Rock failure time prediction method, device and equipment based on machine learning

Abstract

The embodiment of the invention provides a rock failure time prediction method, device and equipment based on machine learning. The method comprises the following steps: constructing a convolutional neural network, a bidirectional long-short-term memory neural network and a physical constraint neural network; obtaining rock mechanical property data for preprocessing, and training the convolutional neural network, the bidirectional long-short-term memory neural network and the physical constraint neural network through a first training set to obtain a first model; inputting the first test set into a first model, outputting peak stress of an optimal solution result, obtaining corresponding strain, and generating a second training set and a second test set; training the first model through a second training set to obtain a second model; and inputting the second test set into the second model, and outputting a second prediction result. The BiLSTM is added on the basis of the CNN, and the PINN is combined, so that the mode can be learned from data, the physical law can be followed, and the accuracy and generalization ability of the model are improved.

Description

Technical Field

[0001] The present invention generally relates to the field of rock engineering, and more specifically, to a method, apparatus and device for predicting rock failure time based on machine learning. Background Art

[0002] Rock failure is crucial to the smooth operation of rock engineering, and is the core scientific issue of natural and engineering disasters, and is particularly critical for early warning of geological disasters in rock engineering and other fields. However, due to the complexity of the on-site environment, it is extremely difficult to directly obtain its various laws. Therefore, how to accurately and quickly predict the failure time of rocks and provide effective reference for the timely reinforcement and long-term stability of rock engineering has important practical and theoretical value.

[0003] For the questions of "how rocks are destroyed" and "how to predict destruction", common detection methods mainly include infrared detection systems, acoustic emission instruments and digital imaging technology. When rocks are destroyed, heat is generated inside. When the infrared radiation temperature effect is the strongest, it means that the destruction boundary has been reached. However, the infrared destruction precursor is later than the acoustic emission destruction precursor prediction in warning time and does not have priority; although the damage constitutive model of acoustic emission has a better prediction, it cannot match the prediction of the whole process of rock destruction, so more influencing factors still need to be considered and in-depth research needs to be continued; digital imaging technology can only measure the surface of rocks, and requires large-scale image processing and data analysis, involving complex algorithms and calculations; it also has certain requirements for imaging conditions, such as lighting conditions and field of view. If the conditions are poor, it will affect the results.

[0004] The application of machine learning methods in the field of rock mechanics has received increasing attention. Machine learning models are widely used to solve complex engineering problems, and they can capture nonlinear correlations between inputs and outputs. Compared with traditional machine learning methods, deep learning has a stronger ability to model nonlinear systems or complex mappings. Convolutional Neural Networks (CNN) have better data feature learning capabilities and improve the ability to recognize complex patterns and data, but they cannot detect long-term dependency problems in time series data. For example, when predicting stress, as the strain continues to increase, the predicted stress value may deviate significantly from the actual value, resulting in low prediction accuracy. Summary of the invention

[0005] According to an embodiment of the present invention, a rock failure time prediction scheme based on machine learning is provided. This scheme adds BiLSTM and PINN on the basis of CNN, which can well learn time series data and solve the long dependency problem that CNN cannot handle.

[0006] In a first aspect of the present invention, a method for predicting rock failure time based on machine learning is provided. The method comprises:

[0007] Constructing a convolutional neural network, a bidirectional long short-term memory neural network and a physical constraint neural network, and defining a loss function of the physical constraint neural network;

[0008] Use Bayesian optimization method to tune hyperparameters and obtain the optimal hyperparameter combination;

[0009] Acquire rock mechanical property data, preprocess the rock mechanical property data, and divide the preprocessed data into a first training set and a first test set, wherein the first training set and the first test set include mass, density, elastic modulus, strain, and stress, The strains of the first test set were increased by 0.008(‰), 0.01(‰), and 0.012(‰) respectively. add ;

[0010] The convolutional neural network, the bidirectional long short-term memory neural network and the physical constraint neural network are trained by using a first training set, and the physical constraint neural network is trained based on a loss function of the physical constraint neural network to obtain a first model;

[0011] The first test set is input into a first model, and a first prediction result is output, where the first prediction result includes three kind Peak stress of the optimal solution , and compare this result ;

[0012] Based on the peak stress of the optimal solution result, the corresponding strain is obtained, and a second training set and a second test set are generated. The second training set and the second test set include mass, density, elastic modulus, strain rate, peak stress of the optimal solution result and its corresponding strain, and time , the second test set strain and stress are divided into three cases, corresponding to the first prediction Peak stress and corresponding strain of the three optimal solutions in the results ;

[0013] Training the first model using a second training set to obtain a second model;

[0014] The second test set is input into the second model, and a second prediction result is output. The second prediction result is Three types The situations are Predicted rock failure time , and compare the three results to find the best result .

[0015] Furthermore, the convolutional neural network comprises:

[0016] A first convolution module, comprising a first convolution layer, a first activation layer and a first pooling layer in sequence; the first convolution layer obtains the input of the convolutional neural network, and the output of the first pooling layer is used as the output of the first convolution module;

[0017] A second convolution module, comprising a second convolution layer, a second activation layer, and a second pooling layer in sequence; the second convolution layer is used to take the output of the first convolution module as input, and the output of the second pooling layer is used as the output of the second convolution module;

[0018] A third convolution module, comprising a third convolution layer, a third activation layer and a third pooling layer in sequence; the third convolution layer is used to take the output of the second convolution module as input, and the output of the third pooling layer is used as the output of the third convolution module;

[0019] A fourth convolution module, comprising a fourth convolution layer, a fourth activation layer and a fourth pooling layer in sequence; the fourth convolution layer is used to take the output of the third convolution module as input, and the output of the fourth pooling layer is used as the output of the fourth convolution module;

[0020] The fifth convolution module includes a fifth convolution layer, a fifth activation layer and a fifth pooling layer in sequence; the fifth convolution layer is used to take the output of the fourth convolution module as input, and the output of the fifth pooling layer is used as the output of the fifth convolution module.

[0021] Furthermore, the method further comprises:

[0022] An Attention layer and a dropout layer are set between the third convolution module and the fourth convolution module; the Attention layer takes the output of the third convolution module as input, the dropout layer takes the output of the Attention layer as input, and the output of the dropout layer is used as the input of the fourth convolution module.

[0023] Furthermore, the physical constraint neural network includes a first fully connected layer, a second fully connected layer, a third fully connected layer and a fourth fully connected layer in sequence; wherein the second fully connected layer and the third fully connected layer contain ReLU activation functions.

[0024] Furthermore, the loss function of the physical constraint neural network includes data fitting loss and physical drive constraint loss equation, namely:

[0025] L=L data +ɑL phy

[0026] Where L is the loss function of the physical constraint neural network; L data is the data fitting loss; L phy is the physics driven constraint loss.

[0027] Furthermore, the data fitting loss is:

[0028]

[0029] Where N is the total number of samples; σpred is the predicted stress; x i is the i-th input sample; σ true is the test stress; T pred is the prediction time; j is the jth input sample; T true For test time.

[0030] Furthermore, the physical drive constraint loss is:

[0031]

[0032] Among them, σ pred is the predicted stress; k is the kth input sample; E is the elastic modulus; ε is the strain; σ3 is the confining pressure; ɑ is the equilibrium coefficient.

[0033] In a second aspect of the present invention, a rock failure time prediction device based on machine learning is provided. The device comprises:

[0034] A network construction module, used to construct a convolutional neural network, a bidirectional long short-term memory neural network and a physical constraint neural network, and define a loss function of the physical constraint neural network;

[0035] Hyperparameter tuning module, which is used to tune hyperparameters using Bayesian optimization method to obtain the optimal hyperparameter combination;

[0036] The first data processing module is used to obtain rock mechanical property data, pre-process the rock mechanical property data, and divide the pre-processed data into a first training set and a first test set, wherein the first test set includes mass, density, elastic modulus, strain, and stress. The strains of the first test set were 0.008(‰), 0.01(‰), and 0.012(‰) respectively. Increase the value ;

[0037] A first training module, used to train the convolutional neural network, the bidirectional long short-term memory neural network and the physical constraint neural network through a first training set, and to train the physical constraint neural network based on a loss function of the physical constraint neural network to obtain a first model;

[0038] A first prediction module is used to input the first test set into a first model and output a first prediction result, wherein the first prediction result includes Three types Peak stress of the optimal solution , and compare this result ;

[0039] The second data processing module is used to obtain the corresponding strain based on the peak stress of the optimal solution result, and generate a second training set and a second test set, wherein the second training set and the second test set include the preprocessed mass, density, elastic modulus, strain rate, the peak stress of the optimal solution result and its corresponding strain, and time , the second test set strain and strain The stress is divided into three cases, corresponding to the peak stress of the three optimal solutions in the first prediction result and their corresponding stress Change ;

[0040] A second training module, used to train the first model using a second training set to obtain a second model;

[0041] The second prediction module is used to input the second test set into the second model and output a second prediction result, wherein the second prediction result is Three situations Predicted rock failure time , and compare the three results to find the best result .

[0042] In a third aspect of the present invention, an electronic device is provided. The electronic device has at least one processor; and a memory connected to the at least one processor in communication; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the method of the first aspect of the present invention.

[0043] In a fourth aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method of the first aspect of the present invention.

[0044] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0046] Figure 1 A flow chart of a rock failure time prediction method based on machine learning according to an embodiment of the present invention is shown;

[0047] Figure 2 A block diagram of a rock failure time prediction device based on machine learning according to an embodiment of the present invention is shown;

[0048] Figure 3 A block diagram of an exemplary electronic device capable of implementing embodiments of the present invention is shown;

[0049] Among them, 300 is an electronic device, 301 is a computing unit, 302 is a ROM, 303 is a RAM, 304 is a bus, 305 is an I / O interface, 306 is an input unit, 307 is an output unit, 308 is a storage unit, and 309 is a communication unit. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 are within the scope of protection of the present invention.

[0051] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0052] Figure 1 A flow chart of a rock failure time prediction method based on machine learning according to an embodiment of the present invention is shown.

[0053] The method includes:

[0054] S101, constructing a convolutional neural network, a bidirectional long short-term memory neural network and a physical constraint neural network, and defining a loss function of the physical constraint neural network.

[0055] In this embodiment, the convolutional neural network CNN is a combination of "convolution + pooling", specifically including 5 convolution modules, namely a first convolution module, a second convolution module, a third convolution module, a fourth convolution module and a fifth convolution module.

[0056] The first convolution module includes a first convolution layer, a first activation layer and a first pooling layer in sequence; the first convolution layer obtains the input of the convolutional neural network, and the output of the first pooling layer is used as the output of the first convolution module.

[0057] Specifically, the first convolution module is represented as: nn.Conv1d(in_channels=input_channels,out_channels=16,kernel_size=3,padding=1). That is, the data with input_channels as input channel is convolved into 16 channels, the convolution kernel size is 3, and the padding is 1 to keep the data size unchanged. Then the nn.ReLU() activation function is used to introduce nonlinearity, and then the nn.MaxPool1d(kernel_size=2) pooling layer reduces the length of the data by half.

[0058] The activation function nn.ReLU() is:

[0059]

[0060] The second convolution module includes a second convolution layer, a second activation layer and a second pooling layer in sequence; the second convolution layer is used to take the output of the first convolution module as input, and the output of the second pooling layer is used as the output of the second convolution module.

[0061] Specifically, the second convolution module is represented as: nn.Conv1d(in_channels=16,out_channels=32,kernel_size=3,padding=1). That is, the data of 16 channels is convolved into 32 channels, the convolution kernel size is 3, the padding is 1, and then activated by the nn.ReLU() activation function and the nn.MaxPool1d(kernel_size=2) pooling operation.

[0062] The third convolution module includes a third convolution layer, a third activation layer and a third pooling layer in sequence; the third convolution layer is used to take the output of the second convolution module as input, and the output of the third pooling layer is used as the output of the third convolution module.

[0063] Specifically, the third convolution module is represented as: nn.Conv1d(in_channels=32, out_channels=64, kernel_size=3, padding=1). That is, the data of 32 channels is convolved into 64 channels, and then activated by the nn.ReLU() activation function and the nn.MaxPool1d(kernel_size=2) pooling operation.

[0064] The fourth convolution module includes a fourth convolution layer, a fourth activation layer and a fourth pooling layer in sequence; the fourth convolution layer is used to take the output of the third convolution module as input, and the output of the fourth pooling layer is used as the output of the fourth convolution module.

[0065] Specifically, the fourth convolution module is represented as: nn.Conv1d(in_channels=64, out_channels=128, kernel_size=3, padding=1). That is, the data of 64 channels are convolved into 128 channels, and then activated by the nn.ReLU() activation function and the nn.MaxPool1d(kernel_size=2) pooling operation.

[0066] The fifth convolution module includes a fifth convolution layer, a fifth activation layer and a fifth pooling layer in sequence; the fifth convolution layer is used to take the output of the fourth convolution module as input, and the output of the fifth pooling layer is used as the output of the fifth convolution module.

[0067] Specifically, the fifth convolution module is represented as: nn.Conv1d(in_channels=128, out_channels=256, kernel_size=3, padding=1): the data of 128 channels are convolved into 256 channels, activated by nn.ReLU() and pooled by nn.MaxPool1d(kernel_size=2).

[0068] The CNN model has good data feature learning capabilities and improves the ability to recognize complex patterns and data, but it cannot discover long-term dependencies in time series data. After the above improvements to the CNN model, a series of convolutional modules can gradually extract local features in the input data, and as the number of layers increases, more complex and abstract features can be captured.

[0069] As an embodiment of the present invention, the attention mechanism Attention can be applied to the third convolution layer, that is, between the third convolution module and the fourth convolution module, and a dropout layer can be set after the Attention layer; the output of the dropout layer is used as the input of the fourth convolution module. The dropout rate of the dropout layer is 0.2. Setting the dropout layer can prevent the model from overfitting the training data, thereby improving the generalization ability of the model.

[0070] In this embodiment, the bidirectional long short-term memory neural network (BiLSTM): the input is the CNN output, the forward LSTM layer: 128 units, the dropout rate is 0.2; the backward LSTM layer: 128 units, the dropout rate is 0.2. The outputs of the forward and backward LSTM layers are concatenated, the dense layer: 128 units, the activation function "tanh". The bidirectional design enables the model to consider both forward and reverse time series information. In the forward method, after lstm processing, the output of the last time step is taken as last_output = lstm_out[:,-1,:], because for regression tasks, the impact of the state at the last moment on the prediction result is usually concerned.

[0071] BiLSTM can learn time series data well and solve the long dependency problem that CNN cannot handle. Therefore, combining CNN and BiLSTM to form a CNN-BiLSTM model can give full play to the advantages of each neural network model, thereby improving the prediction accuracy of the model.

[0072] In this embodiment, the physical constraint neural network (PINN): the input is the BiLSTM output, which consists of 4 fully connected layers. The first fully connected layer self.pinn_fc1 = nn.Linear (in_features = lstm_hidden_size * 2, out_features = 128) receives the features of the BiLSTM output and outputs 128 features. Since it is a bidirectional LSTM, the feature dimension is lstm_hidden_size * 2. The subsequent fully connected layers self.pinn_fc2, self.pinn_fc3 and self.pinn_fc4 gradually reduce the feature dimension to the final output_size, and introduce nonlinearity through the nn.functional.relu activation function in the middle to better fit the complex relationship, and finally obtain the predicted rock stress and failure time. Output_size = 1, because it is a single value in the predicted stress or rock failure time.

[0073] Furthermore, after constructing the physical constraint neural network (PINN), the loss function of the physical constraint neural network is defined. The loss function of the PINN consists of the following parts:

[0074] 1) Data fitting loss: the difference between the predicted stress and failure time and the true value.

[0075] The data fitting loss is:

[0076]

[0077] Where N is the total number of samples; σ pred is the predicted stress; x i is the i-th input sample; σ true is the test stress; T pred is the prediction time; j is the jth input sample; T true For test time.

[0078] 2) Physically driven constraint loss: Physical constraints are imposed according to the rock constitutive equation. The rock constitutive equation can be expressed as: σ = E·∈ + σ3, so the physical driven constraint loss function L phy for:

[0079]

[0080] Among them, σ pred is the predicted stress; k is the kth input sample; E is the elastic modulus; ε is the strain; σ3 is the confining pressure; ɑ is the balance coefficient, which is used to control the influence of different loss terms on the overall loss.

[0081] Combining the above data fitting loss and physical drive constraint loss, the loss function of the physical constraint neural network is obtained as follows:

[0082]

[0083] Where L is the loss function of the physical constraint neural network; L data is the data fitting loss; L phy is the physics driven constraint loss.

[0084] The addition of the attention mechanism can selectively focus on important parts, thereby improving the efficiency and accuracy of the model. The PINN physical neural network is used to integrate physical equations to make the results consistent with physical reality.

[0085] S102. Use the Bayesian optimization method to tune hyperparameters and obtain the optimal hyperparameter combination.

[0086] In machine learning and deep learning, hyperparameter tuning is an important issue that involves finding the best hyperparameter combination for a model in order to achieve the best performance on a given data set. Bayesian optimization is an efficient hyperparameter optimization method that gradually selects the optimal parameters by building a proxy model during model training to effectively find the global optimal solution. It is particularly suitable for black box function optimization problems that are computationally expensive.

[0087] Specifically, first, some initial hyperparameter combinations are selected and their performance in the real model is evaluated. These initial data points are used to build a Gaussian Process Model that can predict the possible performance of other hyperparameter combinations and estimate the uncertainty of the prediction. The Gaussian Process Model is used to predict the performance of different hyperparameter combinations and calculate the predicted value and uncertainty of each combination. The Probability of Improvement (PI) strategy is used to select the next hyperparameter combination to be evaluated. The PI strategy selects combinations with good prediction performance and high uncertainty, which means that these combinations have great potential to become better choices. After selecting a new hyperparameter combination, its performance is evaluated in the real model to generate new evaluation results. The new evaluation results are added to the existing dataset and used to update the Gaussian Process Model. By evaluating new hyperparameter combinations, more information can be obtained, thereby improving the prediction ability of the surrogate model. The Gaussian Process Model is updated with new data points to make the model more accurately reflect the actual situation. The updated model will better predict the performance of other hyperparameter combinations. As the iteration proceeds, the prediction accuracy of the model continues to improve, making it possible to more accurately select the next hyperparameter combination to be evaluated. The above process will be repeated many times until the predetermined stopping criteria (such as the maximum number of iterations) are reached. Each iteration will select new candidate points for evaluation and update the proxy model to gradually approach the optimal hyperparameter combination. Through continuous iteration and optimization, the performance of the model can be gradually improved and the best hyperparameter configuration can be found. In the Bayesian optimization process, 50 iterations are set, 5-fold cross validation is used, accuracy and AUC values ​​are used as scoring criteria, and all available processors are used to accelerate calculations.

[0088] The performance and generalization ability of the model are improved by finding the optimal combination of parameters. The Bayesian optimization method of this embodiment finds the optimal. Bayesian optimization can find an approximate optimal solution within a limited number of searches and does not rely on the derivative information of the function, so it is suitable for solving black box optimization problems.

[0089] S103, obtaining rock mechanical property data, including mass, density, elastic modulus, strain, strain rate, stress, and time, preprocessing the rock mechanical property data, and dividing the preprocessed data into a first training set and a first test set.

[0090] The rock mechanical property data include mass, density, elastic modulus, strain, strain rate, stress, and time. The first training set and the first test set consist of mass, density, elastic modulus, strain, and stress. The first test set should The change was increased by 0.008(‰), 0.01(‰), and 0.012(‰) respectively. .

[0091] In this embodiment, the rock sampled from the coal mine was processed to make standard specimens of 50×100 mm (89 pieces in total), and uniaxial (MST 816 testing machine) and triaxial compression tests (MST 815 testing machine) were performed on them. Finally, the rock mechanical property data such as the mass, density, elastic modulus, strain, strain rate, stress, and failure time of the rock were obtained.

[0092] In this embodiment, the preprocessing includes data processing processes such as normalization and regularization.

[0093] The normalization includes:

[0094]

[0095] Among them, x norm is the normalized value; x i is the original value; x min is the minimum value; x max is the maximum value.

[0096] In this embodiment, the preprocessed data is divided into a first training set and a first test set in a ratio of 8:2, wherein the first training set and the first test set include mass, density, elastic modulus, strain, and stress. The first test set The strains were increased by 0.008(‰), 0.01(‰), and 0.012(‰) respectively. .

[0097] S104. Train the convolutional neural network, the bidirectional long short-term memory neural network and the physical constraint neural network through a first training set, and train the physical constraint neural network based on the loss function of the physical constraint neural network to obtain a first model.

[0098] In each round of training, an early stopping mechanism is used to prevent overfitting. If the validation loss does not improve for several consecutive rounds, the training is stopped early and the first model is finally obtained.

[0099] S105: Input the first test set into a first model and output a first prediction result, where the first prediction result includes Three types Peak stress of the optimal solution.

[0100] S106, obtaining the corresponding strain based on the peak stress of the optimal solution result, generating a second training set and a second test set, wherein the second training set and the second test set include mass, density, elastic modulus, strain rate, peak stress of the optimal solution result and its corresponding strain, and time , the second test set strain and stress are divided into three cases, corresponding to Peak stress and corresponding strain of the three optimal solutions in the prediction results .

[0101] According to the optimal result obtained, the peak stress (maximum stress) of each specimen is found, and the corresponding strain is the peak strain.

[0102] S107: Train the first model using a second training set to obtain a second model.

[0103] Similarly, in each round of training, an early stopping mechanism is used to prevent overfitting. If the validation loss does not improve for several consecutive rounds, the training is stopped early and the second model is finally obtained.

[0104] S108: Input the second test set into the second model and output a second prediction result. The second prediction result is Three situations Predicted rock failure time, i.e. the time of peak point , and compare the three results to find the best result fruit .

[0105] As an embodiment of the present invention, in order to reveal the key features and their interactions that affect the rock failure time, SHAP (SHapley Additive exPlanations) values ​​are used.

[0106] Specifically, first, the SHAP library and other necessary libraries were installed and imported, the trained model was loaded, and the feature names were defined. Next, a SHAP interpreter object was created, and the test data was interpreted to calculate the SHAP value of each test sample. By generating summary graphs and force graphs of SHAP values, it is intuitively shown which features have the greatest impact on the model prediction, and the specific contribution direction (positive or negative) of each feature to the prediction results of a specific sample is analyzed.

[0107] According to an embodiment of the present invention, BiLSTM is added on the basis of CNN, which can well learn time series data and solve the long dependency problem that CNN cannot handle. Therefore, CNN and BiLSTM are combined to form a CNN-BiLSTM model, in which CNN is used to extract local features and BiLSTM is used to extract global features, which can give full play to the advantages of each neural network model, thereby improving the prediction accuracy of the model. And combined with PINN, prior knowledge such as physical laws is integrated into the neural network, which enables the model to not only learn patterns from data, but also follow physical laws, thereby improving the accuracy and generalization ability of the model. When used in engineering, strain gauges can be used to monitor the strain changes for a long time, and then the strain rate can be calculated, and its mass, density, elastic modulus, etc. can be obtained by detection, and the strain rate in the data set is used to predict its failure time, and then the ratio of the actual engineering strain rate to the laboratory strain rate is multiplied by the obtained failure time, and then the failure time of the rock in the actual engineering is obtained.

[0108] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0109] The above is an introduction to a method embodiment. The following is a further explanation of the scheme of the present invention through a device embodiment having the same inventive concept as the method in the aforementioned embodiment.

[0110] like Figure 2 As shown, the device 200 includes:

[0111] A network construction module 210 is used to construct a convolutional neural network, a bidirectional long short-term memory neural network and a physical constraint neural network, and define a loss function of the physical constraint neural network;

[0112] A hyperparameter tuning module 220 is used to perform hyperparameter tuning using a Bayesian optimization method to obtain an optimal hyperparameter combination;

[0113] The first data processing module 230 is used to obtain rock mechanical property data, pre-process the rock mechanical property data, and divide the pre-processed data into a first training set and a first test set, wherein the first test set includes mass, density, elastic modulus, strain, and stress. The strains were increased according to the values ​​of 0.008(‰), 0.01(‰), and 0.012(‰) respectively. add ;

[0114] A first training module 240 is used to train the convolutional neural network, the bidirectional long short-term memory neural network and the physical constraint neural network through a first training set, and to train the physical constraint neural network based on a loss function of the physical constraint neural network to obtain a first model;

[0115] The first prediction module 250 is used to input the first test set into a first model and output a first prediction result. The first prediction result includes Three types Peak stress of the optimal solution;

[0116] The second data processing module 260 is used to obtain the corresponding strain based on the peak stress of the optimal solution result, and generate a second training set and a second test set, wherein the second training set and the second test set include the preprocessed mass, density, elastic modulus, strain rate, the peak stress of the optimal solution result and its corresponding strain, and time , the second test set strain and The stress is divided into three cases, corresponding to the peak stress of the three optimal solutions in the first prediction result and their corresponding stress Change ;

[0117] A second training module 270, configured to train the first model using a second training set to obtain a second model;

[0118] The second prediction module 280 is used to input the second test set into the second model and output a second prediction result. The second prediction result is Three situations Predicted rock failure time , and compare the three results to find the best result .

[0119] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0120] In the technical solution of the present invention, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0121] According to an embodiment of the present invention, the present invention further provides an electronic device and a readable storage medium.

[0122] Figure 3 A schematic block diagram of an electronic device 300 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0123] The electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 can also be stored. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0124] A number of components in the electronic device 300 are connected to the I / O interface 305, including: an input unit 306, such as a keyboard, a mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a disk, an optical disk, etc.; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 309 allows the electronic device 300 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0125] The computing unit 301 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 301 performs the various methods and processes described above, such as methods S101 to S108. For example, in some embodiments, methods S101 to S108 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the methods S101 to S108 described above may be performed. Alternatively, in other embodiments, the computing unit 301 may be configured to execute methods S101 to S108 in any other appropriate manner (eg, by means of firmware).

[0126] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0127] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0128] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0129] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0130] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0131] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0132] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0133] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A rock failure time prediction method based on machine learning, characterized in that: include: Constructing a convolutional neural network, a bidirectional long short-term memory neural network and a physical constraint neural network, and defining a loss function of the physical constraint neural network; Use Bayesian optimization method to tune hyperparameters and obtain the optimal hyperparameter combination; Acquire rock mechanical property data, preprocess the rock mechanical property data, and divide the preprocessed data into a first training set and a first test set, wherein the first training set and the first test set include mass, density, elastic modulus, strain, and stress; The convolutional neural network, the bidirectional long short-term memory neural network and the physical constraint neural network are trained by using a first training set, and the physical constraint neural network is trained based on a loss function of the physical constraint neural network to obtain a first model; Inputting the first test set into a first model, and outputting a first prediction result, wherein the first prediction result includes a peak stress of an optimal solution result; Obtaining the corresponding strain based on the peak stress of the optimal solution result, generating a second training set and a second test set, wherein the second training set and the second test set include mass, density, elastic modulus, strain rate, the peak stress of the optimal solution result and the corresponding strain, and time; Training the first model using a second training set to obtain a second model; The second test set is input into a second model, and a second prediction result is output, where the second prediction result is the predicted rock failure time.

2. The method according to claim 1, characterized in that The convolutional neural network comprises: A first convolution module, comprising a first convolution layer, a first activation layer and a first pooling layer in sequence; the first convolution layer obtains the input of the convolutional neural network, and the output of the first pooling layer is used as the output of the first convolution module; A second convolution module, comprising a second convolution layer, a second activation layer, and a second pooling layer in sequence; the second convolution layer is used to take the output of the first convolution module as input, and the output of the second pooling layer is used as the output of the second convolution module; A third convolution module, comprising a third convolution layer, a third activation layer and a third pooling layer in sequence; the third convolution layer is used to take the output of the second convolution module as input, and the output of the third pooling layer is used as the output of the third convolution module; A fourth convolution module, comprising a fourth convolution layer, a fourth activation layer and a fourth pooling layer in sequence; the fourth convolution layer is used to take the output of the third convolution module as input, and the output of the fourth pooling layer is used as the output of the fourth convolution module; The fifth convolution module includes a fifth convolution layer, a fifth activation layer and a fifth pooling layer in sequence; the fifth convolution layer is used to take the output of the fourth convolution module as input, and the output of the fifth pooling layer is used as the output of the fifth convolution module.

3. The method according to claim 2, characterized in that Also includes: An Attention layer and a dropout layer are set between the third convolution module and the fourth convolution module; the Attention layer takes the output of the third convolution module as input, the dropout layer takes the output of the Attention layer as input, and the output of the dropout layer is used as the input of the fourth convolution module.

4. The method according to claim 1, characterized in that: The physical constraint neural network includes a first fully connected layer, a second fully connected layer, a third fully connected layer and a fourth fully connected layer in sequence; wherein the second fully connected layer and the third fully connected layer contain ReLU activation functions.

5. The method according to claim 1, characterized in that The loss function of the physical constraint neural network includes data fitting loss and physical drive constraint loss equation, namely: L=L data +ɑL phy Where L is the loss function of the physical constraint neural network; L data is the data fitting loss; L phy is the physics driven constraint loss.

6. The method according to claim 5, characterized in that The data fitting loss is: Where N is the total number of samples; σ pred is the predicted stress; x i is the i-th input sample; σ true is the test stress; T pred is the prediction time; j is the jth input sample; T true For test time.

7. The method according to claim 5, characterized in that The physical drive constraint loss is: Among them, σ pred is the predicted stress; k is the kth input sample; E is the elastic modulus; ε is the strain; σ3 is the confining pressure; ɑ is the equilibrium coefficient.

8. A rock failure time prediction device based on machine learning, characterized in that: include: A network construction module, used to construct a convolutional neural network, a bidirectional long short-term memory neural network and a physical constraint neural network, and define a loss function of the physical constraint neural network; Hyperparameter tuning module, which is used to tune hyperparameters using Bayesian optimization method to obtain the optimal hyperparameter combination; A first data processing module is used to obtain rock mechanical property data, preprocess the rock mechanical property data, and divide the preprocessed data into a first training set and a first test set, wherein the first test set includes mass, density, elastic modulus, strain, and stress; A first training module, used to train the convolutional neural network, the bidirectional long short-term memory neural network and the physical constraint neural network through a first training set, and to train the physical constraint neural network based on a loss function of the physical constraint neural network to obtain a first model; A first prediction module, configured to input the first test set into a first model and output a first prediction result, wherein the first prediction result includes a peak stress of an optimal solution result; A second data processing module, configured to obtain the corresponding strain based on the peak stress of the optimal solution result, and generate a second training set and a second test set, wherein the second training set and the second test set include the preprocessed mass, density, elastic modulus, strain rate, the peak stress of the optimal solution result and its corresponding strain, and time; A second training module, used to train the first model using a second training set to obtain a second model; The second prediction module is used to input the second test set into a second model and output a second prediction result, where the second prediction result is a predicted rock failure time.

9. An electronic device comprising at least one processor; and a memory connected in communication with the at least one processor; characterized in that: The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

Images