Well drilling daily report analysis method, device and system based on long short-term memory network

Abstract

The invention relates to the technical field of well drilling data analysis, in particular to a well drilling daily report analysis method, device and system based on a long short-term memory network, and the method comprises the steps: carrying out the keyword extraction of a to-be-analyzed well drilling log, and obtaining keyword information; and inputting the keyword information into the neural network classification model to obtain a complex accident classification result. According to the method, the actual experience of drilling complex accident processing and the natural language processing related model in the drilling implementation process are combined, the neural network classification model is obtained through long-term and short-term memory network training, classification of drilling daily complex accident information is achieved based on the neural network classification model, the classification accuracy is high, and the classification efficiency is high. And the workload of classified statistics of complex accidents and compilation of weekly and monthly report materials is greatly reduced, the oilfield yield management cost is reduced, and a digital support is provided for cost reduction and benefit increase of oilfield productivity construction.

Description

Technical Field

[0001] The present invention relates to the technical field of drilling data analysis, in particular to a drilling daily report analysis method, device and system based on a long short-term memory network. Background Art

[0002] As one of the important links in the process of capacity construction implementation, drilling engineering has a great impact on the overall progress of capacity well commissioning. Therefore, drilling engineering is the top priority for production and operation personnel to track the progress. Complex drilling accidents, as a special situation encountered in drilling engineering, have a great impact on the drilling and completion period. Production and operation personnel will classify and count complex accidents, and extract key information of complex accident information in the preparation of weekly and monthly reports.

[0003] Existing technical solution 1: At present, the method for processing complex accident information in the daily drilling report is mainly to upload it through the information system or to collect and manage it manually after filling it out. In general, the business process is online. Compared with the completely manual processing method, it assists business personnel in the relevant statistics and material compilation of the progress of drilling implementation to a certain extent.

[0004] The disadvantages of the existing technical solution 1: There is a lack of intelligent analysis methods for daily drilling reports. Therefore, the efficiency is low and it is difficult to meet the actual needs of production operators and drilling professional managers. Since complex accidents are usually recorded by construction personnel and the format is not fixed, production operators need to manually and repeatedly process the complex accident information before they can count the accident types and compile relevant materials, resulting in huge labor losses to complete repetitive work.

[0005] Existing technical solution 2: Although there are no application cases of related technologies in the natural language processing of drilling daily reports, using RNN for natural language analysis in other fields is a relatively common solution. The purpose of CNN is to extract features of things with a certain model, and then classify, identify, predict or make decisions based on the features. The most important step is feature extraction, that is, how to extract features that can distinguish things to the greatest extent, and iterative training, so as to perform language processing of daily reports and similar documents through feature extraction.

[0006] The disadvantages of the second existing technical solution are: 1) When the number of network layers is too deep, the use of backpropagation to adjust the internal parameters will make the changes close to the input layer slower; 2) When gradient descent is used for iteration, it is easy for the training results to converge to the local optimum rather than the global optimum; 3) The pooling layer will lose some valuable information and ignore the correlation between the local and the whole; 4) The physical meaning of feature extraction is not very clear, resulting in general interpretability.

[0007] Existing technical solution three: Although there are no application cases of related technologies in the natural processing of drilling daily reports, using RNN for natural language analysis in other fields is also a relatively common solution. Recurrent neural network (RNN) is a neural network used to deal with time-dependent problems and similar studies. In text analysis, it can be mapped to the connection between contexts. Compared with other neural networks, it has a good processing ability for data that is in a fluctuating state in time series. For example, the meaning and content of a certain text will change due to the different derivatives of the previous meaning, and the final result must be different. RNN can solve this kind of problem very well. This is very practical for daily report analysis.

[0008] Disadvantages of the existing technical solution three: RNN fails to pay attention to the problems of gradient explosion (too much stored information) and vanishing (too long a time interval between key information and target units) that may occur during long-term training. Gradient vanishing will cause the network weights of the previous layers in our neural network to be unable to be updated, and learning will stop. Gradient explosion will make learning unstable, and the parameters will change too much to make it impossible to obtain the optimal parameters. In deep multi-layer perceptron networks, gradient explosion will cause network instability. The best result is that it is impossible to learn from the training data. The worst result is that the weights cannot be updated due to the NaN weight value. In recurrent neural networks (RNNs), gradient explosion will cause network instability, making it impossible for the network to learn well from the training data. The best result is that the network cannot learn on long input data sequences. Summary of the invention

[0009] The present invention provides a drilling daily report analysis method, device and system based on long short-term memory network, which overcomes the shortcomings of the above-mentioned prior art and can effectively solve the problems of low efficiency and easy errors in the existing method of manually analyzing complex accidents in drilling daily reports.

[0010] One of the technical solutions of the present invention is achieved by the following measures: A drilling daily report analysis method based on long short-term memory network, comprising: Extract keywords from the drilling log to be analyzed to obtain keyword information; The keyword information is input into a neural network classification model to obtain a complex accident classification result, wherein the neural network classification model is obtained through machine learning training using multiple sets of data, and each set of data in the multiple sets of data includes: a drilling log and a label identifying the complex accident classification result in the drilling log.

[0011] The following are further optimizations and / or improvements to the above technical solutions: The above keyword extraction is performed on the drilling log to be analyzed to obtain keyword information, including: Obtain the drilling log to be analyzed; Use the word segmentation tool to perform word segmentation on the drilling log to be analyzed; Use the word vector model Word2vec to transform each word after word segmentation, obtain the word vector representation corresponding to each word, and form a word segmentation set; The keyword extraction algorithm is used to extract keyword information from the word segmentation set.

[0012] The keyword extraction algorithm used in extracting keyword information from a word segmentation set using a keyword extraction algorithm is a TextRank algorithm.

[0013] The construction process of the above neural network classification model includes: Obtain several historical drilling logs, extract keywords from each historical drilling log and identify the corresponding complex accident classification results to form a sample library; The historical drilling logs in the sample library are divided into training samples and test samples in proportion; The long short-term memory network model is trained using the training samples to obtain a neural network classification model; The neural network classification model is tested using the test samples. If the test result does not meet the set threshold, the parameters of the neural network classification model are adjusted and the training is repeated until the optimal neural network classification model is obtained.

[0014] The above keyword extraction and corresponding complex accident classification result identification for each historical drilling log include: Use the word segmentation tool to perform word segmentation on the drilling log to be analyzed; Use the word vector model Word2vec to transform each word after word segmentation, obtain the word vector representation corresponding to each word, and form a word segmentation set; Use keyword extraction algorithm to extract keyword information from the word segmentation set; The labels are used to identify the miscellaneous accident classification results corresponding to the historical drilling log.

[0015] The second technical solution of the present invention is achieved by the following measures: a drilling daily report analysis device based on long short-term memory network, comprising: A data acquisition unit extracts keywords from the drilling log to be analyzed to obtain keyword information; The classification unit inputs the keyword information into the neural network classification model to obtain a complex accident classification result, wherein the neural network classification model is obtained by machine learning training using multiple sets of data, and each set of data in the multiple sets of data includes: a drilling log and a label identifying the complex accident classification result in the drilling log.

[0016] The following are further optimizations and / or improvements to the above technical solutions: The data acquisition unit includes: The initial data acquisition module acquires the drilling logs to be analyzed; The first preprocessing module uses a word segmentation tool to perform word segmentation on the drilling log to be analyzed; The second preprocessing module uses the word vector model Word2vec to convert each word after word segmentation, obtains the word vector representation corresponding to each word, and forms a word segmentation set; The third preprocessing module uses a keyword extraction algorithm to extract keyword information from the word segmentation set.

[0017] The third technical solution of the present invention is achieved by the following measures: a drilling daily report analysis system based on long short-term memory network, comprising: A daily drilling report analysis device based on a long short-term memory network, wherein the daily drilling report analysis device based on a long short-term memory network is the daily drilling report analysis device based on a long short-term memory network as described above; An interactive device is used for an operator to communicate with a daily drilling analysis device based on a long short-term memory network, so that the operator provides the daily drilling analysis device based on a long short-term memory network with initial data to be predicted and parameter setting instructions through an interactive unit.

[0018] The present invention combines the actual experience of handling complex drilling accidents during the implementation of drilling and the relevant models of natural language processing, and uses long short-term memory network training to obtain a neural network classification model. Based on the neural network classification model, the classification of complex accident information in the daily drilling report is realized. The classification has high accuracy and greatly reduces the workload of classification statistics of complex accidents and preparation of weekly and monthly reporting materials, reduces the cost of oilfield production management, and provides digital support for cost reduction and efficiency improvement of oilfield capacity construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Attached Figure 1 The figure is a flow chart of the method of the present invention.

[0020] Attached Figure 2 This is a flow chart of the method for keyword information in the present invention.

[0021] Attached Figure 3 It is a structural diagram of the LSTM model in the present invention.

[0022] Attached Figure 4 It is a schematic diagram of the device structure of the present invention.

[0023] Attached Figure 5 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0024] The present invention is not limited by the following embodiments, and specific implementation methods can be determined based on the technical solution of the present invention and actual conditions.

[0025] The present invention will be further described below in conjunction with embodiments and drawings: Embodiment 1: As attached Figure 1 As shown, the embodiment of the present invention discloses a drilling daily report analysis method based on a long short-term memory network, comprising: Step S110, extracting keywords from the drilling log to be analyzed to obtain keyword information.

[0026] As attached Figure 2 As shown, the above step S110 includes: Step S111, obtaining the drilling log to be analyzed; Step S112, using a word segmentation tool to perform word segmentation on the drilling log to be analyzed; the word segmentation tool here can be the Jieba Chinese word segmentation tool, which mainly implements efficient word graph scanning based on a prefix dictionary, generates a directed acyclic graph (DAG) consisting of all possible word formation situations of Chinese characters in a sentence, uses dynamic programming to find the maximum probability path, finds the maximum segmentation combination based on word frequency, and for unregistered words, uses an HMM model based on the word formation ability of Chinese characters, and uses the Viterbi algorithm for word segmentation.

[0027] For example, the drilling log to be analyzed is: Second opening work (complex): 08:00-9:10 circulation (adjust mud)-11:30 lifting the drill-12:00 electrical measurement preparation (observation found a small amount of oil floating in the drilling fluid at the wellhead, secondary well cleaning, mud treatment)-13:30 drilling to 407m-18:00 circulation (adjust mud)-8:00 treatment of well leakage is complex (18:00-20:30 lifting drilling fluid, preparing plugging slurry-21:00 plugging slurry, a small amount of slurry returned, gradually no slurry returned, the wellhead liquid level is visible-2:50 static plugging, the wellhead liquid level is visible, and drilling fluid is configured-3:10 small displacement circulation, no slurry returned at the outlet, 4.8m3 of drilling fluid was lost-2:50 static plugging, the wellhead liquid level is visible-5:00 lifting the drill 7 columns, the drill bit position is 178m-6:00 active drilling tools, observe the wellhead liquid level Visible - 8:00 Push method to treat contaminated drilling fluid in the well), use the Jieba Chinese word segmentation tool for word segmentation, the results after word segmentation are: two, open, mid-completion, operation, complex, circulation, adjustment, mud, drill lifting, electrical measurement preparation, observation, discovery, wellhead, drilling fluid, floating, a small amount of oil flowers, secondary, well passing, treatment, mud, drilling, circulation, adjustment mud, treatment, well leakage complex, hanging irrigation, drilling fluid, matching plugging, leaking slurry, plugging, leaking slurry, a small amount, return slurry, gradually, no, return slurry, wellhead, liquid level, visible, static, plugging, wellhead, liquid level, visible, configuration, drilling fluid, small displacement, circulation, outlet, no return slurry, leakage, drilling fluid, static, plugging, wellhead, liquid level, visible, drill lifting, column, drill bit, position, to, activity, drilling tool, observation, wellhead, liquid level, visible, push, method, treatment, well, pollution, drilling fluid.

[0028] Step S113, using the word vector model Word2vec to convert each word after the word segmentation processing, obtain the word vector representation corresponding to each word, and form a word segmentation set; This step converts word vectors. Word vectors are a preprocessing technique used in natural language processing problems. They extract effective classification information (features) and then create a vector library through corresponding feature learning techniques. After that, the text and words are mapped. The purpose is to process the space of vocabulary dimensions, thereby achieving vocabulary embedding with reduced dimensions and more continuous expressions between words in space.

[0029] The word vector model Word2vec used in the present invention helps to better represent data: words that are similar to each other can have more similar vector structures, while words with different parts of speech or different characteristics are very different and have different vectors. Specifically, each word is represented as an appropriate dimension, which actually means that each word has multiple attributes of the dimension to represent the word. For example: the 0th dimension represents the part of speech (verb, noun, adjective), the 1st dimension represents the emotional attribute (negative, positive), and the 2nd dimension represents the commendatory part of speech... In this way, the dimension of the word can be reduced and better put into the model for learning.

[0030] Step S114, extracting keyword information from the word segmentation set using a keyword extraction algorithm.

[0031] The present invention can use the TextRank algorithm to extract keyword information from the word segmentation set. The TextRank algorithm regards text data as nodes of a graph and uses the relationship between texts to establish an adjacency matrix. According to different goals, TextRank can realize two types of tasks: keyword extraction and key sentence extraction. The process of the TextRank algorithm includes: clarifying the task goal, and adding the text unit corresponding to the task as a node (vertice) of the graph; adding the relationship between the text units as an edge (edge) connecting the nodes in the graph, which can be a directed edge or an undirected edge, and can be a weighted edge or an unweighted edge. At this time, the establishment of the adjacency matrix is ​​completed; iterate the TextRank algorithm until convergence, and calculate the score of each node; sort the nodes according to the final score, and extract the top-k as keywords or key sentences according to the sorting result. Generally speaking, for the keyword extraction task, the text unit is each word after the sentence is segmented, and the adjacency matrix is ​​the number of times the word and the word appear in adjacent positions (normalized); for the key sentence extraction task, the adjacency matrix considers the text similarity between sentences.

[0032] For example, the word set is open, mid-completion, operation, complex, circulation, adjustment, mud, drill lifting, electrical measurement preparation, observation, discovery, wellhead, drilling fluid, floating, a small amount of oil flowers, secondary, well passing, treatment, mud, drilling, circulation, adjustment mud, treatment, complex well leakage, hanging irrigation, drilling fluid, matching plugging, leaking slurry, plugging, leaking slurry, a small amount, return slurry, gradually, no, return slurry, wellhead, liquid level, visible, static, plugging, wellhead, liquid level, visible, configuration, drilling fluid, small displacement, circulation, outlet, no return slurry, leakage, drilling fluid, static, plugging, wellhead, liquid level, visible, drill lifting, column, drill bit, position, to, activity, drilling tool, observation, wellhead, liquid level, visible, push, method, treatment, in the well, pollution, drilling fluid. The TextRank algorithm is used to extract keyword information from the word set, and the extracted keyword information includes wellhead, drilling fluid, leaking slurry, complex, liquid level, plugging, treatment, and pollution.

[0033] Step S120, inputting the keyword information into the neural network classification model to obtain a complex accident classification result, wherein the neural network classification model is obtained by machine learning training using multiple sets of data, and each set of the multiple sets of data includes: a drilling log and a label identifying the complex accident classification result in the drilling log.

[0034] In the above steps, the construction process of the neural network classification model includes: (1) Obtain a number of historical drilling logs, extract keywords from each historical drilling log, and identify the corresponding complex accident classification results to form a sample library; Here, keywords are extracted from each historical drilling log and the corresponding complex accident classification results are marked, including: Use the word segmentation tool to perform word segmentation on the drilling log to be analyzed; Use the word vector model Word2vec to transform each word after word segmentation, obtain the word vector representation corresponding to each word, and form a word segmentation set; Use keyword extraction algorithm to extract keyword information from the word segmentation set; The labels are used to identify the miscellaneous accident classification results corresponding to the historical drilling log.

[0035] (2) The historical drilling logs in the sample library are divided into training samples and test samples in proportion.

[0036] (3) Use the training samples to train the long short-term memory network model to obtain the neural network classification model.

[0037] (4) Use the test samples to test the neural network classification model. If the test results do not meet the set threshold, adjust the parameters of the neural network classification model and retrain until the optimal neural network classification model is obtained.

[0038] The present invention uses a bidirectional LSTM to classify the preprocessed information. As a very special RNN, the long short-term memory network LSTM is mainly used to solve the problem of gradient disappearance and corresponding gradient explosion that will occur in the training process of long-term and long-sequence related data. Compared with ordinary RNN, LSTM can achieve better results in longer sequences. Compared with RNN, which has only one parameter to be transmitted, LSTM has two transmission parameters or states, one is the hidden state and the other is the cell state. The state in RNN corresponds to the state in LSTM. Among them, for the cell state transmitted to the next moment, it changes very slowly. Usually, what is output to the next moment is based on the previous moment plus some values. The hidden state is often very different at different nodes. In fact, the cell state is for long-term transmission, while the hidden state is for short-term information transmission.

[0039] As attached Figure 2 As shown, the LSTM model is composed of the cell state at time t C t ,enter X t , hidden state H t , temporary cell state C t , output gate O t , Memory Gate i t , the Forget Gate f t The internal calculation process of a LSTM cell can be summarized as: forgetting the unimportant information contained in the cell state corresponding to the current moment and memorizing the new and important information. Eventually, the information that is more important for the calculation of the next moment will be transmitted, while the useless information will be discarded. At each moment, the cell will output the hidden state, where the output, memory and forgetting are controlled by the output gate, memory gate and forget gate calculated by the current input and the hidden state of the previous moment.

[0040] There are three main stages inside LSTM: (a) Forget stage: This stage is mainly to selectively forget the input from the previous node. In simple terms, it means "forgetting the unimportant and remembering the important". Specifically, the forget gate is calculated to control which parts of the previous state need to be retained and which need to be forgotten.

[0041] (b) Selective memory stage. The work of this stage is to selectively "memorize" the information input at the current moment. In fact, it is to make selective judgments on the input. The important data are recorded, and the unimportant information is forgotten. The valid information obtained at the previous moment, that is, the information calculated at the previous moment, is also input. After that, the gate signal is controlled by the core of this stage: the memory gate. Finally, the corresponding results obtained in the above two steps are added together to get the result transmitted to the next moment.

[0042] (c) Output stage. The core work of this stage is to determine which information (data) will be output as important content of the current state through calculation. The core step of this work is to control it through the output gate. In this stage, the dimension reduction process of the data obtained in the previous stage is also performed (using an activation function: tanh for scaling). The common point with ordinary RNN is that the output is finally obtained by processing the changes of .

[0043] The update of the entire LSTM is very similar to that of the RNN, except that: the LSTM first determines what is not important and discards it, then determines what important information to obtain from the current input, adds it, and then produces the result based on the final memory. In general, the LSTM uses two gates to control the information that needs to be retained in a unit state. One is the forget gate, which can determine how much of each important information retained at the previous moment needs to be passed on to the current moment; the other is the input gate, which directly determines how much data input into the network needs to be saved at the current moment so as not to affect the current unit state.

[0044] The relevant parameter settings are as follows: Set dropout=0.5 Because the parameter set of the neural network is huge in the past, or if there is too little training data or even too many training times, a phenomenon called overfitting may occur. In fact, overfitting means that the assumptions become too strict in order to obtain consistent assumptions. This results in poor performance of generalized data.

[0045] Dropout is the most effective regularization method to avoid overfitting in neural networks. When dropout is used, the neural cell units in each layer will perform deep learning according to different probabilities, so that the trained network obtained each time will be different, that is, the batch_size will change, which is equivalent to training in a new model network. Dropout is essentially a way to fuse multiple models. When dropout is set to 0.5, the effect is the best, because at this time the total number of model categories is the largest, at this time, in addition, when updating related parameters, such as: weight matrix, hidden layer information, etc., only the neural cells that need to be retained are updated, which can also speed up the training.

[0046] It should also be noted that the sample library can be continuously updated. Each time the sample library is updated, the neural network classification model needs to be retrained using the new sample library to ensure the classification accuracy of the model.

[0047] The present invention discloses a daily drilling report analysis method based on a long short-term memory network. The method combines the actual experience of handling complex drilling accidents during the drilling process and the relevant models of natural language processing, and uses the long short-term memory network training to obtain a neural network classification model, so as to realize the classification of complex accident information in the daily drilling report, greatly reduce the workload of classification statistics of complex accidents and preparation of weekly and monthly report materials, reduce the cost of oilfield production management, and provide digital support for cost reduction and efficiency improvement of oilfield capacity construction.

[0048] Embodiment 2: As attached Figure 4 As shown, the embodiment of the present invention discloses a drilling daily report analysis device based on a long short-term memory network, comprising: A data acquisition unit extracts keywords from the drilling log to be analyzed to obtain keyword information; The classification unit inputs the keyword information into the neural network classification model to obtain a complex accident classification result, wherein the neural network classification model is obtained by machine learning training using multiple sets of data, and each set of data in the multiple sets of data includes: a drilling log and a label identifying the complex accident classification result in the drilling log.

[0049] The data acquisition unit includes: The initial data acquisition module acquires the drilling logs to be analyzed; The first preprocessing module uses a word segmentation tool to perform word segmentation on the drilling log to be analyzed; The second preprocessing module uses the word vector model Word2vec to convert each word after word segmentation, obtains the word vector representation corresponding to each word, and forms a word segmentation set; The third preprocessing module uses a keyword extraction algorithm to extract keyword information from the word segmentation set.

[0050] Embodiment 3: As attached Figure 5 As shown, the embodiment of the present invention discloses a daily drilling report analysis system based on a long short-term memory network, comprising: A daily drilling report analysis device based on a long short-term memory network, wherein the daily drilling report analysis device based on a long short-term memory network is the daily drilling report analysis device based on a long short-term memory network described in the above embodiment 2; An interactive device is used for an operator to communicate with a daily drilling analysis device based on a long short-term memory network, so that the operator provides the daily drilling analysis device based on a long short-term memory network with initial data to be predicted and parameter setting instructions through an interactive unit.

[0051] Embodiment 4: The embodiment of the present invention discloses a storage medium, on which a computer program readable by a computer is stored, and the computer program is configured to execute a daily drilling report analysis method based on a long short-term memory network when running.

[0052] The above storage medium may include, but is not limited to: a USB flash drive, a read-only memory, a mobile hard disk, a magnetic disk or an optical disk, and other media that can store computer programs.

[0053] Embodiment 5: The embodiment of the present invention discloses an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement a daily drilling report analysis method based on a long short-term memory network.

[0054] The processor may be a central processing unit (CPU), a general purpose processor, a digital signal processor (DSP), an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. It may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The memory may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory, a mobile hard disk, a magnetic disk or an optical disk.

[0055] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present application may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0056] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0057] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0058] The above technical features constitute the best embodiment of the present invention, which has strong adaptability and best implementation effect. Non-essential technical features can be added or reduced according to actual needs to meet the requirements of different situations.

Claims

1. A drilling daily report analysis method based on long short-term memory network, characterized in that: include: Extract keywords from the drilling log to be analyzed to obtain keyword information; The keyword information is input into a neural network classification model to obtain a complex accident classification result, wherein the neural network classification model is obtained through machine learning training using multiple sets of data, and each set of data in the multiple sets of data includes: a drilling log and a label identifying the complex accident classification result in the drilling log.

2. The drilling daily report analysis method based on long short-term memory network according to claim 1 is characterized in that: The keyword information obtained by extracting keywords from the drilling log to be analyzed includes: Obtain the drilling log to be analyzed; Use the word segmentation tool to perform word segmentation on the drilling log to be analyzed; Use the word vector model Word2vec to transform each word after word segmentation, obtain the word vector representation corresponding to each word, and form a word segmentation set; The keyword extraction algorithm is used to extract keyword information from the word segmentation set.

3. The drilling daily report analysis method based on long short-term memory network according to claim 2 is characterized in that: The keyword extraction algorithm in extracting keyword information from a word segmentation set using a keyword extraction algorithm is a TextRank algorithm.

4. The drilling daily report analysis method based on long short-term memory network according to claim 1, 2 or 3, characterized in that: The construction process of the neural network classification model includes: Obtain several historical drilling logs, extract keywords from each historical drilling log and identify the corresponding complex accident classification results to form a sample library; The historical drilling logs in the sample library are divided into training samples and test samples in proportion; The long short-term memory network model is trained using the training samples to obtain a neural network classification model; The neural network classification model is tested using the test samples. If the test result does not meet the set threshold, the parameters of the neural network classification model are adjusted and the training is repeated until the optimal neural network classification model is obtained.

5. The drilling daily report analysis method based on long short-term memory network according to claim 4 is characterized in that: The keyword extraction and corresponding complex accident classification result identification of each historical drilling log includes: Use the word segmentation tool to perform word segmentation on the drilling log to be analyzed; Use the word vector model Word2vec to transform each word after word segmentation, obtain the word vector representation corresponding to each word, and form a word segmentation set; Use keyword extraction algorithm to extract keyword information from the word segmentation set; The labels are used to identify the miscellaneous accident classification results corresponding to the historical drilling log.

6. A drilling daily report analysis device based on a long short-term memory network using the method as claimed in any one of claims 1 to 5, characterized in that: include: A data acquisition unit extracts keywords from the drilling log to be analyzed to obtain keyword information; The classification unit inputs the keyword information into the neural network classification model to obtain a complex accident classification result, wherein the neural network classification model is obtained by machine learning training using multiple sets of data, and each set of data in the multiple sets of data includes: a drilling log and a label identifying the complex accident classification result in the drilling log.

7. The drilling daily report analysis device based on long short-term memory network according to claim 6 is characterized in that: The data acquisition unit comprises: The initial data acquisition module acquires the drilling logs to be analyzed; The first preprocessing module uses a word segmentation tool to perform word segmentation on the drilling log to be analyzed; The second preprocessing module uses the word vector model Word2vec to convert each word after word segmentation, obtains the word vector representation corresponding to each word, and forms a word segmentation set; The third preprocessing module uses a keyword extraction algorithm to extract keyword information from the word segmentation set.

8. A daily drilling data analysis system based on long short-term memory network, characterized in that: include: A daily drilling report analysis device based on a long short-term memory network, wherein the daily drilling report analysis device based on a long short-term memory network is the daily drilling report analysis device based on a long short-term memory network as described in claims 6 to 7; An interactive device is used for an operator to communicate with a daily drilling analysis device based on a long short-term memory network, so that the operator provides the daily drilling analysis device based on a long short-term memory network with initial data to be predicted and parameter setting instructions through an interactive unit.

9. A storage medium, characterized in that: The storage medium stores a computer program that can be read by a computer, and the computer program is configured to execute the daily drilling report analysis method based on the long short-term memory network as described in any one of claims 1 to 5 when running.

10. An electronic device, characterized in that: It comprises a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the daily drilling report analysis method based on the long short-term memory network as described in any one of claims 1 to 3.

Images