WO2023151072A1 - 砂堵预测方法、装置、设备及存储介质 - Google Patents

砂堵预测方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2023151072A1
WO2023151072A1 PCT/CN2022/076163 CN2022076163W WO2023151072A1 WO 2023151072 A1 WO2023151072 A1 WO 2023151072A1 CN 2022076163 W CN2022076163 W CN 2022076163W WO 2023151072 A1 WO2023151072 A1 WO 2023151072A1
Authority
WO
WIPO (PCT)
Prior art keywords
period
construction data
fracturing construction
fracturing
prediction model
Prior art date
Application number
PCT/CN2022/076163
Other languages
English (en)
French (fr)
Inventor
王海威
邢海平
王振岐
杨浩
Original Assignee
烟台杰瑞石油装备技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 烟台杰瑞石油装备技术有限公司 filed Critical 烟台杰瑞石油装备技术有限公司
Priority to PCT/CN2022/076163 priority Critical patent/WO2023151072A1/zh
Publication of WO2023151072A1 publication Critical patent/WO2023151072A1/zh

Links

Images

Classifications

    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21B—EARTH OR ROCK DRILLING; OBTAINING OIL, GAS, WATER, SOLUBLE OR MELTABLE MATERIALS OR A SLURRY OF MINERALS FROM WELLS
    • E21B43/00—Methods or apparatus for obtaining oil, gas, water, soluble or meltable materials or a slurry of minerals from wells
    • E21B43/25—Methods for stimulating production
    • E21B43/26—Methods for stimulating production by forming crevices or fractures

Definitions

  • the present application relates to the technical field of oil and gas fracturing operations, in particular to a sand plugging prediction method, device, equipment and storage medium.
  • Fracturing refers to the formation of artificial fractures in the oil and gas layer by the pressure of fracturing fluid to improve the flow capacity of fluids (such as oil and natural gas, etc.) in the oil and gas layer.
  • fluids such as oil and natural gas, etc.
  • the fracturing fluid may include proppants (such as quartz sand, ceramsite, and sand-like materials such as walnut shells). Proppants are used to prop up artificial fractures.
  • Sand plugging may occur during fracturing construction. Sand plugging refers to the phenomenon that the proppant in the fracturing fluid is blocked, causing the pressure of the fracturing fluid to suddenly rise during fracturing construction, which in turn makes it difficult to continue the fracturing construction. .
  • the judgment of sand plugging mainly relies on manual observation of the fracturing construction data collected on site, and then makes judgments based on personal experience.
  • the present application provides a sand plugging prediction method, device, equipment and storage medium, which can predict whether sand plugging occurs during fracturing construction.
  • the present application provides a sand plugging prediction method, the method comprising: obtaining the fracturing construction data in the first period; according to the fracturing construction data in the first period and the first prediction model, determining The fracturing construction data in the second cycle; according to the fracturing construction data in the second cycle and the second prediction model, the prediction result of sand plugging in the second cycle is determined.
  • determining the sand plugging prediction result in the second period includes: preprocessing the fracturing construction data in the second period, Obtain the first data; input the first data, and/or, the fracturing construction data in the second period into the second prediction model, and obtain the sand plugging prediction result in the second period; the first data includes the fracturing in the second period The slope of the construction data.
  • the method before determining the fracturing construction data in the second period after the first period according to the fracturing construction data in the first period and the first prediction model, the method further includes: obtaining historical The maximum fracturing construction data in the time period; according to the quotient of the fracturing construction data in the first period and the maximum fracturing construction data in the historical time period, the normalized fracturing construction data in the first period is obtained.
  • the method further includes: obtaining the fracturing construction data in the third period; obtaining the fracturing construction data in the fourth period after the third period; according to the fracturing construction data in the third period
  • the first training sample is obtained from the data; the fracturing construction data in the fourth period is used as the first label of the first training sample; the long-short memory LSTM network is trained according to the first training sample with the first label, and the training is completed.
  • the method further includes: obtaining the first test sample according to the fracturing construction data in the third period; using the fracturing construction data in the fourth period as the second label of the first test sample; The first prediction model is tested according to the first test sample with the second label to obtain the first prediction model that has completed the test.
  • the method further includes: obtaining the fracturing construction data in the fifth period, and the sand plugging states corresponding to the fracturing construction data at each moment in the fifth cycle; the sand plugging states include Sand plugging and non-sand plugging; the second training sample is obtained according to the fracturing construction data in the fifth cycle; the second training sample is obtained according to the sand plugging state corresponding to the fracturing construction data at each moment in the fifth cycle the third label; the second prediction model is trained according to the second training sample with the third label to obtain the second prediction model that has been trained.
  • the method further includes: obtaining the second test sample according to the fracturing construction data in the fifth period; As the fourth label of the second test sample; the second prediction model is tested according to the second test sample with the fourth label to obtain the second prediction model that has completed the test.
  • the sand plugging prediction method provided by this application can obtain the fracturing construction data in the first period, use the first prediction model to predict the fracturing construction data in the first period, and obtain the fracturing construction data in the second period after the first period The fracturing construction data, and then use the second prediction model to predict the fracturing construction data in the second period, and get the sand plugging prediction result of the second cycle, so as to accurately and timely predict whether sand plugging will occur during future fracturing construction. predict. The stagnation of fracturing construction caused by sand plugging is avoided.
  • the present application provides a sand plugging prediction device, which includes various modules for the method described in the above first aspect or any possible implementation manner of the first aspect.
  • the present application provides a computer program product.
  • the computer program product When the computer program product is run on a computer, it causes the computer to execute the steps of the related method described in the first aspect above, so as to realize the method described in the first aspect above.
  • the present application provides an electronic device, which includes: a processor and a memory; the memory stores instructions executable by the processor; when the processor is configured to execute the instructions, the electronic device implements the above-mentioned first aspect. described method.
  • the present application provides a computer-readable storage medium, the computer-readable storage medium comprising: computer software instructions; when the computer software instructions are run in an electronic device, the electronic device is made to implement the method described in the first aspect above .
  • Figure 1 is a schematic diagram of fracturing construction
  • Fig. 2 is the composition diagram of the sand plugging prediction system provided by the embodiment of the present application.
  • Fig. 3 is a schematic flow chart of the sand plugging prediction method provided by the embodiment of the present application.
  • FIG. 4 is a schematic structural diagram of the LSTM unit provided by the embodiment of the present application.
  • Fig. 5 is a schematic diagram of the XGBoost model provided by the embodiment of the present application.
  • Figure 6 is a schematic diagram of the composition of the sand plugging prediction device provided by the embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • words such as “first” and “second” are used to distinguish the same or similar items with basically the same functions and functions. A skilled person can understand that words such as “first” and “second” do not limit the quantity and execution order.
  • Fracturing refers to the formation of artificial fractures in the oil and gas layer by the pressure of fracturing fluid to improve the flow capacity of fluids (such as oil and natural gas, etc.) in the oil and gas layer.
  • Fig. 1 is a schematic diagram of fracturing construction.
  • the fracturing fluid needs to be injected into the oil and gas layer through the well, and the well can be covered with a casing.
  • the pressure of the fracturing fluid is higher than that of the oil and gas layers, the pressure of the fracturing fluid can press the oil and gas layers out of the fractures, thereby forming artificial fractures in the oil and gas layers.
  • the fracturing fluid may also include a proppant (such as one or more of sand-like materials such as quartz sand, artificial ceramsite, and walnut shells).
  • a proppant such as one or more of sand-like materials such as quartz sand, artificial ceramsite, and walnut shells.
  • Proppants can be brought to artificial fractures by fracturing fluid to support artificial fractures.
  • the proppant may be in the artificial fracture, and/or, the bottom of the well is blocked, causing the pressure of the fracturing fluid to suddenly rise during the fracturing operation, which makes it difficult to continue the fracturing operation.
  • This phenomenon can be called sand plugging .
  • the judgment of sand plugging mainly depends on the manual observation of the fracturing construction data collected on site, and then makes judgments based on personal experience.
  • this application provides a sand plugging prediction method, which can obtain the fracturing construction data in a period, and then use the first prediction model to predict the fracturing construction data in the period, and obtain the fracturing construction data after the period The fracturing construction data in another period, and finally use the second prediction model to predict the fracturing construction data in another period after this period, and obtain the sand plugging prediction result of another period after this period, so as to realize timely , Accurately predict sand plugging.
  • Fig. 2 is a schematic diagram of the composition of the sand plugging prediction system provided by the embodiment of the present application.
  • the sand plugging prediction system may include a data collection device 10 and a computing processing device 20, and the data collecting device 10 and the computing processing device 20 may be connected through a wired network or a wireless network.
  • the data acquisition device 10 may include multiple sensors, such as pressure sensors, flow sensors, and concentration sensors.
  • the data acquisition device 10 can collect fracturing construction data in real time, such as one or more of oil pressure, casing pressure, construction displacement, sand ratio, total sand volume, total liquid volume, sand concentration, and fracturing construction time, etc. .
  • the embodiment of the present application does not limit the specific types of fracturing construction data collected by the data collection device 10.
  • the computing processing device 20 may be a device with computing functions such as a server or a computer.
  • the server may be a single server, or may be a server cluster composed of multiple servers. In some implementations, the server cluster may also be a distributed cluster.
  • the embodiment of the present application does not limit the specific form of the computing processing device 20 .
  • the first prediction model and the second prediction model may be preset in the computing processing device 20 .
  • the first prediction model has the function of predicting the fracturing construction data in another period after the period according to the fracturing construction data in one period.
  • the second prediction model has the function of predicting whether sand plugging occurs according to the fracturing construction data.
  • the calculation processing device 20 can acquire the fracturing construction data in one period from the data acquisition device 10, use the first prediction model to predict the fracturing construction data in another period after this period, and then use the second prediction model to predict According to the fracturing construction data in another cycle, the sand plugging prediction is carried out in another cycle after this cycle, and the sand plugging prediction result of another cycle after this cycle is obtained.
  • Fig. 3 is a schematic flowchart of a method for predicting sand plugging provided in an embodiment of the present application.
  • the main body for executing the method may be the computing processing device 20 shown in FIG. 1 above.
  • the method may include S101 to S103.
  • the first period may be preset in the data collection device 10 or the computing processing device 20 by the administrator.
  • the first period is 5 minutes, or 10 minutes, or 20 minutes, etc.
  • the embodiment of the present application does not limit the specific duration of the first period.
  • the fracturing construction data may include one or more of oil pressure, casing pressure, construction displacement, sand ratio, total sand volume, total liquid volume, sand concentration, and fracturing construction time.
  • Oil pressure is the pressure directed at injecting fracturing fluid into the well.
  • Casing pressure refers to the pressure between the well and the casing when the fracturing fluid is injected into the well.
  • Construction displacement refers to the flow rate of fracturing fluid injected into the well per minute.
  • the sand ratio refers to the proportion of proppant in the fracturing fluid injected into the well.
  • Total sand volume refers to the total amount of proppant included in the fracturing fluid.
  • the total fluid volume refers to the total amount of fracturing fluid injected into the well.
  • the sand concentration refers to the concentration of the proppant in the fracturing fluid injected into the well, and the unit is kilogram per cubic meter (kg/m 3 ).
  • the fracturing construction time includes collecting the time stamps corresponding to the above oil pressure, casing pressure, construction displacement, sand ratio, total sand volume, total liquid volume, and sand concentration.
  • the fracturing construction data in the first period may be as shown in Table 1 below.
  • the fracturing construction data in the first period can include oil pressure, casing pressure, construction displacement, sand ratio, total sand volume, total liquid volume, and sand concentration, etc., in total n types.
  • the values of the n types of fracturing construction data at time 1, time 2, time 3, and time m in total m times are acquired as the fracturing construction data in the first period.
  • acquiring the fracturing construction data in the first period may include: the calculation processing device 20 receives the data acquisition device 10 every first A set of fracturing construction data sent periodically.
  • acquiring the fracturing construction data in the first cycle may include: the computing and processing device 20 receiving the data sent by the data acquisition device 10 The fracturing construction data collected by the collection device 10 in real time; the calculation and processing device 20 determines a set of fracturing construction data corresponding to the first period according to the fracturing construction time in the fracturing construction data.
  • the method may also include: normalizing the fracturing construction data. For example, the maximum fracturing construction data in the historical time period is obtained; according to the quotient of the fracturing construction data in the first period and the maximum fracturing construction data in the historical time period, the normalized fracturing construction data in the first period is obtained. crack construction data.
  • the normalization process of the sand ratio at this moment in the first cycle can be performed according to the following formula (1).
  • G sand ratio represents the sand ratio after normalization.
  • the first prediction model may include a prediction model based on a long-short-term memory (LSTM) network.
  • the second period may be preset in the computing processing device 20 by the administrator, and the duration of the second period may be the same as that of the first period, or the duration of the second period may be different from that of the first period.
  • the embodiment of the present application does not limit the duration relationship between the second period and the first period.
  • the second prediction model may include a sand plugging prediction model based on an extreme gradient boosting (eXtreme gradient boosting, XGBoost) model.
  • the prediction results of sand plugging may include sand plugging or non-sand plugging.
  • determining the sand plugging prediction result in the second period may include: preprocessing the fracturing construction data in the second period to obtain the first Data; input the first data, and/or, the fracturing construction data in the second period into the second prediction model, and obtain the sand plugging prediction result in the second period.
  • the preprocessing may include calculating an average value, calculating a maximum value, calculating a slope, and the like.
  • the first data may include the average value of the fracturing construction data in the second period, the maximum value of the fracturing construction data in the second period, and the slope of the fracturing construction data in the second period.
  • FIG. 5 is a schematic diagram of the XGBoost model provided by the embodiment of the present application.
  • the XGBoost model can include multiple decision trees. Each decision tree may include a root node and multiple levels of nodes that are sequentially refined from the root node to leaf nodes.
  • the sand plugging prediction method provided in the embodiment of the present application can obtain the fracturing construction data in the first period, use the first prediction model to predict the fracturing construction data in the first period, and obtain the second period after the first period Then use the second prediction model to predict the fracturing construction data in the second period, and get the prediction result of sand plugging in the second cycle, so as to accurately and accurately determine whether sand plugging will occur during future fracturing operations. predict in time.
  • the fracturing construction data in the second period may also be preprocessed to obtain the slope of the fracturing construction data.
  • the slope represents the change rate of data, which can more intuitively reflect the change of fracturing construction data, and generally improves the accuracy of sand plugging prediction for the second period using the second prediction model.
  • the LSTM network before using the first prediction model to predict the fracturing construction data in the second period after the first period, may also be trained to obtain the first prediction model. That is, according to the fracturing construction data in the first period and the first prediction model, before determining the fracturing construction data in the second period after the first period, the method may further include: obtaining the fracturing construction data in the third period Construction data; obtain the fracturing construction data in the fourth period after the third period; obtain the first training sample according to the fracturing construction data in the third period; use the fracturing construction data in the fourth period as the first training sample The first label; according to the first training sample with the first label, the LSTM network is trained to obtain the first prediction model that has completed the training.
  • the third period may include any period before the first period.
  • the duration of the third period may be the same as that of the first period.
  • the third cycle can be preset by the administrator.
  • the fourth period may also include any period before the first period.
  • the duration of the fourth period may be the same as that of the second period.
  • the fourth cycle can be preset by the management personnel.
  • the first training samples may also be obtained after performing importance analysis on the fracturing construction data in the third period by using the LSTM network. That is, obtaining the first training sample according to the fracturing construction data in the third period may include: using the LSTM network to analyze the importance of the fracturing construction data in the third period to obtain the fracturing construction data in the third period According to the preset first quantity, the fracturing construction data is selected from the importance ranking of the fracturing construction data in the third period as the first training sample.
  • the first quantity can be preset by the management personnel.
  • the first number may be 4, or 5, or 6, etc.
  • the embodiment of the present application does not limit the specific numerical value of the first quantity.
  • the order of importance of the obtained fracturing construction data in the third period from large to small is: construction displacement, oil pressure, sand ratio, sand concentration, total sand volume, total liquid volume, and set pressure.
  • the first training samples may include construction displacement, oil pressure, sand ratio, sand concentration, and total sand volume in the third period.
  • the LSTM network may include an input layer, a hidden layer, and an output layer.
  • the input layer can be used to input training samples into the prediction model.
  • the hidden layer may include multiple layers (for example, 4 layers), and each layer may include one or more LSTM units.
  • the input of the hidden layer is also the input of the LSTM unit, including an n-dimensional vector corresponding to a certain moment in the training sample (that is, a column of fracturing construction data corresponding to a certain moment in Table 1 above) and a q-dimensional array, and q is also That is, the duration of the second cycle, for example, q is 60, that is, to predict the fracturing construction data within 60 seconds after the first cycle.
  • the hidden layer has two directions of output, one is the n-dimensional vector for the hidden layer at the next moment, and the other is the n-dimensional vector for the fully connected unit.
  • the output of the fully connected unit is the n-dimensional vector of the previous LSTM unit, and the output is an n-dimensional vector for the output layer.
  • the output layer can be used to predict the learning results of the hidden layer, and calculate the error by comparing the output fracturing construction data with the fracturing construction data in the label of the training sample.
  • FIG. 4 is a schematic structural diagram of an LSTM unit provided in an embodiment of the present application.
  • the LSTM unit of the hidden layer may include a forget gate, an update gate, an output gate, an activation function 1, and an activation function 2.
  • the activation function 1 is the softsign function
  • the activation function 2 is the relu function.
  • the value of the forget gate can be calculated according to the following formula (2).
  • x t represents the n-dimensional vector of the current input.
  • a t-1 represents the n-dimensional vector output at the previous moment.
  • w f represents the weight matrix of the forget gate.
  • b f represents the bias term of the forget gate.
  • the value of the update gate can be calculated according to the following formula (3).
  • the candidate values of the memory cells can also be calculated according to the following formula (4).
  • c' t represents the candidate value of the memory cell.
  • w c represents the weight matrix of the cell state.
  • b c represents the bias term of the cell state.
  • c t represents the unit state of the LSTM unit at the current moment.
  • c t-1 represents the unit state of the LSTM unit at the previous moment at the current moment.
  • At represents the output of the hidden layer.
  • the hidden layer may output the above output result at to the output layer.
  • the output layer can decode a t to get q-dimensional (predicted) pressure construction data.
  • the trained first prediction model may also be tested. That is, the method may further include: obtaining the first test sample according to the fracturing construction data in the third period; using the fracturing construction data in the fourth period as the second label of the first test sample; The first test sample of the label tests the first prediction model to obtain the first prediction model that has completed the test.
  • the process of testing the first prediction model according to the first test sample with the second label may include: inputting the first test sample with the second label into the trained first prediction model network to obtain the first prediction model network.
  • the process of inputting the test samples into the trained LSTM network to obtain the test results corresponding to the test samples can refer to the training process described in the above formula (2) to formula (7), and will not be repeated here.
  • the error threshold can be preset in the computing processing device 20 by a manager.
  • the error threshold may be 10%, or 20%, and so on.
  • the embodiment of the present application does not limit the specific value of the error threshold.
  • the above-mentioned first prediction model training process can continue to be repeated until the error between the first test result and the second label is less than the preset error threshold.
  • the set error threshold When the error between the first test result and the second label is equal to the preset error threshold, the first prediction model that completes the test can be obtained, or continue to repeat the above-mentioned training process of the first prediction model until the first test result and the second The error of the label is less than the preset error threshold.
  • the embodiment of the present application does not limit this.
  • the first training sample and the first testing sample may also be obtained at the same time. That is, before determining the fracturing construction data in the second period after the first period according to the fracturing construction data in the first period and the first prediction model, the method may further include: obtaining the fracturing construction data in the third period Construction data; obtain the fracturing construction data in the fourth period after the third period; according to the fracturing construction data in the third period, obtain the first training sample and the first test sample according to the first ratio; according to the fracturing construction data in the fourth period
  • the labels of the first training samples and the labels of the first testing samples are obtained from the fracturing construction data.
  • the first ratio can be preset by the management personnel.
  • the first proportion is 80% of training samples, 20% of testing samples, or 90% of training samples, 10% of testing samples, etc.
  • the embodiment of the present application does not limit the specific numerical value of the first ratio.
  • the second prediction model before using the second prediction model to predict the fracturing construction data in the second period, the second prediction model may also be trained to obtain a trained second prediction model. That is, according to the fracturing construction data in the second period and the second prediction model, before determining the sand plugging prediction result in the second period, the method may further include: obtaining the fracturing construction data in the fifth period, and the fifth The sand plugging state corresponding to the fracturing construction data at each moment in the period; obtain the second training sample according to the fracturing construction data in the fifth period; The sand plugging state of is used as the third label of the second training sample; the second prediction model is trained according to the second training sample with the third label, and the second prediction model that has completed the training is obtained. For example, a grid search (grid search) method is used for model parameter adjustment to determine the second prediction model with the highest accuracy.
  • a grid search grid search
  • the fifth cycle may include any cycle before the first cycle.
  • the duration of the fifth cycle may be the same as that of the first cycle.
  • the fifth cycle can be preset by the management personnel.
  • the sand plugging status (that is, the third label above) corresponding to the fracturing construction data at each moment in the fifth period may include sand plugging and non-sand plugging.
  • the second training sample may also be obtained after performing importance analysis on the fracturing construction data in the fifth period by using the LSTM network. That is, obtaining the second training sample according to the fracturing construction data in the fifth period may include: using the LSTM network to perform importance analysis on the fracturing construction data in the fifth period to obtain the fracturing construction data in the fifth period According to the preset second quantity, select the fracturing construction data from the importance ranking of the fracturing construction data in the fifth period as the second training sample.
  • the second quantity can be preset by the manager.
  • the second number may be 4, or 5, or 6, etc.
  • the embodiment of the present application does not limit the specific value of the second quantity.
  • the order of importance of the obtained fracturing construction data in the fifth cycle is as follows: construction displacement, oil pressure, sand ratio, sand concentration, total sand volume, total liquid volume, and casing pressure.
  • the second training samples may include construction displacement, oil pressure, sand ratio, and sand concentration in the fifth period.
  • the second training samples can be obtained simultaneously with the first training samples.
  • the first training sample can be obtained after using the LSTM network to perform importance analysis on the fracturing construction data in the third period.
  • the second training sample may also be obtained after performing importance analysis on the fracturing construction data in the fifth period by using the LSTM network.
  • the duration of the fifth cycle may be the same as that of the third cycle.
  • selecting the fracturing construction data as the second training sample from the importance ranking of the fracturing construction data in the fifth period may include: according to the preset first quantity, selecting the fracturing construction data from the third Select the fracturing construction data as the first candidate training sample in the importance ranking of the fracturing construction data in the cycle; select the fracturing data from the importance ranking of the fracturing construction data in the fifth cycle according to the preset second quantity
  • the construction data is used as the second candidate training sample; the fracturing construction data shared by the first candidate training sample and the second candidate training sample is selected as the first prediction model and the second training sample.
  • the trained second prediction model may also be tested. That is, the method may further include: obtaining a second test sample according to the fracturing construction data in the fifth period; using the sand plugging state corresponding to the fracturing construction data at each moment in the fifth period as the second test sample the fourth label; according to the second test sample with the fourth label, the second prediction model is tested to obtain the second prediction model that has completed the test.
  • the process of testing the second predictive model according to the second test sample may include: inputting the second test sample into the trained second predictive model network to obtain a second test result corresponding to the second test sample; according to the second test sample The second test result and the fourth label determine the accuracy rate of the second prediction model; when the accuracy rate is greater than the preset accuracy rate threshold, the second prediction model that completes the test is obtained.
  • the accuracy rate threshold can be preset by the manager.
  • the accuracy threshold is 70%, or 80%.
  • the embodiment of the present application does not limit the specific value of the accuracy threshold.
  • the above-mentioned second prediction model training process may continue to be repeated until the accuracy rate is greater than the preset accuracy rate threshold.
  • the accuracy rate is equal to the preset accuracy rate threshold, a second prediction model that completes the test can be obtained, or the above-mentioned training process of the second prediction model can be repeated until the accuracy rate is greater than the preset accuracy rate threshold.
  • the embodiment of the present application does not limit this.
  • the second training samples and the second testing samples may also be obtained at the same time. That is, according to the fracturing construction data in the second period and the second prediction model, before determining the sand plugging prediction result in the second period, the method may further include: obtaining the fracturing construction data in the fifth period, and the fifth The sand plugging state corresponding to the fracturing construction data at each moment in the period; according to the fracturing construction data in the fifth period, the second training sample and the second test sample are obtained according to the second ratio; according to the fracturing construction data in the fifth period The fracturing construction data at each moment corresponds to the sand plugging state, and the third label of the second training sample and the fourth label of the second test sample are obtained.
  • the second ratio can be preset by the management personnel.
  • the second ratio may be the same as or different from the first ratio described above.
  • the second proportion is 80% of training samples, 20% of testing samples, or 90% of training samples, 10% of testing samples, and so on.
  • the embodiment of the present application does not limit the specific value of the second ratio.
  • the embodiment of the present application further provides a device for predicting sand plugging, and the device for predicting sand plugging can be applied to the above-mentioned computing and processing device 20 .
  • Fig. 6 is a schematic diagram of the composition of the sand plugging prediction device provided in the embodiment of the present application.
  • the apparatus may include: an acquisition model 601 and a processing module 602 .
  • An acquisition module 601 configured to acquire fracturing construction data in the first period.
  • the processing module 602 is used to determine the fracturing construction data in the second period after the first period according to the fracturing construction data in the first period and the first prediction model;
  • the second prediction model determines the sand plugging prediction results of the second cycle.
  • the processing module 602 is specifically configured to preprocess the fracturing construction data in the second period to obtain the first data; the first data, and/or, the fracturing construction in the second period
  • the data is input into the second prediction model to obtain the prediction result of sand plugging in the second period; the first data includes the slope of the fracturing construction data in the second period.
  • the acquiring module 601 is also configured to acquire the maximum fracturing construction data in a historical time period.
  • the processing module 602 is further configured to obtain the normalized fracturing construction data in the first period according to the quotient of the fracturing construction data in the first period and the maximum fracturing construction data in the historical time period.
  • the acquiring module 601 is further configured to acquire the fracturing construction data in the third period; acquire the fracturing construction data in the fourth period after the third period.
  • the processing module 602 is also used to use the fracturing construction data in the fourth period as the first label of the first training sample; train the long-short memory LSTM network according to the first training sample with the first label, and obtain the completed training first predictive model.
  • the processing module 602 is also configured to obtain the first test sample according to the fracturing construction data in the third period; use the fracturing construction data in the fourth period as the second label of the first test sample ; Test the first prediction model according to the first test sample with the second label to obtain the first prediction model that has completed the test.
  • the acquiring module 601 is also used to acquire the fracturing construction data in the fifth period, and the sand plugging status corresponding to the fracturing construction data at each moment in the fifth period; the sand plugging status Including sand plugging and non-sand plugging.
  • the processing module 602 is also used to obtain the second training sample according to the fracturing construction data in the fifth period; according to the sand plugging state corresponding to the fracturing construction data at each moment in the fifth period, obtain the second training sample a third label; training the second prediction model according to the second training sample with the third label to obtain a trained second prediction model.
  • the processing module 602 is also used to obtain the second test sample according to the fracturing construction data in the fifth period; The state is used as the fourth label of the second test sample; the second prediction model is tested according to the second test sample with the fourth label, and the second prediction model that completes the test is obtained.
  • the embodiment of the present application further provides a computer program product, which, when running on a computer, causes the computer to execute the above-mentioned relevant method steps, so as to implement the methods in the foregoing method embodiments.
  • the embodiment of the present application further provides an electronic device, and the electronic device may be the computing processing device 20 described in the foregoing method embodiments.
  • FIG. 7 is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As shown in FIG. 7 , the electronic device may include: a processor 701 and a memory 702; the memory 702 stores instructions executable by the processor 701; when the processor 701 is configured to execute the instructions, the electronic device implements the aforementioned method embodiment method described in .
  • the embodiment of the present application also provides a computer-readable storage medium, on which computer program instructions are stored; when the computer program instructions are executed by the electronic device, the electronic device realizes the above-mentioned embodiment. method described in .
  • the computer-readable storage medium may be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device wait.

Landscapes

  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Geology (AREA)
  • Mining & Mineral Resources (AREA)
  • Physics & Mathematics (AREA)
  • Environmental & Geological Engineering (AREA)
  • Fluid Mechanics (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Geochemistry & Mineralogy (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种砂堵预测方法、装置、设备及存储介质被公开。该方法包括：获取第一周期内的压裂施工数据；根据第一周期内的压裂施工数据和第一预测模型，确定第一周期之后的第二周期内的压裂施工数据；根据第二周期内的压裂施工数据和第二预测模型，确定第二周期的砂堵预测结果。该方法适用于压裂作业过程中，用于预测压裂作业过程中是否会出现砂堵。

Description

砂堵预测方法、装置、设备及存储介质 技术领域
本申请涉及石油天然气压裂作业技术领域，尤其涉及一种砂堵预测方法、装置、设备及存储介质。
背景技术
压裂是指利用压裂液的压力在油气层中形成人工裂缝，提高油气层中流体(例如石油和天然气等)流动能力的一种油气层改造技术。进行压裂施工(或者称为压裂作业)时，需要将高压的压裂液注入油气层，利用压裂液的压力把油气层压出人工裂缝。压裂液中可以包括支撑剂(例如石英砂、人造陶粒、以及核桃壳等砂状材料)。支撑剂用于支撑人工裂缝。进行压裂施工时可能出现砂堵，砂堵是指压裂液中的支撑剂发生了堵塞，使压裂施工时的压裂液的压力突然升高，进而导致压裂施工难以进行下去的现象。
目前，对于砂堵的判断主要依靠对现场采集到的压裂施工数据进行人工观察，然后根据个人经验进行判断。
但是，根据个人经验进行判断容易受到工作人员的个人经验的水平差异的限制，无法及时、准确地对砂堵进行预测。
发明内容
本申请提供一种砂堵预测方法、装置、设备及存储介质，可以对压裂施工过程中是否出现砂堵进行预测。
第一方面，本申请提供一种砂堵预测方法，该方法包括：获取第一周期内的压裂施工数据；根据第一周期内的压裂施工数据和第一预测模型，确定第一周期之后的第二周期内的压裂施工数据；根据第二周期内的压裂施工数据和第二预测模型，确定第二周期的砂堵预测结果。
一种可能的实现方式中，根据第二周期内的压裂施工数据和第二预测模型，确定第二周期的砂堵预测结果，包括：对第二周期内的压裂施工数据进行预处理，得到第一数据；将第一数据，和/或，第二周期内的压裂施工数据输入第二预测模型，得到第二周期的砂堵预测结果；第一数据包括第二周期内的压裂施工数据的斜率。
另一种可能的实现方式中，在根据第一周期内的压裂施工数据和第一预测模型，确定第一周期之后的第二周期内的压裂施工数据之前，该方法还包括：获取历史时间段内的最大压裂施工数据；根据第一周期内的压裂施工数据和历史时间段内的最大压裂施工数据的商，得到归一化之后的第一周期内的压裂施工数据。
又一种可能的实现方式中，该方法还包括：获取第三周期内的压裂施工数据；获取第三周期之后的第四周期内的压裂施工数据；根据第三周期内的压裂施工数据获得第一训练样本；将第四周期内的压裂施工数据作为第一训练样本的第一标签；根据带有第一标签的第一训练样本对长短记忆LSTM网络进行训练，得到完成训练的第一预测模型。
又一种可能的实现方式中，该方法还包括：根据第三周期内的压裂施工数据获得第一测试样本；将第四周期内的压裂施工数据作为第一测试样本的第二标签；根据带有第二标签的第一测试样本对第一预测模型进行测试，得到完成测试的第一预测模型。
又一种可能的实现方式中，该方法还包括：获取第五周期内的压裂施工数据、以及第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态；砂堵状态包括已砂堵和未砂堵；根据第五周期内的压裂施工数据获得第二训练样本；根据第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态，获得第二训练样本的第三标签；根据带有第三标签的第二训练样本对第二预测模型进行训练，得到完成训练的第二预测模型。
又一种可能的实现方式中，该方法还包括：根据第五周期内的压裂施工数据获得第二测试样本；将第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态作为第二测试样本的第四标签；根据带有第四标签的第二测试样本对第二预测模型进行测试，得到完成测试的第二预测模型。
本申请提供的砂堵预测方法，可以获取第一周期内的压裂施工数据，利用第一预测模型对第一周期内的压裂施工数据进行预测，得到第一周期之后的第二周期内的压裂施工数据，再利用第二预测模型预测第二周期内的压裂施工数据，得到第二周期的砂堵预测结果，从而可以对未来压裂施工时是否会发生砂堵进行准确、及时地预测。避免了砂堵导致的压裂施工停滞。
第二方面，本申请提供一种砂堵预测装置，该砂堵预测装置包括用于之上第一方面或第一方面中任一种可能的实现方式所述的方法的各个模块。
第三方面，本申请提供一种计算机程序产品，当该计算机程序产品在计算机上运行时，使得计算机执行上述第一方面所述相关方法的步骤，以实现上述第一方面所述的方法。
第四方面，本申请提供一种电子设备，该电子设备包括：处理器和存储器；存储器存储有处理器可执行的指令；处理器被配置为执行指令时，使得电子设备实现上述第一方面所述的方法。
第五方面，本申请提供一种计算机可读存储介质，该计算机可读存储介质包括：计算机软件指令；当计算机软件指令在电子设备中运行时，使得电子设备实现上述第一方面所述的方法。
上述第二方面至第五方面的有益效果可以参考第一方面所述，不再赘述。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为压裂施工示意图；
图2为本申请实施例提供的砂堵预测系统的组成示意图；
图3为本申请实施例提供的砂堵预测方法的流程示意图；
图4为本申请实施例提供的LSTM单元的结构示意图；
图5为本申请实施例提供的XGBoost模型示意图；
图6为本申请实施例提供的砂堵预测装置的组成示意图；
图7为本申请实施例提供的电子设备的结构示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
需要说明的是，本申请实施例中，“示例性地”或者“例如”等词用于表示作例子、例证或说明。本申请实施例中被描述为“示例性地”或者“例如”的任何实施例或设计方案不应被解释为比其它实施例或设计方案更优选或更具优势。确切而言，使用“示例性地”或者“例如”等词旨在以具体方式呈现相关概念。
为了便于清楚描述本申请实施例的技术方案，在本申请的实施例中，采用了“第一”、“第二”等字样对功能和作用基本相同的相同项或相似项进行区分，本领域技术人员可以理解“第一”、“第二”等字样并不是在对数量和执行次序进行限定。
压裂是指利用压裂液的压力在油气层中形成人工裂缝，提高油气层中流体(例如石油和天然气等)流动能力的一种油气层改造技术。
示例性地，图1为压裂施工示意图。如图1所示，进行压裂施工时，需要将压裂液通过井道注入油气层，井道外可以套有套管。当压裂液的压力高于油气层的压力时，压裂液的压力可以将油气层压出裂缝，从而在油气层形成人工裂缝。
压裂液中还可以包括支撑剂(例如石英砂、人造陶粒、以及核桃壳等砂状材料中的一种或多种)。支撑剂可以被压裂液带至人工裂缝处，支撑人工裂缝。但是支撑剂可能在人工裂缝处，和/或，井道底部发送堵塞，使压裂施工时的压裂液的压力突然升高，进而导致压裂施工难以进行下去，这种现象可以称作砂堵。目前，对于砂堵的判断主要依靠对现场采集到的压裂施工的数据进行人工观察，然后根据个人经验进行判断。
但是，根据个人经验进行判断容易受到工作人员的个人经验的水平差异的限制，无法及时、准确地对砂堵进行预测。
在此背景技术下，本申请提供一种砂堵预测方法，可以获取一个周期内的压裂施工数据，然后利用第一预测模型对该周期内的压裂施工数据进行预测，得到该周期之后的另一个周期内的压裂施工数据，最后利用第二预测模型对该周期之后的另一个周期内的压裂施工数据进行预测，得到该周期之后的另一个周期的砂堵预测结果，从而实现及时、准确地对砂堵进行预测。
下面将结合附图对本申请实施例提供的砂堵预测方法进行详细描述。
图2为本申请实施例提供的砂堵预测系统的组成示意图。如图2所示，该砂堵预测系统可以包括数据采集设备10和计算处理设备20，数据采集设备10和计算处理设备20可以通过有线网络或者无线网络连接。
数据采集设备10可以包括多个传感器，例如压力传感器、流量传感器、以及浓度传感器等。数据采集设备10可以实时采集压裂施工数据，例如，油压、套压、施工排量、砂比、总砂量、总液量、砂浓度、以及压裂施工时间等的一种或多种。本申请实 施例对数据采集设备10采集到的压裂施工数据的具体种类不作限制。
计算处理设备20可以是服务器或计算机等具有计算功能的设备。其中，服务器可以是单独的一个服务器，或者，也可以是由多个服务器构成的服务器集群。部分实施方式中，服务器集群还可以是分布式集群。本申请实施例对计算处理设备20的具体形态不作限制。计算处理设备20中可以预设有第一预测模型和第二预测模型。第一预测模型具有根据一个周期内的压裂施工数据预测数该周期之后的另一个周期内的压裂施工数据的功能。第二预测模型具有根据压裂施工数据预测是否发生砂堵的功能。
计算处理设备20可以从数据采集设备10获取一个周期内的压裂施工数据，利用第一预测模型预测该周期之后的另一个周期内的压裂施工数据，然后利用第二预测模型根据该周期之后的另一个周期内的压裂施工数据，对该周期之后的另一个周期进行砂堵预测，得到该周期之后的另一个周期的砂堵预测结果。
图3为本申请实施例提供的砂堵预测方法的流程示意图。执行该方法的主体可以是上述图1所示的计算处理设备20。如图3所示，该方法可以包括S101至S103。
S101、获取第一周期内的压裂施工数据。
其中，第一周期可以由管理人员预设在上述数据采集设备10或者计算处理设备20中。例如，第一周期是5分钟、或者10分钟、又或者20分钟等。本申请实施例对第一周期的具体时长不作限制。如上所述，压裂施工数据可以包括油压、套压、施工排量、砂比、总砂量、总液量、砂浓度、以及压裂施工时间等的一种或多种。油压是指向井道中注入压裂液的压力。套压是指向井道中注入压裂液时井道和套管之间的压力。施工排量是指每分钟向井道中注入的压裂液的流量。砂比是指向井道中注入的压裂液中，支撑剂所占的比例。总砂量是指压裂液中包括的支撑剂的总量。总液量是指向井道中注入压裂液的总量。砂浓度是指向井道注入的压裂液中，支撑剂的浓度，单位为千克每立方米(kg/m 3)。压裂施工时间包括采集上述油压、套压、施工排量、砂比、总砂量、总液量、以及砂浓度分别对应的时间戳。
示例性地，以第一周期的时长等于时刻m减去时刻1为例，第一周期内的压裂施工数据可以如下述表1所示。
表1
序号 压裂施工数据 时刻1 时刻2 时刻3 … 时刻m
1 油压 油压1 油压2 油压3 … 油压m
2 套压 套压1 套压2 套压3 … 套压m
3 施工排量 排量1 排量2 排量3 … 排量m
4 砂比 比例1 比例2 比例3 … 比例m
5 总砂量 砂量1 砂量2 砂量3 … 砂量m
6 总液量 液量1 液量2 液量3 … 液量m
… … … … … … …
n 砂浓度 浓度1 浓度2 浓度3 … 浓度m
如表1所示，第一周期内的压裂施工数据可以包括油压、套压、施工排量、砂比、总砂量、总液量、以及砂浓度等共计n种。获取该n种压裂施工数据在时刻1、时刻2、时刻3、以及时刻m等共计m个时刻的数值作为第一周期内的压裂施工数据。
一种可能的实现方式中，当第一周期是预设在数据采集设备10中时，获取第一周期内的压裂施工数据，可以包括：计算处理设备20接收数据采集设备10每隔第一周期发送的一组压裂施工数据。
另一种可能的实现方式中，当第一周期是预设在计算处理设备20中时，获取第一周期内的压裂施工数据，可以包括：计算处理设备20接收数据采集设备10发送的数据采集设备10实时采集的压裂施工数据；计算处理设备20根据压裂施工数据中的压裂施工时间，确定第一周期对应的一组压裂施工数据。
可选地，在获取第一周期内的压裂施工数据之后，根据第一周期内的压裂施工数据和第一预测模型，确定第一周期之后的第二周期内的压裂施工数据之前，该方法还可以包括：对压裂施工数据进行归一化处理。例如，获取历史时间段内的最大压裂施工数据；根据第一周期内的压裂施工数据和历史时间段内的最大压裂施工数据的商，得到归一化之后的第一周期内的压裂施工数据。
示例性地，以压裂施工数据中的砂比为例，假设第一周期内的某一时刻的砂比为20％，在第一周期之前的历史时间段内的砂比的最大值为40％。则对第一周期内的该时刻的砂比进行归一化处理可以按照下述公式(1)进行。
Figure PCTCN2022076163-appb-000001
公式(1)中，G 砂比表示归一化处理后的砂比。
S102、根据第一周期内的压裂施工数据和第一预测模型，确定第一周期之后的第二周期内的压裂施工数据。
其中，第一预测模型可以包括基于长短记忆(long short-term memory，LSTM)网络的预测模型。第二周期可以由管理人员预设在计算处理设备20中，第二周期的时长可以和第一周期的时长相同，或者，第二周期的时长与第一周期的时长不同。本申请实施例对第二周期和第一周期的时长关系不做限制。
S103、根据第二周期内的压裂施工数据和第二预测模型，确定第二周期的砂堵预测结果。
其中，第二预测模型可以包括基于极限梯度提升(eXtreme gradient boosting，XGBoost)模型的砂堵预测模型。砂堵预测结果可以包括已砂堵、或者未砂堵。
可选地，根据第二周期内的压裂施工数据和第二预测模型，确定第二周期的砂堵预测结果，可以包括：对第二周期内的压裂施工数据进行预处理，得到第一数据；将第一数据，和/或，第二周期内的压裂施工数据输入第二预测模型，得到第二周期的砂堵预测结果。
其中，预处理可以包括求平均值、求最大值、以及求斜率等。相对应地，第一数据可以包括第二周期内的压裂施工数据的平均值、第二周期内的压裂施工数据的最大值、以及第二周期内的压裂施工数据的斜率。
示例性地，图5为本申请实施例提供的XGBoost模型示意图。如图5所示，XGBoost模型可以包括多棵决策树。每棵决策树可以包括一个根节点、以及由根节点依次细化到叶子节点的多个层级的节点。
本申请实施例提供的砂堵预测方法，可以获取第一周期内的压裂施工数据，利用第一预测模型对第一周期内的压裂施工数据进行预测，得到第一周期之后的第二周期 内的压裂施工数据，再利用第二预测模型预测第二周期内的压裂施工数据，得到第二周期的砂堵预测结果，从而可以对未来压裂施工时是否会发生砂堵进行准确、及时地预测。
另外，本申请实施例提供的砂堵预测方法中，还可以对第二周期内的压裂施工数据进行预处理，获得压裂施工数据的斜率。斜率代表了数据的变化速率，更能直观体现压裂施工数据的变化，整体上提高了利用第二预测模型对第二周期进行砂堵预测的准确率。
一些可能的实施例中，在利用第一预测模型预测第一周期之后的第二周期内的压裂施工数据之前，还可以对LSTM网络进行训练，得到第一预测模型。也即，根据第一周期内的压裂施工数据和第一预测模型，确定第一周期之后的第二周期内的压裂施工数据之前，该方法还可以包括：获取第三周期内的压裂施工数据；获取第三周期之后的第四周期内的压裂施工数据；根据第三周期内的压裂施工数据获得第一训练样本；将第四周期内的压裂施工数据作为第一训练样本的第一标签；根据带有第一标签的第一训练样本对LSTM网络进行训练，得到完成训练的第一预测模型。
其中，第三周期可以包括第一周期之前的任意一个周期。第三周期的时长可以和第一周期的时长相同。第三周期可以由管理人员预设。类似地，第四周期也可以包括第一周期之前的任意一个周期。第四周期的时长可以和第二周期相同。第四周期可以由管理人员预设。
可选地，第一训练样本还可以是利用LSTM网络对第三周期内的压裂施工数据进行重要性分析之后得到的。也即，根据第三周期内的压裂施工数据获得第一训练样本，可以包括：利用LSTM网络对第三周期内的压裂施工数据进行重要性分析，得到第三周期内的压裂施工数据的重要性排序；根据预设的第一数量，从该第三周期内的压裂施工数据的重要性排序中选择压裂施工数据作为第一训练样本。
其中，第一数量可以由管理人员预设。例如，第一数量可以是4、或者5、又或者6等。本申请实施例对第一数量的具体数值不作限制。
示例性地，以获得的第三周期内的压裂施工数据的重要性排序从大到小依次为：施工排量、油压、砂比、砂浓度、总砂量、总液量、以及套压。假设预设的第一数量为5，则第一训练样本可以包括第三周期内的施工排量、油压、砂比、砂浓度、以及总砂量。
需要说明的是，LSTM网络可以包括输入层、隐藏层、以及输出层。其中，输入层，可以用于向预测模型中输入训练样本。隐藏层，可以包括多层(例如4层)，每层可以包括一个或多个LSTM单元。隐藏层的输入也即LSTM单元的输入，包括训练样本中某一时刻对应的一个n维向量(也即上述表1中某一时刻对应的一列压裂施工数据)和一个q维数组，q也即上述第二周期的时长，例如q为60，也即预测第一周期之后的60秒内的压裂施工数据。隐藏层有两个方向的输出，一个是给下一时刻隐藏层的n维向量，另一个是给全连接单元的n维向量，全连接单元的输出为上一个LSTM单元的n维向量，输出为给输出层的n维向量。输出层，可以用于对隐藏层的学习结果进行预测，通过输出的压裂施工数据与训练样本的标签中的压裂施工数据进行比较，计算误差。
示例性地，图4为本申请实施例提供的LSTM单元的结构示意图。如图4所示，隐藏层的LSTM单元可以包括遗忘门、更新门、输出门、激活函数1、以及激活函数2。以下以激活函数1为softsign函数，激活函数2为relu函数进行介绍。
其中，遗忘门的值可以按照下述公式(2)进行计算。
Figure PCTCN2022076163-appb-000002
公式(2)中，
Figure PCTCN2022076163-appb-000003
表示遗忘门的值。x t表示当前输入的n维向量。a t-1表示上一时刻输出的n维向量。w f表示遗忘门的权重矩阵。b f表示遗忘门的偏置项。
更新门的值可以按照下述公式(3)进行计算。
Figure PCTCN2022076163-appb-000004
公式(3)中，
Figure PCTCN2022076163-appb-000005
表示更新门的值。w u表示更新门的权重矩阵。b u表示更新门的偏置项。
计算得到遗忘门的值和更新门的值之后，还可以按照下述公式(4)计算记忆细胞候选值。
c′ t＝tanh(w c[a t-1,x t]+b c)       公式(4)
公式(4)中，c′ t表示记忆细胞候选值。w c表示单元状态的权重矩阵。b c表示单元状态的偏置项。
然后按照下述公式(5)计算LSTM单元当前时刻的单元状态。
Figure PCTCN2022076163-appb-000006
公式(5)中，c t表示LSTM单元当前时刻的单元状态。c t-1表示LSTM单元当前时刻的前一时刻的单元状态。
再然后，按照下述公式(6)计算LSTM单元当前时刻的输出门的值。
Figure PCTCN2022076163-appb-000007
公式(6)中，
Figure PCTCN2022076163-appb-000008
表示LSTM单元当前时刻的输出门的值。w o表示输出门的权重矩阵。b o表示输出门的偏置项。
若t<m,则令t＝t+1,重复按照上述公式(2)至公式(6)计算
Figure PCTCN2022076163-appb-000009
c′ t、c t、以及
Figure PCTCN2022076163-appb-000010
直至最后一个时刻t＝m时，按照下述公式(7)计算隐藏层的输出结果。
Figure PCTCN2022076163-appb-000011
公式(7)中，a t表示隐藏层的输出结果。
隐藏层可以将上述输出结果a t输出给输出层。输出层可以对a t进行解码，得到q维的(预测)压力施工数据。
可选地，在利用第一训练样本对LSTM网络进行训练之后，还可以对训练过后的第一预测模型进行测试。也即，该方法还可以包括：根据第三周期内的压裂施工数据获得第一测试样本；将第四周期内的压裂施工数据作为第一测试样本的第二标签；根据带有第二标签的第一测试样本对第一预测模型进行测试，得到完成测试的第一预测模型。
可选地，根据带有第二标签的第一测试样本对第一预测模型进行测试的过程可以包括：将带有第二标签的第一测试样本输入完成训练的第一预测模型网络，得到第一测试样本对应的第一测试结果；当该第一测试结果和第二标签的误差小于预设的误差阈值时，得到完成测试的第一预测模型。
需要说明的是，将测试样本输入完成训练的LSTM网络，得到测试样本对应的测试结果的过程，可以参照上述公式(2)至公式(7)所述的训练过程，此处不再赘述。 误差阈值可以由管理人员预设在计算处理设备20中。例如，误差阈值可以是10％、或者20％等。本申请实施例对误差阈值的具体数值不作限制。
同样需要说明的是，当第一测试结果和第二标签的误差大于预设的误差阈值时，还可以继续重复上述第一预测模型训练过程，直至第一测试结果和第二标签的误差小于预设的误差阈值。当第一测试结果和第二标签的误差等于预设的误差阈值时，可以得到完成测试的第一预测模型，或者，继续重复上述第一预测模型的训练过程，直至第一测试结果和第二标签的误差小于预设的误差阈值。本申请实施例对此不作限制。
可选地，第一训练样本和第一测试样本还可以是同时获得。也即，根据第一周期内的压裂施工数据和第一预测模型，确定第一周期之后的第二周期内的压裂施工数据之前，该方法还可以包括：获取第三周期内的压裂施工数据；获取第三周期之后的第四周期内的压裂施工数据；根据第三周期内的压裂施工数据，按照第一比例获得第一训练样本和第一测试样本；根据第四周期内的压裂施工数据获得第一训练样本的标签和第一测试样本的标签。
其中，第一比例可以由管理人员预设。例如，第一比例为80％的训练样本、20％的测试样本，或者90％的训练样本、10％的测试样本等。本申请实施例对第一比例的具体数值不作限制。
另一些可能的实施例中，在利用第二预测模型预测第二周期内的压裂施工数据之前，还可以对第二预测模型进行训练，得到完成训练的第二预测模型。也即，根据第二周期内的压裂施工数据和第二预测模型，确定第二周期的砂堵预测结果之前，该方法还可以包括：获取第五周期内的压裂施工数据、以及第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态；根据第五周期内的压裂施工数据获得第二训练样本；将第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态作为第二训练样本的第三标签；根据带有第三标签的第二训练样本对第二预测模型进行训练，得到完成训练的第二预测模型。例如，使用网格搜索(grid search)法进行模型调参，确定准确度最高的第二预测模型。
其中，第五周期可以包括第一周期之前的任意一个周期。第五周期的时长可以和第一周期的时长相同。第五周期可以由管理人员预设。第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态(也即上述第三标签)可以包括已砂堵和未砂堵。
可选地，第二训练样本也可以是利用LSTM网络对第五周期内的压裂施工数据进行重要性分析之后得到的。也即，根据第五周期内的压裂施工数据获得第二训练样本，可以包括：利用LSTM网络对第五周期内的压裂施工数据进行重要性分析，得到第五周期内的压裂施工数据的重要性排序；根据预设的第二数量，从该第五周期内的压裂施工数据的重要性排序中选择压裂施工数据作为第二训练样本。
其中，第二数量可以由管理人员预设。例如，第二数量可以是4、或者5、又或者6等。本申请实施例对第二数量的具体数值不作限制。
示例性地，以获得的第五周期内的压裂施工数据的重要性排序从大到小依次为：施工排量、油压、砂比、砂浓度、总砂量、总液量、以及套压。假设预设的第二数量为4，则第二训练样本可以包括第五周期内的施工排量、油压、砂比、以及砂浓度。
可选地，第二训练样本可以和第一训练样本同时获得。如上所述，第一训练样本 可以是利用LSTM网络对第三周期内的压裂施工数据进行重要性分析之后得到的。第二训练样本也可以是利用LSTM网络对第五周期内的压裂施工数据进行重要性分析之后得到的。第五周期的时长可以和第三周期的时长相同。上述根据预设的第二数量，从该第五周期内的压裂施工数据的重要性排序中选择压裂施工数据作为第二训练样本，可以包括：根据预设的第一数量，从第三周期内的压裂施工数据的重要性排序中选择压裂施工数据作为第一候选训练样本；根据预设的第二数量，从第五周期内的压裂施工数据的重要性排序中选择压裂施工数据作为第二候选训练样本；选择第一候选训练样本和第二候选训练样本中共有的压裂施工数据作为第一预测模型和第二训练样本。
可选地，在利用第二训练样本对第二预测模型进行训练之后，还可以对训练过后的第二预测模型进行测试。也即，该方法还可以包括：根据第五周期内的压裂施工数据获得第二测试样本；将第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态作为第二测试样本的第四标签；根据带有第四标签的第二测试样本对第二预测模型进行测试，得到完成测试的第二预测模型。
可选地，根据第二测试样本对第二预测模型进行测试的过程可以包括：将第二测试样本输入完成训练的第二预测模型网络，得到第二测试样本对应的第二测试结果；根据第二测试结果和第四标签，确定第二预测模型的准确率；当准确率大于预设的准确率阈值时，得到完成测试的第二预测模型。
其中，准确率阈值可以由管理人员预设。例如，准确率阈值为70％、或者80％等。本申请实施例对准确率阈值的具体数值不作限制。
需要说明的是，当准确率小于预设的准确率阈值时，还可以继续重复上述第二预测模型的训练过程，直至准确率大于预设的准确率阈值。当准确率等于预设的准确率阈值时，可以得到完成测试的第二预测模型，或者，重复上述第二预测模型的训练过程，直至准确率大于预设的准确率阈值。本申请实施例对此不作限制。
可选地，第二训练样本和第二测试样本也可以是同时获得。也即，根据第二周期内的压裂施工数据和第二预测模型，确定第二周期的砂堵预测结果之前，该方法还可以包括：获取第五周期内的压裂施工数据、以及第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态；根据第五周期内的压裂施工数据，按照第二比例获得第二训练样本和第二测试样本；根据第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态，获得第二训练样本的第三标签和第二测试样本的第四标签。
其中，第二比例可以由管理人员预设。第二比例可以和上述第一比例相同，或者不同。例如，第二比例为80％的训练样本、20％的测试样本，或者90％的训练样本、10％的测试样本等。本申请实施例对第二比例的具体数值不作限制。
在示例性的实施例中，本申请实施例还提供了一种砂堵预测装置，该砂堵预测装置可以应用于上述计算处理设备20。图6为本申请实施例提供的砂堵预测装置的组成示意图。如图6所示，该装置可以包括：获取模型601和处理模块602。获取模块601，用于获取第一周期内的压裂施工数据。处理模块602，用于根据第一周期内的压裂施工数据和第一预测模型，确定第一周期之后的第二周期内的压裂施工数据；根据第二周期内的压裂施工数据和第二预测模型，确定第二周期的砂堵预测结果。
一些可能的实施例中，处理模块602，具体用于对第二周期内的压裂施工数据进 行预处理，得到第一数据；将第一数据，和/或，第二周期内的压裂施工数据输入第二预测模型，得到第二周期的砂堵预测结果；第一数据包括第二周期内的压裂施工数据的斜率。
另一些可能的实施例中，获取模块601，还用于获取历史时间段内的最大压裂施工数据。处理模块602，还用于根据第一周期内的压裂施工数据和历史时间段内的最大压裂施工数据的商，得到归一化之后的第一周期内的压裂施工数据。
又一些可能的实施例中，获取模块601，还用于获取第三周期内的压裂施工数据；获取第三周期之后的第四周期内的压裂施工数据。处理模块602，还用于将第四周期内的压裂施工数据作为第一训练样本的第一标签；根据带有第一标签的第一训练样本对长短记忆LSTM网络进行训练，得到完成训练的第一预测模型。
又一些可能的实施例中，处理模块602，还用于根据第三周期内的压裂施工数据获得第一测试样本；将第四周期内的压裂施工数据作为第一测试样本的第二标签；根据带有第二标签的第一测试样本对第一预测模型进行测试，得到完成测试的第一预测模型。
又一些可能的实施例中，获取模块601，还用于获取第五周期内的压裂施工数据、以及第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态；砂堵状态包括已砂堵和未砂堵。处理模块602，还用于根据第五周期内的压裂施工数据获得第二训练样本；根据第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态，获得第二训练样本的第三标签；根据带有第三标签的第二训练样本对第二预测模型进行训练，得到完成训练的第二预测模型。
又一些可能的实施例中，处理模块602，还用于根据第五周期内的压裂施工数据获得第二测试样本；将第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态作为第二测试样本的第四标签；根据带有第四标签的第二测试样本对第二预测模型进行测试，得到完成测试的第二预测模型。
在示例性的实施例中，本申请实施例还提供了一种计算机程序产品，当该计算机程序产品在计算机上运行时，使得计算机执行上述相关方法步骤，以实现前述方法实施例中的方法。
在示例性的实施例中，本申请实施例还提供了一种电子设备，该电子设备可以是上述方法实施例中的所述的计算处理设备20。图7为本申请实施例提供的电子设备的结构示意图。如图7所示，该电子设备可以包括：处理器701和存储器702；存储器702存储有处理器701可执行的指令；处理器701被配置为执行指令时，使得电子设备实现如前述方法实施例中所述的方法。
在示例性的实施例中，本申请实施例还提供一种计算机可读存储介质，其上存储有计算机程序指令；当所述计算机程序指令被电子设备执行时，使得电子设备实现如前述实施例中所述的方法。计算机可读存储介质可以是非临时性计算机可读存储介质，例如，所述非临时性计算机可读存储介质可以是ROM、随机存取存储器(RAM)、CD-ROM、磁带、软盘和光数据存储设备等。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何在本申请揭露的技术范围内的变化或替换，都应涵盖在本申请的保护范围之内。因 此，本申请的保护范围应该以权利要求的保护范围为准。

Claims (10)

  1. 一种砂堵预测方法，其特征在于，所述方法包括:
    获取第一周期内的压裂施工数据；
    根据所述第一周期内的压裂施工数据和第一预测模型，确定所述第一周期之后的第二周期内的压裂施工数据；
    根据所述第二周期内的压裂施工数据和第二预测模型，确定所述第二周期的砂堵预测结果。
  2. 根据权利要求1所述的方法，其特征在于，所述根据所述第二周期内的压裂施工数据和第二预测模型，确定所述第二周期的砂堵预测结果，包括：
    对所述第二周期内的压裂施工数据进行预处理，得到第一数据；
    将所述第一数据，和/或，所述第二周期内的压裂施工数据输入所述第二预测模型，得到所述第二周期的砂堵预测结果；所述第一数据包括所述第二周期内的压裂施工数据的斜率。
  3. 根据权利要求1所述的方法，其特征在于，所述根据所述第一周期内的压裂施工数据和第一预测模型，确定所述第一周期之后的第二周期内的压裂施工数据之前，所述方法还包括：
    获取历史时间段内的最大压裂施工数据；
    根据所述第一周期内的压裂施工数据和所述历史时间段内的最大压裂施工数据的商，得到归一化之后的所述第一周期内的压裂施工数据。
  4. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    获取第三周期内的压裂施工数据；
    获取所述第三周期之后的第四周期内的压裂施工数据；
    根据所述第三周期内的压裂施工数据获得第一训练样本；
    将所述第四周期内的压裂施工数据作为所述第一训练样本的第一标签；
    根据带有所述第一标签的第一训练样本对长短记忆LSTM网络进行训练，得到完成训练的所述第一预测模型。
  5. 根据权利要求4所述的方法，其特征在于，所述方法还包括：
    根据所述第三周期内的压裂施工数据获得第一测试样本；
    将所述第四周期内的压裂施工数据作为所述第一测试样本的第二标签；
    根据带有所述第二标签的第一测试样本对所述第一预测模型进行测试，得到完成测试的所述第一预测模型。
  6. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    获取第五周期内的压裂施工数据、以及所述第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态；所述砂堵状态包括已砂堵和未砂堵；
    根据所述第五周期内的压裂施工数据获得第二训练样本；
    根据所述第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态，获得第二训练样本的第三标签；
    根据带有第三标签的第二训练样本对第二预测模型进行训练，得到完成训练的第二预测模型。
  7. 根据权利要求6所述的方法，其特征在于，所述方法还包括：
    根据所述第五周期内的压裂施工数据获得第二测试样本；
    将所述第五周期内的每个时刻的压裂施工数据分别对应的砂堵状态作为所述第二测试样本的第四标签；
    根据带有所述第四标签的第二测试样本对第二预测模型进行测试，得到完成测试的第二预测模型。
  8. 一种砂堵预测装置，其特征在于，所述装置包括：获取模块和处理模块；
    所述获取模块，用于获取第一周期内的压裂施工数据；
    所述处理模块，用于根据所述第一周期内的压裂施工数据和第一预测模型，确定所述第一周期之后的第二周期内的压裂施工数据；根据所述第二周期内的压裂施工数据和第二预测模型，确定所述第二周期的砂堵预测结果。
  9. 一种电子设备，其特征在于，所述电子设备包括：处理器和存储器；
    所述存储器存储有所述处理器可执行的指令；
    所述处理器被配置为执行所述指令时，使得所述电子设备实现如权利要求1-7任一项所述的方法。
  10. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质包括：计算机软件指令；
    当所述计算机软件指令在电子设备中运行时，使得所述电子设备实现如权利要求1-7任一项所述的方法。
PCT/CN2022/076163 2022-02-14 2022-02-14 砂堵预测方法、装置、设备及存储介质 WO2023151072A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/076163 WO2023151072A1 (zh) 2022-02-14 2022-02-14 砂堵预测方法、装置、设备及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/076163 WO2023151072A1 (zh) 2022-02-14 2022-02-14 砂堵预测方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2023151072A1 true WO2023151072A1 (zh) 2023-08-17

Family

ID=87563421

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/076163 WO2023151072A1 (zh) 2022-02-14 2022-02-14 砂堵预测方法、装置、设备及存储介质

Country Status (1)

Country Link
WO (1) WO2023151072A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117408168A (zh) * 2023-12-15 2024-01-16 中国石油大学(华东) 一种融合物理条件约束的压裂裂缝扩展快速预测方法
CN117408165A (zh) * 2023-12-14 2024-01-16 中国石油大学(华东) 一种基于机器学习的压裂过程复杂事件智能预警优化方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108573076A (zh) * 2017-03-09 2018-09-25 中国石油化工股份有限公司 一种页岩气压裂施工事故的预测方法
CN109740210A (zh) * 2018-12-21 2019-05-10 中国石油大学(北京) 一种页岩气压裂井下砂堵事故实时风险评估方法及装置
CN109815542A (zh) * 2018-12-21 2019-05-28 东软集团股份有限公司 压裂砂堵的识别方法、装置、存储介质及电子设备
CN110533163A (zh) * 2019-08-02 2019-12-03 中国石油集团长城钻探工程有限公司 基于长短时记忆神经网络的压裂砂堵预警方法
US20200301043A1 (en) * 2019-03-22 2020-09-24 Chengdu University Of Technology Numerical simulation and parameter optimization method for volumetric fracturing of unconventional dual medium reservoir

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108573076A (zh) * 2017-03-09 2018-09-25 中国石油化工股份有限公司 一种页岩气压裂施工事故的预测方法
CN109740210A (zh) * 2018-12-21 2019-05-10 中国石油大学(北京) 一种页岩气压裂井下砂堵事故实时风险评估方法及装置
CN109815542A (zh) * 2018-12-21 2019-05-28 东软集团股份有限公司 压裂砂堵的识别方法、装置、存储介质及电子设备
US20200301043A1 (en) * 2019-03-22 2020-09-24 Chengdu University Of Technology Numerical simulation and parameter optimization method for volumetric fracturing of unconventional dual medium reservoir
CN110533163A (zh) * 2019-08-02 2019-12-03 中国石油集团长城钻探工程有限公司 基于长短时记忆神经网络的压裂砂堵预警方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117408165A (zh) * 2023-12-14 2024-01-16 中国石油大学(华东) 一种基于机器学习的压裂过程复杂事件智能预警优化方法
CN117408165B (zh) * 2023-12-14 2024-03-15 中国石油大学(华东) 一种基于机器学习的压裂过程复杂事件智能预警优化方法
CN117408168A (zh) * 2023-12-15 2024-01-16 中国石油大学(华东) 一种融合物理条件约束的压裂裂缝扩展快速预测方法
CN117408168B (zh) * 2023-12-15 2024-03-15 中国石油大学(华东) 一种融合物理条件约束的压裂裂缝扩展快速预测方法

Similar Documents

Publication Publication Date Title
WO2023151072A1 (zh) 砂堵预测方法、装置、设备及存储介质
CN103942457B (zh) 基于关联向量机回归的水质参数时间序列预测方法
CN109711636A (zh) 一种基于混沌萤火虫与梯度提升树模型的河流水位预测方法
CN104766259A (zh) 一种基于单病种模型的医疗临床质量监测与评价系统
CA2640725A1 (en) Methods, systems, and computer-readable media for fast updating of oil and gas field production models with physical and proxy simulators
CN110952978A (zh) 一种基于神经网络数据挖掘的钻井漏失裂缝宽度预测方法
CN110489844B (zh) 一种适用于软岩隧道不均匀大变形等级预测方法
CN109740210A (zh) 一种页岩气压裂井下砂堵事故实时风险评估方法及装置
CN110084435A (zh) 一种油气藏参数解释方法及系统
CN112948932A (zh) 一种基于TSP预报数据与XGBoost算法的围岩等级预测方法
CN105869100A (zh) 一种基于大数据思维的滑坡多场监测数据的融合及预测方法
CN111859814B (zh) 基于lstm深度学习的岩石时效变形预测方法及其系统
CN113236228A (zh) 一种单井产量快速预测方法及系统
CN110232119B (zh) 基于元分析的通用智能度量模型构建方法和系统
CN109252855A (zh) 确定气井最终累积产量的方法及装置
Wang et al. Hybrid application of unsupervised and supervised learning in forecasting absolute open flow potential for shale gas reservoirs
CN106875052A (zh) 一种冰川融水补给区干旱预警方法
CN108843296A (zh) 一种基于多因素影响下的单井重复压裂效果预测方法
CN118095878A (zh) 油井解堵双级选井方法、装置、计算设备及存储介质
CN111751878A (zh) 横波速度的预测方法和装置
Matemilayo et al. Design and implementation of an integrated result processing system in a networked environment
CN115511341A (zh) 一种滑坡时变失效概率评估方法及装置
CN118152937B (zh) 岩性识别模型的训练方法、装置、电子设备及存储介质
CN115470664B (zh) 一种裂缝油藏裂缝敏感性分析方法
CN114358441B (zh) 一种致密气产量的智能分段预测方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22925434

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE