A deep-migratable multi-factor accident warning method and system for offshore oil drilling

Abstract

The present invention relates to the technical field of offshore oil drilling safety monitoring, and more specifically, to a deep transferable multi-factor accident warning method and system for offshore oil drilling. The accident warning method includes the following specific steps: S1: obtaining the correlation score of each single attribute factor of drilling data; S2: determining the key factor according to the correlation score; S3: obtaining the data depth feature by training the key factor; S4: obtaining the prediction result through the data depth feature. The purpose of the present invention is to overcome the shortcomings of the prior art and provide a warning method that is applied to offshore oil drilling and can screen multiple key factors that cause accidents and then be used to predict the occurrence of accidents.

Description

Technical Field

[0001] The present invention relates to the technical field of offshore oil drilling safety monitoring, and more specifically, to a deep-migratable multi-factor accident early warning method and system for offshore oil drilling. Background Art

[0002] Offshore oil drilling is a complex underground project, and the drilling process is accompanied by a large number of random and uncontrollable factors. Drilling accidents caused by many reasons such as stratum structure, hydrology, drilling technology, oil and gas distribution, etc. seriously threaten drilling safety, and serious accidents will cause huge economic losses. Real-time detection and accurate warning of drilling accidents have always been a strong demand in the field of offshore oil drilling. Research on this technology is of great significance for ensuring drilling safety, reducing costs and improving efficiency. At present, most of the existing methods of this technology are based on anomaly detection. This type of method uses the abnormal fluctuation of the value of a single or a few single attribute factors within a period of time to define anomalies and give warnings. The method based on this principle has the following disadvantages: (1) Only considering the abnormal fluctuation of the value of a single or a few key factors for accident warning. This simple method of describing abnormal characteristics is prone to high false alarm rate or low warning confidence; (2) It is impossible to effectively link the abnormal fluctuation of a single or a few attribute factors with the occurrence of accidents. The estimation of the possibility of accidents still requires human intervention and judgment; 3) Different parameter systems need to be designed in different data environments, and the portability is poor.

[0003] Chinese patent CN109508827A discloses a drilling accident early warning model based on a time recursive neural network. By constructing a time recursive neural network model, a portion of the labeled time series data is randomly selected as a training set. The specific input is a combination of features and the selection of a time window, and the model is trained to predict the probability of an accident and the type of accident one minute later. In this scheme, multiple attribute factors are considered simultaneously for accident early warning, but the multiple attribute factors are random, not selected attribute factors with a high correlation with the accident. Then, the early warning is obtained by training and predicting these random attribute factors. Therefore, the early warning result lacks data pertinence, which makes the entire prediction model too noisy and easily leads to a high false alarm rate. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide an early warning method for offshore oil drilling that can screen multiple key factors that lead to accidents and then be used to predict the occurrence of accidents.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] Provided is an accident warning method for offshore oil drilling, comprising the following specific steps:

[0007] S1: Based on the correlation analysis between offshore oil drilling accidents and the single attribute factors of each drilling data in the time series dimension, the correlation degree between the accident and each single attribute factor is determined;

[0008] S2: According to the degree of correlation between offshore oil drilling accidents and each single attribute factor, multiple factors with high correlation are selected as key factors;

[0009] S3: Using the long short-term memory network, predict the data trends of multiple key factors selected in step S2;

[0010] S4: Comprehensively consider the relationship between multiple key factors and the information of multiple key factors in the time series dimension, combine the predicted data in step S3, and use the deep learning network to express the deep features of the data;

[0011] S5: Based on the deep features extracted in step S4, a transferable multi-factor detector is trained to achieve deep transferable multi-factor accident prediction for offshore oil drilling.

[0012] The present invention selects a plurality of single-attribute factors related to the stuck drill accident as key factors, then combines these key factors, performs data trend prediction analysis, and finally obtains the prediction results to realize accident prediction for offshore oil drilling; the accident warning method of the present invention integrates a plurality of key factors for the occurrence of offshore oil drilling accidents, can avoid the high false alarm rate caused by the abnormality of a single-attribute factor, and further improves the accuracy of the prediction. Moreover, in reality, the occurrence of offshore oil drilling accidents is also caused by multiple factors rather than a single factor, and the present invention has practical significance; at the same time, the present invention accurately extracts a plurality of key factors, and can also serve as the basis for subsequent in-depth research on these key factors, and explores the connection between multiple key factors and the occurrence of offshore oil drilling accidents.

[0013] Furthermore, in step S1, the correlation score is used to measure the correlation degree between the offshore oil drilling accident and each single factor. The specific steps are as follows:

[0014] S11: Obtaining data distribution before and after the offshore oil drilling accident and classifying them to obtain a data set before and after the offshore oil drilling accident;

[0015] S12: Based on the principle that the data distribution before and after the offshore oil drilling accident is different, the time distance is used as the measurement standard to determine the distinguishability of the data set before the offshore oil drilling accident and the data set after the offshore oil drilling accident;

[0016] S13: Determine the correlation score between offshore oil drilling accidents and the single attribute factors of each drilling data through time series and distinguishability, and then determine the degree of correlation.

[0017] Furthermore, the distinguishability determination formula of the data set is as follows:

[0018]

[0019] I r (x,A1,A2)=1 when x∈A i &&NN r (x,A)∈A i ,A=A1∪A2,

[0020] I r (x,A1,A2)=0 when otherwise,

[0021] Where A1 is the set of time series before or after an offshore oil drilling accident, A2 is the set of subsequences randomly sampled from the entire time series data, x is the time series in the data set before or after an offshore oil drilling accident, for A = A1 ∪ A2, x∈A, NN r (x,A) represents the rth element in A-{x} that is closest to x; I r (x, A1, A2) indicates whether x and the rth closest element to x are in the same subset; T r,p represents the probability that A1 and A2 do not belong to the same data distribution, where p is the size of set A, r represents the index of the rth nearest element of x, if T r,p If T is small, it means that the two types of data A1 and A2 have no distinction; r,p If it is large, it means that the two types of data A1 and A2 are distinguishable.

[0022] Furthermore, the specific steps for obtaining the key factors in step S2 are as follows:

[0023] S21: sorting the correlation scores of the single attribute factors of the drilling data from high to low;

[0024] S22: Setting a correlation score threshold, and then filtering out all single attribute factors of the drilling data that are higher than the correlation score threshold as key factors.

[0025] Furthermore, the specific steps for obtaining the prediction results in step S3 are as follows:

[0026] S31: Divide memory units: Divide the long and short time series memory network into multiple memory units.

[0027] S32: Predicting the memory unit: predicting each memory unit of each key factor;

[0028] S33: Obtain prediction results: The prediction information of each key factor can be obtained by collecting the prediction information of each memory unit, and then the prediction results are obtained by summarizing them.

[0029] Furthermore, the prediction formula of the memory unit in step S32 is as follows:

[0030] (h t ,C t )=H(h t-1 ,x t ,C t-1 )

[0031] where h t-1 is the output data of the memory unit at the previous moment, C t-1 is the memory state of the memory unit at the last moment, x t is the data of each key factor input at the current moment, h t is the predicted data of the memory unit at the next moment, C t The current memory state of the memory unit.

[0032] Furthermore, the h t , C t The specific calculation formula is:

[0033] f t =σ(W f [h t-1 ,x t ]+b f ),

[0034] i t =σ(W i [h t-1 ,x t ]+b i ),

[0035] o t =σ(W o [h t-1 ,x t ]+b o ),

[0036]

[0037]

[0038] h t =o t *tanh(C t ),

[0039] Where W f , W i , W o , W C They represent the weights learned by individual neurons at different positions during the calculation process, and b f , b i , b o , b C They represent the offset of a single neuron at different positions during the calculation process. Tanh is the activation function of the neural network, and its function is defined as follows:

[0040]

[0041]

[0042] Furthermore, the specific steps of acquiring the data depth feature in step S4 are as follows:

[0043] S41: Construct training samples and feature extraction network;

[0044] S42: Bring the training sample data into the feature extraction network to obtain the data deep features.

[0045] Furthermore, it also includes an optimization step for the next prediction result of the deep migratable multi-factor accident warning for offshore oil drilling. The optimization step optimizes and improves the next prediction result through the current prediction result, as follows:

[0046] S61: Construct multi-factor detector;

[0047] S62: Perform detector training and calculate the loss function of the prediction result using the deep features;

[0048] S63: In the loss function, a coefficient function for dynamically adjusting the sample loss weight is added;

[0049] S64: Optimizing the prediction model through a coefficient function to improve the accuracy of the prediction result.

[0050] The present invention also provides an accident warning system for offshore oil drilling, comprising a correlation analysis module, a key factor screening module, a depth feature description module, and a data prediction module which are electrically connected in sequence.

[0051] Preferably, it also includes a multi-factor detection module that can be used to optimize the prediction model, and the multi-factor detection module is electrically connected to the data prediction module and the correlation analysis module.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] (1) Comprehensively consider the relationship between multiple single attribute factors in offshore oil drilling data and the information of multiple single attribute factors in the time series dimension, and use deep learning methods to express the three-dimensional, deep, and highly semantic features of the data, which greatly reduces the false alarm rate and improves the credibility of the warning;

[0054] (2) Using key semantic information near the point where an offshore oil drilling accident occurs, we extract the abnormal data features before the offshore oil drilling accident occurs, effectively link the abnormal fluctuations of key factors with the occurrence of offshore oil drilling accidents, and realize intelligent early warning without human intervention;

[0055] (3) Based on the prediction and detection of multiple key factors, a deeply transferable multi-factor accident prediction model was trained using deep learning methods, which improved the transferability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of an accident warning method for offshore oil drilling according to the present invention;

[0057] Figure 2 This is a structural block diagram of an accident warning for offshore oil drilling according to the present invention;

[0058] Figure 3 This is a schematic diagram of the results of the correlation analysis between offshore oil drilling accidents and each single attribute factor;

[0059] Figure 4 This is the experimental result diagram of model migration on Well B;

[0060] Figure 5 This is the experimental result diagram of model migration on Well C; DETAILED DESCRIPTION

[0061] The present invention is further described below in conjunction with specific implementation methods. The accompanying drawings are only used for exemplary descriptions and are only schematic diagrams, not actual drawings, and cannot be understood as limiting this patent; in order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the accompanying drawings may be omitted.

[0062] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0063] Example 1

[0064] like Figure 1 The first embodiment of the accident warning method for offshore oil drilling of the present invention is shown, which includes the following specific steps:

[0065] S1: Based on the correlation analysis between offshore oil drilling accidents and the single attribute factors of each drilling data in the time series dimension, the correlation degree between the accident and each single attribute factor is determined;

[0066] S11: Obtaining data distribution before and after the offshore oil drilling accident and classifying them to obtain a data set before and after the offshore oil drilling accident;

[0067] S12: Based on the principle that the data distribution before and after the offshore oil drilling accident is different, the time distance is used as the measurement standard to determine the distinguishability of two data sets;

[0068] The statement based on the principle that the data distribution before and after the offshore oil drilling accident is different is as follows:

[0069]

[0070] Among them, S is the time series, e i For an offshore oil drilling accident, represents the time series S in the offshore oil drilling accident e i The previous subsequence of length k; Indicates that the time series S is in the event e i The following subsequence of length k, Γ front is the time subseries data set before the offshore oil drilling accident, Γ rear It is a time subseries data set after an offshore oil drilling accident occurs, that is, the data distribution before and after the offshore oil drilling accident should be different.

[0071] The criterion for determining whether two data distributions are the same is distance. Based on this criterion, the following formula for determining the distinguishability of two data sets is defined:

[0072]

[0073] I r (x,A1,A2)=1 when x∈A i &&NN r (x,A)∈A i ,A=A1∪A2,

[0074] I r (x,A1,A2)=0 when otherwise;

[0075] Where A1 is the set of time series before or after an offshore oil drilling accident, A2 is the set of subsequences randomly sampled from the entire time series data, x is the time series in the data set before or after an offshore oil drilling accident, for A = A1 ∪ A2, x∈A, NN r (x,A) represents the rth element in A-{x} that is closest to x; I r (x, A1, A2) indicates whether x and the rth closest element to x are in the same subset; T r,p represents the probability that A1 and A2 do not belong to the same data distribution, where p is the size of set A, r represents the index of the rth nearest element of x, if T r,p If T is small, it means that the two types of data A1 and A2 have no distinction; r,p If it is large, it means that the two types of data A1 and A2 are distinguishable.

[0076] S13: Determine the correlation score between offshore oil drilling accidents and the single attribute factors of each drilling data through time series and distinguishability, and then determine the degree of correlation.

[0077] S2: According to the degree of correlation between offshore oil drilling accidents and each single attribute factor, multiple factors with high correlation are selected as key factors;

[0078] S21: sorting the correlation scores of the single attribute factors of each drilling data from high to low;

[0079] S22: Setting a correlation score threshold, and then filtering out all single attribute factors of the drilling data that are higher than the correlation score threshold as key factors.

[0080] Select K single attribute factors with the highest correlation scores as key factors;

[0081] S3: Using the long short-term memory network, predict the data trends of multiple key factors selected in step S2;

[0082] S31: Divide memory units: Divide the long and short time series memory network into multiple memory units;

[0083] The long short sequence memory network is used to predict the data trends of multiple key factors. The long short sequence memory network is composed of multiple memory units. The number of memory units in the network varies according to the amount of data predicted at each moment.

[0084] S32: Predict the memory unit: predict each memory unit of each key factor, the formula is as follows:

[0085] (h t ,C t )=H(h t-1 ,x t ,C t-1 ),

[0086] where h t-1 is the output data of the memory unit at the previous moment, C t-1 is the memory state of the memory unit at the last moment, x t is the data of each key factor input at the current moment, h t is the predicted data of the memory unit at the next moment, C t is the current memory state of the memory unit;

[0087] And h t , C t The specific calculation formula is:

[0088] f t =σ(W f [h t-1 ,x t ]+b f ),

[0089] i t =σ(W i [h t-1 ,x t ]+b i ),

[0090] o t =σ(W o [h t-1 ,x t ]+b o ),

[0091]

[0092]

[0093] h t =o t *tanh(C t ),

[0094] Where W f , W i , W o , W C They represent the weights learned by individual neurons at different positions during the calculation process, and b f , b i , b o , b C They represent the offset of a single neuron at different positions during the calculation process. Tanh is the activation function of the neural network. The function f is defined as follows:

[0095]

[0096]

[0097] S33: Obtaining the data trend of key factors: The prediction results of each memory unit are collected to obtain the prediction information of each key factor, and then the data trends of all key factors are obtained by summarizing them.

[0098] S4: Comprehensively consider the relationship between multiple key factors and the information of multiple key factors in the time series dimension, combine the predicted data in step S3, and use the deep learning network to express the deep features of the data;

[0099] S41: construct training samples;

[0100] Using the K key factors determined in step S22, for each key factor, w consecutive data points are selected in the time series to construct a sample, and the sample size is K*w, wherein the data of each w consecutive time points of each key factor is taken as a sample, and the time interval between data points is determined according to the sampling frequency of the source data;

[0101] Construct feature extraction network;

[0102] The feature extraction network consists of a two-layer convolutional layer with convolution kernel sizes of 7 and 3 respectively. The output of each convolutional layer passes through an activation function. Finally, the output of the two convolutions passes through a maximum pooling layer with a convolution kernel of 2 to complete the extraction of multiple key factors - time series depth features;

[0103] S42: Bring the training sample data into the feature extraction network to obtain the data depth features;

[0104] Data depth features refer to three-dimensional semantic features that comprehensively consider the connections between multiple key factors and the depth of temporal information.

[0105] S5: Based on the deep features extracted in step S4, a transferable multi-factor detector is trained to achieve deep transferable multi-factor accident prediction for offshore oil drilling.

[0106] Furthermore, the method also includes steps for optimizing the next prediction result of the deep migratable multi-factor accident warning for offshore oil drilling, which are as follows:

[0107] S61: Construct multi-factor detector;

[0108] The transferable multi-factor detector in this scheme consists of two layers of continuous neural networks, with 784 and 64 neurons respectively;

[0109] S62: Perform detector training and use deep features to calculate the loss function of the prediction results;

[0110] The detector is trained using training samples, and back propagation is used to guide the network to extract deep features for detecting accidents before they occur. The loss function used to calculate the prediction results during back propagation is defined as:

[0111]

[0112] where y i represents the label of the sample, y i The value is 0 or 1, (0 is a negative sample, indicating data containing accidents, 1 is a positive sample,) and p i It represents the probability that a sample is predicted to be a positive sample. The value of pi is a decimal between 0 and 1. N represents the number of predicted samples.

[0113] S63: In the loss function, add a coefficient function for dynamically adjusting the sample loss weight;

[0114] The purpose of dynamically adjusting the coefficient function of sample loss weight is to balance the extreme imbalance of samples. The function is defined as:

[0115] R = {r i |1≤i≤n},r max =max R

[0116]

[0117] R represents the number of samples of different categories, where r i represents the number of samples in the i-th category, max is the function for finding the maximum value, B represents the weight coefficient of the loss of samples in different categories, L i Represents the loss of the prediction result of the i-th class sample;

[0118] S64: Optimize the prediction model through the coefficient function to improve the accuracy of the prediction result.

[0119] The above optimization steps are applied in the early training mode to adjust the accuracy of the prediction results, reduce the deviation between the prediction results and the actual results, optimize the entire prediction model, and further reduce the deviation between the prediction results and the actual results in subsequent actual use.

[0120] Example 2

[0121] The following is a second embodiment of an accident warning method for offshore oil drilling of the present invention. This embodiment is similar to embodiment 1, except that the method for extracting key factors in step S1 is different. The following method can be used to replace steps S11 to S13 to extract key factors.

[0122] The formula is as follows:

[0123]

[0124] Among them, X t is the time series data used to calculate autocorrelation, X t-1 represents the time series at time t-1, n is the length of the time series, μ is the mean of the time series, σ 2 is the variance of the time series, l represents the lag value, and the value of the entire equation is the correlation score of the single attribute factor; if the autocorrelation calculated by this formula in the time series near the occurrence of the offshore oil drilling accident is large, it means that the correlation level with the offshore oil drilling accident is low; if the autocorrelation calculated by this formula in the time series near the occurrence of the offshore oil drilling accident is small, it means that the correlation level with the offshore oil drilling accident is high.

[0125] It should be noted that this embodiment is described on the basis of the above-mentioned embodiment 1. In this embodiment 2, the correlation between the offshore oil drilling accident mentioned in embodiment 1 and each single attribute factor is determined. For data on known offshore oil drilling accidents, the time series data before and after the offshore oil drilling accident is taken for autocorrelation analysis. The autocorrelation coefficient of each single attribute factor can be calculated according to the above formula, thereby determining the correlation between the offshore oil drilling accident and each single attribute factor.

[0126] Example 3

[0127] The following is a third embodiment of an accident warning method for offshore oil drilling of the present invention. This embodiment is similar to the first embodiment, except that in step S2, the method for determining the key factor is different. The following method can be used to replace steps S21 to S22. Specifically, different weights are assigned to different single attribute factors in combination with human experience, and a higher weight is assigned to the single attribute factor with a higher credibility associated with the offshore oil drilling accident, and a lower weight is assigned to the single attribute factor with an uncertain credibility associated with the offshore oil drilling accident. Then, the weight value of the sum of different single attribute factor sets is calculated. The key factor can be determined by the following formula:

[0128]

[0129] Among them, H is the set sum of all single attribute factors of data, H k is a subset of H, k represents the number of single attribute factors in the subset, which is a fixed constant, and h i is a single attribute factor, p is a weight function, p(h i ) represents the single attribute factor h i The weight of , max is the function to find the maximum value.

[0130] Example 4

[0131] The following is a fourth embodiment of an accident warning method for offshore oil drilling of the present invention. This embodiment is similar to the first embodiment, except that the method for obtaining the data trend of the key factor in step S3 is different. The following method can be used to replace steps S31 to S33:

[0132] Sa: Use recurrent neural networks to predict the data trends of various key factors. Recurrent neural networks are composed of neurons. The number of memory units in the network varies according to the amount of data predicted at each moment. The prediction formula for a single memory unit is defined as:

[0133] h t =F(h t-1 ,x t ),

[0134] where h t-1 is the output of the neuron at the previous moment, x t is the data of each key factor input at the current moment, h t The data of the next moment predicted for the memory unit;

[0135] Among them, h t The specific calculation formula is:

[0136] h t =tanh(W h [h t-1 ,x t]+b h )

[0137] Where W h , b h is the weight and offset learned by a single neuron, and tanh is the activation function of the neural network, which is defined as follows:

[0138]

[0139] Sb: Collect the outputs of each neuron in step Sa to get the data trends of all key factors.

[0140] The following is the effectiveness of the technical solution of the present invention using real offshore oil drilling data. The stuck pipe accident was selected from the more types of accidents that occurred in offshore oil. There are three drilling data of A, B, and C with stuck pipe accidents. Each well data contains 39 different single attribute factors. The correlation analysis results between the stuck pipe accident and each single attribute factor are shown in the attached figure. Figure 3 As shown:

[0141] Select the 14 single attribute factors {x1,x2,…,x 14}As the key factors, they mainly include TVA mud pool volume, MFIA inlet mud flow, MFOP mud pool flow, SPPA pump pressure, BPOS big hook height, MTIA inlet mud temperature, TVCA mud pool volume change, HKLA big hook hanging weight, TQA torque, RPM A turntable speed, DBTM drill bit depth and other factors.

[0142] The 14 selected single attribute factors are used to construct data samples on three wells A, B, and C. Then the model is trained on the data of well A, and the model is directly transferred to wells B and C for experiments. The experimental results are as follows: Figure 4 , Figure 5 As shown, Figure 4 and Figure 5 The horizontal axis in the middle represents the number of times the model is trained on the training set, and the vertical axis represents the test accuracy on the test set after each model training.

[0143] In the experiments on data B and C, a period of data before a real offshore oil drilling accident occurred was selected to test the model. The experimental results are shown in the figure. The model detected the occurrence of the accident before the real offshore oil drilling accident occurred and gave a warning, which illustrates the effectiveness of the present invention.

[0144] Example 5

[0145] like Figure 2The figure shows an embodiment of an accident warning system for offshore oil drilling of the present invention, which includes a correlation analysis module, a key factor screening module, a deep feature description module, and a data prediction module which are electrically connected in sequence.

[0146] The correlation analysis module is used to perform correlation analysis on offshore oil drilling accidents and single attribute factors of drilling data in the time series dimension to determine the degree of correlation between offshore oil drilling accidents and each single attribute factor; the key factor screening module is used to screen multiple single attribute factors according to the degree of correlation between offshore oil drilling accidents and each single attribute factor to determine the key factors; the deep feature description module is used to use the deep learning network to construct three-dimensional, deep, and highly semantic features between key factors and multiple key factors in the time series dimension; the data prediction module is used to predict the data trend of key factors with the help of long and short time series memory networks and obtain prediction results;

[0147] As an embodiment of the present invention, it also includes a multi-factor detection module that can be used to optimize the prediction model. The multi-factor detection module is electrically connected to the data prediction module and the correlation analysis module.

[0148] The multi-factor detection module is used to train a transferable multi-factor detector to further improve the accuracy of the prediction results and reduce the deviation between the prediction results and the actual results. The data prediction module feeds the prediction results back to the multi-factor detection module. The multi-factor detection module compares the prediction results with the actual results, and then adjusts them through the coefficient function of the loss function and the loss weight. The adjustment results are fed back to the correlation analysis module, so that when the data is predicted next time, the key factors are screened more strictly from the source, and the correlation scores of the factors are redefined to improve the accuracy of subsequent prediction results. In this process, the prediction model can be optimized to achieve deep transferable multi-factor accident prediction for offshore oil drilling. Through the multi-factor detection module, a stable detector with good migration ability that can detect data anomaly features can be trained to improve the accuracy of the prediction results, and the function of the prediction model can be further improved, so that the prediction model can detect anomalies before offshore oil drilling accidents in various complex environments.

[0149] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. An accident warning method for offshore oil drilling, characterized in that: The specific steps include: S1: Based on the correlation analysis between offshore oil drilling accidents and the single attribute factors of each drilling data in the time series dimension, the correlation degree between the accident and each single attribute factor is determined; In step S1, the correlation score is used to measure the correlation degree between offshore oil drilling accidents and each single factor. The specific steps are as follows: S11: Obtaining data distribution before and after the offshore oil drilling accident and classifying them to obtain a data set before and after the offshore oil drilling accident; S12: Based on the principle that the data distribution before and after the accident is different, the temporal distance is used as the measurement standard to determine the distinguishability of the data set before the offshore oil drilling accident and the data set after the offshore oil drilling accident; The distinguishability determination formula of the data set in step S12 is as follows: I r (x,A1,A2)=1 when x∈A i &&NN r (x,A)∈A i ,A=A1∪A2, I r (x,A1,A2)=0 when otherwise, Among them, A1 is the set of time series before or after the offshore oil drilling accident, A2 is the set of subsequences randomly sampled from the entire time series data, x is the time series in the data set before the offshore oil drilling accident or the data set after the offshore oil drilling accident, for A = A1A2, x∈A, NN r (x,A) represents the rth element in A-{x} that is closest to x; I r (x, A1, A2) indicates whether x and the rth closest element to x are in the same subset; T r,p represents the probability that A1 and A2 do not belong to the same data distribution, where p is the size of set A, r represents the index of the rth nearest element of x, if T r,p If T is small, it means that the two types of data A1 and A2 have no distinction; r,p If it is large, it means that the two types of data A1 and A2 are distinguishable; S13: Determine the correlation score between offshore oil drilling accidents and single attribute factors of each drilling data through time series and distinguishability, and then determine the degree of correlation; S2: According to the degree of correlation between offshore oil drilling accidents and each single attribute factor, multiple factors with high correlation are selected as key factors; The specific steps for obtaining the key factors in step S2 are as follows: S21: sorting the correlation scores of the single attribute factors of the drilling data from high to low; S22: setting a correlation score threshold, and then screening out all single attribute factors of the drilling data that are higher than the correlation score threshold as key factors; S3: Using the long short-term memory network, predict the data trends of multiple key factors selected in step S2; The specific steps for obtaining the data trend of the key factors in step S3 are as follows: S31: Divide memory units: Divide the long and short time series memory network into multiple memory units; S32: Predicting the memory unit: predicting each memory unit of each key factor; S33: Obtaining the data trend of the key factors: The prediction information of each key factor can be obtained by collecting the prediction information of each memory unit, and then the data trend of all key factors is obtained by summarizing; S4: Comprehensively consider the relationship between multiple key factors and the information of multiple key factors in the time series dimension, combine the predicted data in step S3, and use the deep learning network to express the deep features of the data; S5: Based on the deep features extracted in step S4, a transferable multi-factor detector is trained to achieve deep transferable multi-factor accident prediction for offshore oil drilling.

2. The accident warning method for offshore oil drilling according to claim 1, characterized in that: The prediction formula of the memory unit in step S32 is as follows: (h t ,C t )=Η(h t-1 ,x t ,C t-1 ) where h t-1 is the output data of the memory unit at the previous moment, C t-1 is the memory state of the memory unit at the last moment, x t is the data of each key factor input at the current moment, h t is the predicted data of the memory unit at the next moment, C t The current memory state of the memory unit.

3. The accident warning method for offshore oil drilling according to claim 2 is characterized in that: The h t , C t The specific calculation formula is: f t =σ(W f [h t-1 ,x t ]+b f ), i t =σ(W i [h t-1 ,x t ]+b i ), the t =σ(W o [h t-1 ,x t ]+b o ), h t =o t *tanh(C t ), Where W f , W i , W o , W C They represent the weights learned by individual neurons at different positions during the calculation process, and b f 、b i 、b o 、b C They represent the offset of a single neuron at different positions during the calculation process. Tanh is the activation function of the neural network, and its function is defined as follows:

4. The accident warning method for offshore oil drilling according to claim 3 is characterized in that: The specific steps of acquiring data depth features in step S4 are as follows: S41: Construct training samples and feature extraction network; S42: Substitute the training sample data into the feature extraction network to obtain data depth features.

5. The accident warning method for offshore oil drilling according to claim 1, characterized in that: It also includes an optimization step for the next prediction result of the deep migratable multi-factor accident warning for offshore oil drilling. The optimization step optimizes and improves the next prediction result through the current prediction result, as follows: S61: Construct multi-factor detector; S62: Perform detector training and calculate the loss function of the prediction result using the deep features; S63: In the loss function, a coefficient function for dynamically adjusting the sample loss weight is added; S64: Optimize the prediction model through the coefficient function to improve the accuracy of the prediction result.

6. An accident warning system for offshore oil drilling, characterized in that: The accident warning method for offshore oil drilling as described in any one of claims 1 to 5, wherein the warning system includes a correlation analysis module, a key factor screening module, a deep feature description module, and a data prediction module which are electrically connected in sequence.

Images