and mechanistic models. The main approach is to integrate a mechanistic model with a data-driven model. The mechanistic model ensures the reliability of the overall model, while the data-driven model uses convolutional neural networks (CNNs) to extract highly complex nonlinear relationships, thereby improving prediction accuracy. Furthermore, an adaptive fusion update mechanism is proposed. When the test error exceeds a set threshold, the tested data is returned to both the mechanistic and data-driven models in a batch learning manner to update the model parameters, thus improving accuracy.

[0007] The technical means employed in this invention are as follows:

[0008] This invention provides a cement strength prediction method based on the adaptive fusion of data-driven and mechanistic models, comprising:

[0009] Acquire data, which includes process variables, including linearly correlated variables and non-linearly correlated variables;

[0010] The linearly correlated variables are input into the trained mechanistic model to obtain the first predicted value;

[0011] The trained data-driven model obtains a second predicted value based on the nonlinear correlation variables;

[0012] The first predicted value and the second predicted value are added together to obtain the final predicted value.

[0013] Further, training the data-driven model includes:

[0014] Obtain a training set, which includes multiple training samples, and the training samples include the process variable and the quality variable;

[0015] The first predicted value is obtained by inputting the linearly correlated variables into the trained mechanistic model.

[0016] The difference between the quality variable and the first predicted value is used as the label of the data-driven model. The nonlinear correlation variable is input into the data-driven model for training, and the trained data-driven model is obtained.

[0017] Furthermore, before training the data-driven model by inputting the nonlinear correlation variable into the data-driven model, the method further includes using the difference between the quality variable and the first predicted value as the label of the data-driven model.

[0018] The difference between the quality variable and the first predicted value, and the nonlinear correlation variable are normalized and calculated in the following manner:

[0019]

[0020] Where, x i Let x be the i-th nonlinearly correlated variable, and let x be x i The corresponding vector, y i During the training of the data-driven model, y is the difference between the quality variable and the first predicted value. i For the corresponding vectors, min() is the function to find the minimum value, and max() is the function to find the maximum value. For x i The normalized value, For y i The normalized value.

[0021] Furthermore, it also includes:

[0022] The error limit is obtained based on the average cement strength from day 1 to day n and the preset accuracy.

[0023] When the deviation between the final predicted value and the actual value is less than or equal to the error limit, the actual value and the data are saved as a data group.

[0024] When the deviation between the final predicted value and the true value is greater than the error limit, the saved data set is updated to the training set, and the updated training set is used to train the mechanism model and the data-driven model.

[0025] Further, the trained data-driven model obtains the second predicted value based on the nonlinear correlation variable, including:

[0026] The linearly correlated variables are normalized and then input into the trained data-driven model to obtain the output value.

[0027] The output value is then denormalized to obtain the second predicted value.

[0028] Furthermore, the output value is inversely normalized and calculated as follows:

[0029]

[0030] in, The output value is... Let be the second predicted value, min() be the minimum value function, max() be the maximum value function, and y be the maximum value function. i The corresponding vector, y i The difference between the quality variable and the first predicted value during the training of the data-driven model.

[0031] Furthermore, the data-driven model includes a first Block module, a second Block module, a third Block module, a first fully connected layer, and a second fully connected layer connected in sequence.

[0032] The first Block module, the second Block module, and the third Block module have the same structure, including sequentially connected convolutional layers, batch normalization layers, and non-linear activation function layers;

[0033] The first fully connected layer includes interconnected linear layers and the nonlinear activation function layer;

[0034] The second fully connected layer includes the connected Linear layer and Sigmoid activation function layer.

[0035] Furthermore, the second fully connected layer is calculated as follows:

[0036]

[0037] Where, x Block The output value of the third Block module is represented by Linear(), which is processed by the Linear layer, and Sigmoid() is processed by the Sigmoid activation function layer. The output value is denoted as .

[0038] Furthermore, the error limit is obtained based on the average cement strength from day 1 to day n and the preset accuracy, and is calculated as follows:

[0039]

[0040] in, y represents the average cement strength from day 1 to day n. j Let be the cement strength value on day j, s be the preset accuracy, and R be the error limit.

[0041] Furthermore, when training the data-driven model using the training set, the loss function is calculated as follows:

[0042]

[0043] Where y′ is the normalized value of the true value. λ is the output value, λ is the regularization coefficient, and w is the parameter of the data-driven model.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] 1. The cement strength prediction method based on adaptive fusion of data-driven and mechanistic models provided by this invention proposes a method that integrates mechanistic models and data-driven models. The mechanistic model addresses qualitative issues, while the data-driven model addresses quantitative issues, and the two complement each other. The mechanistic model improves the reliability of the model; the data-driven model, through learning the nonlinear characteristics of complex processes using CNNs, establishes a data-driven model and improves prediction accuracy.

[0046] 2. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model provided by this invention proposes an adaptive fusion update mechanism and defines the error limit between the predicted value and the actual value. When the test error is greater than the set threshold, the tested data will be returned to the mechanism model and the data-driven model to update the model parameters, so as to achieve the purpose of adaptively updating the model and improving the accuracy.

[0047] 3. The cement strength prediction method based on data-driven and mechanism model adaptive fusion provided by this invention takes into account the unique production conditions and processing differences of different cement manufacturers, and can continuously update the model to adaptively correct prediction deviations in a batch learning mode. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of a cement strength prediction method based on adaptive fusion of data-driven and mechanism model provided by the present invention.

[0050] Figure 2 This is another flowchart illustrating the cement strength prediction method based on the adaptive fusion of data-driven and mechanistic models provided by the present invention.

[0051] Figure 3 This is another flowchart illustrating the cement strength prediction method based on the adaptive fusion of data-driven and mechanism model provided by the present invention.

[0052] Figure 4 This is a schematic diagram of a data-driven model. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0054] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0055] Combination Figure 1 , Figure 1 This is a flowchart illustrating a cement strength prediction method based on adaptive fusion of data-driven and mechanistic models provided by the present invention, to illustrate a specific embodiment of the cement strength prediction method based on adaptive fusion of data-driven and mechanistic models provided by the present invention, including:

[0056] S1: Obtain data, which includes process variables, including linearly correlated variables and non-linearly correlated variables;

[0057] S2: Input the linearly correlated variables into the trained mechanistic model to obtain the first predicted value;

[0058] S3: The trained data-driven model obtains a second predicted value based on non-linearly correlated variables;

[0059] S4: The first predicted value and the second predicted value are added together to obtain the final predicted value.

[0060] It is understandable that one set of data includes 12 linearly correlated variables and 11 non-linearly correlated variables, but this is not the only possibility. Since the relationship between cement strength and some cement components is a linear function, a mechanistic model is used, and the calculation is performed as follows:

[0061]

[0062] Where x is a vector of linearly related variables, x = [x1, x2, ..., xn]. 12 ] T , x1 to x 12 They are SiO2, MgO, R2O, f-CaO, KH, KH-, n, p, C3S, C2S, C3A, and C4AF, respectively. Let K = [k1, k2, ..., k] be the first predicted value. 12 ], b are the parameters of the mechanism model. K and b are obtained by least squares during the training of the mechanism model. During the prediction process, linearly related variables are input, and the mechanism model obtains the first predicted value by least squares.

[0063] The 11 nonlinearly correlated variables include Al2O3, Fe2O3, CaO, SO3, loss on ignition, standard consistency, initial setting, final setting, specific surface area, grinding time, and fineness. However, this list is not exhaustive. These 11 nonlinearly correlated variables are used as input values ​​to the data-driven model to obtain the second predicted value. The first and second predicted values ​​are then added together to obtain the final predicted value, calculated as follows:

[0064]

[0065] in, The first predicted value, The second predicted value, y p This is the final predicted value.

[0066] This invention improves the reliability of the model through a mechanistic model; and improves prediction accuracy by establishing a data-driven model through learning the nonlinear characteristics of complex processes using CNN.

[0067] In some alternative embodiments, refer to Figure 2 , Figure 2 This is another flowchart illustrating the cement strength prediction method based on adaptive fusion of data-driven and mechanistic models provided by the present invention. Training the data-driven model includes:

[0068] Obtain the training set, which includes multiple training samples, including process variables and quality variables;

[0069] The first predicted value is obtained by inputting the linearly correlated variables into the trained mechanistic model.

[0070] The difference between the quality variable and the first predicted value is used as the label of the data-driven model. Non-linear correlation variables are input into the data-driven model for training, resulting in the trained data-driven model.

[0071] Understandably, the mass variable is cement strength.

[0072] Specifically, the difference between the quality variable and the first predicted value is used as the label for the data-driven model. Before training the data-driven model by inputting non-linear correlation variables, the following steps are also included:

[0073] The difference between the quality variable and the first predicted value, as well as the nonlinear correlation variable, are normalized and calculated as follows:

[0074]

[0075] Where, x i Let x be the i-th nonlinearly correlated variable, and let x be x i The corresponding vector, y i During the training of the data-driven model, y represents the difference between the quality variable and the first predicted value. i For the corresponding vectors, min() is the function to find the minimum value, and max() is the function to find the maximum value. For x i The normalized value, For y i The normalized value.

[0076] Understandably, in order to speed up training and avoid the model being disturbed by features with large or small distribution ranges, the features are scaled to the [0, 1] interval through normalization.

[0077] In some optional embodiments, when training the data-driven model using a training set, the loss function is calculated as follows:

[0078]

[0079] Where y′ is the normalized value of the true value. λ is the output value, λ is the regularization coefficient, and w is the parameter of the data-driven model.

[0080] It is understood that the data-driven model is a convolutional neural network model. During training, the Adaptive Moment Estimation (Adam) optimization algorithm is used, with a learning rate of 0.0001, 40 training iterations, a batch size of 20, and a mean squared error loss function. To avoid overfitting, an L2 regularization term is also added. Industry professionals typically consider 28 days as the age at which cement strength basically stabilizes. In this embodiment, y′ is obtained by normalizing the true value of the cement on day 28. λ helps smooth the convolutional neural network to avoid overfitting; in this embodiment, λ is set to 0.0005. Of course, the specific value of λ can be adjusted according to actual needs, and this embodiment does not impose specific limitations on it. The trained model is saved as a file, which can be deployed on-site for testing.

[0081] In some alternative embodiments, refer to Figure 3 , Figure 3 This is another flowchart illustrating the cement strength prediction method based on adaptive fusion of data-driven and mechanistic models provided by the present invention. The cement strength prediction method based on adaptive fusion of data-driven and mechanistic models provided by the present invention further includes:

[0082] The error limit is obtained based on the average cement strength from day 1 to day n and the preset accuracy.

[0083] When the deviation between the final predicted value and the actual value is less than or equal to the error limit, the actual value and the data are grouped together and saved.

[0084] When the deviation between the final predicted value and the true value exceeds the error limit, the saved data set is updated to the training set, and the updated training set is used to train the mechanistic model and the data-driven model.

[0085] Specifically, based on the average cement strength from day 1 to day n and the preset accuracy, the error limit is obtained and calculated as follows:

[0086]

[0087] in, y represents the average cement strength from day 1 to day n. j Let be the cement strength value on day j, s be the preset accuracy, and R be the error limit.

[0088] It is understood that this embodiment provides an adaptive fusion update mechanism. The specific value of the preset accuracy can be set according to actual needs. When the accuracy of the final predicted value is lower than the preset accuracy compared with the historical average, it is considered that the model needs to be updated. During testing, when the deviation between the final predicted value and the true value is greater than the calculated error limit, the model will add the newly tested data to the training set in a batch learning manner to recalculate the parameters of the linear model and the nonlinear model, that is, retrain the mechanistic model and the data-driven model, and re-fuse them to achieve the purpose of adaptively updating the fusion model.

[0089] In some alternative embodiments, reference continues to be made to... Figure 1 The trained data-driven model obtains a second predicted value based on non-linearly correlated variables, including:

[0090] After normalizing the linearly correlated variables, the data is input into the trained data to drive the model and obtain the output value.

[0091] The output value is denormalized to obtain the second predicted value.

[0092] Furthermore, the output value is inversely normalized, calculated as follows:

[0093]

[0094] in, For the output value, The second predicted value is given by , min() is the function to find the minimum value, max() is the function to find the maximum value, and y is the value of y. i The corresponding vector, y i This refers to the difference between the quality variable and the first predicted value during the training of the data-driven model.

[0095] In some alternative embodiments, refer to Figure 4 , Figure 4 This is a schematic diagram of a data-driven model, which includes a first block module, a second block module, a third block module, a first fully connected layer, and a second fully connected layer connected in sequence.

[0096] The first block, second block, and third block have the same structure, including sequentially connected convolutional layers, batch normalization layers, and non-linear activation function layers;

[0097] Specifically, the function of the convolutional layer is to extract features, and the formula for calculating the input and output feature sizes is as follows:

[0098]

[0099] Where H and W represent the height and width of the feature, respectively, in represents the input, out represents the output, F represents the size of the convolution kernel on the corresponding width or height, P represents the size of the feature boundary to be padded, and S represents the stride of the convolution kernel.

[0100] The purpose of the batch normalization layer is to accelerate convergence and avoid the gradient vanishing problem. The calculation formula is as follows:

[0101]

[0102] Where X' is the input data that needs to be normalized, E[X'] and Var[X'] are the mean and variance of the batch data, ò is the variable added to prevent the denominator from being zero, and γ and β are the learning parameters of the data-driven model, which are affine operations on the input values, i.e. linear transformations.

[0103] Non-linear activation function layers can map input features to a non-linear feature space, thereby enabling more complex representations. The formula for the ReLU activation function is as follows:

[0104] ReLu(L) = max(0,L);

[0105] Here, max() represents the function to find the maximum value.

[0106] The first fully connected layer consists of interconnected linear layers and non-linear activation function layers;

[0107] The second fully connected layer consists of interconnected linear layers and sigmoid activation function layers.

[0108] It should be noted that the second fully connected layer is calculated as follows:

[0109]

[0110] Where, x Block This is the output value of the third block module. Linear() is processed by the Linear layer, and Sigmoid() is processed by the Sigmoid activation function layer. This is the output value.

[0111] It is understood that in this embodiment, the convolution kernel size is (1×3), the maximum number of channels per convolution layer is (64, 128, 64), and the predicted value is output through two fully connected layers after convolution. The Sigmoid activation function maps the predicted value to the interval [0, 1] to obtain the output value.

[0112] In some optional embodiments, the root mean square error (RMSE) is used as the metric. Refer to Tables 1 and 2 below:

[0113] Table 1. Comparison of Root Mean Square Error Results between Mechanism Model and Fusion Model

[0114]

[0115] Table 2. Comparison of root mean square error results before and after adding adaptive fusion update mechanism to the fusion model.

[0116]

[0117] It is understandable that the fusion model refers to a model that combines a mechanism model and a neural network model. The experimental results in Table 1 show that the mechanism and data-driven fusion model can effectively improve the accuracy of cement prediction. At the same time, the proposed mechanism and CNN fusion model achieved the best results. The results in Table 2 show that adding an adaptive fusion update mechanism module to the fusion model can further improve the prediction accuracy.

[0118] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0119] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0120] The technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cement strength prediction method based on adaptive fusion of data-driven and mechanism model, characterized in that, The method comprises: obtaining data, the data comprising process variables, the process variables comprising linearly correlated variables and nonlinearly correlated variables; a set of data comprising 12 linearly correlated variables and 11 nonlinearly correlated variables, the linearly correlated variables being input into a trained mechanism model to obtain first predicted values; the mechanism model being used to calculate in the following manner: ； wherein is a vector of linearly dependent variables, , to are , MgO, , f-CaO, KH, KH-, n, p, , , , , is a first prediction, , is a parameter of the mechanistic model, and are obtained by a least squares method during training of the mechanistic model. 11 nonlinearly dependent variables include , , CaO, , loss on ignition, standard consistency, initial setting, final setting, specific surface area, fineness, and small grinding time. a trained data-driven model obtaining second predicted values according to the nonlinearly correlated variables; the first predicted values and the second predicted values being added to obtain final predicted values.

2. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model according to claim 1, characterized in that, training the data-driven model, comprising: obtaining a training set, the training set comprising a plurality of training samples, the training samples comprising the process variables and quality variables; inputting the linearly correlated variables into the trained mechanism model to obtain the first predicted values; the difference between the quality variables and the first predicted values being used as labels of the data-driven model, the nonlinearly correlated variables being input into the data-driven model for training to obtain the trained data-driven model.

3. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model according to claim 2, characterized in that, Before the difference between the quality variables and the first predicted values is used as the labels of the data-driven model and the nonlinearly correlated variables are input into the data-driven model for training, further comprising: normalizing the difference between the quality variables and the first predicted values and the nonlinearly correlated variables in the following manner: ； wherein, is the i-th non-linearly dependent variable, is is the corresponding vector, is the difference between the quality variable and the first predicted value during the data-driven model training process, is is the corresponding vector, is the min function, is the max function, is is the normalized value, is is the normalized value.

4. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model according to claim 2, characterized in that, Further comprising: obtaining an error limit according to the average cement strength of the cement from the first day to the nth day and a preset accuracy; when the deviation of the final predicted values from true values is less than or equal to the error limit, saving the true values and the data to form a data set; when the deviation of the final predicted values from the true values is greater than the error limit, updating the saved data set to the training set, and training the mechanism model and the data-driven model using the updated training set.

5. The method of claim 1, wherein, The trained data-driven model obtaining the second predicted values according to the nonlinearly correlated variables, comprising: inputting the normalized linearly correlated variables into the trained data-driven model to obtain output values; de-normalizing the output values to obtain the second predicted values.

6. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model according to claim 5, characterized in that, De-normalizing the output values in the following manner: ； wherein, is the output value, is the second prediction value, is a min function, is a max function, is is a corresponding vector, is a difference between the quality variable and the first prediction value during the data driven model training process.

7. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model according to claim 5, characterized in that, The data-driven model comprises a first Block module, a second Block module, a third Block module, a first fully connected layer and a second fully connected layer connected in sequence; The first Block module, the second Block module and the third Block module have the same structure, comprising a convolution layer, a batch normalization layer and a nonlinear activation function layer connected in sequence; The first fully connected layer comprises a Linear layer and the nonlinear activation function layer connected; The second fully connected layer comprises the Linear layer and a Sigmoid activation function layer connected.

8. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model according to claim 7, characterized in that, The second fully connected layer is calculated in the following manner: ； wherein, is the output value of the third Block module, is processed by the Linear layer, is processed by the Sigmoid activation function layer, is the output value.

9. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model of claim 4, wherein, The error limit is obtained according to the average cement strength of the cement from the first day to the nth day and the preset accuracy in the following manner: ； wherein, is the average of the cement strength of the cement from day 1 to day n, is the cement strength value of day j, is the pre-set accuracy, is the error limit.

10. The cement strength prediction method based on adaptive fusion of data-driven and mechanism model of claim 2, wherein, When training the data-driven model using the training set, a loss function is computed in the following manner: ； wherein, is a true value normalized value, is an output value, is a regularization coefficient, is a parameter of the data-driven model.

Images