Ocean Drilling Cycle Prediction Method Based on One-Dimensional Convolutional Neural Network

Abstract

The present invention discloses a method for predicting the ocean drilling cycle based on a one-dimensional convolutional neural network, which comprises the following steps: constructing a training data set according to historical data; building and training a one-dimensional convolutional neural network model; constructing a prediction data set according to the basic data of ocean drilling design; and using the trained one-dimensional convolutional neural network model to process the prediction data set to obtain the predicted ocean drilling cycle. The present invention does not require the empirical knowledge of domain experts, can automatically predict the ocean drilling cycle according to the ocean drilling design parameters, realizes an end-to-end training and prediction model framework. This method can effectively solve the problem that the prediction of the ocean drilling cycle depends on the data of adjacent wells, and solve the problem that the statistical probability method cannot consider the non-linear relationship in the drilling parameters and has limited accuracy. By utilizing the superiority of the neural network in fitting non-linear relationships and combining the characteristics of local perception of the one-dimensional convolutional neural network, the accuracy of the prediction result is finally significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of drilling cycle prediction, and in particular to a method for predicting an ocean drilling cycle based on a one-dimensional convolutional neural network. Background Art

[0002] Over the past few decades, rapid industrialization around the world has significantly increased energy demand, making it more common to explore and develop oil and gas reserves. Exploratory wells are drilled in search of oil and gas, with the main purpose of gathering information about underground conditions and confirming whether geological formations contain hydrocarbons.

[0003] As a key parameter for calculating investment in offshore oil drilling projects, the drilling cycle is of great significance for ensuring that the overall exploration and development goals of oil and gas fields are achieved as planned and that they are put into production in a timely manner and that investments are recovered as soon as possible. Therefore, accurately predicting the offshore drilling cycle is an important task in oil and gas exploration and development planning. The current methods for predicting the drilling cycle are:

[0004] (1) Typical case method: This is a method that uses the geological data of adjacent wells or adjacent blocks to infer the drilling cycle of the new well by drawing on their actual drilling cycle data. This method is often used for the cycle design of exploration wells in new areas, but the number of adjacent wells is limited, and the prediction accuracy is greatly affected by the quality of adjacent well data. In addition, the drilling process, completion depth and other data of adjacent wells may be different from those of the new well, so the reference for predicting the drilling cycle of the new well is limited.

[0005] (2) Empirical estimation method: The empirical estimation method is a method of formulating target procedures or drilling operation time through seminars based on the on-site experience of construction technicians. This method is time-saving, labor-saving and easy to implement, but its accuracy depends largely on the experience of the evaluator. It is usually suitable for on-site operation plan discussions and is not realistic in the drilling plan preparation or drilling design stage.

[0006] (3) Historical level method: This is a method of determining the drilling cycle based on the actual production data of the same type of wells in the same region in previous years and the statistical average. This method is relatively general and simple and is a commonly used method, but it is based on statistical methods and has a large error.

[0007] (4) Cycle quota method: The cycle quota method refers to the statistical time required to complete a task. This method has a reasonable calculation structure and the data is close to reality, but the task of data statistics and analysis is heavy and it is not suitable for oil fields with a large number of wells. It is only suitable for oil and gas fields with a small number of wells.

[0008] (5) Learning curve method: This method proposes that when drilling a series of wells with similar wellbore structures and operation steps in a certain area, the drilling cycle can be fitted with a learning curve. By analyzing the learning curve, the management ability and technical level of the operator can be evaluated, and the problems encountered in the drilling operation can be described. However, the learning curve method has many restrictive conditions for well selection and low prediction accuracy, and is not suitable for predicting the offshore drilling cycle.

[0009] In summary, the existing methods for predicting the drilling cycle of target wells are difficult to guarantee prediction accuracy, the prediction process is cumbersome, relying on adjacent well data, and unable to consider the non-linear relationship in drilling parameters, resulting in low prediction accuracy. Summary of the Invention

[0010] In view of the above deficiencies in the prior art, the offshore drilling cycle prediction method based on a one-dimensional convolutional neural network provided by the present invention solves the problems of low prediction accuracy and cumbersome prediction process in the existing methods for predicting the drilling cycle of target wells.

[0011] In order to achieve the above invention purpose, the technical solution adopted by the present invention is as follows:

[0012] Provide an offshore drilling cycle prediction method based on a one-dimensional convolutional neural network, which includes the following steps:

[0013] S1. Obtain and construct a training data set based on the drilling historical data of offshore drilling with one drilling time for each well section; wherein the drilling data includes the drilling start year, the basin where it is located, the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, the drilling depth of each well section, and the drilling cycle.

[0014] S2. Build a one-dimensional convolutional neural network model and train the one-dimensional convolutional neural network model with the training data set.

[0015] S3. Obtain and construct a prediction data set based on the design basic data of the offshore drilling whose drilling cycle is to be predicted; wherein the design basic data includes the drilling start year, the basin where it is located, the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, and the drilling depth of each well section.

[0016] S4. Use the trained one-dimensional convolutional neural network model to process the prediction data set to obtain the predicted offshore drilling cycle.

[0017] Further, the specific method for constructing the training data set in step S1 includes the following steps:

[0018] S1-1. Zero-fill the data with less than six openings in the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, and the drilling depth of each well section to six openings to complete data preprocessing.

[0019] S1-2. Perform one-hot encoding on the drilling year and the basin where the well is located, and perform global normalization on the preprocessed data according to the type;

[0020] S1-3. Concatenate the vectors obtained by one-hot encoding and the vectors obtained by global normalization corresponding to the same well to obtain training samples;

[0021] S1-4. Use the known drilling cycle as the label of the corresponding training sample to obtain a training dataset.

[0022] Furthermore, the one-dimensional convolutional neural network model in step S2 includes an input layer, a one-dimensional convolutional layer C1, a one-dimensional convolutional layer C2, a one-dimensional max pooling layer, a Dropout layer, a fully connected layer, and an output layer connected in sequence; where:

[0023] The activation functions of the one-dimensional convolutional layer C1 and the one-dimensional convolutional layer C2 are both ReLU functions, the activation function of the fully connected layer is a linear activation function, and the padding methods of the one-dimensional convolutional layer C1, the one-dimensional convolutional layer C2, and the one-dimensional max pooling layer are valid;

[0024] The length of the convolutional kernel of the one-dimensional convolutional layer C1 is 3, the number of convolutional kernels is 64, the stride is 1, and the length of the output feature subsequence is 31 and the number is 64;

[0025] The length of the convolutional kernel of the one-dimensional convolutional layer C2 is 3, the number of convolutional kernels is 64, the stride is 1, and the length of the output feature subsequence is 29 and the number is 64;

[0026] The length of the downsampling region of the one-dimensional max pooling layer is 4, and the length of the output feature subsequence is 7 and the number is 64;

[0027] The random inactivation probability of the Dropout layer is 0.4 and it includes 448 nodes;

[0028] The output node of the output layer is 1.

[0029] Furthermore, the specific method for training the one-dimensional convolutional neural network model with the training dataset in step S2 includes the following sub-steps:

[0030] S2-1. Divide the training dataset into a training set and a validation set;

[0031] S2-2. Input the samples in the training set into the current one-dimensional convolutional neural network model, and use the output corresponding to the current one-dimensional convolutional neural network model as its predicted value;

[0032] S2-3. Based on the predicted value and the true label, use the MSE loss function to obtain the cost error;

[0033] S2-4. Determine whether the cost error reaches the preset iteration requirement. If yes, go to step S2-6; otherwise, go to step S2-5;

[0034] S2-5. Perform backpropagation based on the cost error, update the weights and biases of each network layer in the one-dimensional convolutional neural network model, and return to step S2-2;

[0035] S2-6. Input the M samples in the validation set into the current one-dimensional convolutional neural network model respectively to obtain the corresponding M predicted values;

[0036] S2-7. Statistically analyze the prediction errors of the M predicted values, and determine whether the prediction errors reach the preset requirements. If yes, output the current one-dimensional convolutional neural network model to complete the training; otherwise, adjust the hyperparameters of the one-dimensional convolutional neural network model and return to step S2-2.

[0037] Further, the specific method for constructing the prediction data set in step S3 is: process the design basic data by the same method as constructing the training samples to obtain the prediction data set.

[0038] The beneficial effects of the present invention are as follows: This method does not require the empirical knowledge of domain experts, can automatically predict the ocean drilling cycle according to ocean drilling design parameters, and realizes an end-to-end training and prediction model framework. Experiments show that this method can effectively solve the problem that the prediction of the ocean drilling cycle depends on adjacent well data, solve the problem that the statistical probability method has limited accuracy because it cannot consider the non-linear relationship in drilling parameters, utilize the superiority of the neural network in fitting non-linear relationships, and combine the local perception characteristics of the one-dimensional convolutional neural network, and finally significantly improve the accuracy of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a schematic flow chart of the present invention;

[0040] Figure 2 is a graph showing the relationship between the loss values of the training set and the validation set and the number of iterations during the training process of the one-dimensional convolutional neural network with the MSE as the loss function;

[0041] Figure 3 is the prediction effect diagram of the method in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The following describes the specific embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

[0043] As Figure 1 shown, the method for predicting the ocean drilling cycle based on a one-dimensional convolutional neural network includes the following steps:

[0044] S1. Obtain and construct a training data set based on the drilling history (log) data of ocean drilling where the number of drilling times for each well section is one; the drilling data includes the starting year of drilling, the basin where it is located, the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, the drilling depth of each well section, and the drilling cycle; the number of drilling times is the number of times for the complete process of drilling, pure drilling, and tripping out for completing one well section, and there are at most six well sections for drilling one well;

[0045] S2. Build a one-dimensional convolutional neural network model and train the one-dimensional convolutional neural network model with the training data set;

[0046] S3. Obtain and construct a prediction data set based on the design basic data of the ocean drilling for which the drilling cycle is to be predicted; the design basic data includes the starting year of drilling, the basin where it is located, the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, and the drilling depth of each well section;

[0047] S4. Use the trained one-dimensional convolutional neural network model to process the prediction data set and obtain the predicted ocean drilling cycle.

[0048] The specific method for constructing the training data set in step S1 includes the following steps:

[0049] S1-1. Zero-fill the data with less than six openings in the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, and the drilling depth of each well section to six openings to complete data preprocessing;

[0050] S1-2. Perform one-hot encoding on the starting year of drilling and the basin where it is located, and perform global normalization on the preprocessed data according to the type respectively;

[0051] S1-3. Concatenate the vector obtained by one-hot encoding and the vector obtained by global normalization corresponding to the same well to obtain a training sample;

[0052] S1-4. Use the known drilling cycle as the label of the corresponding training sample to obtain the training data set.

[0053] The formula for global normalization is:

[0054]

[0055] where represents the i-th data x in a certain type of data of the sample iThe result after global normalization, where min(x) represents the minimum value of the data of this type in the sample, and max(x) represents the maximum value of the data of this type in the sample. For example, when performing global normalization on the first drilling depth, the first drilling depth data of all wells are used to perform normalization on each well, and the maximum value is the maximum value of all first drilling depths.

[0056] In step S2, the one-dimensional convolutional neural network model includes an input layer, a one-dimensional convolutional layer C1, a one-dimensional convolutional layer C2, a one-dimensional max pooling layer, a Dropout layer, a fully connected layer, and an output layer connected in sequence; where:

[0057] The activation functions of the one-dimensional convolutional layer C1 and the one-dimensional convolutional layer C2 are both ReLU functions, the activation function of the fully connected layer is a linear activation function, and the padding methods of the one-dimensional convolutional layer C1, the one-dimensional convolutional layer C2, and the one-dimensional max pooling layer are valid;

[0058] The convolution kernel length of the one-dimensional convolutional layer C1 is 3, the number of convolution kernels is 64, the stride is 1, and the length of the output feature subsequence is 31, and the number is 64;

[0059] The convolution kernel length of the one-dimensional convolutional layer C2 is 3, the number of convolution kernels is 64, the stride is 1, and the length of the output feature subsequence is 29, and the number is 64;

[0060] The downsampling region length of the one-dimensional max pooling layer is 4, and the length of the output feature subsequence is 7, and the number is 64;

[0061] The random inactivation probability of the Dropout layer is 0.4, and it includes 448 nodes;

[0062] The output node of the output layer is 1, and the final output is a row vector or column vector with 1 element in one dimension.

[0063] The specific method for training the one-dimensional convolutional neural network model with the training data set in step S2 includes the following sub-steps:

[0064] S2-1. Divide the training data set into a training set and a validation set;

[0065] S2-2. Input the samples in the training set into the current one-dimensional convolutional neural network model, and use the output corresponding to the current one-dimensional convolutional neural network model as its predicted value;

[0066] S2-3. Based on the predicted value and the true label, use the MSE loss function to obtain the cost error;

[0067] S2-4. Determine whether the cost error reaches the preset iteration requirement. If so, enter step S2-6; otherwise, enter step S2-5;

[0068] S2-5. Perform backpropagation based on the cost error, update the weights and biases of each network layer in the one-dimensional convolutional neural network model, and return to step S2-2;

[0069] S2-6. Input the M samples in the validation set into the current one-dimensional convolutional neural network model respectively, and obtain the corresponding M predicted values;

[0070] S2-7. Statistically analyze the prediction errors of the M predicted values, and determine whether the prediction errors meet the preset requirements. If so, output the current one-dimensional convolutional neural network model to complete the training; otherwise, adjust the hyperparameters of the one-dimensional convolutional neural network model and return to step S2-2. The hyperparameters of the one-dimensional convolutional neural network model include the learning rate, dropout, and the number of iterations;

[0071] The expression of the MSE loss function is:

[0072]

[0073] where y i represents the true label, f(x i ) represents the predicted label, and MSE represents the cost error. Taking MSE as the loss function, the relationship diagram between the loss values of the training set and the validation set and the number of iterations during the training process of the one-dimensional convolutional neural network is as Figure 2 shown.

[0074] The specific method for constructing the prediction data set in step S3 is: Process the design basic data using the same method as constructing the training samples to obtain the prediction data set.

[0075] In the specific implementation process, the drilling cycle in this method is the footage operation time and the cementing operation time used for the drilling to reach the designed depth, excluding the weather waiting time and the non-production time caused by shutdowns; among them:

[0076] Footage operation: The footage operation refers to the process of drilling different formations into the specified holes using mechanical equipment. The footage operation cycle includes the footage time and the footage auxiliary time. The footage time includes the pure drilling, reaming, tripping, and connection time; the footage auxiliary time includes the circulation, surveying, short tripping, etc. time.

[0077] Cementing operation: Cementing is the construction operation of lowering the casing into the well and injecting cement into the annular space between the wellbore and the casing. The cementing operation cycle consists of the preparation work, casing running, inner pipe tripping, cement injection, waiting for setting, wellhead disassembly and assembly, drilling the cement plug, and cementing quality survey time.

[0078] After convolution calculation, in order to increase the non-linearity of the neural network model, an activation function needs to be introduced. The rectified linear unit (ReLU) function can accelerate the convergence of the network and prevent gradient disappearance. Its expression is ReLU(x) = max(0, x). Therefore, the final output of each neuron in the convolutional layer is the input of the k-th neuron in the t-th layer.

[0079] The pooling layer can accelerate the calculation speed, reduce the calculation cost, and prevent overfitting problems, and can also maintain the translational invariance of features. This method uses max pooling, that is, taking the maximum value in an adjacent region H as the final output of this region. The expression is

[0080] Dropout means that during the deep learning training process, to prevent the model from overfitting, neurons in the hidden layer have a certain probability of failing in each iteration (including forward and backward propagation).

[0081] In the one-dimensional convolutional neural network adopted in this method, the one-dimensional convolutional layer extracts local one-dimensional sequence segments (i.e., subsequences) from the sequence. Each output time step is obtained by using a small segment of the input sequence in the time dimension, that is, it can identify local patterns in the sequence. Because the same input transformation is performed on each sequence segment, the patterns learned at a certain position in the sequence can be recognized at other positions later, which makes the one-dimensional convolutional neural network have translational invariance (for time translation). The one-dimensional pooling layer is used to perform time downsampling on the time series in the one-dimensional convolutional neural network. That is, it extracts one-dimensional sequence segments (i.e., subsequences) from the input, and then outputs their maximum values (max pooling) or average values (average pooling). This operation is used to reduce the length of the one-dimensional input (subsampling), reduce the computational complexity of the network, and at the same time has the function of maintaining the feature scale invariant and reducing the overfitting phenomenon.

[0082] In an embodiment of the present invention, in order to verify the accuracy and effectiveness of the method, 50 wells in the validation set are selected for model verification. Some of the data used are shown in Tables 1, 2, and 3. One-hot encoding converts the attribute values into binary values, and the corresponding bits are represented by "1" for a certain category or attribute. In this embodiment, if the drilling start year is before 2015 (including 2015), it is encoded as 01, and if the drilling start year is after 2016 (including 2016), it is encoded as 10. The basins where it is located specifically include Basin A, Basin B, Basin C, Basin D, Basin E, and Basin F. When it is in Basin A, it is encoded as 100000, and when it is in Basin B, it is encoded as 010000.

[0083] Table 1

[0084] Before 2015 (including 2015) After 2016 (including 2016) Basin A Basin B Basin C Basin D Basin E Basin F Total depth of completed well Bit size of the first spud Casing size of the first spud 0 1 0 0 0 0 0 1 0.610351967 1 1 1 0 0 0 0 0 1 0 0.438716356 1 1 0 1 0 1 0 0 0 0 0.129192547 0.7 0.63898917 0 1 1 0 0 0 0 0 0.584679089 0.075 0 1 0 0 0 0 0 1 0 0.402070393 1 1

[0085] Table 2

[0086] Top depth of the first spud Drilling depth of the first spud Bit size of the second spud Casing size of the second spud Top depth of the second spud Drilling depth of the second spud Bit size of the third spud Casing size of the third spud Top depth of the third spud Drilling depth of the third spud Bit size of the fourth spud 0.078424172 0.010218088 0.428571429 0.669 0.077447336 0.254269126 0.7 0.717488789 0.301694915 0.769883423 0.693877551 0.041179248 0.022944944 1 1 0.049406237 0.164040301 1 1 0.203543914 0.48905317 1 0.014163563 0.026384017 0.428571429 0.669 0.025144568 0.127663934 0.7 0 3.15624037 0.283764572 0 0.013638987 0.619490621 0.214285714 0.48 0.426218505 0.646857923 0.485714286 0 0.864098613 0.256184248 0 0.037769501 0.017080982 1 1 0.042079719 0.178278689 1 1 3.21201849 0.391526869 1

[0087] Table 3

[0088] Casing size of the fourth spud Top depth of the fourth spud Drilling depth of the fourth spud Bit size of the fifth spud Casing size of the fifth spud Top depth of the fifth spud Drilling depth of the fifth spud Bit size of the sixth spud Casing size of the sixth spud Top depth of the sixth spud Drilling depth of the sixth spud 0 0.767759227 0.056119874 0 0 0 0 0 0 0 0 1 0.495744366 0.166206625 0.985882353 1 0.54030081 0.06923626 0.98 0 0.605535248 0.227677464 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0.430040797 0.223974763 0.985882353 0 0.507713074 0.135617416 0 0 0 0

[0089] Determine the trend of the goodness of fit according to the following relational expressions:

[0090]

[0091] where R 2 represents the correlation coefficient, which is used to characterize the trend of the goodness of fit, SSR is the sum of squares of regression, SSE is the sum of squares of residuals, SST is the total sum of squares of deviations, is the average value of the true labels, and m is the number of training samples.

[0092] Determine the mean absolute percentage error according to the following relational expressions:

[0093]

[0094] Determine the root mean square error according to the following relational expressions:

[0095]

[0096] In the specific prediction process, the prediction effect of this method is as Figure 3 shown. The comparison of the results of this method with the schemes using BPNN and XGBoost is shown in Table 4.

[0097] Table 4

[0098] Method MAE RMSE MAPE / % <![CDATA[R 2 / %]]> The method of the present invention 0.99 1.65 12.13 91.97 BPNN 2.51 4.25 16.10 83.44 XGBoost 2.41 3.96 16.65 85.60

[0099] It can be seen that this method is superior to the schemes using BPNN (BP neural network) and XGBoost model in terms of MAE, RMSE, MAPE, R 2 indexes.

[0100] To sum up, this method does not require the empirical knowledge of domain experts and can automatically predict the marine drilling cycle according to the marine drilling design parameters, realizing an end-to-end training and prediction model framework. Experiments show that this method can effectively solve the problem that the prediction of the marine drilling cycle depends on the data of adjacent wells, solve the problem that the statistical probability method has limited accuracy because it cannot consider the nonlinear relationship in the drilling parameters, and utilize the superiority of the neural network in fitting nonlinear relationships, combined with the characteristics of local perception of the one-dimensional convolutional neural network, and finally significantly improve the accuracy of the prediction results.

Claims

1. A method for predicting the ocean drilling cycle based on a one-dimensional convolutional neural network, characterized in that, it includes the following steps: S1. Obtain and construct a training data set based on the drilling historical data of ocean drilling with one drilling time for each well section; wherein the drilling data includes the drilling year, the basin where it is located, the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, the drilling depth of each well section, and the drilling cycle; the drilling cycle is the footage operation time and the cementing operation time used for the drilling to reach the designed depth; S2. Build a one-dimensional convolutional neural network model and train the one-dimensional convolutional neural network model with the training data set; S3. Obtain and construct a prediction data set based on the design basic data of the ocean drilling whose drilling cycle is to be predicted; wherein the design basic data includes the drilling year, the basin where it is located, the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, and the drilling depth of each well section; S4. Use the trained one-dimensional convolutional neural network model to process the prediction data set and obtain the predicted ocean drilling cycle; The one-dimensional convolutional neural network model includes an input layer, a one-dimensional convolutional layer C1, a one-dimensional convolutional layer C2, a one-dimensional max pooling layer, a Dropout layer, a fully connected layer, and an output layer connected in sequence; wherein: The activation functions of the one-dimensional convolutional layer C1 and the one-dimensional convolutional layer C2 are both ReLU functions, the activation function of the fully connected layer is a linear activation function, and the padding methods of the one-dimensional convolutional layer C1, the one-dimensional convolutional layer C2, and the one-dimensional max pooling layer are valid; The convolutional kernel length of the one-dimensional convolutional layer C1 is 3, the number of convolutional kernels is 64, the stride is 1, and the length of the output feature subsequence is 31 and the number is 64; The convolutional kernel length of the one-dimensional convolutional layer C2 is 3, the number of convolutional kernels is 64, the stride is 1, and the length of the output feature subsequence is 29 and the number is 64; The downsampling region length of the one-dimensional max pooling layer is 4, and the length of the output feature subsequence is 7 and the number is 64; The random inactivation probability of the Dropout layer is 0.4 and it includes 448 nodes; The output node of the output layer is 1; The specific method for constructing the training data set in step S1 includes the following steps: S1-1. Pad the data with less than six openings in the total depth of the completed well, the bit size of each well section, the casing size of each well section, the top depth of each well section, and the drilling depth of each well section with zeros to six openings to complete data preprocessing; S1-2. Perform one-hot encoding on the drilling year and the basin where it is located, and perform global normalization on the preprocessed data according to the type respectively; S1-3. Concatenate the vector obtained by one-hot encoding and the vector obtained by global normalization corresponding to the same well to obtain a training sample; S1-4. Use the known drilling cycle as the label of the corresponding training sample to obtain a training data set.

2. The method for predicting the ocean drilling cycle based on a one-dimensional convolutional neural network according to claim 1, characterized in that, the specific method for training the one-dimensional convolutional neural network model with the training data set in step S2 includes the following sub-steps: S2-1. Divide the training data set into a training set and a validation set; S2-2. Input the samples in the training set into the current one-dimensional convolutional neural network model, and take the output corresponding to the current one-dimensional convolutional neural network model as its predicted value; S2-3. Based on the predicted value and the true label, use the MSE loss function to obtain the cost error; S2-4. Determine whether the cost error reaches the preset iteration requirement. If so, go to step S2-6; otherwise, go to step S2-5; S2-5. Perform backpropagation based on the cost error, update the weights and biases of each network layer in the one-dimensional convolutional neural network model, and return to step S2-2; S2-6. Input the M samples in the validation set into the current one-dimensional convolutional neural network model respectively to obtain the corresponding M predicted values; S2-7. Statistically analyze the prediction errors of the M predicted values, determine whether the prediction errors reach the preset requirements. If so, output the current one-dimensional convolutional neural network model to complete the training; otherwise, adjust the hyperparameters of the one-dimensional convolutional neural network model and return to step S2-2.

3. The method for predicting the ocean drilling cycle based on a one-dimensional convolutional neural network according to claim 1, characterized in that the specific method for constructing the prediction data set in step S3 is: process the design basic data by the same method as constructing the training samples to obtain the prediction data set.

Images