Pumping unit transmission flexible connection fault diagnosis method based on big data and artificial intelligence

Abstract

The invention discloses an oil pumping unit transmission flexible connection fault diagnosis method based on big data and artificial intelligence. The oil pumping unit transmission flexible connection fault diagnosis method sequentially comprises the following steps of S1, collecting monitoring data related to oil pumping unit transmission flexible connection; s2, preprocessing the collected monitoring data; s3, a traditional machine learning method and a deep learning technology are fused, and an oil pumping unit transmission flexible connection fault early warning model is established; s4, analyzing the real-time monitoring data of the flexible connection by using the fault early-warning model, quickly judging abnormal conditions in the real-time monitoring data, making a fault early-warning and analysis report, and providing the fault early-warning and analysis report to a manager; and S5, continuously monitoring and optimizing the early warning model. According to the fault diagnosis method, through data management, an intelligent algorithm and model analysis, the transmission flexible connection state of the oil pumping unit is monitored in real time, potential faults are predicted and diagnosed in advance, and an alarm is given in time when a belt of the oil pumping unit or a steel wire rope of a beam hanger is broken, so that the production efficiency and safety of an oil well are improved, and construction of an intelligent oil field is assisted.

Description

Technical Field

[0001] The present invention relates to a method for diagnosing a failure of a pumping unit transmission, and in particular to a method for diagnosing a failure of a pumping unit transmission soft connection based on big data and artificial intelligence, and belongs to the technical field of intelligent diagnosis of oil fields. Background Art

[0002] As the main oil extraction equipment in oil fields, the walking beam pumping unit plays an important role in oil production. The whole unit is like a balance. The middle part of the walking beam is hinged to the top of the bracket through the middle bearing. During the operation of the walking beam, the middle bearing is used as a fulcrum to reciprocate at a certain angle. The head end of the walking beam is connected to a donkey head, which bears the oil pumping load; the tail shaft bearing seat is fixed below the tail end of the walking beam, and a crossbeam is fixed below the tail shaft bearing seat. The axis of the crossbeam is perpendicular to the axis of the walking beam. Connecting rods are hinged at the front and rear ends of the crossbeam, and connecting rod flanges are provided at the lower ends of the two connecting rods. The two connecting rod flanges are connected to the crank mechanism through crank pins and are provided with a balancing counterweight load. During operation, the motor drives the reduction box through the belt, and the reduction box drives one end of the crank to rotate. The other end of the crank drives the tail end of the walking beam to swing up and down through two connecting rods and a crossbeam, and the donkey head at the head end of the walking beam also moves up and down. The donkey head drives the plunger of the deep well pump down and up through the rope braid, the bare rod, and the sucker rod, thereby continuously pumping the crude oil in the well out of the wellbore.

[0003] The two power transmission soft connection parts of the beam pumping unit are the belt and the wire rope, which are key equipment for the normal operation of the pumping unit. Since the production equipment such as temperature and pressure instruments, dynamometers, RTUs, etc. in the oil field well site are in the harsh environment of the field for a long time, they are prone to failure, resulting in the inability to accurately and timely feedback the production data to the central control platform, which brings difficulties to production management.

[0004] The existing pumping unit transmission soft connection fault diagnosis technology mainly relies on manual on-site inspection or video inspection through camera observation, which cannot guarantee the real-time fault detection, affects oilfield production, and cannot meet the needs of smart oilfield construction. In terms of data collection and processing, the existing pumping unit transmission soft connection relies on manual methods to analyze and diagnose data, which is time-consuming and labor-intensive, making it difficult to monitor a large number of pumping units in real time, and traditional data analysis methods lack sufficient intelligence to process massive data and conduct in-depth analysis. In terms of fault diagnosis, the existing pumping unit transmission soft connection monitoring system failed to fully consider future data growth and analysis needs when designing, making it difficult for the system to cope with the growing amount of data and the need to analyze complex problems.

[0005] Under the background of rapid development of informatization and intelligence, constructing advanced fault diagnosis and early warning model for oil pump transmission soft connection is of great significance to reducing on-site labor and improving production efficiency. Summary of the invention

[0006] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0007] In view of the above problems and / or the problems existing in the prior art, the present invention is proposed to realize the intelligence and systematization of oil field development and improve the working efficiency in the oil-related fields.

[0008] The purpose of the present invention is to overcome the problems existing in the prior art and provide a method for diagnosing faults of oil pump transmission soft connections based on big data and artificial intelligence. Through efficient data management, intelligent algorithms and model analysis, the status of oil pump transmission soft connections can be monitored in real time, potential faults can be predicted and diagnosed in advance, and an alarm can be issued in time when the oil pump belt breaks or the wire rope of the rope hanger breaks, thereby improving the efficiency and safety of oil well production and facilitating the construction of smart oil fields.

[0009] In order to solve the above technical problems, a method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence is provided in the present invention, which comprises the following steps in sequence:

[0010] S1. Collect monitoring data related to the soft connection of the oil pumping unit transmission;

[0011] S2. Preprocessing the collected monitoring data;

[0012] S3. Integrate traditional machine learning methods and deep learning technology to create a fault warning model for the soft connection of the oil pumping unit;

[0013] S4. Use the fault warning model to analyze the real-time monitoring data of the soft connection, quickly determine the abnormal situation, and prepare a fault warning and analysis report to provide to the management personnel;

[0014] S5. Continuously monitor and optimize the early warning model.

[0015] Furthermore, in step S1, the monitoring data related to the oil pump transmission soft connection includes static data and dynamic data, wherein the static data mainly includes the oil pump model and motor power, and the dynamic data includes current data and dynamometer data.

[0016] Furthermore, in step S1, the amount of data collected needs to be greater than 20,000, and must cover various soft connection failures including "broken braid" and "broken belt".

[0017] Furthermore, in step S2, the preprocessing of the monitoring data includes the following sub-steps in sequence:

[0018] S2.1, Data cleaning: Detect and remove abnormal values ​​in current data and dynamometer data through statistical methods;

[0019] S2.2, Data normalization: Normalize the data to eliminate the dimensional effects between different features;

[0020] S2.3, Feature extraction: Extract key features that are helpful for model training from the cleaned and normalized data;

[0021] S2.4. Dataset classification: Classify the dataset according to the model and motor power of the pumping unit to improve the pertinence and accuracy of the model.

[0022] Furthermore, in step S2.1, for missing values, interpolation or the mean is used to fill in to ensure the integrity and continuity of the data set.

[0023] Furthermore, in step S2.2, the method for normalizing the data includes minimum-maximum normalization, scaling the data to between [0, 1]; and Z-score normalization, converting the data to a distribution with a mean of 0 and a standard deviation of 1.

[0024] Furthermore, in step S2.3, the key features extracted include statistical features, time domain features and / or frequency domain features.

[0025] Furthermore, step 2 also includes: S2.5, checking the quality of the preprocessed data. If the data quality meets the requirements, it is determined to be indicator data that can be used by the early warning model and can be used for training and testing the early warning model; if the data quality does not meet the requirements, it is discarded, and the data is collected again and preprocessed to avoid the impact of low-quality data on model performance.

[0026] Furthermore, step S3 creates a pumping unit transmission soft connection fault warning model, which includes the following sub-steps in sequence:

[0027] S3.1. For the obtained index data, apply the K-means algorithm to divide the transmission soft connection data into clusters with similar characteristics, and find out the abnormal points and regular patterns of transmission soft connection operation;

[0028] S3.2. Use the CNN network to extract high-dimensional feature vectors of the data related to the above transmission soft connection data set, construct the obtained high-dimensional feature vectors into a time series, and use them as input data of the LSTM network;

[0029] S3.3. Apply the LSTM algorithm to the modeling of time series data. By training the LSTM model, the data in the future can be predicted, so as to realize the early warning of abnormal conditions of transmission soft connection.

[0030] S3.4. Build a CNN-LSTM network.

[0031] Furthermore, the K-means algorithm in step S3.1 includes the following steps in sequence:

[0032] S3.1.1. Select k initialized transmission soft connection data samples as initial cluster centers a=a1, a2, ...a k ;

[0033] S3.1.2. For each sample x in the transmission soft connection dataset i Calculate its distance to k cluster centers and divide it into the class corresponding to the cluster center with the smallest distance;

[0034] S3.1.3 For each category a j , recalculate its cluster center That is, the centroid of all samples belonging to this class;

[0035] S3.1.4. Repeat steps S3.1.2 and S3.1.3 above until the number of iterations is reached or the algorithm terminates when the minimum error changes.

[0036] Furthermore, the CNN-LSTM network in step S3.4 includes the following layers:

[0037] S3.4.1, the first layer is the input layer;

[0038] S3.4.2, the second layer is the convolutional neural network layer;

[0039] S3.4.3, the third layer is a multi-layer LSTM;

[0040] S3.4.4, the fourth layer is the attention layer;

[0041] S3.4.5. The fifth layer is the output layer.

[0042] Furthermore, in step S3.4.1, the format of the input data is specified, the batch size is set to 1 by default, the number of time steps is t, and the feature dimension is n. Then a sample is represented as a real number sequence matrix Rt×n, and xi is recorded as the vector representation of the i-th time step data in Rt×n;

[0043] In step S3.4.2, the sample data enters the CNN layer and performs convolution, pooling and dimensionality reduction operations in sequence;

[0044] In step S3.4.3, the output of the previous LSTM layer is the input of the next layer, and it is passed down layer by layer. The output of the last LSTM hidden layer enters the attention layer for further processing;

[0045] S3.4.4, the LSTM hidden layer output vector is used as the input of the attention layer, trained through the fully connected layer, and then the output of the fully connected layer is normalized using the softmax function to obtain the distribution weight of each hidden layer vector. The weight size indicates the importance of the hidden state of each time step to the prediction result;

[0046] S3.4.5, the output layer specifies the prediction time step 0 t , and finally output 0 t The prediction results of the step.

[0047] Further, in step S3.4.2,

[0048] One-dimensional convolution is adopted, and the convolution kernel is convolved only in a single time domain direction; the number of convolution kernels is r, the size is set to k, and X1:i+k-1 is the real number matrix from the i-th time step to the i+k-1-th time step in Rt×n, and the sliding step size is 1;

[0049] The weight matrix W1 is a k×n real number matrix; a feature extraction is performed on the sequence vector of each k time step to obtain a feature 0 i , the calculation formula is as follows:

[0050]

[0051] f is a nonlinear activation function, b1∈R is a bias; when a convolution kernel extracts a sample sequence data, a feature map of (t-k+1)×1 shape is obtained, and the calculation formula is as follows:

[0052] 0=[01,02,…,0 t-k+1 ] T

[0053] CNN has a total of r convolution kernels, so r feature maps will be obtained in the end; after convolution, the maximum pooling operation is performed, the pooling size is 2, the sliding step is 2, and r feature maps of the shape of [(t-k+1) / 2]×1 are obtained. The calculation formula is as follows:

[0054] 0=max{01,0 1+1}(1=1,3,5,…,tk)

[0055] These r feature maps are the features extracted by the CNN layer. They are reduced in dimension into a real vector of length r*(t-k+1) / 2, which stores the spatial connection between different eigenvalues ​​in the sample data and is then input into the LSTM layer for further processing.

[0056] Furthermore, in step S3.4.4, the weight training process is as follows:

[0057] Si =tanh(WH1+b1)

[0058] α i =softmax(S i )

[0059] Then use the trained weights to calculate the weighted average sum of the hidden layer output vectors. The calculation results are as follows:

[0060]

[0061] Among them, H i is the output of the last LSTM hidden layer, S i is the score for each hidden layer output, α i is the weight coefficient, C1 is the result of weighted summation, and softmax is the activation function.

[0062] Furthermore, step S4 includes the following sub-steps in sequence:

[0063] S4.1. Identify abnormal situations: Use the fault warning model to analyze the real-time monitoring data of the transmission soft connection in real time. When an abnormal situation is found, the fault warning model outputs a warning result;

[0064] S4.2. Determine the warning level: Determine the corresponding warning level for each warning situation;

[0065] S4.3. Formulate response measures: formulate corresponding response measures according to the warning level;

[0066] S4.4. Send warning notification: After determining the warning level and response measures, send a warning notification to relevant personnel.

[0067] Furthermore, the specific scheme of step S5 for continuously monitoring and optimizing the fault warning model includes:

[0068] S5.1. Monitor the performance of the early warning model for the soft connection failure of the oil pumping unit transmission;

[0069] S5.2. Evaluate the effectiveness of the early warning model for the soft connection failure of the oil pumping unit transmission;

[0070] S5.3. Optimize the parameters and algorithm of the pumping unit transmission soft connection fault warning model:

[0071] S5.4. Introduce advanced models and algorithms.

[0072] Compared with the prior art, the present invention has achieved the following beneficial effects: 1. Improved production efficiency of oilfield well stations: This method effectively manages transmission soft connection equipment and data, greatly improves production efficiency of oilfield well stations, and ensures the continuity and stability of the production process.

[0073] 2. Improved the production and economic benefits of oilfield well stations: This method uses big data analysis technology to process the monitoring data of the pumping unit transmission soft connection, which can help optimize the production process; by better understanding and utilizing production data, oil fields can manage production in a refined manner and improve production and benefits.

[0074] 3. Real-time monitoring and predictive maintenance of oil pumps: This method can detect abnormal conditions before equipment failures by real-time monitoring and analysis of oil pump transmission soft connection data, which helps to implement predictive maintenance, repair and replacement in time, and reduce the impact of equipment failures on production.

[0075] 4. Reduce maintenance costs: Through predictive maintenance, this method can reduce the frequency of emergency repairs and equipment replacement, thereby reducing maintenance costs; accurate fault prediction and effective maintenance strategies help extend equipment life and reduce the maintenance burden of enterprises.

[0076] 5. Reduced oil field operating costs: By achieving fine management of pumping unit transmission soft connection equipment and data, the maintenance and management costs of the oil field are reduced, thereby reducing the operating costs of the oil field.

[0077] 6. Enhanced the safety of oilfield well stations: Through real-time collection and analysis of pumping unit transmission soft connection data, combined with traditional machine learning and deep learning technologies, a multi-level pumping unit transmission soft connection early warning model is constructed, which can predict and identify abnormal situations in real time, effectively identify and predict the failure of well site transmission soft connections, so as to take corresponding response measures in time, effectively avoid potential safety risks, and ensure the production safety of oilfield well stations.

[0078] 7. Improved oil field reliability and reduced production downtime: Through predictive maintenance and timely fault warning, this method can reduce production downtime caused by pumping unit transmission soft connection failure. This is crucial for the oil field production industry because any production interruption may lead to economic losses.

[0079] 8. It also has continuous monitoring and optimization functions: by introducing advanced models and algorithms, optimizing model parameters and algorithms, etc., the performance and stability of the early warning model are continuously improved. Continuous performance improvement ensures the reliability of oilfield pumping unit production and further reduces failure rate and downtime.

[0080] 9. Due to the huge number of pumping units in the domestic petroleum industry and the huge workload of manual inspections, the application of this intelligent diagnosis method for pumping unit transmission soft connection faults can greatly improve production efficiency and reduce maintenance costs. It has extremely broad application prospects and provides an excellent technical approach to accelerate the construction of smart oil fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor. The accompanying drawings are only provided for reference and explanation, and are not used to limit the present invention.

[0082] in:

[0083] Figure 1 This is a general flow chart of the method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence of the present invention;

[0084] Figure 2 Preprocessing the collected monitoring data in the present invention;

[0085] Figure 3 A flow chart of creating a pumping unit transmission soft connection fault warning model in the present invention;

[0086] Figure 4 A flow chart for sending an early warning for the fault diagnosis and analysis of the transmission soft connection in the present invention;

[0087] Figure 5 A flowchart of system monitoring and optimization in the present invention;

[0088] Figure 6 This is an interface diagram of the system of the present invention diagnosing the faults such as "broken belt" in Yang 53-2 well, etc.;

[0089] Figure 7 When the "belt break fault" occurs in Yang 53-2 Well, click the "fault occurrence time" cell to select the cycle average current graph to be displayed;

[0090] Figure 8 When the "belt break fault" occurs in Yang 53-2 well, click the "fault occurrence time" cell to select the cycle active power diagram to be displayed;

[0091] Fig. 9 The diagnostic feedback displayed when clicking the "Diagnostic Notes" cell when a "Belt Break Fault" occurs in Yang 53-2 Well;

[0092] Fig.10 When the "braid break fault" occurs in Min 15-13 well, click the "fault occurrence time" cell to select the cycle average current diagram to be displayed;

[0093] Fig.11 When the "braid break fault" occurs in Min 15-13 well, click the "fault occurrence time" cell to select the cycle active power diagram to be displayed;

[0094] Fig.12 This is the diagnostic feedback displayed when clicking the "Diagnostic Notes" cell when the "braid break fault" occurs in Min 15-13 well. DETAILED DESCRIPTION

[0095] like Figures 1 to 5 As shown, the method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence of the present invention comprises the following steps in sequence:

[0096] S1. Collect monitoring data related to the soft connection of the oil pump transmission, including static data and dynamic data. The static data mainly includes the oil pump model and motor power. The oil pump model is the key information for identifying the design and working characteristics of the oil pump, which is usually found in the equipment manual and related technical data; the motor power refers to the output power of the motor driving the oil pump, which is also a key parameter obtained from the equipment manual.

[0097] Dynamic data mainly comes from the monitoring system and historical database of the oil pump, including: current data and dynamometer data. Current data: records the current changes of the oil pump in normal working and fault conditions, and is used to analyze the workload and efficiency of the equipment. Dynamometer data: is the instrument data that measures the displacement and load during the working process of the oil pump, and is used to analyze the performance status of the oil pump. For example:

[0098] Fault name Braids broken Braids broken Belt broken Belt broken Oil well name Europe and North 17 Ma 33-19 Side Europe 2 Min 24-27 Failure time 2023 / 9 / 18 19：48 2023 / 9 / 11 6：39 2023 / 9 / 2 12：18 2023 / 9 / 1 8：05 Phase A current during normal operation (A) 11.91 6.58 10.85 10.07 Forward active power in normal operation (kWh) 5.48 4.02 4.25 6.2 Reverse active power in normal operation (kWh) 5.04 0.98 5.74 1.66 Phase A current during fault (A) 30.36 20.48 6.95 4.62 Forward active power at fault (kwh) 8.21 13.83 0.84 1.05 Reverse active power during fault (kwh) 10.26 3.21 6.89 2.86

[0099] To ensure the adequacy of the data sample, the amount of data collected needs to be greater than 20,000 and must cover different types of soft connection failures, including "broken braids" and "broken belts".

[0100] S2. Preprocessing of collected monitoring data

[0101] like Figure 2 As shown in Figure 1, data preprocessing is an important step to ensure data quality and improve model training results. The acquired historical data is preprocessed according to the following steps:

[0102] S2.1. Data cleaning: Detect and remove outliers in current data and dynamometer data using statistical methods. For missing values, interpolation or filling with the mean is used to ensure the integrity and continuity of the data set.

[0103] S2.2, Data normalization: In order to eliminate the dimensionality effects between different features, the data is normalized. Common methods include minimum-maximum normalization, which scales the data to between [0, 1]; and Z-score normalization, which converts the data to a distribution with a mean of 0 and a standard deviation of 1.

[0104] S2.3, Feature extraction: Extract key features that are helpful for model training from the cleaned and normalized data. These features include statistical features (such as mean, variance), time domain features (such as peaks, valleys, cycles, etc. of the waveform), and / or frequency domain features (such as frequency components after Fourier transform).

[0105] S2.4. Dataset classification: The data set is classified according to the model and motor power of the pumping unit. This allows different models to be trained and tested for pumping units of different models and powers, thereby improving the pertinence and accuracy of the model.

[0106] S2.5. Check the quality of the preprocessed data. If the data quality meets the requirements, it is determined to be indicator data that can be used by the early warning model and can be used for training and testing the early warning model. If the data quality does not meet the requirements, it is discarded, and the data is collected again and preprocessed to avoid the impact of low-quality data on model performance.

[0107] like Figure 3 As shown in the figure, creating a pumping unit transmission soft connection fault warning model includes the following steps in sequence:

[0108] S3.1. For the obtained index data, the K-means algorithm is applied to divide the transmission soft connection data into clusters with similar characteristics, and find out the abnormal points and regular patterns of transmission soft connection operation. These abnormal points may be transmission soft connections with problems, while the regular patterns can be used to identify transmission soft connections that are operating normally.

[0109] The algorithm steps of K-means are:

[0110] S3.1.1. Select k initialized transmission soft connection data samples as initial cluster centers a=a1, a2, ...a k ;

[0111] S3.1.2. For each sample x in the transmission soft connection dataset i Calculate its distance to k cluster centers and divide it into the class corresponding to the cluster center with the smallest distance;

[0112] S3.1.3 For each category a j , recalculate its cluster center That is, the centroid of all samples belonging to this class;

[0113] S3.1.4. Repeat steps S3.1.2 and S3.1.3 above until the number of iterations is reached or the algorithm terminates when the minimum error changes.

[0114] S3.2. Use the CNN network to extract high-dimensional feature vectors of the above transmission soft connection data set: By performing convolution and pooling operations on the text data, the CNN network extracts feature information of different scales and angles to form high-dimensional feature vectors. The obtained high-dimensional feature vectors are constructed into a time series sequence and used as input data for the LSTM network. When constructing a time series sequence, it is necessary to consider the setting of the time step and sliding window size, as well as maintaining the continuity and consistency of the data. The LSTM network uses the high-dimensional feature vectors extracted by the previous CNN network to model and predict the transmission soft connection data.

[0115] S3.3. The LSTM algorithm is used to effectively capture the long-term dependencies in time series data, and has good modeling and prediction capabilities. In the transmission soft connection warning model, the LSTM algorithm is applied to the modeling of time series data. By training the LSTM model, the data in the future period is predicted, thereby realizing the early warning of abnormal transmission soft connection conditions.

[0116] S3.4. Construct a CNN-LSTM network. The CNN-LSTM network includes the following layers:

[0117] S3.4.1. The first layer is the input layer. The format of the input data (batch size, time steps, feature dimension) is specified. The batch size is set to 1 by default, the time step is t, and the feature dimension is n. Then a sample is represented by a real number sequence matrix Rt×n, and xi is the vector representation of the i-th time step data in Rt×n.

[0118] S3.4.2, the second layer is the convolutional neural network layer, namely the CNN layer. The CNN layer can extract the spatial connection between different eigenvalues ​​in the data, thereby making up for the shortcomings of the LSTM network that cannot capture the spatial components of the data. At the same time, the features it extracts are still temporal. The sample data enters the CNN layer and performs convolution, pooling and node expansion (dimensionality reduction) operations in sequence. For sequence data, this model adopts one-dimensional convolution, and the convolution kernel only performs convolution in a single time domain direction. The number of convolution kernels is r, the size is set to k, and X1:i+k-1 is the real number matrix from the i-th time step to the i+k-1-th time step in Rt×n, and the sliding step size is 1.

[0119] The weight matrix W1 is a k×n real number matrix. A feature extraction is performed on the sequence vector of each k time step to obtain a feature 0 i , the calculation formula is as follows:

[0120]

[0121] f is a nonlinear activation function, and b1∈R is a bias. When a convolution kernel extracts a sample sequence data, a feature map of (t-k+1)×1 shape is obtained, and the calculation formula is as follows:

[0122] 0=[01,02,…,0 t-k+1 ] T

[0123] CNN has a total of r convolution kernels, so it will eventually get r feature maps. After convolution, the maximum pooling operation is performed, the pooling size is 2, the sliding step is 2, and r feature maps of the shape [(t-k+1) / 2]×1 are obtained. The calculation formula is as follows:

[0124] 0=max{01,0 1+1}(1=1,3,5,…,tk)

[0125] These r feature maps are the features extracted by the CNN layer. They are reduced in dimension into a real vector of length r*(t-k+1) / 2, which stores the spatial connection between different eigenvalues ​​in the sample data and is then input into the LSTM layer for further processing.

[0126] S3.4.3, the third layer is a multi-layer LSTM. LSTM has a memory function and can extract the time series change information of the nonlinear data of the transmission soft connection fault. It introduces the input gate, forget gate, and output gate, and also adds candidate states, cell states, and hidden states. The cell state stores long-term memory, which can alleviate the gradient disappearance, and the hidden state stores short-term memory. This model uses a multi-layer LSTM. The output of the previous LSTM layer is the input of the next layer, which is passed down layer by layer. The output of the last LSTM hidden layer enters the attention layer for further processing.

[0127] S3.4.4, the fourth layer is the attention layer. Attention can enhance the role of important time steps in LSTM, thereby further reducing the model prediction error. Attention is essentially to find the weighted average sum of the output vectors of the last layer of LSTM. The output vector of the LSTM hidden layer is used as the input of the attention layer, trained through the fully connected layer, and then the output of the fully connected layer is normalized using the softmax function to obtain the assigned weight of each hidden layer vector. The weight size indicates the importance of the hidden state of each time step to the prediction result. The weight training process is as follows:

[0128] S i =tanh(WH1+b1)

[0129] α i =softmax(S i )

[0130] Then use the trained weights to calculate the weighted average sum of the hidden layer output vectors. The calculation results are as follows:

[0131]

[0132] Among them, H i is the output of the last LSTM hidden layer, S i is the score for each hidden layer output, α i is the weight coefficient, C1 is the result of weighted summation, and softmax is the activation function.

[0133] S3.4.5, the fifth layer is the output layer. This layer specifies the prediction time step 0 t , and finally output 0 t The prediction results of the step.

[0134] By combining the characteristics of K-means and CNN-LSTM methods, a comprehensive transmission soft connection fault warning model can be constructed.

[0135] like Figure 4 As shown in the figure, the fault warning model is used to analyze and diagnose the real-time monitoring data of the transmission soft connection of the oil pumping unit, quickly identify abnormal conditions in the transmission soft connection data, and produce fault warning and analysis reports to provide to management personnel. The specific work steps are as follows:

[0136] S4.1. Identify abnormal conditions: Input the real-time monitoring data of the transmission soft connection of the oil pump into the fault warning model for analysis and diagnosis. When an abnormal condition is found, the fault warning model outputs a warning result. These warning results include abnormal status of the transmission soft connection, trend changes, or indicator data exceeding the set threshold.

[0137] S4.2, determine the warning level: according to the output and real-time data of the pumping unit transmission soft connection fault warning model, determine the corresponding warning level for each warning situation. The warning level can be evaluated and classified according to the severity, impact range and duration of the abnormality. Generally, the warning level can be divided into different levels, such as low, medium and high, to facilitate subsequent response and processing.

[0138] S4.3. Develop response measures: Develop corresponding response measures according to the warning level. These measures may include adjusting transmission soft connection parameters, starting backup equipment, performing maintenance and repair, and dispatching additional personnel. The warning decision module can select the most appropriate response measures to deal with abnormal situations based on historical data and experience knowledge, combined with the current situation and warning level, and ensure the timeliness and effectiveness of the actions taken.

[0139] S4.4. Send warning notifications: After determining the warning level and response measures, send warning notifications to relevant personnel. These notifications can be sent via email, SMS, mobile applications, etc. to ensure that relevant personnel can receive them in time and take necessary actions. The notification content includes the warning level, description of the abnormal situation, recommended response measures, and contact information, etc., so that relevant personnel can conduct further processing and communication.

[0140] Through the operation of the early warning decision module, oilfield managers can be notified in time when an early warning occurs and take appropriate measures to deal with abnormal situations, thereby reducing risks and losses. At the same time, the early warning decision module can also record early warning information and response processes for subsequent analysis and optimization. This real-time early warning and decision-making capability will help improve the production efficiency, safety and reliability of oilfields.

[0141] like Figure 5 As shown in the figure, the specific process of the system monitoring and optimization module to continuously monitor and optimize the pumping unit transmission soft connection fault warning model includes:

[0142] S5.1. Monitor the performance of the pumping unit transmission soft connection fault warning model: By monitoring the various indicators and operating status of the pumping unit transmission soft connection fault warning model, the performance of the system can be understood. For example, the accuracy, recall rate, false alarm rate and other indicators of the pumping unit transmission soft connection fault warning model, as well as the real-time response time, resource utilization and other operating status of the system can be monitored to timely discover potential problems and performance bottlenecks of the system, and take corresponding measures to optimize them.

[0143] S5.2. Evaluate the effectiveness of the pumping unit transmission soft connection fault warning model: The system monitoring and optimization module will regularly evaluate the effectiveness and accuracy of the pumping unit transmission soft connection fault warning model. By comparing with actual abnormal conditions and events, analyze the consistency between the prediction results of the pumping unit transmission soft connection fault warning model and the actual situation, and evaluate the accuracy and reliability of the model. If deviations or errors are found in the pumping unit transmission soft connection fault warning model, make corresponding adjustments and improvements based on the evaluation results.

[0144] S5.3. Optimize the parameters and algorithms of the pumping unit transmission soft connection fault warning model: Based on the evaluation and monitoring results of the warning model effect, optimize the model parameters and algorithms as needed. Improve the model's prediction ability and stability by adjusting the model's parameter settings, selecting a more appropriate algorithm, or introducing new feature engineering methods. This module can also use historical data and feedback information, and use machine learning technology to automatically optimize and update the pumping unit transmission soft connection fault warning model.

[0145] S5.4. Introducing advanced models and algorithms: In order to continuously improve the performance and effectiveness of the early warning system, the system monitoring and optimization module can gradually introduce more advanced models and algorithms. For example, deep learning models, reinforcement learning algorithms, etc. can be introduced to better explore the potential laws and characteristics of the data. At the same time, combined with the experience and knowledge of experts in the field, model integration and fusion are carried out to further improve the prediction accuracy and stability of the early warning system. Through the operation of the system monitoring and optimization module, the early warning model can detect and correct problems in a timely manner to ensure the stability and accuracy of the system. The process of optimizing model parameters and algorithms will continuously improve the performance of the early warning model, enabling it to better adapt to actual application scenarios. At the same time, the introduction of advanced models and algorithms will further promote the development of the early warning system, improve its prediction accuracy and ability to cope with complex situations.

[0146] This technical solution was tested on site in multiple wells in Jiangsu Oilfield and achieved the expected results. It can realize intelligent fault diagnosis of oilfield pumping unit transmission soft connections and improve the safety and efficiency of production operations.

[0147] like Figure 6 As shown, the system of the present invention selects to display faults in a certain time period. During this time period, the system displays that "broken belt" faults occur in wells such as Yang 53-2 and Min 20-40, and "broken belt" faults occur in wells such as Min 15-13 and Min 15-1.

[0148] like Figure 7 As shown, click the "Fault Occurrence Time" cell on the right side of the row where Yang 53-2 Well is located, and select to display the cycle average current graph. It can be seen from the graph that when the belt breaks, the current drops sharply and approaches zero;

[0149] like Figure 8 As shown, click the "Fault Occurrence Time" cell on the right side of the row where Yang 53-2 Well is located, and select Display Period Active Power Diagram. It can be seen from the figure that when the belt breaks, the power drops drastically and approaches zero.

[0150] like Fig. 9 As shown, click the "Diagnosis Notes" cell on the right side of the row where Yang 53-2 Well is located to obtain diagnostic feedback within 500 words, which shows that the belt is broken in the on-site video and the diagnosis is correct.

[0151] like Fig.10 As shown, click the "Fault Occurrence Time" cell on the right side of the row where the Min 15-13 well is located, and select to display the cycle average current graph. It can be seen from the figure that when the braid is broken, the current drops precipitously and approaches zero;

[0152] like Fig.11As shown, click the "Fault Occurrence Time" cell on the right side of the row where the Min 15-13 well is located, and select Display Period Active Power Diagram. It can be seen from the figure that when the braid is broken, the power drops drastically and approaches zero.

[0153] like Fig.12 As shown, click the "Diagnosis Notes" cell on the right side of the row where the Min 15-13 well is located to obtain diagnostic feedback within 500 words, showing that the video is confirmed, the hair braid is broken, and the diagnosis is correct.

[0154] The above is only a preferred embodiment of the present invention, which shows and describes the basic principles, main features and advantages of the present invention, but does not limit the scope of patent protection of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. In addition to the above embodiments, the present invention may have other implementation modes without departing from the spirit and scope of the present invention. The present invention may also have various changes and improvements, and all technical solutions formed by equivalent substitution or equivalent transformation fall within the scope of protection required by the present invention. The scope of protection required by the present invention is defined by the attached claims and their equivalents. The technical features not described in the present invention can be realized by or using existing technologies, which will not be repeated here.

Claims

1. A method for diagnosing faults of oil pump transmission soft connections based on big data and artificial intelligence, characterized in that: The steps are as follows: S1. Collect monitoring data related to the soft connection of the oil pumping unit transmission; S2. Preprocessing the collected monitoring data; S3. Integrate traditional machine learning methods and deep learning technology to create a fault warning model for the soft connection of the oil pumping unit; S4. Use the fault warning model to analyze the real-time monitoring data of the soft connection, quickly determine the abnormal situation, and prepare a fault warning and analysis report to provide to the management personnel; S5. Continuously monitor and optimize the early warning model.

2. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 1 is characterized in that: In step S1, the monitoring data related to the oil pump transmission soft connection includes static data and dynamic data. The static data mainly includes the oil pump model and motor power, and the dynamic data includes current data and dynamometer data.

3. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 2 is characterized in that: In step S1, the amount of data collected needs to be greater than 20,000, and must cover various soft connection failures including "broken braid" and "broken belt".

4. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 1 is characterized in that: In step S2, the preprocessing of the monitoring data includes the following sub-steps in sequence: S2.1, Data cleaning: Detect and remove abnormal values ​​in current data and dynamometer data through statistical methods; S2.2, Data normalization: Normalize the data to eliminate the dimensional effects between different features; S2.3, Feature extraction: Extract key features that are helpful for model training from the cleaned and normalized data; S2.

4. Dataset classification: Classify the dataset according to the model and motor power of the pumping unit to improve the pertinence and accuracy of the model.

5. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 4 is characterized in that: In step S2.1, missing values ​​are filled by interpolation or by using the mean to ensure the integrity and continuity of the data set.

6. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 4 is characterized in that: In step S2.2, the method for normalizing the data includes minimum-maximum normalization, scaling the data to between [0,1]; and Z-score normalization, converting the data to a distribution with a mean of 0 and a standard deviation of 1.

7. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 4 is characterized in that: In step S2.3, the key features extracted include statistical features, time domain features and / or frequency domain features.

8. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 4 is characterized in that: Step 2 also includes: S2.

5. Check the quality of the preprocessed data. If the data quality meets the requirements, it is determined to be indicator data that can be used by the early warning model and can be used for training and testing the early warning model. If the data quality does not meet the requirements, it is discarded, and the data is collected again and preprocessed to avoid the impact of low-quality data on model performance.

9. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 1 is characterized in that: Step S3 creates a pumping unit transmission soft connection fault warning model, which includes the following sub-steps in sequence: S3.

1. For the obtained index data, apply the K-means algorithm to divide the transmission soft connection data into clusters with similar characteristics, and find out the abnormal points and regular patterns of transmission soft connection operation; S3.

2. Use the CNN network to extract high-dimensional feature vectors of the data related to the above transmission soft connection data set, construct the obtained high-dimensional feature vectors into a time series, and use them as input data of the LSTM network; S3.

3. Apply the LSTM algorithm to the modeling of time series data. By training the LSTM model, the data in the future can be predicted, so as to realize the early warning of abnormal conditions of transmission soft connection. S3.

4. Build a CNN-LSTM network.

10. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 9 is characterized in that: The K-means algorithm in step S3.1 includes the following steps in sequence: S3.1.

1. Select k initialized transmission soft connection data samples as initial cluster centers a=a1, a2, ...a k ; S3.1.

2. For each sample x in the transmission soft connection dataset i Calculate its distance to k cluster centers and divide it into the class corresponding to the cluster center with the smallest distance; S3.1.3 For each category a j , recalculate its cluster center That is, the centroid of all samples belonging to this class; S3.1.

4. Repeat steps S3.1.2 and S3.1.3 above until the number of iterations is reached or the algorithm terminates when the minimum error changes.

11. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 9, characterized in that: The CNN-LSTM network in step S3.4 includes the following layers: S3.4.1, the first layer is the input layer; S3.4.2, the second layer is the convolutional neural network layer; S3.4.3, the third layer is a multi-layer LSTM; S3.4.4, the fourth layer is the attention layer; S3.4.

5. The fifth layer is the output layer.

12. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 11, characterized in that: In step S3.4.1, the format of the input data is specified, the batch size is set to 1 by default, the number of time steps is t, and the feature dimension is n. Then a sample is represented by a real sequence matrix Rt×n, and xi is the vector representation of the i-th time step data in Rt×n; In step S3.4.2, the sample data enters the CNN layer and performs convolution, pooling and dimensionality reduction operations in sequence; In step S3.4.3, the output of the previous LSTM layer is the input of the next layer, and it is passed down layer by layer. The output of the last LSTM hidden layer enters the attention layer for further processing; S3.4.4, the LSTM hidden layer output vector is used as the input of the attention layer, trained through the fully connected layer, and then the output of the fully connected layer is normalized using the softmax function to obtain the distribution weight of each hidden layer vector. The weight size indicates the importance of the hidden state of each time step to the prediction result; S3.4.5, the output layer specifies the prediction time step 0 t , and finally output 0 t The prediction results of the step.

13. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 12, characterized in that: In step S3.4.2, One-dimensional convolution is adopted, and the convolution kernel is convolved only in a single time domain direction; the number of convolution kernels is r, the size is set to k, X 1: i+k-1 is the real number matrix from the i-th time step to the i+k-1-th time step in Rt×n, and the sliding step length is 1; The weight matrix W1 is a k×n real number matrix; a feature extraction is performed on the sequence vector of each k time step to obtain a feature 0 i , the calculation formula is as follows: f is a nonlinear activation function, b1∈R is a bias; when a convolution kernel extracts a sample sequence data, a feature map of (t-k+1)×1 shape is obtained, and the calculation formula is as follows: 0＝[01,02,…,0 t-k+1 ] T CNN has a total of r convolution kernels, so r feature maps will be obtained in the end; after convolution, the maximum pooling operation is performed, the pooling size is 2, the sliding step is 2, and r feature maps of the shape of [(t-k+1) / 2]×1 are obtained. The calculation formula is as follows: 0=max{01,0 1+1 }(1=1,3,5,…,t-k) These r feature maps are the features extracted by the CNN layer. They are reduced in dimension into a real vector of length r*(t-k+1) / 2, which stores the spatial connection between different eigenvalues ​​in the sample data and is then input into the LSTM layer for further processing.

14. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 12, characterized in that: In step S3.4.4, the weight training process is as follows: S i =tanh(WH1+b1) α i =softmax(S i ) Then use the trained weights to calculate the weighted average sum of the hidden layer output vectors. The calculation results are as follows: Among them, H i is the output of the last LSTM hidden layer, S i For each hidden layer output score, α i is the weight coefficient, C1 is the result of weighted summation, and softmax is the activation function.

15. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 12, characterized in that: Step S4 includes the following sub-steps in sequence: S4.

1. Identify abnormal situations: Use the fault warning model to analyze the real-time monitoring data of the transmission soft connection in real time. When an abnormal situation is found, the fault warning model outputs a warning result; S4.

2. Determine the warning level: Determine the corresponding warning level for each warning situation; S4.

3. Formulate response measures: formulate corresponding response measures according to the warning level; S4.

4. Send warning notification: After determining the warning level and response measures, send a warning notification to relevant personnel.

16. The method for diagnosing oil pump transmission soft connection faults based on big data and artificial intelligence according to claim 12, characterized in that: The specific scheme of step S5 for continuously monitoring and optimizing the fault warning model includes: S5.

1. Monitor the performance of the early warning model for the soft connection failure of the oil pumping unit transmission; S5.

2. Evaluate the effectiveness of the early warning model for the soft connection failure of the oil pumping unit transmission; S5.

3. Optimize the parameters and algorithm of the pumping unit transmission soft connection fault warning model: S5.

4. Introduce advanced models and algorithms.

Images