CN115222065A - Wellhead pressure online multi-step prediction method based on Stacking ensemble learning - Google Patents
Wellhead pressure online multi-step prediction method based on Stacking ensemble learning Download PDFInfo
- Publication number
- CN115222065A CN115222065A CN202210722350.3A CN202210722350A CN115222065A CN 115222065 A CN115222065 A CN 115222065A CN 202210722350 A CN202210722350 A CN 202210722350A CN 115222065 A CN115222065 A CN 115222065A
- Authority
- CN
- China
- Prior art keywords
- data
- prediction
- online
- learner
- time
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 93
- 238000012549 training Methods 0.000 claims description 55
- 230000010354 integration Effects 0.000 claims description 19
- 238000012795 verification Methods 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 17
- 238000006243 chemical reaction Methods 0.000 claims description 11
- 238000006073 displacement reaction Methods 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 9
- 230000007246 mechanism Effects 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 7
- 239000004576 sand Substances 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000009826 distribution Methods 0.000 abstract description 12
- 230000008859 change Effects 0.000 abstract description 7
- 238000011161 development Methods 0.000 abstract description 3
- 238000010276 construction Methods 0.000 abstract description 2
- 238000002474 experimental method Methods 0.000 description 14
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 8
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 8
- 238000011160 research Methods 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000013136 deep learning model Methods 0.000 description 5
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N Silicium dioxide Chemical compound O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 230000015654 memory Effects 0.000 description 4
- 238000007637 random forest analysis Methods 0.000 description 4
- 238000009825 accumulation Methods 0.000 description 3
- 239000007788 liquid Substances 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000001934 delay Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 241000288105 Grus Species 0.000 description 1
- 239000006004 Quartz sand Substances 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000007636 ensemble learning method Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012856 packing Methods 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 238000007430 reference method Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Development Economics (AREA)
- Computing Systems (AREA)
- Quality & Reliability (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Agronomy & Crop Science (AREA)
- Animal Husbandry (AREA)
- Marine Sciences & Fisheries (AREA)
- Mining & Mineral Resources (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wellhead pressure online multi-step prediction method based on Stacking ensemble learning, relating to the technical field of shale gas development, and comprising the following steps: the method comprises the steps of constructing an online Stacking integrated multi-step prediction model, acquiring fracturing operation data in real time, inputting the fracturing operation data into the online Stacking integrated multi-step prediction model, outputting multi-time-step prediction data based on the online Stacking integrated multi-step prediction model, and updating the online Stacking integrated multi-step prediction model by utilizing the fracturing operation data. The method integrates online learning and Stacking, continuously learns newly-appeared data distribution and forgets outdated concepts, improves the adaptability and the prediction performance of an integrated model OSE-MS to the distribution change of the data flow, is applied to wellhead pressure multi-step prediction of fracturing construction, and can help field constructors to adjust fracturing operation parameters in time, thereby improving the operation efficiency and avoiding safety accidents.
Description
Technical Field
The invention relates to the technical field of shale gas development, in particular to a wellhead pressure online multi-step prediction method based on Stacking ensemble learning.
Background
In shale gas exploitation, fracturing operation has obvious effects of increasing storage and increasing production. In the fracturing operation process, a large amount of high-viscosity liquid is required to be pressed into a stratum, after enough cracks are formed, propping agents such as quartz sand are added to fill the cracks, and therefore the permeability of an oil-gas layer is effectively improved. The existing fracturing operation control system mainly displays real-time wellhead pressure and other data about fracturing operation through a monitoring system, and a field expert adjusts fracturing operation parameters according to experience, so that effective expansion of cracks is controlled, wellhead pressure is manually pre-judged in advance, wellhead pressure is prevented from exceeding a safety threshold, and the existing fracturing operation control system has great limitation.
Real-time operation data (namely fracturing operation data) acquired in a fracturing operation process is typical time series data, and investigation finds that single-step prediction is considered in the conventional time series prediction research, the research on multi-step prediction is less, and the real-time operation data is mainly concentrated in special fields such as wind power prediction, network flow prediction and power prediction. For example, zhou et al propose a deep multi-output model to predict the air quality 4 steps ahead, and the number of predicted steps is short; guojia et al introduce a full attention mechanism into the time series multi-step prediction problem to predict the flow at the second level within the future minute-level time window; galicia and the like propose a big data time sequence multi-step prediction method based on ensemble learning, the method integrates random forests, decision trees and gradient enhancement trees, and the research of Guojia and Galicia has the problem of error accumulation without considering the time dependence characteristics of predicted values in a prediction window; sahoo et al compared the effect of different multi-step prediction strategies on multi-step prediction of network bandwidth utilization, and experiments prove that the best effect can be obtained by performing multi-step prediction with a multi-input multi-output strategy. Chen et al propose a time series prediction nonlinear learning integration, which integrates a long-and-short-term memory network, a support vector machine and an extreme optimization algorithm to perform wind speed multi-step prediction, and a result display integration model can effectively improve the prediction accuracy. The method assumes that the prediction data is stable and concept drift does not occur, and actually, data distribution always changes in various special fields, and for the technical field of shale gas development, because the data distribution of different oil and gas wells and different physical sections has great difference, a model obtained by using fracturing operation data training of a certain well cannot be directly applied to other well sections, so that the conventional multi-step prediction method is difficult to be directly applied to well head multi-time step pressure prediction. Moreover, research and study find that research on online learning is mostly focused on classification tasks, and that a small amount of research on prediction tasks is limited to single-step prediction, so that an online learning method for multi-step prediction tasks is not found at all.
In addition, ensemble learning is also widely used for time series prediction, and an ensemble model adopts a plurality of base learners for sequence prediction, so that the ensemble model has better generalization capability compared with a single model. The Stacking ensemble learning is a typical ensemble mode, a plurality of training subsets are obtained on the whole training data set through a cross validation or self-help sampling method, then a series of base learners are obtained through training of the sub-training subsets, finally, the outputs of the base learners are fused, then a meta-learner is trained, and the meta-learner is used for prediction. Over time, the data distribution of the non-steady-state data stream may change somewhat. In order to solve the problem, oza provides online ensemble learning, and provides an online boosting algorithm and an online bagging algorithm, and realizes effective integration of a base learner through the change of online learning distribution, so that the performance of an integral model can be improved to a certain extent. The main idea of online ensemble learning is to update the parameters of some base learners or replace some of them in the prediction process. However, no researchers have studied online Stacking integration.
The fracturing operation data is collected according to time, the method has the characteristics of large distribution change, strong data relevance and the like, and the well head pressure can be predicted in advance as a multi-time step sequence prediction (namely multi-step prediction) problem according to the characteristics of the fracturing operation data. The multi-step prediction is a plurality of observed values of the estimated future time, compared with the single-step prediction which only estimates a future value based on historical data, the multi-step prediction is more uncertain, the accumulation of errors and the lack of information make the multi-step prediction more difficult, and the current research aiming at the multi-step prediction has the defects of short prediction time and difficult adaptation to the change of the wellhead pressure data flow at any moment. Therefore, how to improve the accuracy of predicting the wellhead pressure based on multi-step prediction is an urgent problem to be solved in order to guide safe and efficient fracturing operation.
Disclosure of Invention
The invention aims to: in order to solve the problem that the accuracy of predicting wellhead multi-time step pressure in fracturing operation data which is large in fluctuation and easy to generate concept drift is low in the existing method, a wellhead pressure online multi-step prediction method based on Stacking ensemble learning is provided.
The technical scheme adopted by the invention is as follows:
a wellhead pressure online multi-step prediction method based on Stacking ensemble learning comprises the following steps: an Online Stacking integrated multi-step prediction model (OSE-MS model for short) is constructed, fracturing operation data are obtained in real time and input into the OSE-MS model, multi-time-step prediction data are output based on the OSE-MS model, and the OSE-MS model is updated by the aid of the fracturing operation data. Some of the symbols involved in the process and their meanings are shown in table 1 below:
TABLE 1 symbols and their meanings
In the invention, the fracturing operation data comprises a wellhead pressure value and a related factor data set thereof, the related factor data set comprises the characteristics of a fracturing physical section, a stage, casing pressure, discharge displacement, stage discharge displacement, accumulated discharge displacement and sand concentration, the characteristic number contained in the related factor data set is d, and d is a variable and is a positive integer. The fracturing operation data is single time step data, and the formula is as follows:
the OSE-MS model comprises three parts of feature engineering, an online training mechanism and online prediction, and the method for constructing the OSE-MS model comprises the following steps:
The method for converting the characteristic engineering data comprises the following steps: and dividing the fracturing operation data through a sliding time window based on the size w of the predicted output window, and converting the divided fracturing operation data into multi-time step data by utilizing characteristic engineering.
step 2.1, online learning of a base learner: dynamically constructing a base learner integration SE in an online mode, wherein the method specifically comprises the following steps:
and (3) hot start: firstly, off-line data B is processed 0 Initial base learner BM trained as a first subset of data 0 {lstm 0 ,gru 0 Add to the base learner integration SE. The number of basis learners included in the basis learner set SE is 1 (i.e. only the initial basis learner BM is included) 0 {lstm 0 ,gru 0 ) } directly using the initial base learner BM 0 {lstm 0 ,gru 0 Predicting the wellhead pressure value, and enabling lstm 0 ，gru 0 The predicted average value of (a) is used as the wellhead pressure predicted value.
With the continuous increase of the fracturing operation data flow, the OSE-MS model continuously delays and verifies the current prediction effect in the whole prediction process, and the specific operation is as follows: after the wellhead pressure value is predicted every time, if the prediction error of the current OSE-MS model is larger than the prediction loss threshold value T h And triggering a data flow dynamic acquisition method. The data flow dynamic acquisition method comprises the following steps: let the collection time point be t 1 Time period (t) 0 ,t 1 ]Carrying out characteristic engineering data conversion on a plurality of internally acquired fracturing operation data (the number of the fracturing operation data is more than the minimum sample number M of model training) to obtain a data subset B 1 Using the data subset B 1 Get new base learner BM on-line training 1 And new base learning BM 1 To the base learner integration SE. Data subset B continuously obtained in the step n And B n-1 Are not overlapped with each other, therefore, the differentiated base learning machine BM can be continuously obtained n {lstm n ,gru n To adapt to the current data distribution.
Preferably, when the initialization setting in step 1 is performed, the maximum number N of the base learners is also initialized, where N is a constant, and is used to limit the number of the base learners in the SE integrated by the base learners in the whole prediction process, so as to control the size of the memory space occupied by the OSE-MS model in the prediction process. After the maximum number N of the base learners is set, the value range of the maximum number N of the base learners in the integrated SE of the base learners is [0, N ].
Step 2.2, the meta-learner learns online: data subset B 0 ,...,B n The method includes the steps that (1) a verification set of an integrated SE (sequence element) of a base learner is used for obtaining intermediate prediction data; and converting the acquired wellhead pressure true value by using a characteristic project to obtain a true data label, and splicing the intermediate prediction data and the true data label to construct the on-line training data of the element learner MM.
When the number of base learners in the base learner set SE changes, assume that there are n data subsets B at this time 0 ,B 1 ,...,B n And n base learners obtained by corresponding training (BM) 0 ,BM 1 ,...,BM n Wherein, the data corresponding to the n sub-data sets are
The kth (k is a variable, and 1. Ltoreq. K.ltoreq.n-1) data subset is taken as the verification set of the (k + 1) th base learner, and the initial base learner BM 0 Verification set B of current time n And fitting to obtain corresponding intermediate prediction data. Finally, the obtained intermediate prediction data
After being sorted according to the time stamp, the real data label is associated with the real data label
(each element in the real data tag is a pass t n Real tags obtained after the real values of the wellhead pressure obtained after the time and the characteristic engineering data conversion are carried out on the real values) are spliced to obtain the training data of the meta-learner MM.
Step 2.3, dynamic integration: judging whether a base learner is added or not based on the precision real-time evaluation method and the quantity of fracturing operation data which are not subjected to on-line training in the fracturing operation data stream; and judging whether the base learner integrated in the SE needs to be deleted or not based on the prediction errors before and after the base learner is added. The method of the step is as follows:
the precision real-time evaluation method comprises the following steps: let the last data subset collection time point be t n-1 Calculating t every time the prediction of wellhead pressure value is completed n-1 All wellhead pressure predicted values from moment to current moment t
And all the obtained actual values of the wellhead pressure
The prediction error in between.
Dynamically adding a base learner: if the prediction error of the multi-time-step prediction data of the current OSE-MS model is larger than the prediction loss threshold value T h And time period t n-1 +1,t]When the quantity of the fracturing operation data in the fracturing flow dynamic acquisition method is larger than the minimum sample quantity M of model training, the dynamic acquisition method of the data flow is triggered as follows: first for the time period t n-1 +1,t]Performing characteristic engineering data conversion on the fracturing operation data to obtain a new data subset, performing online training by using the data subset to obtain a new base learner, adding the new base learner into a base learner integrated SE (selected element), and finally performing element learner MM (selected element) by using the base learner integrated SE; and setting a variable delta equal to the prediction error of the current OSE-MS model for subsequent judgment of whether the base learner needs to be deleted dynamically.
Dynamic delete base learner: in order to limit the scale of the OSE-MS model and effectively improve the overall prediction performance, the weak base learning device is required to be deleted in the dynamic integration process: if the prediction error after the new base learner is larger than the prediction error delta before the new base learner, intercepting the time period t-N verify ,t]The prediction error of each base learner in the verification set is calculated as the verification set, and the base learner corresponding to the maximum calculated prediction error (i.e., the base learner)Weak base learner) and retrain the meta-learner MM. If the maximum number N of the learners with the base is initialized, judging whether the weak base learners need to be deleted or not according to the following judgment conditions: and when the condition is met, the weak base learners in the SE integrated by the base learners need to be deleted.
Online prediction: the OSE-MS model converts fracturing operation data acquired in real time into multi-time step data by utilizing characteristic engineering, then obtains intermediate prediction data through a base learner integrated SE, averages the intermediate prediction data, inputs the average intermediate prediction data into a meta-learner MM, and finally predicts the average intermediate prediction data by the meta-learner MM to obtain multi-time step prediction data, namely a plurality of wellhead pressure prediction values.
Preferably, the calculation method of the prediction error adopts average absolute error, root mean square error or average absolute percentage error.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
1. because the fracturing operation is usually carried out according to well depth subsection, the geology of each subsection has large difference, and complete training data is difficult to acquire, the method firstly trains and obtains an initial base learner based on a small amount of off-line data, and realizes predicted hot start; continuously evaluating the prediction error of the current OSE-MS model prediction result along with the continuous arrival of real-time fracturing operation data, when the prediction error exceeds a set prediction loss threshold value, training by using the previously collected fracturing operation data to obtain a new base learner, adding the new base learner into the OSE-MS model in a Stacking integrated mode on line, updating the meta-learner MM, and finally obtaining multi-time-step prediction data. The invention integrates online learning and Stacking, continuously learns newly-appeared data distribution and forgets outdated concepts, and improves the adaptability and the prediction performance of an integrated model OSE-MS to the distribution change of the data flow.
2. Compared with the existing multiple reference methods, the method improves the prediction performance of the wellhead pressure in the fracturing construction by more than 63.8%, 56.0% and 64.6% on the indexes of MAE, RMSE and MAPE respectively, has high prediction accuracy and is highly consistent with the actual wellhead pressure.
3. The method takes a plurality of cyclic neural network models as a base learner, so that error accumulation caused by overlong prediction output window is reduced; the circulating neural network is integrated in a packing mode, the advantage of strong stability of an integrated model is fully exerted, the defect of large fluctuation of the predicting performance of the circulating neural network is overcome, good stability can be still kept when the predicting performance is 30 steps ahead, reliable multi-step predicting results of wellhead pressure can be provided for field constructors in field application, the constructors are helped to adjust fracturing operation parameters in time, and the method has positive significance for improving operation efficiency and avoiding safety accidents.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a graph A comparing the predicted performance of the present invention with a prior art method;
FIG. 2 is a graph B comparing the predicted performance of the present invention with a prior art method;
FIG. 3 is a characteristic engineering schematic of the present invention;
FIG. 4 is a schematic diagram of the OSE-ME model on-line training in the present invention;
FIG. 5 is a graph comparing the MAE trend in different steps predicted by the present invention and the prior art;
FIG. 6 is a graph comparing the RMSE variation trend in different prediction steps of the present invention and the prior art;
FIG. 7 is a graph comparing the MAPE trend in different prediction steps of the present invention with that of the prior art.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments.
A wellhead pressure online multi-step prediction method based on Stacking ensemble learning includes the steps of obtaining fracturing operation data in real time by building an OSE-MS model, inputting the fracturing operation data into the OSE-MS model, outputting multi-time-step prediction data based on the OSE-MS model, and updating the OSE-MS model by utilizing the fracturing operation data.
The fracturing operation data comprises a wellhead pressure value and a related factor data set, the related factor data set comprises the characteristics of a fracturing physical section, a stage, casing pressure, discharge displacement, stage discharge displacement, accumulated discharge displacement and sand concentration, and the characteristic number contained in the related factor data set is d which is a variable and is a positive integer. The fracturing operation data is single time step data, and is shown in formula (1):
the OSE-MS model comprises three parts of feature engineering, an online training mechanism and online prediction, and the method for constructing the OSE-MS model comprises the following steps:
In particular, a predictive loss threshold T h Is related to the application scenario and controls the time for updating the OSE-MS model; the minimum number of model training samples M is the minimum number of training samples for the base learner units LSTM and GRU, and is determined by the OSE-MS model pre-obtained by training the M fracturing job dataWhether the measuring precision can meet the application requirement or not is determined; the predicted window size w controls the predicted time span, and the larger the window size is, the influence analysis of the parameter on the prediction precision can be seen in section 3.2 below; the maximum number N of the base learners controls the number of the base learners contained in the base learners in the integrated SE of the base learners, the larger N is, the higher the memory space occupied by the OSE-MS model is, and the determination of N needs to be combined with hardware equipment in an application scene and the requirement of prediction time.
The method for converting the characteristic engineering data comprises the following steps: and dividing the fracturing operation data through a sliding time window based on the size w of the predicted output window, and converting the divided fracturing operation data into multi-time step data by utilizing characteristic engineering.
Specifically, the characteristic engineering uses fracturing operation data at a plurality of continuous moments to model wellhead pressure values at a plurality of moments in the future, so that a mapping relation between historical fracturing operation data and future wellhead pressure changes is constructed.
In order to convert single time step data into multi-time step data, the method divides fracturing operation data through a sliding time window, converts the fracturing operation data into a series of time segments, and sets the time segment step length to be 1s. The process is shown in fig. 3, in which fig. 3, the horizontal axis represents time, and the vertical axis represents the true value of the wellhead pressure. Assuming that the current time is t, the sliding time window size is set equal to the predicted output window size w, and t>w, then the white boxes represent the time periods t-w, t-1]And (4) splicing the corresponding characteristic vector and the actual wellhead pressure value to form characteristic data, wherein the characteristic data is used as { X' t-w ,X' t-w+1 ,...,X' t-1 Represents it. The gray boxes represent the OSE-MS model versus time period [ t, t + w-1 ] from time t]Predicting wellhead pressure prediction value
By using
And (4) showing. Because the OSE-MS model also continuously obtains the true value of the wellhead pressure in the prediction process, namely y t The actual value y of the wellhead pressure at the time of t +1 and so on t+w-1 May be obtained at time t + w. Thus, starting at time t, the sequence of true values of wellhead pressure { y ] within a sliding time window of width w t ,y t+1 ,..,,y t+w-1 Will have to be obtained after the lapse of time w. Converting historical wellhead pressure data into real labels in subsequent online training by the characteristic engineering, and using Y t-w And (4) showing. The characteristic engineering data conversion is shown in formulas (2) and (3).
The multi-time step data obtained by converting the single-time step data into the feature engineering data is (X' 1 ,Y 1 ),(X' 2 ,Y 2 ),...,(X' t-w ,Y t-w ) Wherein, X' t-w The eigenvalues, Y, of the OSE-MS model will be input for time t-w t-w The real label is a real label obtained by acquiring a real value of the wellhead pressure after w time and performing characteristic engineering data conversion.
As shown in fig. 4, the online training mechanism includes two sub-modules, namely, a base learner online learning module and a meta learner online learning module, which are respectively represented by sub-diagrams (a) and (b). Each base learner BM is composed of an LSTM unit and a GRU unit, and a base learner integration SE is formed from 0 to n in an online manner by using { BM 0 ,...,BM n Represents it. Meanwhile, to prevent overfitting, intermediate prediction data obtained by integrating SE by the basis learner using a multiple linear regressor as the meta learner MM
Constructing the on-line training data of the meta-learner MM, and finally obtaining the multi-time step prediction through the prediction of the meta-learner MMAnd (6) measuring data.
Step 2.1, the base learner learns online: dynamically constructing a base learner integration SE in an online mode, wherein the method comprises the following steps:
and (3) hot start: firstly, off-line data B is processed 0 Initial base learner BM trained as a first subset of data 0 {lstm 0 ,gru 0 Add to the base learner integration SE; the number of basis learners included in the basis learner set SE is 1 (i.e. only the initial basis learner BM is included) 0 {lstm 0 ,gru 0 ) } directly using the initial base learner BM 0 {lstm 0 ,gru 0 Predicting the wellhead pressure value, and enabling lstm 0 ，gru 0 The predicted average value of (a) is used as the wellhead pressure predicted value.
With the continuous increase of the wellhead pressure data flow, the OSE-MS model continuously delays and verifies the current prediction effect in the whole prediction process, and the specific operation is as follows: after the wellhead pressure value is predicted every time, if the average absolute error value of the current OSE-MS model is larger than the prediction loss threshold T h And triggering a data flow dynamic acquisition method. The data flow dynamic acquisition method comprises the following steps: let the collection time point be t 1 Time period (t) 0 ,t 1 ]Carrying out characteristic engineering data conversion on a plurality of internally acquired fracturing operation data (the number of the internally acquired fracturing operation data is more than the minimum sample number M of model training) to obtain a data subset B 1 Using the data subset B 1 On-line training to obtain new base learning machine BM 1 And new base learning BM 1 To the base learner integration SE. Data subset B continuously obtained in the step n And B n-1 Are not overlapped with each other, therefore, the differentiated base learning machine BM can be continuously obtained n {lstm n ,gru n To adapt to the current data distribution.
Step 2.2, online learning of the meta learner: data subset B 0 ,...,B n The method comprises the steps of (1) taking a verification set of a base learner integrated SE (SE) to obtain intermediate prediction data; converting the obtained wellhead pressure true value by using feature engineering to obtain a true data label, splicing the intermediate prediction data and the true data label to construct the MM of the meta-learnerTraining data online.
When the number of base learners in the base learner set SE changes, assume that there are n data subsets B at this time 0 ,B 1 ,...,B n And n base learners obtained by corresponding training (BM) 0 ,BM 1 ,...,BM n Wherein, the data corresponding to the n sub-data sets are
Taking the kth (k is variable, and k is more than or equal to 1 and less than or equal to n-1) data subset as the verification set of the (k + 1) th base learner (i.e. the off-line data B 0 BM as a new base learner 1 Of the verification set, data subset B n-2 BM as a new base learner n-1 Of the verification set, data subset B n-1 BM as a new base learner n Verification set of) and the initial base learner BM 0 Verification set B of current time n And fitting to obtain corresponding intermediate prediction data. Finally, the obtained intermediate prediction data
After being sorted according to the time stamp, the data are labeled with real data
(each element in the real data tag is a pass t n Real tags obtained after the real values of the wellhead pressure obtained after the time and the characteristic engineering data conversion are carried out on the real values) are spliced to obtain the training data of the meta-learner MM. In the whole prediction process, the meta-learner MM is continuously updated in an iterative manner along with the dynamic integration of the base learner, so that the meta-learner MM can obtain better prediction performance.
Step 2.3, dynamic integration: judging whether a base learner is added or not based on the precision real-time evaluation method and the quantity of fracturing operation data which are not trained on line in the fracturing operation data flow; and judging whether the base learner integrated in the SE needs to be deleted or not based on the average absolute error values before and after the base learner is added. The method of the step is as follows:
the precision real-time evaluation method comprises the following steps:let the last data subset collection time point be t n-1 Calculating t every time the prediction of wellhead pressure value is completed n-1 All wellhead pressure predicted values from moment to current moment t
And all the obtained actual values of the wellhead pressure
The average absolute error value δ, δ is calculated as shown in equation (4):
in the above formula, i is an independent variable, y i And
and respectively representing the real value of the wellhead pressure and the predicted value of the wellhead pressure at the moment i.
Dynamically adding a base learner: if the average absolute error value delta of the multi-time-step prediction data of the current OSE-MS model is larger than the prediction loss threshold value T h And time period [ t ] n-1 ,t]When the quantity of the fracturing operation data is larger than the minimum sample number M of the model training, the dynamic data flow acquisition method is triggered by the following steps: first for the time period t n-1 +1,t]Performing characteristic engineering data conversion on the fracturing operation data to obtain a new data subset, performing online training by using the data subset to obtain a new base learner, adding the new base learner into a base learner integrated SE (selected element), and finally performing element learner MM (selected element) by using the base learner integrated SE; and setting a variable delta equal to the average absolute error delta of the current OSE-MS model for subsequent judgment of whether the base learner needs to be deleted dynamically.
Dynamic delete base learner: in order to limit the scale of the OSE-MS model and effectively improve the overall prediction performance, the weak base learner is also required to be deleted in the dynamic integration process: if the average absolute error value delta after the new base learner is larger than the average absolute error value delta before the new base learner is addedWhen the error value delta or the number of the base learners is larger than the maximum number N of the base learners, intercepting a time period t-N verify ,t]The average absolute error value delta of each base learner in the verification set is calculated, the base learner (the base learner is the weak base learner) corresponding to the calculated maximum average absolute error value is deleted, and the meta-learner MM is retrained. The average absolute error δ of the dynamic erasure basis learner stage is shown as:
in the above formula, N verify Is the total number of data, i is an argument, y i And
and respectively representing the real value of the wellhead pressure and the predicted value of the wellhead pressure at the moment i.
Online prediction: the OSE-MS model converts fracturing operation data acquired in real time into multi-time step data by utilizing characteristic engineering, then obtains intermediate prediction data through a base learner integrated SE, averages the intermediate prediction data, inputs the average intermediate prediction data into a meta-learner MM, and finally predicts the average intermediate prediction data by the meta-learner MM to obtain multi-time step prediction data, namely a plurality of wellhead pressure prediction values.
The method is tested as follows:
1. description of data
The experimental data are fracture operation data, the data set is sampled in time sequence by taking the second as the unit, and comprises 139,800 fracture operation data examples and 8 characteristics, and the characteristics of the data set and the related description thereof are shown in the following table 2.
Table 2 fracturing data description
| Characteristic name | Unit of | Maximum value | Minimum value | Mean value of | Standard deviation of |
| Physical section of fracture | - | - | - | - | - |
| Phases | - | - | - | - | - |
| Casing pressure | MPa | 19.73 | 3.98 | 9.64 | 2.24 |
| Discharge capacity | m 3 /min | 12.54 | 2.04 | 11.05 | 0.98 |
| Amount of stage discharge liquid | m 3 | 262.34 | 0.04 | 22.05 | 20.05 |
| Cumulative amount of discharged liquid | m 3 | 2326.29 | 94.29 | 1058.24 | 545.64 |
| Concentration of sand | Kg/m 3 | 190.80 | 0.00 | 41.58 | 50.91 |
| Pressure at well head | MPa | 88.74 | 57.41 | 75.86 | 4.09 |
2. Experimental setup
The experimental hardware environment of the test is Intel I9-9900 CPU and NVIDIAGeforce 2080s display card.
In order to evaluate the performance of the multi-step prediction of the method, the method is compared with LSTM, GRU, a Multilayer Perceptron (MLP for short), random forest (RF for short) and Bagging algorithm (Bagging for short) in an ensemble learning method in an experiment. The neural network model is realized through a Keras framework, and the integrated learning method is realized based on a Scikit-Learn package.
In the experiment, the base learner units LSTM and GRU are both set to be simple two-layer structures, and each layer comprises 200 neurons; adjusting model parameters by using a random gradient descent method of an ADAM optimizer; through experimental analysis, a loss threshold T is set h When the value is 0.7, a more stable result can be obtained; meanwhile, in order to enable the base learner to have good prediction performance, the lower limit M of the minimum sample number of model training is set to be 500; and the maximum number of the base learners is set to be 8 by combining the opinions of field experts, so that the memory occupation requirement of field operation can be met. The parameter settings for the LSTM and GRU methods in the comparison method are the same as those in the OSE-MS model. MLP employs an ADAM optimizer and the activation function is the Relu function. The number of the base learners in the random forest and Bagging algorithms is set to be 100.
The prediction performance is measured by three evaluation indexes, namely, average Absolute Error (MAE), root Mean Square Error (RMSE) and average Absolute Percentage Error (MAPE), wherein the definitions of the evaluation indexes are respectively shown in the following formulas (6) to (8):
wherein n is the total number of wellhead pressure prediction data, y t And
respectively represented as the true value and predicted value at time t.
3. Results and analysis of the experiments
The method is used for testing and analyzing the multi-step prediction effect under the fracturing operation scene, and comprises multi-step prediction performance analysis, stability analysis, time efficiency analysis and parameter experiments.
3.1 prediction of Performance
Single step prediction: at present, many models can only realize single-step prediction, and in order to illustrate the effectiveness of the invention, the performance of single-step prediction on fracture data of an OSE-MS model and other regression models is firstly compared. The results of the experiment are shown in table 3.
TABLE 3 Single step prediction Performance comparison
| Model | MAE | RMSE | MAPE/％ |
| LSTM | 0.907 | 1.125 | 1.208 |
| GRU | 1.271 | 1.487 | 1.726 |
| MLP | 2.177 | 2.688 | 2.924 |
| RF | 3.955 | 4.962 | 5.032 |
| Bagging | 3.982 | 4.961 | 5.015 |
| OSE-MS | 0.603 | 0.883 | 0.823 |
The results show that the deep learning model has better prediction accuracy than the integrated learning model, and the OSE-MS model integrating LSTM and GRU has significant performance improvement compared with other models, namely the performance improvement of MAE, RMSE and MAPE is improved by 33.5%, 21.5% and 31.9%, respectively. It is fully demonstrated that the OSE-MS model still has good prediction effect in the single-step prediction.
Multi-step prediction: a series of predicted steps are designed for comparison in the experiment, wherein the step interval is [2-20] steps, and the step interval is 2. Three experiments were performed for each prediction step for a total of 30 experiments. Table 4 below shows the average performance comparison of the multi-step predictions of the present invention and the baseline method.
TABLE 4 comparison of average Performance for Multi-step prediction
| Model | MAE | RMSE | MAPE/％ |
| LSTM | 1.944 | 2.367 | 2.630 |
| GRU | 1.819 | 2.158 | 2.487 |
| MLP | 3.793 | 4.723 | 5.013 |
| RF | 2.711 | 3.884 | 3.457 |
| Bagging | 2.712 | 3.884 | 3.776 |
| OSE-MS | 0.657 | 0.950 | 0.881 |
From the above table, it can be seen that GRUs in the deep learning model perform better than LSTM in the multi-step prediction task. Compared with a GRU model, the invention respectively improves the MAE index, the RMSE index and the MAPE index by 63.8 percent, 56 percent and 64.6 percent, and the improvement range is greatly increased compared with that in a single-step prediction task. Therefore, the method integrates the deep learning model by an online Stacking method, combines the advantages of LSTM and GRU in joint modeling of the prediction output window, and can adapt to the distribution change of the data stream in real time.
In order to compare the prediction performance of the method with that of the prior art more intuitively, fig. 1 and 2 show the prediction curves of [0,4000] time interval and [14000,18000] time interval in the wellhead pressure data flow test data respectively, and the experiment only uses 2000 pieces of offline data as training data, so as to obtain an initial OSE-MS model. Compared with other methods, the method has the advantage that the predicted value and the true value of the wellhead pressure are kept with the minimum prediction error all the time.
3.2 multistep prediction of stability
In order to analyze the stability of the OSE-MS method, the section expands the range of the prediction steps, statistical analysis is carried out in the interval of [2-30], the step length is set to be 2, and three experiments are carried out in each prediction step for 45 experiments. Fig. 5 to 7 show the index changes of MAE, RMSE, MAPE, respectively. As can be seen from the figure, the fluctuation of the prediction performance of the MLP is most obvious, which shows that the correlation between the prediction effect of the MLP and the prediction step number is not great. With the increase of the prediction steps, the fluctuation of the prediction performance of the random forest and Bagging algorithm is minimum, which shows that the integrated learning has extremely strong stability in multi-step prediction, but the prediction accuracy is far lower than that of a deep learning model, and the integrated learning is not suitable for scenes with high accuracy requirements. LSTM and GRU have smaller fluctuation range, and the prediction precision is in a fluctuation descending trend along with the increase of the step number, thereby conforming to the general rule of multi-step prediction. The invention combines the characteristics of high prediction precision and strong stability of an integrated model of a deep learning model, integrates an LSTM and a GRU base learner in a Stacking integrated mode, and always keeps the optimal prediction precision in the experiment; when the prediction step number is in the range of [2-20], the stability of the OSE-MS model is strong, after 20 steps are exceeded, the prediction precision is changed greatly, but the overall stability is only secondary to ensemble learning, and the effectiveness and the reliability of the OSE-MS model are fully verified.
3.3 time efficiency
In order to evaluate the effectiveness of the invention, the experiment in this section counts the average unit training time of the newly added base learner integrated with SE of the base learner and the average times of the newly added base learner on line within the range of the predicted step number of [2-30 ].
In the actual fracturing operation, the fracturing operation time of one well is usually about 1-2 days, while in the method, the average training time of every hundred data of the base learner is about 0.044 second when the base learner is trained on line, and the average number of times of adding the base learner on line is about 9.33 times in the whole operation process. The method can fully meet the timeliness requirement of fracturing operation and has higher practical application value.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (10)
1. A wellhead pressure online multi-step prediction method based on Stacking ensemble learning is characterized by comprising the steps of constructing an online Stacking ensemble multi-step prediction model, obtaining fracturing operation data in real time, inputting the fracturing operation data into the online Stacking ensemble multi-step prediction model, outputting multi-time-step prediction data based on the online Stacking ensemble multi-step prediction model, and updating the online Stacking ensemble multi-step prediction model by using the fracturing operation data.
2. The wellhead pressure online multi-step prediction method based on the Stacking ensemble learning as claimed in claim 1, wherein the fracturing operation data comprises wellhead pressure values and relevant factor data sets thereof, and the relevant factor data sets comprise characteristic fracturing physical sections, stages, casing pressure, discharge displacement, stage discharge displacement, accumulated discharge displacement and sand concentration.
3. The wellhead pressure online multi-step prediction method based on Stacking ensemble learning as claimed in claim 1, wherein the building of the online Stacking ensemble multi-step prediction model comprises the following steps:
step 1, initialization setting: initializing parameters, training an initial basis learner based on offline data, wherein the parameters comprise a predictive loss threshold T h Minimum sample number M of model training, size w of prediction output window and number N of verification sets verify ，T h 、M、w、N verify Are all constant; the off-line data is characteristic data obtained by converting real fracturing operation data in a time period in the fracturing operation process through characteristic engineering data;
and 2, step: and (3) online training: based on an online training mechanism, dynamically constructing a base learner integrated SE and a meta learner MM in an online mode, and specifically operating as follows:
step 2.1, the base learner learns online: adding an initial base learner into a base learner integrated SE, continuously delaying and verifying the prediction effect of the online Stacking integrated multi-step prediction model, namely after the wellhead pressure value prediction is finished each time, if the prediction error of the current online Stacking integrated multi-step prediction model is larger than the prediction loss threshold T h If so, acquiring a data subset based on a data flow dynamic acquisition method, training a new base learner on line by using the data subset, and adding the new base learner into the base learner integration SE;
step 2.2, online learning of the meta learner: all data subsets are used as a verification set of a base learner integrated SE to obtain intermediate prediction data; converting the obtained wellhead pressure true value by using a characteristic project to obtain a true data label, splicing the intermediate prediction data and the true data label, and constructing the on-line training data of the MM of the meta-learner;
step 2.3, dynamic integration: judging whether a base learner is added or not based on the precision real-time evaluation method and the quantity of fracturing operation data which are not subjected to on-line training in the fracturing operation data stream; and judging whether the base learner integrated with the base learner in the SE needs to be deleted or not based on the prediction errors before and after the addition of the base learner.
4. The wellhead pressure online multi-step prediction method based on Stacking ensemble learning as claimed in claim 3, characterized in that the method for feature engineering data conversion is as follows: and based on the size w of the prediction output window, dividing fracturing operation data through a sliding time window, and converting the divided fracturing operation data into multi-time step data by utilizing characteristic engineering.
5. The wellhead pressure online multi-step prediction method based on Stacking ensemble learning as claimed in claim 3, wherein in the step 2.1, the dynamic data flow acquisition method comprises: and (3) carrying out characteristic engineering data conversion on a plurality of fracturing operation data (the number of the fracturing operation data is more than the minimum sample number M of model training) according to the time sequence of the fracturing operation data which is not trained online to obtain a data subset.
6. The wellhead pressure online multi-step prediction method based on Stacking ensemble learning as claimed in claim 3, wherein in the step 2.3, the precision real-time assessment method comprises: each time the prediction of the wellhead pressure value is completed, the acquisition time point (marked as t) of the last data subset is calculated n-1 ) Until the current moment (marked as t), all wellhead pressure predicted values and all acquired wellhead pressure true values are predicted to have errors;
the method for judging whether the base learner is added or not comprises the following steps: if the prediction error of the multi-time-step prediction data of the current online Stacking integrated multi-step prediction model is larger than the prediction loss threshold T h And time period [ t ] n-1 +1,t]When the number of the fracturing operation data in the new base learner is larger than the minimum sample number M of the model training of the data subset, the new base learner based on the data flow dynamic acquisition method comprises the following steps: first for a time period t n-1 +1,t]Performing characteristic engineering data conversion on the fracturing operation data to obtain a new data subset, performing online training by using the data subset to obtain a new base learner, adding the new base learner into a base learner integrated SE (selected element) and finally performing element learner MM (selected element) by using the base learner integrated SE;
the method for judging whether the base learner is deleted comprises the following steps: if the prediction error after the new base learner is larger than the prediction error before the new base learner, the base learner needs to be deleted: intercepting a time period [ t-N verify ,t]The fracturing operation data is used as a verification set, the prediction error of each base learner in the verification set is calculated, the base learner corresponding to the maximum calculated prediction error is deleted, the meta-learner MM is retrained, and the deleted base learner is the weak base learner.
7. The wellhead pressure online multi-step prediction method based on Stacking ensemble learning as claimed in claim 3, wherein the method for outputting multi-time-step prediction data based on the online Stacking ensemble multi-step prediction model comprises: the online Stacking integrated multi-step prediction model converts fracturing operation data acquired in real time into multi-time-step data by using feature engineering, then obtains intermediate prediction data based on a base learner integrated SE (selected element), averages the intermediate prediction data, inputs the average intermediate prediction data into a meta-learner MM, and obtains the multi-time-step prediction data through the meta-learner MM prediction.
8. The wellhead pressure online multi-step prediction method based on the Stacking ensemble learning as claimed in claim 3 or 6, wherein the calculation method of the prediction error adopts average absolute error, root mean square error or average absolute percentage error.
9. The wellhead pressure online multi-step prediction method based on Stacking ensemble learning as claimed in claim 3, wherein in the step 1, a maximum number N of base learners is initialized, wherein N is a constant for limiting the number of base learners in the base learner ensemble SE, and the value range of the number of base learners in the base learner ensemble SE is [0, N ].
10. The wellhead pressure online multi-step prediction method based on Stacking ensemble learning according to claim 9, wherein the judgment condition for judging whether the weak-based learner needs to be deleted is as follows: and when the condition is met, the weak base learners in the SE integrated by the base learners need to be deleted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210722350.3A CN115222065A (en) | 2022-06-20 | 2022-06-20 | Wellhead pressure online multi-step prediction method based on Stacking ensemble learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210722350.3A CN115222065A (en) | 2022-06-20 | 2022-06-20 | Wellhead pressure online multi-step prediction method based on Stacking ensemble learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115222065A true CN115222065A (en) | 2022-10-21 |
Family
ID=83610593
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210722350.3A Pending CN115222065A (en) | 2022-06-20 | 2022-06-20 | Wellhead pressure online multi-step prediction method based on Stacking ensemble learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115222065A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116303626A (en) * | 2023-05-18 | 2023-06-23 | 西南石油大学 | Well cementation pump pressure prediction method based on feature optimization and online learning |
| CN117349610A (en) * | 2023-12-04 | 2024-01-05 | 西南石油大学 | Fracturing operation multi-time-step pressure prediction method based on time sequence model |
-
2022
- 2022-06-20 CN CN202210722350.3A patent/CN115222065A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116303626A (en) * | 2023-05-18 | 2023-06-23 | 西南石油大学 | Well cementation pump pressure prediction method based on feature optimization and online learning |
| CN116303626B (en) * | 2023-05-18 | 2023-08-04 | 西南石油大学 | Well cementation pump pressure prediction method based on feature optimization and online learning |
| CN117349610A (en) * | 2023-12-04 | 2024-01-05 | 西南石油大学 | Fracturing operation multi-time-step pressure prediction method based on time sequence model |
| CN117349610B (en) * | 2023-12-04 | 2024-02-09 | 西南石油大学 | Fracturing operation multi-time-step pressure prediction method based on time sequence model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Guo et al. | Remaining useful life prediction for rolling bearings using EMD-RISI-LSTM | |
| CN112101220B (en) | Rolling bearing service life prediction method based on unsupervised model parameter migration | |
| CN115222065A (en) | Wellhead pressure online multi-step prediction method based on Stacking ensemble learning | |
| CN109766583A (en) | Based on no label, unbalanced, initial value uncertain data aero-engine service life prediction technique | |
| CN111985610B (en) | Oil pumping well pump efficiency prediction system and method based on time sequence data | |
| CN109583565A (en) | Forecasting Flood method based on the long memory network in short-term of attention model | |
| CN114325395B (en) | Battery state determining method and device | |
| CN111325403B (en) | Method for predicting residual life of electromechanical equipment of highway tunnel | |
| CN113487855B (en) | Traffic flow prediction method based on EMD-GAN neural network structure | |
| Kocoglu et al. | Application of Bayesian optimized deep Bi-LSTM neural networks for production forecasting of gas wells in unconventional shale gas reservoirs | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN115760484A (en) | Method, device and system for improving potential danger identification capability of power distribution station area and storage medium | |
| CN112257914B (en) | Aviation safety causal prediction method based on random forest | |
| Ye et al. | A deep learning-based method for automatic abnormal data detection: Case study for bridge structural health monitoring | |
| CN116957331A (en) | Risk passenger flow range prediction method and device | |
| CN113837266B (en) | Software defect prediction method based on feature extraction and Stacking ensemble learning | |
| CN109669017A (en) | Refinery's distillation tower top based on deep learning cuts water concentration prediction technique | |
| CN113674807A (en) | Molecular screening method based on deep learning technology qualitative and quantitative model | |
| CN112200251A (en) | Monitoring system data evaluation method based on multi-scale convolutional neural network | |
| CN116663742A (en) | Regional capacity prediction method based on multi-factor and model fusion | |
| CN114463978B (en) | Data monitoring method based on track traffic information processing terminal | |
| CN113762791B (en) | Railway engineering cost management system | |
| CN115906639A (en) | Line operation fault rate prediction method and device based on line operation working conditions | |
| CN114372640A (en) | Wind power prediction method based on fluctuation sequence classification correction | |
| Zhang et al. | Data regression framework for time series data with extreme events |
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 |