An intelligent control method for full-well flow of oilfield separate layer water injection

Abstract

The application discloses an oilfield layered water injection full-well flow intelligent regulation and control method. In the initial stage of water injection of a new well, a historical expert experience model is used to regulate and control the flow of each layer of the well, so that the actual real-time injection flow of each layer is matched with the set injection allocation, automatic identification of a steady-state working point is carried out based on adjustment process data to extract steady-state adjustment working condition data, a full-well prediction neural network model is trained and constructed by using the steady-state adjustment working condition data, and the prediction accuracy thereof is judged in real time; when the prediction accuracy reaches the set control prediction accuracy, the full-well prediction neural network model is combined with a feedforward controller and a feedback PID controller to realize intelligent control of the full-well flow prediction; the feedforward controller and the feedback PID controller are used to process the time lag and nonlinearity of the system, reduce mutual interference during the regulation of the flow of each layer, realize rapid deployment of the flow of multiple layers, improve the regulation rate of the full-well flow, and reduce the action time and frequency of the downhole actuator.

Description

TECHNICAL FIELD

[0001] The present application relates to an oilfield layered water injection full-well flow intelligent regulation method, belonging to the layered injection technology field of oilfield water injection wells. BACKGROUND

[0002] With the continuous development of oilfields, the formation energy will be continuously consumed, and at present, most of the oilfields have entered the middle and late stages of water injection development. Oilfield water injection mining technology can maintain oil reservoir pressure and reduce the decline rate of crude oil mining, and is the most economical and effective technical means to achieve long-term high yield and stable yield of oilfields. In the layered water injection process, how to accurately and quickly obtain the required injection allocation of each layer is one of the key technologies to ensure efficient water injection of oilfields. However, due to the variety of factors affecting the injection allocation of each layer, and the non-linear and uncertain influence between each factor, it greatly increases the difficulty of accurate injection downhole. Another key technology for efficient water injection of oilfields is the accurate measurement and accurate control of the flow of each layer. In the process of regulating the flow of the downhole layered water injection, the flow regulation is affected by many adverse factors such as the change of water nozzle opening, pressure change, and mutual interference between different injection layers. Therefore, it is necessary to design a controller with strong anti-interference ability, excellent dynamic performance and strong stability to realize accurate control of the flow.

[0003] At present, most of the layered water injection flow regulation adopts repeated manual adjustment of the water nozzle opening of the downhole injection allocator to match the actual injection flow with the injection allocation. However, there are problems such as great difficulty in flow regulation, large workload, and low efficiency. With the installation of layered water injection systems in a large number of water injection wells in oilfields, fast, automatic and intelligent regulation of the full-well water injection volume has become an urgent technical requirement of oilfields. In order to effectively improve the efficiency of multi-layer flow regulation, patent CN114357852A proposes to use long short-term memory neural network and particle swarm optimization to obtain the injection flow of each injection layer. However, a large amount of data is required for model training, and it is difficult to collect data on site, and there is a lack of automatic data acquisition and analysis process. CN110533344A provides an automatic regulation method for layered water injection system, which evaluates the injection layer by using injection parameters, and determines the regulation sequence of each injection layer according to the evaluation result. This method can improve the regulation efficiency to a certain extent, but still requires repeated manual adjustment, which is a large amount of work. SUMMARY

[0004] The purpose of the present application is to provide an oilfield layered water injection full-well flow intelligent regulation method. This method solves the problem of large workload, low efficiency and poor effect in the process of layered water injection flow regulation, and further improves the efficiency and accuracy of multi-layer flow measurement and regulation.

[0005] The technical scheme of the present application: an oilfield layered water injection full-well flow intelligent regulation method, comprising the following steps:

[0006] S1: In the initial stage of flow regulation in a new well, use the historical expert experience model to regulate the flow of each layer, and adjust the flow of each layer step by step according to the set injection allocation of each layer.

[0007] S2: Collect adjustment process data, including actual injection real-time flow, tubing pressure, formation pressure, set injection allocation and injection nozzle opening degree of each layer; based on the steady-state identification method, the adjustment process data is adaptively identified and extracted to obtain steady-state adjustment working condition data under different adjustment states;

[0008] S3: Design and build a full-well prediction neural network model, and use the steady-state adjustment working condition data obtained in S2 for data preprocessing, and use the preprocessed data to train the full-well prediction neural network model;

[0009] S4: Evaluate the prediction accuracy of the full-well prediction neural network model, when the prediction accuracy is less than the set control prediction accuracy, use the historical expert experience model to stabilize the regulation of the actual injection real-time flow of each layer, and continue to collect steady-state adjustment working condition data of the adjustment process to continue to optimize the training of the full-well prediction neural network model, when the prediction accuracy reaches the set control prediction accuracy, use the neural network intelligent control module to intelligently regulate the actual injection real-time flow of each layer in the well, and realize the optimization calculation of the combination of the injection nozzle opening degree of each layer;

[0010] S5: Roll S2 process, real-time collect, identify and extract steady-state adjustment working condition data of the adjustment process, real-time judge the prediction accuracy of the full-well prediction neural network model, when the prediction accuracy is less than the set control prediction accuracy, start the model rolling optimization training mechanism, introduce the latest steady-state adjustment working condition data for model training, and ensure that the real-time prediction accuracy of the model is greater than the set control prediction accuracy.

[0011] The foregoing method for intelligent regulation of full-well flow in oilfield layered water injection, wherein the step S1 specifically comprises:

[0012] S11: The flow regulation method using the historical expert experience model is: first, use the total injection amount constant flow control method at the wellhead to ensure that the total injection flow of the well is the same as the sum of the set injection allocation of each layer, and determine the under-injection layer and the over-injection layer according to the positive and negative deviation between the actual injection real-time flow, the set injection allocation and the allowable error E of each layer; if the sum of the set injection allocation and the allowable error E of a layer minus the actual injection real-time flow of the layer is negative, it indicates that the layer is an over-injection layer, and if the difference between the set injection allocation and the allowable error E of a layer minus the actual injection real-time flow of the layer is positive, it indicates that the layer is an under-injection layer; record the number of under-injection layers and over-injection layers of the well as and , determine the adjustment direction according to the number of under-injection layers and over-injection layers; when Greater than When adjusting, the direction is to first adjust the under-injected layer, then adjust the over-injected layer, and vice versa. Less than At that time, the adjustment direction is to first adjust the over-injection layer, and then adjust the under-injection layer;

[0013] S12: Based on the adjustment direction determined in step S11, the flow rate of the segments that meet the adjustment direction is automatically adjusted in order from top to bottom. When the deviation between the actual real-time injection flow rate of a certain segment and the set injection amount of that segment is less than the allowable deviation E, the adjustment of that segment is stopped, and the water injection nozzle opening of that segment is maintained at the current opening. The flow rate of the next segment that meets the adjustment direction is automatically adjusted until the deviation between the actual real-time injection flow rate of that segment and the set injection amount of that segment is less than the allowable deviation E. After all segments in the adjustment direction are adjusted to be qualified, the adjustment direction is switched to another adjustment direction, that is, the under-injection segment is adjusted after the over-injection segment is adjusted to be qualified, or the over-injection segment is adjusted after the under-injection segment is adjusted to be qualified, until all segments are adjusted to be qualified, that is, the number of under-injection segments and over-injection segments in the whole well is 0.

[0014] In the aforementioned intelligent control method for the total well flow rate of layered water injection in oilfields, step S2 specifically includes:

[0015] S21: The steady-state identification method is as follows: Set the change thresholds for tubing pressure and formation pressure as follows: and The stable threshold for actual injected real-time flow parameters is: The stability threshold for tubing pressure and formation pressure parameters is: The maximum data length of the steady-state data judgment data window is M, and the variable sliding window for steady-state judgment is N;

[0016] S22: Read the set injection volume, actual real-time injection flow rate, water injection nozzle opening, tubing pressure and formation pressure parameters for each layer, clean the data, and remove abnormal zero-value data or out-of-point data.

[0017] S23: When the formation pressure change in a certain layer exceeds the formation pressure change threshold. Or the change in tubing pressure across all sections of the well exceeds the tubing pressure change threshold. Or when the opening degree of the water injection nozzle of a certain layer segment changes by more than 5%, the N continuous data from the data change point as the starting point are read and put into the stable judgment variable sliding window, the read data parameters include the actual injection real-time flow, the tubing pressure and the formation pressure, the mean values of each data parameter obtained in the stable judgment variable sliding window are calculated, the proportion of the number of data whose deviation from the mean value is less than the stable threshold value of each parameter in N data is calculated, when the proportion is less than the set judgment threshold value, the stable judgment variable sliding window slides backward by Z data points, Z < N, and the judgment is performed again, until the proportion of the number of data whose deviation from the mean value is less than the stable threshold value in the stable judgment variable sliding window is greater than the set judgment threshold value;

[0018] S24: The stable judgment variable sliding window continues to slide backward by N data points, the deviations of all data in the stable judgment variable sliding window from the mean value are all less than the stable threshold value, then all data in the stable judgment variable sliding window are saved to the steady state data judgment data window, and the stable judgment variable sliding window continues to slide backward by N data points for continuous judgment, when the proportion of the number of data whose deviation from the mean value is less than the stable threshold value in the stable judgment variable sliding window is less than the set judgment threshold value or the amount of all data recorded in the steady state data judgment data window is greater than M, the stable judgment variable sliding window sliding judgment process is exited, the mean value of all data in the steady state data judgment data window is taken as the final steady state regulation working condition data, and the value is stored in the database.

[0019] The foregoing method for intelligently regulating and controlling full-well flow of oilfield separate layer water injection specifically comprises the following steps:

[0020] S31: The full-well prediction neural network model adopts a long short-term memory convolutional neural network to predict the time series data of the water injection nozzle opening degree and the water injection flow, the preprocessed steady state regulation working condition data are taken as the input of the front-end convolutional neural network, the automatic extraction of multi-dimensional input data features is realized through the convolutional layer and the pooling layer of the convolutional neural network, the convolutional neural network feature extraction result is taken as the input of the back-end bidirectional long short-term memory network layer, the output result of the bidirectional long short-term memory network layer obtains different weights through the attention mechanism, and the output result is output to the full connection layer after normalization processing to obtain the final prediction output result. The output of the full-well prediction neural network model is the water injection nozzle opening degree and the water injection flow prediction value of each layer segment;

[0021] S32: The steady state regulation working condition data preprocessing method adopts linear normalization processing, and the steady state regulation working condition data of the pressure, the actual injection real-time flow and the water injection nozzle opening degree of each layer segment are subjected to linear normalization processing respectively, and the processing formula is as follows:

[0022] ,

[0023] The original signal is represented as follows: represents the normalized result; represents the minimum value of the original signal; represents the maximum value of the original signal.

[0024] The oilfield layered water injection full-well flow intelligent control method has the following advantages:

[0025] S41: The prediction accuracy calculation method calculates the mean square error by taking K consecutive points, and when the mean square error result is less than a given value , it indicates that the prediction accuracy meets the training requirements;

[0026] ,

[0027] represents the original signal containing noise, represents the model prediction signal, represents the data length, represents the mean square error;

[0028] S42: The neural network intelligent control module combines the full-well prediction neural network model, the feedforward controller and the feedback PID controller, the product of the prediction value of the water injection nozzle opening degree of each layer of the full-well prediction neural network model and the output result of the input set value judgment module is taken as the output of the feedforward controller, the deviation between the prediction value of the water injection flow of each layer of the full-well prediction neural network model and the set injection allocation of each layer is taken as the input value of the feedback PID controller, the feedback adjustment amount of the water injection nozzle opening degree of each layer is calculated, and the sum of the feedback adjustment amount of the water injection nozzle opening degree of each layer and the output of the feedforward controller is taken as the final control output and controls the water injection nozzle opening degree of each layer of the multi-layer layered water injection control object;

[0029] S43: The input set value judgment module is used to judge whether the set injection allocation of each layer has a step change, and the judgment amplitude is , that is, when the change amplitude of the input set injection allocation is greater than or equal to , it indicates that a step change occurs, and the input set value judgment module outputs 1; when the change amplitude of the input value is less than , the input set value judgment module outputs 0.

[0030] The beneficial effects of the present application are as follows:

[0031] 1. The injection device is placed downhole for extended periods, and the harsh environment of high temperature and pressure downhole may damage the injection device sensor. However, due to high operating costs, when one or two flow sensors in a multi-layer injection well malfunction, the faulty injection device is not immediately retrieved. The whole-well predictive neural network model designed in this invention can achieve real-time flow prediction using historical data from that layer and flow opening data from other layers when a sensor in a particular layer malfunctions, ensuring the normal operation of the system.

[0032] 2. This invention combines empirical methods with artificial intelligence neural networks. First, it uses an empirical method to gradually approximate the flow rate of the entire well to meet the flow control requirements of each layer in the field. At the same time, it collects as much steady-state control data as possible under different control conditions. Then, it trains a whole-well predictive neural network model using the steady-state control data under different control conditions. The whole-well predictive neural network model is used to overcome the interference caused by changes in formation pressure and inter-layer flow rate, further improving the efficiency of whole-well flow control and ensuring the accuracy and speed of flow control.

[0033] 3. The model has self-learning capabilities. As process data accumulates, the whole-well prediction neural network model undergoes rolling training and optimization, resulting in increasingly higher prediction accuracy. This enables precise intelligent control of a single well, effectively improving the efficiency of layered water injection throughout the well, reducing the number of times the downhole injector needs adjustment, and extending the service life of the downhole instrument. Attached Figure Description

[0034] Figure 1 Block diagram of the whole-well intelligent control method system;

[0035] Figure 2 Flowchart for automatic identification and extraction of steady-state data;

[0036] Figure 3 This is a diagram of the neural network prediction model for the total well flow rate.

[0037] Figure 4 This is a block diagram of the neural network intelligent control module system. Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0039] The embodiment of the application discloses an intelligent control method for full-well flow of oilfield separate layer water injection, which comprises a full-well flow multi-step approximation adjustment method based on a historical expert experience model. The method comprises the following steps: first, the historical expert experience model is used to adjust the flow of each layer of a well in the initial stage of water injection, so that the actual injection real-time flow of each layer matches the set injection allocation; the adjustment process data of the well is collected; the steady-state working point is automatically identified based on the adjustment process data to obtain the steady-state adjustment working condition data; the steady-state adjustment working condition data is used to train and construct a full-well prediction neural network model; the prediction accuracy of the full-well prediction neural network model is determined in real time; when the prediction accuracy of the full-well prediction neural network model reaches the set control prediction accuracy, the full-well prediction neural network model is combined with a feedforward controller and a feedback PID controller to realize intelligent control of the full-well flow prediction; the time lag and nonlinearity of the system are solved by using the feedforward controller for coarse adjustment and the feedback PID controller for fine adjustment, the mutual interference during the adjustment of the flow of each layer is reduced, the multi-layer flow is quickly adjusted, the full-well flow adjustment rate is improved, the action time and frequency of the downhole actuator are reduced, and the service life of the downhole instrument is prolonged.

[0040] Specifically, the method comprises the following steps:

[0041] S1: in the initial stage of flow adjustment of a new well, a historical expert experience model is used to adjust the flow of each layer of the well, and a single-layer step-by-step adjustment method is used to adjust the flow of each layer of the well according to the set injection allocation of each layer;

[0042] S2: adjustment process data is collected, including the actual injection real-time flow of each layer, the tubing pressure, the formation pressure, the set injection allocation and the injection nozzle opening degree; the adjustment process data is adaptively identified and extracted based on a steady-state identification method to obtain steady-state adjustment working condition data under different adjustment states;

[0043] S3: a full-well prediction neural network model is designed and constructed, and the steady-state adjustment working condition data obtained in S2 is used for data preprocessing, and the preprocessed data is used to train the full-well prediction neural network model;

[0044] S4: the prediction accuracy of the full-well prediction neural network model is evaluated; when the prediction accuracy is less than the set control prediction accuracy, the historical expert experience model is used to stably adjust the actual injection real-time flow of each layer, and the steady-state adjustment working condition data of the adjustment process is continuously collected to continuously optimize and train the full-well prediction neural network model; when the prediction accuracy reaches the set control prediction accuracy, the neural network intelligent control module is used to intelligently adjust the actual injection real-time flow of each layer of the well, the injection nozzle opening degree of each layer is optimized and calculated, and the adjustment effect is quickly and stably achieved.

[0045] S5: rolling execution of S2 process, real-time acquisition, identification and extraction of steady-state regulation working condition data of the regulation process, real-time judgment of the prediction accuracy of the full-well prediction neural network model, when the prediction accuracy is less than the set control prediction accuracy, starting the model rolling optimization training mechanism, introducing the latest steady-state regulation working condition data for model training, ensuring that the real-time prediction accuracy of the model is greater than the set control prediction accuracy.

[0046] The step S1 specifically comprises:

[0047] S11: the flow regulation method using the historical expert experience model is: first, the wellhead adopts total water injection amount constant flow control mode, to ensure that the total injection flow of the full well is the same as the sum of the set injection allocation of each layer, and to determine the under-injection layer and the over-injection layer according to the positive and negative deviation between the actual real-time injection flow of each layer, the set injection allocation and the allowable error E; if the sum of the set injection allocation of a layer and the allowable error E minus the actual real-time injection flow of the layer is negative, it indicates that the layer is an over-injection layer; if the difference between the set injection allocation of a layer and the allowable error E minus the actual real-time injection flow of the layer is positive, it indicates that the layer is an under-injection layer; the number of under-injection layers and over-injection layers of the full well is recorded as and , and the regulation direction is determined according to the number of under-injection layers and over-injection layers; when is greater than , the regulation direction is to regulate the under-injection layer first, and then regulate the over-injection layer; otherwise, when is less than , the regulation direction is to regulate the over-injection layer first, and then regulate the under-injection layer.

[0048] S12: according to the regulation direction determined in step S11, the flow of the layer meeting the regulation direction is automatically regulated in the order from the layer position to the top down; when the deviation between the actual real-time injection flow of a layer and the set injection allocation of the layer is less than the allowable deviation E, the regulation of the layer is stopped, the opening degree of the water injection nozzle of the layer is kept at the current opening degree, and the flow of the next layer meeting the regulation direction is automatically regulated, until the deviation between the actual real-time injection flow of the layer (i.e. the layer being regulated) and the set injection allocation of the layer (i.e. the layer being regulated) is less than the allowable deviation E, and all layers in the regulation direction are regulated qualified, then the regulation direction is switched to the under-injection layer, or the over-injection layer, until all layers are regulated qualified, i.e. the number of under-injection layers and over-injection layers of the full well is 0.

[0049] The step S2 specifically comprises:

[0050] S21: the steady-state identification method is: the change threshold values of the tubing pressure and the formation pressure are set as and ; the stable threshold value of the actual injection real-time flow parameter is , the stable threshold value of the tubing pressure and the formation pressure parameter is , the maximum data length of the stable data judgment data window is M, and the stable judgment variable sliding window is N;

[0051] S22: The set injection allocation, the actual injection real-time flow, the injection water nozzle opening degree, the tubing pressure and the formation pressure parameter of each layer section are read, and the data is cleaned to eliminate abnormal zero value data or flying point data;

[0052] S23: When the formation pressure change of a layer section is greater than the formation pressure change threshold value or the tubing pressure change of each layer section of the whole well is greater than the tubing pressure change threshold value or the injection water nozzle opening degree change of a layer section is greater than 5%, the N continuous data from the data change point as the starting point are read and put into the stable judgment variable sliding window, the read data parameters include the actual injection real-time flow, the tubing pressure and the formation pressure, the mean value of each data parameter obtained in the stable judgment variable sliding window is calculated, that is, the mean value of N injection real-time flow data, the mean value of N tubing pressure data and the mean value of N formation pressure data are calculated, the data number ratio whose deviation from the mean value is less than the stable threshold value of each parameter in N data is calculated, when the ratio is less than the set judgment threshold value, the stable judgment variable sliding window is slid backward by Z data points, Z

[0053] For example, when N injection real-time flow data are divided into A1, A2, A3……A N , and the mean value of N injection real-time flow data is A 均 , the absolute values of A1-A 均 , A2-A 均 , A3-A 均 ……A N -A 均 are calculated respectively, thereby obtaining N absolute values, and then it is observed that how many absolute values are less than the stable threshold value of the injection real-time flow data , thereby calculating the data number ratio whose deviation from the mean value is less than the stable threshold value of the real-time flow data in N data. The tubing pressure and the formation pressure data are also calculated in the same way.

[0054] S24: The stable judgment variable sliding window continues to slide backward by N data points. If the deviations of all data in the stable judgment variable sliding window from the mean value thereof are less than the stability threshold value, all data in the stable judgment variable sliding window are saved to the steady state data judgment data window. Then, the stable judgment variable sliding window continues to slide backward by N data points to continue the judgment. When the proportion of data in the stable judgment variable sliding window whose deviation from the mean value thereof is less than the stability threshold value is less than the set judgment threshold value or the amount of all data recorded in the steady state data judgment data window is greater than M, the stable judgment variable sliding window sliding judgment process is exited. The mean value of all data in the steady state data judgment data window is taken as the final steady state regulation working condition data, and the value is stored in the database.

[0055] The step S3 specifically comprises:

[0056] S31: The full-well prediction neural network model adopts a long short-term memory convolutional neural network to predict the water nozzle opening degree and water injection flow time series data. The preprocessed steady state regulation working condition data is taken as the input of the front-end convolutional neural network. The automatic extraction of multi-dimensional input data features is realized through the convolutional layer and the pooling layer of the convolutional neural network. The convolutional neural network feature extraction result is taken as the input of the back-end bidirectional long short-term memory network layer. The output result of the bidirectional long short-term memory network layer obtains different weights through the attention mechanism. After normalization processing, the output is output to the full connection layer to obtain the final prediction output result. The output of the full-well prediction neural network model is the water nozzle opening degree and water injection flow prediction value of each layer. The bidirectional long short-term memory network layer sets the input gate, the forgetting gate, the output gate and the memory unit, so that the model can selectively retain, forget and output effective information, and ensure the real-time performance and accuracy of the full-well prediction neural network model;

[0057] S32: The steady state regulation working condition data preprocessing method adopts linear normalization processing. The steady state regulation working condition data of the pressure, the actual injection real-time flow and the water nozzle opening degree of each layer is linearly normalized, and the processing formula is as follows:

[0058]

[0059] represents the original signal; represents the normalized result; represents the minimum value of the original signal; represents the maximum value of the original signal.

[0060] The step S4 specifically comprises:

[0061] S41: The prediction accuracy calculation method calculates the mean square error of K consecutive points. When the mean square error result is less than a given value , it indicates that the prediction accuracy meets the training requirements; ​

[0062] ,

[0063] represents a noisy original signal, represents a model predicted signal, K represents a data length, represents a mean square error;

[0064] S42: The neural network intelligent control module combines the full-well prediction neural network model, the feedforward controller and the feedback PID controller, the product of the full-well prediction neural network model of each layer segment injection nozzle opening prediction value and the output result of the input set value judgment module as the output of the feedforward controller, the deviation between the full-well prediction neural network model of each layer segment injection flow prediction value and the set injection allocation of each layer segment as the input value of the feedback PID controller, the output of each layer segment injection nozzle opening feedback adjustment amount is calculated, and the sum of the feedback adjustment amount of each layer segment injection nozzle opening and the output of the feedforward controller as the final control output and control the injection nozzle opening of each layer segment of the multi-layer stratified injection control object;

[0065] S43: The input set value judgment module is used to judge whether the injection allocation set value of each layer segment has a step change, and the judgment amplitude is , that is, when the input value set injection allocation change amplitude is greater than or equal to , it indicates that a step change occurs, and the input set value judgment module outputs 1; when the input value change amplitude is less than , the input set value judgment module outputs 0.

[0066] In the specific implementation process, a field test application is carried out in a four-layer new well of Daqing Oilfield No. 3 Plant, according to the full-well intelligent control method shown in Figure 1 , first, the historical expert experience model is used to control the flow stratification of the full-well flow, and the single-layer step-by-step adjustment method is used to adjust the full-well flow according to the injection allocation of each layer, and the real-time adjustment process data is collected, and the steady-state data automatic identification and extraction logic shown in Figure 2 is used to automatically identify the steady-state adjustment working condition data, and the identified data is automatically added to the model database, and the model database collects 763 valid data samples. 500 valid sample data are used as model training input data set, and 263 data are used as model verification data set, and the model structure shown in Figure 3 is used to train the full-well prediction neural network model, and the model verification data set is used to test the model after 1000 iterations, and the test results show that the prediction accuracy is 96.08%, 96.73% and 98.1% respectively. The control prediction accuracy is set to 96%, and when the full-well prediction neural network model prediction accuracy reaches the control prediction accuracy, the Figure 4The neural network intelligent control module shown intelligently controls the actual injection real-time flow of the whole well.

[0067] Due to the change of formation characteristics, the actual injection real-time flow of each layer of the well and the set injection allocation requirement are shown in Table 1: the historical expert experience model single-layer step-by-step adjustment method and the neural network intelligent control module intelligent control method are respectively used for well flow regulation, and the actual injection real-time flow and the total adjustment time after adjustment are shown in Table 2:

[0068] Table 1 Injection allocation requirement

[0069] Horizon Layer 1 Layer 2 Layer 3 Layer 4 Actual injection real-time flow before adjustment (m 3 / d) 19.6 34.3 15.7 44.0 Set the injection rate (m 3 / d) 40 15 30 30

[0070] Table 2 Adjustment effect table

[0071] Regulation method Layer 1 Layer 2 Layer 3 Layer 4 Total regulation time Historical expert experience model method 38.2 14 32.6 28.4 8.2 minutes Neural network intelligent control method 39.3 14.7 29.8 29.1 4 minutes

[0072] As shown in Table 2, compared with the historical expert model experience method, the neural network intelligent control module has shorter adjustment time, higher regulation efficiency, and better regulation accuracy.

[0073] In order to solve the problems of large workload, low efficiency and poor effect in the process of layered injection flow regulation, the layered injection well flow intelligent regulation method is used, which has important significance for improving the efficiency and accuracy of multi-layer flow measurement and regulation.

Claims

1. A method for intelligent control of the entire well flow rate in stratified water injection in oilfields, characterized in that: Comprise the following steps: S1: in the initial stage of flow regulation in a new well, the flow is stratified and regulated by using a historical expert experience model, and the flow is adjusted by using a single layer step-by-step adjustment method according to the set injection allocation of each layer; S2: collect the adjustment process data, including the actual injection real-time flow, tubing pressure, formation pressure, set injection allocation and water injection nozzle opening degree of each layer; based on the steady-state identification method, the adjustment process data is adaptively identified and extracted to obtain the steady-state adjustment working condition data under different adjustment states; S3: design and build a full well prediction neural network model, and use the steady-state adjustment working condition data obtained in S2 for data preprocessing, and use the preprocessed data to train the full well prediction neural network model; S4: evaluate the prediction accuracy of the full well prediction neural network model, when the prediction accuracy is less than the set control prediction accuracy, use the historical expert experience model to stabilize the regulation of the actual injection real-time flow of each layer, and continue to collect the steady-state adjustment working condition data of the adjustment process, and continue to optimize the training of the full well prediction neural network model, when the prediction accuracy reaches the set control prediction accuracy, use the neural network intelligent control module to intelligently regulate the actual injection real-time flow of each layer of the well, and realize the combination optimization calculation of the injection nozzle opening degree of multiple layers; S5: roll out S2 process, real-time collect, identify and extract the steady-state adjustment working condition data of the adjustment process, real-time judge the prediction accuracy of the full well prediction neural network model, when the prediction accuracy is less than the set control prediction accuracy, start the model rolling optimization training mechanism, introduce the latest steady-state adjustment working condition data for model training, and ensure that the real-time prediction accuracy of the model is greater than the set control prediction accuracy; The step S2 specifically comprises: S21: The steady-state identification method is: setting the change threshold of tubing pressure and formation pressure respectively as and ; The stable threshold of the real-time flow parameter actually injected is The stable threshold of the tubing pressure and the formation pressure parameter is The maximum data length of the stable data judgment data window is M, and the stable judgment variable sliding window is N. S22: read the set injection allocation, actual injection real-time flow, water injection nozzle opening degree, tubing pressure and formation pressure parameters of each layer, clean the data, and eliminate abnormal zero value data or flying point data; S23: When the formation pressure change of a layer segment is greater than the formation pressure change threshold Or the tubing pressure change of each layer segment of the whole well is greater than the tubing pressure change threshold Or the water injection nozzle opening change of a layer segment is greater than 5%, continuously read N continuous data from the data change point as the starting point and put them into the stable judgment variable sliding window, read the data parameters including the actual injection real-time flow, tubing pressure and formation pressure, calculate the mean value of each data parameter obtained in the stable judgment variable sliding window, calculate the data quantity ratio of the data in N data whose deviation from the mean value is less than the stable threshold of each parameter, when the ratio is less than the set judgment threshold, the stable judgment variable sliding window slides Z data points backward, Z < N, and then judges again, until the data quantity ratio of the data whose deviation from the mean value in the stable judgment variable sliding window is less than the stable threshold is greater than the set judgment threshold. S24: continue to slide N data points to the rear of the stable judgment variable sliding window, if the deviation of all data in the stable judgment variable sliding window from the mean value is less than the stable threshold, save all data in the stable judgment variable sliding window to the steady-state data judgment data window; continue to slide N data points to the rear, and continue to judge, when the data in the stable judgment variable sliding window from the mean value is less than the stable threshold, the data proportion is less than the set judgment threshold, or the amount of all data recorded in the steady-state data judgment data window is greater than M, exit the stable judgment variable sliding window sliding judgment process, take the mean value of all data in the steady-state data judgment data window as the final steady-state adjustment working condition data, and store the value in the database.

2. The intelligent control method for full well flow of oilfield separate layer water injection according to claim 1, characterized in that: The step S1 specifically comprises: S11: The flow regulation method using the historical expert experience model is as follows: first, the wellhead adopts constant flow control of total injection volume, to ensure that the total injection flow of the well is the same as the sum of the set injection allocation of each layer, and the under-injection and over-injection layers are determined according to the positive and negative deviation between the actual real-time injection flow, the set injection allocation and the allowable error E of each layer; if the sum of the set injection allocation and the allowable error E of a layer minus the actual real-time injection flow of the layer is negative, it indicates that the layer is an over-injection layer; if the difference between the set injection allocation and the allowable error E of a layer minus the actual real-time injection flow of the layer is positive, it indicates that the layer is an under-injection layer; the number of under-injection layers and over-injection layers of the well is recorded as and respectively, and the adjustment direction is determined according to the number of under-injection layers and over-injection layers; when is greater than , the adjustment direction is to adjust the under-injection layer first, and then adjust the over-injection layer, otherwise when is less than , the adjustment direction is to adjust the over-injection layer first, and then adjust the under-injection layer; S12: According to the adjustment direction determined in step S11, the automatic flow adjustment of the layer section meeting the adjustment direction is carried out in the order from the layer section layer position to the top down, when the deviation between the actual injection real-time flow of a layer section and the set injection allocation of the layer section is less than the allowable deviation E, the adjustment of the layer section is stopped, the injection nozzle opening degree of the layer section is kept at the current opening degree, the automatic flow adjustment of the next layer section meeting the adjustment direction is carried out, until the deviation between the actual injection real-time flow of the layer section and the set injection allocation of the layer section is less than the allowable deviation E, after the adjustment of all layer sections in the adjustment direction is qualified, another adjustment direction is turned to, that is, after the adjustment of the over-injection layer section is qualified, the under-injection layer section is adjusted, or after the adjustment of the under-injection layer section is qualified, the over-injection layer section is adjusted, until all layer sections are adjusted qualified, that is, the number of under-injection layer sections and over-injection layer sections of the whole well is 0.

3. The intelligent control method for full well flow of oilfield separate layer water injection according to claim 1, characterized in that: The step S3 specifically comprises: S31: The long short-term memory convolutional neural network is used for the injection nozzle opening degree and injection flow time series data prediction of the whole well prediction neural network model, the pre-processed steady state adjustment working condition data is used as the input of the front-end convolutional neural network, the automatic extraction of multi-dimensional input data features is realized through the convolutional layer and the pooling layer of the convolutional neural network, the convolutional neural network feature extraction result is used as the input of the back-end bidirectional long short-term memory network layer, the output result of the bidirectional long short-term memory network layer obtains different weights through the attention mechanism, and after the normalization processing, the output is obtained to the full connection layer to obtain the final prediction output result, and the output of the whole well prediction neural network model is the injection nozzle opening degree and injection flow prediction value of each layer section; S32: The linear normalization processing is used as the pre-processing method of the steady state adjustment working condition data, and the linear normalization processing is respectively carried out on the steady state adjustment working condition data of the pressure, the actual injection real-time flow and the injection nozzle opening degree of each layer section, and the processing formula is as follows: ， represents the original signal; represents the normalized result; represents the original signal minimum value; represents the original signal maximum value.

4. The intelligent control method for full well flow of oilfield separate layer water injection according to claim 1, characterized in that: The step S4 specifically comprises: S41: The prediction accuracy calculation method involves calculating the mean square error using K consecutive points. The result of the mean square error is less than a given value. This indicates that the prediction accuracy has met the training requirements; ， denotes a noisy original signal, denotes a model predicted signal, denotes a data length, denotes a mean square error; S42: The neural network intelligent control module combines the whole well prediction neural network model, the feedforward controller and the feedback PID controller, the product of the injection nozzle opening degree prediction value of each layer section of the whole well prediction neural network model and the output result of the input set value judgment module is used as the output of the feedforward controller, the deviation between the injection flow prediction value of each layer section of the whole well prediction neural network model and the set injection allocation of each layer section is used as the input value of the feedback PID controller, the feedback adjustment amount of the injection nozzle opening degree of each layer section is calculated, and the sum of the feedback adjustment amount of the injection nozzle opening degree of each layer section and the output of the feedforward controller is used as the final control output and controls the injection nozzle opening degree of each layer section of the multi-layer layered water injection control object. S43: the input setting value judgment module is used for judging whether the injection rate setting value of each layer section has a step change, and the judgment amplitude is , that is, when the input value setting injection rate change amplitude is greater than or equal to , it indicates that a step change occurs, and the output of the input setting value judgment module is 1; when the input value change amplitude is less than , the output of the input setting value judgment module is 0.

Images