CN106677763B - Dynamic integrated modeling-based oil well working fluid level prediction method - Google Patents
Dynamic integrated modeling-based oil well working fluid level prediction method Download PDFInfo
- Publication number
- CN106677763B CN106677763B CN201611248368.5A CN201611248368A CN106677763B CN 106677763 B CN106677763 B CN 106677763B CN 201611248368 A CN201611248368 A CN 201611248368A CN 106677763 B CN106677763 B CN 106677763B
- Authority
- CN
- China
- Prior art keywords
- model
- new
- training
- sub
- prediction
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 239000003129 oil well Substances 0.000 title claims abstract description 46
- 239000012530 fluid Substances 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 84
- 238000004519 manufacturing process Methods 0.000 claims description 22
- 239000007788 liquid Substances 0.000 claims description 18
- 238000012360 testing method Methods 0.000 claims description 14
- 238000005457 optimization Methods 0.000 claims description 10
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 230000010354 integration Effects 0.000 abstract description 8
- 238000002474 experimental method Methods 0.000 description 9
- 238000005259 measurement Methods 0.000 description 8
- 230000008901 benefit Effects 0.000 description 7
- 230000003068 static effect Effects 0.000 description 7
- 230000007547 defect Effects 0.000 description 6
- 238000012843 least square support vector machine Methods 0.000 description 6
- 238000005086 pumping Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000000691 measurement method Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000012163 sequencing technique Methods 0.000 description 4
- 230000005499 meniscus Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000009530 blood pressure measurement Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
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
- E21B47/00—Survey of boreholes or wells
- E21B47/04—Measuring depth or liquid level
- E21B47/047—Liquid level
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Geology (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mining & Mineral Resources (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Fluid Mechanics (AREA)
- Environmental & Geological Engineering (AREA)
- Geochemistry & Mineralogy (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Geophysics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
An oil well working fluid level prediction method based on dynamic integrated modeling belongs to the technical field of information; the method comprises the following steps: obtaining samples and sorting according to weight; carrying out model training on the sample, retraining the submodel with the error exceeding the threshold value, and calculating the weight of the submodel without exceeding the threshold value; outputting an integrated model by the sub-model in a weighting manner; calculating whether the error of the integrated model exceeds a threshold value by using the patrol data, if so, forming a new training set, retraining the original submodel with the error exceeding the threshold value into a new submodel, otherwise, keeping the atomic model as the new submodel, and outputting the new submodel in a weighting manner into a new integrated model; acquiring auxiliary variables in real time, and inputting the auxiliary variables into an integrated model to predict the working fluid level of the integrated oil well; the method adopts iterative integration modeling, has higher prediction precision than a single model, reduces the influence of individual errors by sub-model weighted output, and has strong model generalization performance; the change of the sample weight in the weak learning machine submodel can improve the adaptability of the new model to the sample.
Description
Technical Field
The invention belongs to the technical field of information, and particularly relates to an oil well working fluid level prediction method based on dynamic integrated modeling.
Background
In the actual production process of the oil field, in order to realize the maximum liquid production amount, the pumping unit needs to adjust the pumping frequency according to the constantly changing parameters of the oil well, so that the pumping unit reaches a reasonable working state. The working fluid level of the oil well is the fluid level depth of the annular space of an oil casing of the oil well in the production process, is one of important production guidance data of the oil field, and is an important reference index for reflecting the working state of the oil pumping unit. The pumping unit can operate in a reasonable operation mode only by timely grasping the data information of the working fluid level. At present, most of oil well working fluid level measurement still adopts a traditional manual measurement method, such as echo measurement, a pressure measurement method, a buoy method and the like, the traditional manual measurement has the defects of large error, low efficiency, poor real-time performance and the like, and the current production requirements cannot be met. In recent years, with the wide application of soft measurement methods in various industries, people propose a single-model intelligent algorithm soft measurement method, however, the method has the defects of poor generalization capability, easy occurrence of overfitting, low precision and the like in practical application, and in the actual production process, along with the reduction of the prediction precision of a produced dynamic operation model, the model needs to be updated to improve the dynamic performance of the model. In the conventional updating method, the model is updated by adding training sample data, so that the performance of the model is gradually deteriorated along with the training, and the production requirement cannot be met.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides an oil well working fluid level prediction method based on dynamic integrated modeling.
The technical scheme of the invention is as follows:
an oil well dynamic liquid level prediction method based on dynamic integrated modeling comprises the following steps:
step 1: acquiring working fluid level data and auxiliary variables corresponding to the working fluid level data in the production process of an oil well to form historical data, namely samples, dividing the samples into training samples and testing samples, and respectively forming a training set and a testing set;
step 2: setting the number of weak learning machines, an error overrun index threshold value and a sample prediction error threshold value;
and step 3: calculating sample weight, and sequencing the training set samples according to the weight;
and 4, step 4: carrying out optimization assignment on LSSVM model parameters by using a black hole optimization algorithm, carrying out model training on sequenced training samples, calculating whether an error overrun index of a submodel obtained by training exceeds a threshold value, if so, abandoning the submodel, and executing the step 3, otherwise, calculating a weight corresponding to the submodel, and updating the sample weight, wherein the sample weight is calculated according to a prediction error of the sample passing through the submodel, the weight of the submodel is calculated according to the combination of the error overrun index and a prediction root mean square error of the submodel, the sample prediction error is a relative error of a prediction result of the sample carrying out dynamic liquid level prediction through the submodel, and the error overrun index is the sum of the sample weights of which the sample prediction error is greater than the threshold value; the error overrun index is the sum of the weights of training samples with sample prediction errors exceeding a threshold value or the sum of the weights of patrol data with patrol data prediction errors exceeding the threshold value.
And 5: judging whether the number of the submodels reaches a set value, if so, finishing the training of the models, and outputting all the submodels in a weighting manner to obtain an integrated model, otherwise, executing the step 3;
step 6: periodically surveying the working fluid level, calculating whether the error overrun rate index of the integrated model exceeds a threshold value by using collected surveying data, if so, executing the step 7, otherwise, executing the step 12; the patrol data have the same weight;
and 7: searching similar data of the patrol data in historical data by utilizing a similarity principle according to the patrol data to form an updated subset, partially replacing dissimilar samples in the original training set to form a new training set, and distributing the same weight to the new training set;
and 8: predicting the new training set by using the original submodel, calculating whether the error overrun index of the submodel exceeds a threshold value, if so, executing the step 9, otherwise, keeping the atomic model as a new submodel and only updating the weight, and executing the step 11;
and step 9: calculating the weights of the samples in the new training set, and sequencing the samples according to the weights;
step 10: carrying out model training on the training samples in the new training set, calculating whether the error overrun rate of the new sub-model obtained by training exceeds a threshold value, if so, discarding the new sub-model, and executing the step 9, otherwise, executing the step 11;
step 11: judging whether the error overrun rate of the new submodel is zero, if so, giving the current maximum value of the weight of the new submodel according to the prediction level index limit, and otherwise, calculating the weight of the new submodel; and the new submodel comprises a new submodel or a reserved atom model serving as a new submodel, and the weight of the new submodel is calculated according to the error overrun rate index of the training samples in the new training set and the root mean square error of the submodel.
Step 12: judging whether the sub-models which are not updated exist, if so, executing the step 8, otherwise, outputting all new sub-models to obtain new integrated models;
step 13: and in the actual production working process of the oil well, acquiring auxiliary variables in real time, and inputting the auxiliary variables into the integrated model to predict the working fluid level of the integrated oil well.
Has the advantages that: compared with the prior art, the dynamic modeling-based oil well working fluid level prediction method has the following advantages:
1. compared with the traditional manual measurement method, the soft measurement modeling method based on iterative integration has the remarkable advantages of high prediction precision and high efficiency;
2. the method of iterative integrated modeling is adopted, and the prediction precision is higher than that of a single model algorithm, and because the influence on the overall prediction precision due to large individual error is reduced to the maximum degree by the weighted output of the sub-models, the generalization capability of the models is obviously improved, and the output is more scientific;
3. the adaptability of the new model to the samples can be improved by changing the sample weight in the weak learning machine submodel;
4. the change of the weight of the submodel of the weak learning machine reflects the direct balance of the overall error level of the model and the sample qualification rate, and the submodel with good effect is more effectively highlighted;
5. compared with the traditional method that training data are only continuously added, the new updating method enables the updating time of the model to be obviously shortened, and can also retain data information before the model, and has great advantages for practical application.
Drawings
FIG. 1 is a flow chart of an integration model establishing method in an oil well working fluid level prediction method based on dynamic integration modeling according to an embodiment of the invention;
FIG. 2 is a flow chart of a dynamic model updating method in the dynamic integrated modeling-based oil well working fluid level prediction method according to the embodiment of the invention;
FIG. 3 is a graph of single model prediction error indicators according to an embodiment of the present invention;
FIG. 4 is a diagram showing the result of predicting the working fluid level of the sub-model 1 according to the embodiment of the present invention;
FIG. 5 is a diagram of the result of predicting the dynamic liquid level of the oil well of the submodel 2 according to the embodiment of the invention;
FIG. 6 is a diagram of the result of predicting the dynamic liquid level of the oil well of the sub-model 3 according to the embodiment of the present invention;
FIG. 7 is a diagram of the sub-model 4 oil well working fluid level prediction results according to an embodiment of the present invention;
FIG. 8 is a diagram of the result of the sub-model 5 oil well working fluid level prediction according to an embodiment of the present invention;
FIG. 9 is a graph of integrated model oil well working fluid level prediction results in accordance with an embodiment of the present invention;
FIG. 10 is a comparison graph of the dynamic model and static model oil well working fluid level prediction effect of the embodiment of the invention.
Detailed Description
An embodiment of the present invention will be described in detail below with reference to the accompanying drawings.
In one embodiment of the present invention, the first and second electrodes are,
as shown in fig. 1-2, a dynamic modeling based oil well dynamic liquid level prediction method specifically includes the following steps:
step 1: as shown in figure 1, the working fluid level data y in the oil well production process is obtainediAnd working fluid level data yiCorresponding auxiliary variables, the auxiliary variables constituting an auxiliary vector xiThe dynamic liquid level data and its auxiliary vector constitute historical data (x)i，yi) Where, i is 1,2, …, m is the number of history data, and m is history data (x)i，yi) Assigning a weighted median value
Dividing historical data into training sets TR for training modelstAnd a test set TE for testing the modeltWherein, TR ist+TEtM, T represents that the current model is the tth sub-model, T is 1,2, …, T; the auxiliary variables include: oil well casing pressure, pumping efficiency and flow rate;
step 2: setting the number T of weak learning machines and the sample prediction error threshold e0Sum error overrun indicator threshold ε0；
And step 3: calculating historical data (x)i，yi) Weight of
Sequencing the training set samples according to the historical data weight;
and 4, step 4: performing optimization and assignment on model parameters of a Least Square Support Vector Machine (LSSVM) by using a black hole optimization algorithm according to the samples, performing model training on the sequenced training samples, and obtaining a sub-model H after the model training is completedt: x → y, calculating the sample prediction error of the submodel by using the test set and the training set, and calculating the error overrun index epsilontWhether or not a threshold value epsilon is exceeded0If yes, the submodel H is discardedtStep 3 is executed, otherwise, the weight corresponding to the submodel, namely the model prediction level index β is calculatedtAnd calculating an updated sample weight median wt+1(i)，t＝t+1；
Sample prediction error ARE of the submodelt(i) The calculation method comprises the following steps:
calculating error overrun index epsilontThat is, for each sample tested by the model, the weights of the samples whose prediction errors exceed the threshold are accumulated:
the model prediction level indicator βtThat is, the model weight can balance the influence of the submodel with large individual prediction error on the overall prediction level, and is specifically calculated by adopting the following formula:
wherein, RMSE is sqrt (sum (Yt-train _ test _ out). Lambda 2)/M) is the root mean square error of the model, and Lambda is the adjustment factor of the model;
the updating of the sample weight intermediate value can increase the chance that a sample with a large prediction error enters a training set, and improve the adaptability of a new model to the sample, and the sample weight intermediate value is calculated by adopting the following formula:
and 5: judging whether the number of the submodels, namely weak learning machines, reaches a set value, if so, finishing the training of the models, and outputting all the submodels in a weighting manner to obtain an integrated model, otherwise, executing the step 3;
the integrated model obtained by weighting and outputting all the sub-models is as follows:
step 6: as shown in FIG. 2, the updating of the original model is performed by gradually updating each weak learning machine sub-model, the working fluid level is periodically measured, and the collected measurement data (x'j，y'j) Calculating a sample prediction error of the integrated model, and calculating whether an error overrun rate index exceeds a threshold value, if so, executing a step 7, otherwise, executing a step 12, wherein j is 1,2, …, n and n are the number of the patrol data; the patrol data have the same weight;
the sample prediction error is calculated by the following formula:
wherein k is the dynamic updating times;
and 7: searching similar data of the patrol data in historical data by utilizing a similarity principle according to the patrol data to form an updated subset, replacing dissimilar samples in the original training set to form a new training set, and distributing a weight intermediate value to the samples in the new training set
And 8: using the original submodel HtPredicting the new training set, calculating the sample prediction error of the submodel, and calculating whether the error overrun index exceeds the threshold value, if yes, executing the step 9, otherwise,the atomic model is reserved as a new sub-model, only the weight is updated, and step 11 is executed;
the sample prediction error of the sub-model is as follows:
the error overrun indexes of the submodels are as follows:
and step 9: updating the weights of the samples in the new training set, and sequencing the samples according to the weights;
step 10: carrying out model training on the training samples in the new training set to obtain a new sub-model after the model training is finished, calculating whether the error overrun rate of the new sub-model exceeds a threshold value by using the test set and the new training set, if so, discarding the new sub-model, calculating the weight of the samples in the new training set, and executing the step 9, otherwise, executing the step 11;
step 11: judging whether the error overrun rate of the new sub-model is zero, if so, giving the current maximum value of the weight of the new sub-model according to the prediction level index limit, and otherwise, calculating the weight of the new sub-model, namely the prediction level index of the model;
the prediction level index of the submodel is calculated by adopting the following formula:
step 12: judging whether the sub-model is not updated, if so, executing the step 8, otherwise, outputting all new sub-models to obtain a new integrated model
After the model is dynamically updated, a new integrated model is obtained;
step 13: in the actual production working process of the oil well, acquiring auxiliary variables in real time, and inputting the auxiliary variables into an integrated model to predict the working fluid level of the integrated oil well;
the invention mainly solves the process variables which are difficult to be directly measured in the actual production process by using the actual production process of the oil field as an application background through a dynamic integrated modeling soft measurement method, and the key point is to predict the working fluid level data of the oil well, so that the oil well reaches a reasonable operation state and the economic benefit maximization is realized. In this embodiment, historical data of actual production in a certain oil field in the Liaohe river is selected as a sample to perform an experiment, which explains the reliability of the experiment and the actual utility of the invention.
And (3) sorting historical working fluid level data in the actual production process, and selecting the casing pressure, flow and pump efficiency of the oil well as auxiliary variables to realize the estimation of the measured variable working fluid level data. 200 sets of sample data are selected from the historical data, wherein 150 sets are used as training samples of a training set, 50 sets are used for checking the generalization of the model, and 200 sets of data are selected to check the prediction level of the model. And acquiring patrol data and searching 30 groups of samples with high similarity in the dynamic process to update the sub-models, and verifying whether the models are suitable under the dynamic condition by adopting 100 groups of data under different working conditions. The first group of experiments adopt a single model to predict the working fluid level; the second group of experiments adopt an iterative integration model to predict the working fluid level data; and selecting patrol data under different working conditions from the initial integrated model sample for training in the third group of experiments, and predicting the working fluid level by adopting a dynamic updating integrated model and a static integrated model. The prediction level of the model was evaluated in the experiment as Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) indicators.
The first group of experiments adopt a single model for prediction, namely an LSSVM model is adopted as a training model, and a black hole optimization algorithm is utilized for carrying out optimization assignment on LSSVM model parameters. And (4) predicting 200 groups of test set sample data by using the model, and finally calculating the Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE) of the model.
TABLE 1 error index for single model meniscus prediction
| Index | MAE | RMSE |
| Test set | 29.0890 | 43.1890 |
As shown in fig. 3, by comparing the real sample curve and the predicted sample curve, when the model is checked by using the test set, the prediction accuracy of the model for new sample data is obviously reduced, the real sample curve and the predicted sample curve are obviously deviated at multiple positions, and the prediction error is large. And then, the error indexes RMSE and MAE of the observation model are used for evaluating the prediction performance of the model, and both the error indexes corresponding to the model are large, so that the prediction effect is not ideal. The experimental result is analyzed, and the defects that the working fluid level prediction of a single model is low in obvious prediction precision, low in model generalization capability and the like exist.
In the second group of experiments, an LSSVM model is used as a training model, and the parameters of the model are subjected to optimization and assignment by using a black hole optimization algorithm to improve the prediction level of the model. The difference from the single model is that the idea of the iterative integration model is to train a plurality of weak learning machines through iteration and then to perform weighted output on all the weak learning machines to obtain the final integration model. Firstly, weighting processing is carried out on training samples, the training samples are ranked according to sample weights, and the ranked training samples are added into a model for training. And after the model error meets the requirement, calculating the weight corresponding to the sub-model, and simultaneously calculating and updating the sample weight. And the new sample weight is utilized to reorder the training samples, so as to increase the opportunity that the sample with large prediction error enters the training set, and then the next model continuously trains the samples, thereby increasing the generalization capability of the model. Until the number of the trained submodels reaches a set value, carrying out weighting input according to weights corresponding to all the submodels, and obtaining a final integrated model, wherein error indexes of the integrated model dynamic liquid level prediction are shown in figure 2:
TABLE 2 error index for integrated model meniscus prediction
| Index | MAE | RMSE |
| Test set | 13.0689 | 16.7410 |
The number of submodels in the integrated model is 5, and the weight value of each submodel is shown in table 3:
TABLE 3 weights corresponding to sub-models in an integrated model
| Integrated model submodel | Submodel 1 | Submodel 2 | Submodel 3 | Submodel 4 | Submodel 5 |
| Sub-model weights βt | 1.857 | 2.149 | 2.624 | 3.098 | 2.971 |
As shown in fig. 4-9, by comparing the prediction error indexes of the models, it can be clearly seen that the prediction error index of the integrated model is much smaller than the error index corresponding to the single model, which intuitively reflects that the integrated model has better prediction accuracy. The simulation image is analyzed, so that the error between the prediction curve of the integrated model and the real curve of the sample data is very small, and the prediction precision is very high. The reason for this is that, although the prediction accuracy of the submodel may not be high, and the error of the data point may be large, the prediction error of the integrated model is small when the entire error is balanced by a method in which the weight corresponding to the model having a small prediction error is large and the weight corresponding to the model having a large prediction error is small by the weighted output of the submodel. The analysis proves that the integrated model has the advantages of very high prediction precision compared with a single model, strong generalization capability of the model and the like.
Along with the production, the related parameters of the oil well can change along with the change of the well condition or the production process, so that the output of the original measurement model can be gradually deteriorated, a third group of experiments uses 100 groups of production data under the other working condition of the same well group, the method for dynamically updating the sample by using the inspection data and the static model which is not updated are respectively adopted for prediction, the prediction effect is shown in figure 10, and the error indexes of dynamic model and static model dynamic liquid level prediction are shown in table 4:
TABLE 4 error index for dynamic model and static model working fluid level prediction
| Index | MAE | RMSE |
| Dynamic model | 7.1599 | 10.8067 |
| Static model | 33.7971 | 38.3507 |
By comparing the prediction error indexes of the models, it can be obviously seen that the prediction error index of the dynamic integrated model is much smaller than the error index corresponding to the static model, and the effectiveness of dynamic update is intuitively reflected.
In conclusion, the method adopts the iterative integration soft measurement dynamic modeling method to predict the oil well working fluid level data, overcomes the defects of large error, low efficiency, poor real-time property and the like of the traditional method, and overcomes the defects of low prediction precision, easy generation of overfitting phenomenon and poor generalization capability of a single model to the working fluid level prediction. And weighting the samples during model training, and increasing the opportunity that the samples with large errors enter a training set so that the model can train the samples for multiple times, thereby improving the prediction accuracy of the model and enhancing the generalization capability of the model. Meanwhile, a plurality of submodels are adopted for weighted output, the submodels have small prediction errors, and the submodels have large occupied weight in the output of the integrated model, and the submodels have small occupied weight with large errors, so that the integrated model has higher prediction precision. Compared with the traditional updating method, the model dynamic updating method has the advantages of short updating time and no loss of the previous data information, and provides effective guarantee for the online application of the model. If the method is applied to the actual production process of the oil field, the problems faced by the conventional oil well dynamic liquid level measurement can be well solved, the operation mode of the oil well of the oil field is well guided, and the method has important practical significance for improving the liquid yield of the oil field.
Claims (9)
1. An oil well dynamic liquid level prediction method based on dynamic integrated modeling is characterized by comprising the following steps:
step 1: acquiring historical data, namely samples, in the production process of an oil well, dividing the samples into training samples and testing samples, and respectively forming a training set and a testing set; the historical data comprises working fluid level data and auxiliary variables corresponding to the working fluid level data;
wherein the working fluid level data y in the production process of the oil welliAnd working fluid level data yiCorresponding auxiliary variables, the auxiliary variables constituting an auxiliary vector xiThe dynamic liquid level data and its auxiliary vector constitute historical data (x)i，yi) I is 1,2, …, m is the number of historical data;
step 2: setting the number of weak learning machines and an error overrun index threshold;
and step 3: performing model training on the training sample to obtain a sub-model;
and 4, calculating whether the error overrun index of the submodel exceeds a threshold value by using the sample, if so, abandoning the submodel, executing the step 3, otherwise, calculating the weight corresponding to the submodel, namely the model prediction level index β t, and calculating the intermediate value w of the updated sample weightt+1(i) T +1, where T denotes that the current model is the tth sub-model, T1, 2, …, T;
the model prediction level index β t is a model weight, which can balance the influence of the submodel with large individual prediction error on the overall prediction level, and is calculated by the following formula:
wherein epsilontFor the error overrun indicator, i.e. for each sample tested by the model, the weights of the samples for which the sample prediction error exceeds a threshold are accumulated:
AREt(i) for the sample prediction error of the sub-model,e0predicting error threshold for samples, historical data weight
Wherein, RMSE is sqrt (sum ((Yt-train _ test _ out). Lambda 2)/m) is the model root mean square error, and Lambda is the model adjusting factor;
the updated sample weight median is calculated by using the following formula:
and 5: judging whether the number of the submodels reaches the number of weak learning machines, if so, outputting all the submodels as an integrated model, otherwise, executing the step 3;
the integrated model obtained by weighting and outputting all the sub-models is as follows:
wherein HtObtaining a sub-model after the model training in the step 4 is completed;
step 6: periodically surveying the working fluid level, calculating whether the error overrun rate index of the integrated model exceeds a threshold value by using collected surveying data, if so, executing the step 7, otherwise, executing the step 12;
and 7: constructing a new training set;
and 8: calculating whether the error overrun index of the atomic model exceeds a threshold value by using the training samples in the new training set, if so, executing the step 9, otherwise, keeping the atomic model as a new sub-model, and executing the step 11;
and step 9: carrying out model training on the new training concentrated training samples to obtain a new sub-model;
step 10: calculating whether the error overrun rate of the new sub-model obtained by training exceeds a threshold value, if so, discarding the new sub-model, and executing the step 9, otherwise, executing the step 11;
step 11: judging whether the error overrun rate of the new sub-model is zero, if so, giving the current maximum value of the weight of the new sub-model according to the prediction level index limit, and otherwise, calculating the weight of the new sub-model, namely the prediction level index of the model;
the prediction level index of the submodel is calculated by adopting the following formula:
step 12: judging whether the sub-model is not updated, if so, executing the step 8, otherwise, outputting all new sub-models to obtain a new integrated model
Wherein, Fk(x) Obtaining a new integrated model for the obtained new integrated model, namely, after the model is dynamically updated;
step 13: and in the actual production working process of the oil well, acquiring auxiliary variables in real time, and inputting the auxiliary variables into the integrated model to predict the working fluid level of the integrated oil well.
2. The method of claim 1, wherein the auxiliary variables include oil pressure, pump efficiency, and flow rate.
3. The method for predicting the dynamic liquid level of the oil well based on the dynamic integrated modeling as claimed in claim 1, wherein the training samples in the step 3 and the training samples in the new training set in the step 8 are training samples sorted according to sample weights, the sample weights are calculated according to the relative errors of prediction results of the dynamic liquid level prediction of the samples through a sub-model, namely prediction errors, and the larger the prediction errors are, the larger the training sample weights are.
4. The method for predicting the working fluid level of the oil well based on the dynamic integrated modeling as claimed in claim 1, wherein the specific method for performing model training on the training samples in the step 3 is as follows: and performing optimization and assignment on the model parameters by using a black hole optimization algorithm according to the sample data, and performing model training on the training sample according to the obtained model parameters, wherein the model is an LSSVM model.
5. The method for predicting the dynamic liquid level of the oil well based on the dynamic integrated modeling as claimed in claim 1, wherein the method for outputting the integrated model in the step 5 is as follows: and performing weighted output on all the submodels according to the weight of the submodels, wherein the weight of the submodels is calculated according to the combination of the error overrun rate index and the root mean square error of the submodels.
6. The method for predicting the working fluid level of the oil well based on the dynamic integrated modeling according to claim 1, wherein the patrol data in the step 6 are patrol data with the same weight.
7. The method for predicting the dynamic liquid level of the oil well based on the dynamic integrated modeling as claimed in claim 1, wherein the specific method for constructing the new training set in the step 7 is as follows:
according to the patrol data, similar data of the patrol data are searched in historical data by utilizing a similarity principle to form an updated subset, and dissimilar samples in an original training set are partially replaced to form a new training set.
8. The method for predicting the dynamic liquid level of the oil well based on the dynamic integrated modeling as claimed in claim 1, wherein the specific method for obtaining the new integrated model by all the new sub-models output in the step 12 is as follows: and weighting and outputting a new integrated model according to the weight of each new sub-model, wherein the new sub-model weight is the updated sub-model weight when the new sub-model is generated or the atom model is reserved as the new sub-model, and the sub-model weight is calculated according to the error overrun rate index and the sub-model root mean square error of the training sample containing the patrol data in the new training set.
9. The method of claim 3, wherein the error overrun indicator is the sum of the weights of training samples with sample prediction errors exceeding a threshold value and the sum of the weights of patrol data with patrol data prediction errors exceeding a threshold value, and the patrol data are patrol data with the same weight.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611248368.5A CN106677763B (en) | 2016-12-29 | 2016-12-29 | Dynamic integrated modeling-based oil well working fluid level prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611248368.5A CN106677763B (en) | 2016-12-29 | 2016-12-29 | Dynamic integrated modeling-based oil well working fluid level prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106677763A CN106677763A (en) | 2017-05-17 |
| CN106677763B true CN106677763B (en) | 2020-02-07 |
Family
ID=58872538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611248368.5A Active CN106677763B (en) | 2016-12-29 | 2016-12-29 | Dynamic integrated modeling-based oil well working fluid level prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106677763B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108596415B (en) * | 2017-12-15 | 2023-11-24 | 创新先进技术有限公司 | Model integration method and device |
| CN109472285B (en) * | 2018-09-29 | 2020-12-22 | 北京中油瑞飞信息技术有限责任公司 | Lost circulation identification method and device and computer equipment |
| CN109611077A (en) * | 2018-11-14 | 2019-04-12 | 渤海大学 | A kind of pumpingh well oil liquid gas-oil ratio online soft sensor method based on multi-model |
| CN109740661B (en) * | 2018-12-28 | 2022-10-28 | 沈阳工业大学 | Oil well moving liquid level self-adaptive prediction method based on fuzzy evaluation |
| CN112395730A (en) * | 2019-08-12 | 2021-02-23 | 北京国双科技有限公司 | Method and device for determining working fluid level depth parameter of pumping well |
| CN110750858B (en) * | 2019-09-10 | 2020-07-21 | 太原理工大学 | 4-NP reduction catalyst modeling prediction method based on ECSA Gaussian process regression |
| CN110751987A (en) * | 2019-10-17 | 2020-02-04 | 北京化工大学 | Method for establishing soft measurement model of monitoring variable for ethylene preparation through naphtha cracking |
| CN112943224B (en) * | 2019-12-11 | 2023-02-07 | 中国石油化工股份有限公司 | Method for calculating dynamic liquid level of heavy oil well |
| CN111126963A (en) * | 2019-12-24 | 2020-05-08 | 微创(上海)网络技术股份有限公司 | Dynamic integration method of heterogeneous multi-source data |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040163806A1 (en) * | 2003-02-20 | 2004-08-26 | Hadley James P. | Well monitoring system |
| CN101382556B (en) * | 2008-09-28 | 2012-07-25 | 山西潞安矿业(集团)有限责任公司 | Soft measurement method for floating fine coal ash from slurry based on data drive |
| CN102270309B (en) * | 2011-07-27 | 2013-05-01 | 华北电力大学 | Short-term electric load prediction method based on ensemble learning |
-
2016
- 2016-12-29 CN CN201611248368.5A patent/CN106677763B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN106677763A (en) | 2017-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106677763B (en) | Dynamic integrated modeling-based oil well working fluid level prediction method | |
| CN109740661B (en) | Oil well moving liquid level self-adaptive prediction method based on fuzzy evaluation | |
| GB2547816B (en) | Actually-measured marine environment data assimilation method based on sequence recursive filtering three-dimensional variation | |
| CN102819663B (en) | Method for forecasting ship wake based on optimized support vector regression parameter | |
| CN110363337B (en) | Oil measuring method and system of oil pumping unit based on data driving | |
| CN103927412A (en) | Real-time learning debutanizer soft measurement modeling method on basis of Gaussian mixture models | |
| CN105740984A (en) | Product concept performance evaluation method based on performance prediction | |
| CN104504475A (en) | AR*-SVM (support vector machine) hybrid modeling based haze time series prediction method | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| CN112861436A (en) | Real-time prediction method for engine emission | |
| CN108830405B (en) | Real-time power load prediction system and method based on multi-index dynamic matching | |
| Chen et al. | Combining fuzzy iteration model with dynamic programming to solve multiobjective multistage decision making problems | |
| CN114169091A (en) | Method for establishing prediction model of residual life of engineering mechanical part and prediction method | |
| CN115982141A (en) | Characteristic optimization method for time series data prediction | |
| CN117474173B (en) | Multi-water source dynamic allocation device and system for plain river network area | |
| CN103605493A (en) | Parallel sorting learning method and system based on graphics processing unit | |
| CN115472233A (en) | Semi-supervised integrated industrial process soft measurement modeling method and system based on thermal diffusion label propagation | |
| CN114970698B (en) | Metering equipment operation performance prediction method based on improved LWPS | |
| CN114819107B (en) | Mixed data assimilation method based on deep learning | |
| CN116822566A (en) | Short-term photovoltaic power generation power prediction method | |
| CN116108963A (en) | Electric power carbon emission prediction method and equipment based on integrated learning module | |
| CN113035363A (en) | Probability density weighted genetic metabolic disease screening data mixed sampling method | |
| Dong et al. | Hydraulic system fault diagnosis based on EMD and improved PSO-Elman ANN | |
| CN104516858A (en) | Phase diagram matrix method for nonlinear dynamic behavior analysis | |
| CN118333129B (en) | Identification model training method, nonlinear system identification method and nonlinear system identification system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |