CN111461551B - Deep learning and SPC criterion-based electric submersible pump fault early warning method - Google Patents
Deep learning and SPC criterion-based electric submersible pump fault early warning method Download PDFInfo
- Publication number
- CN111461551B CN111461551B CN202010252734.4A CN202010252734A CN111461551B CN 111461551 B CN111461551 B CN 111461551B CN 202010252734 A CN202010252734 A CN 202010252734A CN 111461551 B CN111461551 B CN 111461551B
- Authority
- CN
- China
- Prior art keywords
- submersible pump
- electric submersible
- value
- parameter
- data
- 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 60
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 238000012544 monitoring process Methods 0.000 claims abstract description 65
- 230000036541 health Effects 0.000 claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000012216 screening Methods 0.000 claims abstract description 15
- 230000002159 abnormal effect Effects 0.000 claims abstract description 12
- 230000007246 mechanism Effects 0.000 claims abstract description 11
- 230000015654 memory Effects 0.000 claims description 43
- 210000004027 cell Anatomy 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000004140 cleaning Methods 0.000 claims description 7
- 210000002569 neuron Anatomy 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 5
- 238000009826 distribution Methods 0.000 claims description 5
- 230000009471 action Effects 0.000 claims description 4
- 238000005065 mining Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000001105 regulatory effect Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000003672 processing method Methods 0.000 claims description 2
- 238000011425 standardization method Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 abstract description 5
- 230000005856 abnormality Effects 0.000 abstract description 2
- 238000013527 convolutional neural network Methods 0.000 description 10
- 238000004519 manufacturing process Methods 0.000 description 7
- 230000008901 benefit Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000001960 triggered effect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000003129 oil well Substances 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Educational Administration (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Control Of Non-Positive-Displacement Pumps (AREA)
Abstract
An electric submersible pump fault early warning method based on deep learning and SPC criteria comprises the following steps: acquiring a data set; storing a standardized model of the training set; constructing a CNN-LSTM model; before the model is applied, screening monitoring parameters of the electric submersible pump to be evaluated, and standardizing; estimating the monitoring parameter value of the electric submersible pump to be evaluated by applying a CNN-LSTM training model; calculating the gap between the current state and the normal state of the monitoring parameters of the electric submersible pump; calculating the health degree of the electric submersible pump; and judging whether the operation of the electric submersible pump is abnormal according to the health degree of the electric submersible pump, so as to trigger an alarm mechanism. The health degree is calculated on the basis of comprehensively considering the real-time residual errors of a plurality of parameters, and a simple mode of evaluating the health condition of the electric submersible pump by only checking the parameter abnormality of a voucher is avoided; the health degree extraction process of the method does not need any expert experience, does not need to manually set up labels, greatly reduces manual participation, and saves a large amount of manpower.
Description
Technical Field
The invention relates to an electric submersible pump fault early warning method, in particular to an electric submersible pump fault early warning method based on deep learning and SPC criteria.
Background
The electric submersible pump is oil extraction equipment of the offshore drilling platform, the performance of the electric submersible pump is the guarantee of the productivity of an oil well, the electric submersible pump needs to be maintained or replaced in time when abnormal or fault occurs, the operation and maintenance efficiency of enterprises is improved, the fault rate is reduced, the well laying time is shortened, the production cost and the time cost are saved, the running state of the electric submersible pump needs to be detected in real time, the abnormal state is found in time, and the electric submersible pump is effectively early-warned before the fault occurs. The current mature fault monitoring method of the electric submersible pump mainly comprises two types of methods:
model based on electric submersible pump operation mechanism: the method is tightly combined with a control theory, a mathematical model is established through an operation mechanism of the electric submersible pump to predict the output of oil, gas and the like, and the output is compared with an actual measured value to obtain a residual error; analyzing the residual error to determine whether the process fails, further identifying the type of failure, wherein the method has the advantages of: the fault early warning is carried out by analyzing residual errors by combining physical knowledge and parameter monitoring, so that the fault early warning is easy to understand; disadvantages: most of the mechanism models are simplified linear systems; in the actual industrial process, a system with nonlinear, higher degree of freedom and multivariable coupling is often adopted; the use effect is not ideal.
A method based on expertise: the method is based on human experience knowledge, fault characteristics are deduced, namely after the electric submersible pump pumps out of faults, an expert discovers problems through a planning device, connection relations among elements, fault propagation modes and the like in the oil extraction process of the electric submersible pump are qualitatively or quantitatively described by combining historical monitoring parameter change conditions before the faults, qualitative or quantitative characteristics of the faults of the electric submersible pump are summarized, and early warning and monitoring of the faults of the electric submersible pump are completed through the characteristics; wherein, the advantage: when the monitored object is simpler, the process knowledge and the production experience are more sufficient, and the use effect is better; disadvantages: the early warning accuracy has strong dependence on the richness of expert knowledge and the level of expert knowledge; many experiences are difficult to describe in a reasonably formalized expression.
Disclosure of Invention
In order to solve the defects of the technology, the invention provides an electric submersible pump fault early warning method based on deep learning and SPC criteria.
In order to solve the technical problems, the invention adopts the following technical scheme: an electric submersible pump fault early warning method based on deep learning and SPC criteria comprises the following steps:
step one, acquiring a data set formed by all operation monitoring parameters of the electric submersible pump, analyzing the operation mechanism and the data missing condition of the electric submersible pump, determining the monitoring parameters capable of reflecting the performance of the electric submersible pump, and storing the names of the parameter items;
intercepting data of part of monitoring parameters from historical operation data of the electric submersible pump as a training set; performing parameter screening, data cleaning and parameter standardization processing on the training set, and storing a standardized model of the training set; in the parameter screening process, the selection of the monitoring parameters is obtained from the data items stored in the first step; in the data cleaning process, only data in a normal operation period are reserved, and data in a fault period are removed; the data are in accordance with standard normal distribution by the standardization process;
thirdly, constructing a CNN-LSTM model, taking each monitoring parameter value of the electric submersible pump in the training set as a target of the model, taking other monitoring parameter values as input of the model, performing supervised training, and storing a training result model;
step four, screening monitoring parameters of the electric submersible pump to be evaluated before the model is applied, and standardizing parameter values; the method comprises the steps of obtaining monitoring data of a current electric submersible pump to be evaluated, screening monitoring parameter items of the pump, and standardizing a screening result set of the pump, wherein the two steps of treatment are carried out according to parameter item lists and standardized models which are stored in the first step and the second step;
step five, according to the step four, a CNN-LSTM training model is applied, and the monitoring parameter value of the electric submersible pump to be evaluated is estimated;
step six, calculating the difference between the current state and the normal state of the monitoring parameters of the electric submersible pump; the calculation of the gap between the current state and the normal state of the monitoring parameter is shown in the formula (1):
wherein y is the observation value of the monitoring parameter, y' is the estimated value of the monitoring parameter, and r is the difference of the monitoring parameter;
step seven, calculating the health degree of the electric submersible pump according to the difference between the current state and the normal state of the electric submersible pump in the step six; the health calculation is shown in formula (2):
wherein ,
for monitoring the mean value of the parameter differences, i.e. +.>
S is the health degree of the electric submersible pump;
and step eight, judging whether the operation of the electric submersible pump is abnormal according to the health degree of the electric submersible pump obtained in the step seven, so as to trigger an alarm mechanism.
Further, the processing method of parameter standardization in the second step is a Z-score standardization method, wherein the variance of the new data set is 1, the mean value is 0, and the processed data accords with standard normal distribution.
Further, the method for constructing the CNN-LSTM model in the third step comprises the following steps:
a. setting the number of layers of CNN, the number of filters and the size parameters of convolution size, taking N-1 of N effective monitoring parameters of the electric submersible pump as CNN input, traversing the whole input data sequence by using a convolution layer, an activation function and a pooling layer in the CNN, extracting local information of the N-1 monitoring parameters, and mining deep features;
b. and c, performing supervised training on the deep features obtained in the step a through an LSTM network, and storing a result model.
Further, the operation of the convolution layer in the step a is as shown in the formula (3):
wherein ,
the j' th weight of the ith convolution kernel of the first layer, +.>
For the j-th convolved local region in the first layer, V is the width of the convolution kernel.
Further, the operation of the activation function in step a is as shown in formulas (4) to (6):
a l(i,j) ＝f(y l(i,j) )＝max{0,y l(i,j) equation (6)
wherein ,al(i,j) To activate value, y l(i,j) Is the convolutional layer output value, where l (i, j) is the j-th convolved local region of the i-th convolutional kernel of the first layer.
Further, the calculation of the pooling layer is shown in formulas (7) and (8):
wherein ,al(i,t) The activation value, p, of the output of the t-th neuron for the ith feature map of the first layer l(i,j) And pooling the jth neuron of the ith feature map of the first layer to output a feature value, wherein W is the width of a pooling window.
Further, the computing update status of the LSTM network is divided into the following steps:
i, temporary memory status information c t The method comprises the steps of carrying out a first treatment on the surface of the In the refresh memory cell c t Before, a temporary memory cell c is generated t The method comprises the steps of carrying out a first treatment on the surface of the And c t The input of the current moment t and the hidden layer unit output of the last moment t-1 are combined together and are respectively combined with the respective weight matrix linearly to obtain candidate memory unit values of the current moment, and the state information of the memory units is updated, as shown in a formula (9):
c t ＝tanh(W xc x t +W hc h t-1 +b c ) Formula (9)
II, calculating input threshold value i t The method comprises the steps of carrying out a first treatment on the surface of the The current data information is selectively stored into the memory unit through the input gate, so that the state value of the current memory unit is influenced; as shown in formula r:
i t ＝σ(W xi x t +W hi h t-1 +b i ) Equation c
III, calculating the value f of the forgetting door t The method comprises the steps of carrying out a first treatment on the surface of the The forgetting gate mainly processes which information in the memory unit needs to be discarded; such as formula
The following is shown:
IV, calculating the state value c of the memory unit at the current moment t Such as formula
The following is shown:
in the formula ,
representing a point-wise product; it can be seen that the memory cell state is updated from the cell value c at the previous time t-1 And temporary memory status information c t The information is selected and regulated by the combined action of the forgetting door and the input door;
v, calculating output gate o t The method comprises the steps of carrying out a first treatment on the surface of the The output gate mainly acts on the output of the memory cell state value; such as formula
The following is shown:
VI, LSTM unit memory output h t Such as formula
The following is shown:
wherein ：Wxc 、W xi 、W xf 、W xo Output layer x at time t respectively t And hidden layer h t Connection weight between W hc 、W hi 、W hf 、W ho The hidden layer connection weight value between the time t-1 and the time t is b c 、b i 、b f 、b o Bias of input node, input gate, forget gate, output gate, h t-1 For the previous output, σ is a sigmoid function, and the value is (0, 1).
Further, the evaluation method for determining whether the operation of the electric submersible pump is abnormal in the step eight is to fit the health degree variation trend by using the SPC criterion or using a polynomial.
The invention has the beneficial effects that: the invention considers LSTM good sequence structure analysis, CNN good characteristic extraction and transformation, so a deep learning model based on CNN-LSTM network is provided, the model can quantify the health state of the electric submersible pump, namely the health degree estimation, and judges the change trend of the health degree by combining with SPC criterion, and then the electric submersible pump health state is pre-warned, and the invention has the advantages that:
the health degree of the method is calculated on the basis of comprehensively considering a plurality of parameter real-time residual errors, and a simple mode of evaluating the health condition of the electric submersible pump by only using abnormal voucher parameters is avoided;
the health degree extraction process of the method does not need any expert experience, does not need to manually set up labels, greatly reduces manual participation, and saves a large amount of manpower.
Drawings
Fig. 1 is a logic flow diagram of a technical solution of the present invention.
FIG. 2 is a schematic diagram of the structure of the CNN-LSTM model.
FIG. 3 is a schematic diagram of LSTM memory cell structure.
Detailed Description
The present invention will be described in further detail with reference to the following embodiments.
An electric submersible pump fault early warning method based on deep learning and SPC criteria as shown in fig. 1 comprises the following steps:
step one: and collecting the original data of the submersible pump, analyzing the operation mechanism and the data missing condition of the submersible pump, determining monitoring parameters capable of reflecting the performance of the submersible pump, and storing the names of the parameter items. The specific contents are as follows: and acquiring a data set formed by all electric submersible pump operation monitoring parameters, analyzing an electric submersible pump operation mechanism in the data set, selecting the monitoring parameters capable of reflecting the electric submersible pump performance, and storing the names of the parameter items for a training set.
Step two: intercepting part of monitoring parameter data from historical operation data of the electric submersible pump as a training set, performing operations such as data cleaning, standardization and the like on the training set, and storing a standardized model. The specific contents are as follows: performing operations such as parameter screening, data cleaning, parameter standardization and the like on the training set, and storing a standardized model of the training set, wherein in the parameter screening process, the selection of monitoring parameters is obtained from the data items stored in the step one; in the data cleaning process, only the data in the normal operation period are reserved, and the data in the fault period need to be removed; the normalization process uses the method Z-score, i.e., applies the formula x= (X- μ)/σ such that the new X dataset variance is 1 and the mean is 0, so that the processed data will fit the standard normal distribution.
Step three: and constructing a CNN-LSTM model, taking each monitoring parameter value of the electric submersible pump in the training set as a target of the model, taking other monitoring parameter values such as the monitoring parameter values outside the training set as input of the model, performing supervised training, and storing a training result model, wherein the complete CNN-LSTM model structure is shown in figure 2.
Suppose a training set is shaped like D trc ＝{X t ，y t } T, wherein yt ∈R 1 Characteristic of a monitored parameter p at time t, X t ∈R N-1 Representing an N-1 dimensional feature outside the parameter p at time t, the specific modeling process is as follows:
step a: setting parameters such as the number of CNN layers, the number of filters, the size of convolution dimensions and the like, taking N-1 of N effective monitoring parameters of the electric submersible pump as CNN input, traversing the whole input data sequence by using formulas (3) to (8) of convolution layers, activation functions and pooling layers in the CNN, extracting local information of the N-1 monitoring parameters, and mining deep features. The convolutional neural network layer (CNN) may attempt to be replaced with a deep learning layer (DNN), but the advantages of CNN are apparent, it can capture local features (convolutional layer) and global features (pooling layer) of data, and share convolutional kernels, with no stress on high-dimensional data processing.
Convolution layer operation:
in the formula ：
the j' th weight of the ith convolution kernel of the first layer, +.>
For the j-th convolved local region in the first layer, V is the width of the convolution kernel.
Activation function:
a l(i,j) ＝f(y l(i,j) )＝max{0,y l(i,j) equation (6)
in the formula ：al(i,j) To activate value, y l(i,j) Is the convolutional layer output value, where l (i, j) is the j-th convolved local region of the i-th convolutional kernel of the first layer.
Pooling function:
in the formula ：al(i,t) The activation value, p, of the output of the t-th neuron for the ith feature map of the first layer l(i,j) And pooling the jth neuron of the ith feature map of the first layer to output a feature value, wherein W is the width of a pooling window.
Step b: inputting the deep features obtained in the step a into an LSTM network, and applying a formula
And the LSTM memory unit has the advantage of long-term and short-term memory on the time series data, performs supervised training and stores a result model.
The LSTM network is modified from RNN, and includes three gate controllers with cell memory unit structures added to the hidden layer of RNN: the input gate i, the forgetting gate f and the output gate o can allow the network to forget the history information, and can also update the memory state by using the new information, so that the model has a certain ability of learning the long-term dependent information, and the gradient disappearance or explosion problem is effectively overcome. In the training and recognition process, the state value of the LSTM hidden layer depends on the state value of the hidden layer at the current input and the previous moment, and the process is continuously circulated until the input is completed. The effect of the new information on the neurons is controlled through the action of three gates, so that the LSTM network can store and transmit the information for a long time, and the sequence data can be effectively processed. The LSTM memory cell structure is shown in FIG. 3.
It can be seen from the figure that the three gates all use a sigmoid function and all are nonlinear summing units, while the activation functions for the inside and outside of the module are all included, with multiplication operations to control the activation functions of the units. The method comprises the following steps of:
i, temporary memory status information c t . In the refresh memory cell c t Before, temporary record is generatedMemristor c t . And c t The input of the current moment t and the hidden layer unit output of the last moment t-1 are combined together and are respectively combined with the weight matrixes linearly to obtain candidate memory unit values of the current moment, and the state information of the memory unit is updated.
c t ＝tanh(W xc x t +W hc h t-1 +b c ) Formula (9)
II, calculating input threshold value i t . The current data information is selectively stored into the memory cell through the input gate, thereby affecting the current memory cell state value.
i t ＝σ(W xi x t +W hi h t-1 +b i ) Equation c
III, calculating the value f of the forgetting door t . The forget gate mainly processes which information in the memory unit needs to be discarded.
IV, calculating the state value c of the memory unit at the current moment t 。
in the formula ,
representing a point-wise product. It can be seen that the memory cell state update is mainly based on the cell value c at the previous time t-1 And temporary memory status information c t And the information is selected and regulated by the combined action of the forgetting gate and the input gate.
V, calculating output gate o t . The output gate mainly acts on the output of the memory cell state value.
VI, LSTM unit memory output h t 。
wherein ：Wxc 、W xi 、W xf 、W xo Output layer x at time t respectively t And hidden layer h t Connection weight between W hc 、W hi 、W hf 、W ho The hidden layer connection weight value between the time t-1 and the time t is b c 、b i 、b f 、b o Bias of input node, input gate, forget gate, output gate, h t-1 For the previous output, σ is a sigmoid function, and the value is (0, 1).
Step four: before the model is applied, monitoring parameters of the electric submersible pump to be evaluated are screened, and parameter values are standardized. The specific contents are as follows: and (3) acquiring monitoring data of the current electric submersible pump to be evaluated, screening monitoring parameter items of the pump, and standardizing a screening result set of the pump, wherein the two steps of treatment are carried out according to parameter item lists and standardized models stored in the step (1) and the step (2).
Step five: and estimating the monitoring parameter value of the electric submersible pump to be evaluated by using the CNN-LSTM model. The specific contents are as follows: according to step 4, a CNN-LSTM training model is applied to estimate the values of all monitoring parameters of the current electric submersible pump.
Step six: and calculating the difference between the current state and the normal state of each monitoring parameter of the electric submersible pump. The calculation formula of the difference between the current state and the normal state of the monitoring parameter is as follows:
wherein y is the observation value of the monitoring parameter, y' is the estimated value of the monitoring parameter, and r is the difference of the monitoring parameter.
Step seven: and D, calculating the health degree of the electric submersible pump according to the difference between the current state and the normal state of the electric submersible pump in the step six. The health degree calculation formula is as follows:
in the formula ,
for the mean value of the differences of the respective monitored parameters, i.e. +.>
S is the health degree of the electric submersible pump.
Step eight: referring to the eight major judgment criteria of the SPC criteria, SPC rules capable of monitoring the change of the health degree of the electric submersible pump in real time are defined as follows:
the SPC mainly monitors the production process in real time by using a statistical analysis technology, scientifically distinguishes random fluctuation and abnormal fluctuation of the product quality in the production process, and accordingly gives an early warning to the abnormal trend of the production process, so that production management staff can take measures in time, eliminate the abnormality, and recover the stability of the process, and the purposes of improving and controlling the quality are achieved. The eight major criteria for the SPC criteria are as follows:
rule 1:1 point falls outside three times the standard deviation from the centerline;
rule 2: the continuous 9 points fall on the same side of the central line;
rule 3: continuously increasing or decreasing at 6 points;
rule 4: adjacent points in the continuous 14 points are alternately up and down;
rule 5: 2 points in the continuous 3 points fall outside twice standard deviation of the same measurement of the central line;
rule 6: 4 points in the continuous 5 points fall outside one standard deviation of the same side of the central line;
rule 7: the 15 continuous points fall within one standard deviation of two sides of the central line;
rule 8: the continuous 8 points fall on both sides of the central line, and none of the continuous 8 points is within one standard deviation;
the electrical submersible pump health degree sequence calculated in the step seven can reflect the health condition of the electrical submersible pump, so that the applicable SPC criterion is screened and used for judging whether the operation of the electrical submersible pump is abnormal or not, and the specific selected rule is as follows:
rule 2. The values of 9 consecutive points in the sequence are triggered when they are below the centerline;
rule 3, triggering when 6 continuous points in the sequence steadily descend;
rule 4. The value of 2 points out of 3 consecutive points in the sequence is triggered when 2 standard deviations below the center line;
rule 5. The value of 4 points out of the consecutive 5 points in the sequence triggers when 1 standard deviation below the centerline;
rule 6, triggering when the values of the continuous 14 points in the sequence alternate;
rule 7. Values of consecutive 8 points in the sequence are on either side of the centerline and trigger when outside 1 standard deviation from the centerline.
And (3) monitoring the electrical submersible pump health degree sequence value calculated in the step seven in real time by applying the defined SPC criterion, and judging whether the operation of the electrical submersible pump is abnormal or not, thereby triggering an alarm mechanism.
In this step, the health condition of the electric submersible pump may not be evaluated by adopting the SPC criterion, and since the LSTM layer already considers the time characteristics of the sequence, the trend characteristics of the sequence may be represented in the output result, the health change trend may be fitted by adopting a polynomial, and then early warning is performed by a threshold value.
The above embodiments are not intended to limit the present invention, and the present invention is not limited to the above examples, but is also intended to be limited to the following claims.
Claims (8)
1. An electric submersible pump fault early warning method based on deep learning and SPC criteria is characterized in that: the method comprises the following steps:
step one, acquiring a data set formed by all operation monitoring parameters of the electric submersible pump, analyzing the operation mechanism and the data missing condition of the electric submersible pump, determining the monitoring parameters capable of reflecting the performance of the electric submersible pump, and storing the names of the parameter items;
intercepting data of part of monitoring parameters from historical operation data of the electric submersible pump as a training set; performing parameter screening, data cleaning and parameter standardization processing on the training set, and storing a standardized model of the training set; in the parameter screening process, the selection of the monitoring parameters is obtained from the data items stored in the first step; in the data cleaning process, only data in a normal operation period are reserved, and data in a fault period are removed; the data are in accordance with standard normal distribution by the standardization process;
thirdly, constructing a CNN-LSTM model, taking each monitoring parameter value of the electric submersible pump in the training set as a target of the model, taking other monitoring parameter values as input of the model, performing supervised training, and storing a training result model;
step four, screening monitoring parameters of the electric submersible pump to be evaluated before the model is applied, and standardizing parameter values; the method comprises the steps of obtaining monitoring data of a current electric submersible pump to be evaluated, screening monitoring parameter items of the pump, and standardizing a screening result set of the pump, wherein the two steps of treatment are carried out according to parameter item lists and standardized models which are stored in the first step and the second step;
step five, according to the step four, a CNN-LSTM training model is applied, and the monitoring parameter value of the electric submersible pump to be evaluated is estimated;
step six, calculating the difference between the current state and the normal state of the monitoring parameters of the electric submersible pump; the calculation of the gap between the current state and the normal state of the monitoring parameter is shown in the formula (1):
wherein y is the observation value of the monitoring parameter, y' is the estimated value of the monitoring parameter, and r is the difference of the monitoring parameter;
step seven, calculating the health degree of the electric submersible pump according to the difference between the current state and the normal state of the electric submersible pump in the step six; the health calculation is shown in formula (2):
wherein ,
for monitoring the mean value of the parameter differences, i.e. +.>
S is the health degree of the electric submersible pump;
and step eight, judging whether the operation of the electric submersible pump is abnormal according to the health degree of the electric submersible pump obtained in the step seven, so as to trigger an alarm mechanism.
2. The deep learning and SPC criteria based electric submersible pump fault warning method of claim 1, wherein: the parameter standardization processing method in the second step is a Z-score standardization method, wherein the variance of the new data set is 1, the mean value is 0, and the processed data accords with standard normal distribution.
3. The deep learning and SPC criteria based electric submersible pump fault warning method of claim 1, wherein: the method for constructing the CNN-LSTM model in the third step comprises the following steps:
a. setting the number of layers of CNN, the number of filters and the size parameters of convolution size, taking N-1 of N effective monitoring parameters of the electric submersible pump as CNN input, traversing the whole input data sequence by using a convolution layer, an activation function and a pooling layer in the CNN, extracting local information of the N-1 monitoring parameters, and mining deep features;
b. and c, performing supervised training on the deep features obtained in the step a through an LSTM network, and storing a result model.
4. The electric submersible pump fault warning method based on deep learning and SPC criteria of claim 3, wherein: the operation of the convolution layer in the step a is shown in the formula (3):
wherein ,
the j' th weight of the ith convolution kernel of the first layer, +.>
For the j-th convolved local region in the first layer, V is the width of the convolution kernel.
5. The method for early warning of electric submersible pump failure based on deep learning and SPC criteria of claim 4, wherein: the operation of the activation function in the step a is as shown in formulas (4) to (6):
a l(i，j) ＝f(y l(i，j) )＝max{0，y l(i，j) equation (6)
wherein ,al(i，j) To activate value, y l(i，j) Is the convolutional layer output value, where l (i, j) is the j-th convolved local region of the i-th convolutional kernel of the first layer.
6. The method for early warning of electric submersible pump failure based on deep learning and SPC criteria of claim 5, wherein: the calculation of the pooling layer is shown in formulas (7) and (8):
wherein ,al(i，t) The activation value, p, of the output of the t-th neuron for the ith feature map of the first layer l(i，j) And pooling the jth neuron of the ith feature map of the first layer to output a feature value, wherein W is the width of a pooling window.
7. The electric submersible pump fault warning method based on deep learning and SPC criteria of claim 3, wherein: the calculating and updating state of the LSTM network comprises the following steps:
i, temporary memory status information c t The method comprises the steps of carrying out a first treatment on the surface of the In the refresh memory cell c t Before, a temporary memory cell c is generated t The method comprises the steps of carrying out a first treatment on the surface of the And c t The input of the current moment t and the hidden layer unit output of the last moment t-1 are combined together and are respectively combined with the respective weight matrix linearly to obtain candidate memory unit values of the current moment, and the state information of the memory units is updated, as shown in a formula (9):
c t ＝tanh(W xc x t +W hc h t-1 +b c ) Formula (9)
II, calculating input threshold value i t The method comprises the steps of carrying out a first treatment on the surface of the The current data information is selectively stored into the memory unit through the input gate, so that the state value of the current memory unit is influenced; as shown in formula r:
i t ＝σ(W xi x t +W hi h t-1 +b i ) Equation c
III, calculating the value f of the forgetting door t The method comprises the steps of carrying out a first treatment on the surface of the The forgetting gate mainly processes which information in the memory unit needs to be discarded; such as formula
The following is shown:
IV, calculating the state value c of the memory unit at the current moment t Such as formula
The following is shown:
in the formula ,
representing a point-wise product; it can be seen that the memory cell state is updated from the cell value c at the previous time t-1 And temporary memory status information c t The information is selected and regulated by the combined action of the forgetting door and the input door;
v, calculating output gate o t The method comprises the steps of carrying out a first treatment on the surface of the The output gate mainly acts on the output of the memory cell state value; such as formula
The following is shown:
VI, LSTM unit memory output h t Such as formula
The following is shown:
wherein ：Wxc 、W xi 、W xf 、W xo Output layer x at time t respectively t And hidden layer h t Connection weight between W hc 、W hi 、W hf 、W ho The hidden layer connection weight value between the time t-1 and the time t is b c 、b i 、b f 、b o Bias of input node, input gate, forget gate, output gate, h t-1 For the previous output, σ is a sigmoid function, and the value is (0, 1).
8. The deep learning and SPC criteria based electric submersible pump fault warning method of claim 1, wherein: and in the step eight, judging whether the operation of the electric submersible pump is abnormal or not by adopting an SPC rule or adopting a polynomial fit health degree change trend.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010252734.4A CN111461551B (en) | 2020-04-01 | 2020-04-01 | Deep learning and SPC criterion-based electric submersible pump fault early warning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010252734.4A CN111461551B (en) | 2020-04-01 | 2020-04-01 | Deep learning and SPC criterion-based electric submersible pump fault early warning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111461551A CN111461551A (en) | 2020-07-28 |
| CN111461551B true CN111461551B (en) | 2023-05-02 |
Family
ID=71679338
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010252734.4A Active CN111461551B (en) | 2020-04-01 | 2020-04-01 | Deep learning and SPC criterion-based electric submersible pump fault early warning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111461551B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112132394B (en) * | 2020-08-21 | 2024-03-29 | 西安交通大学 | Power plant circulating water pump predictive state evaluation method and system |
| CN112506687B (en) * | 2020-11-24 | 2022-03-01 | 四川长虹电器股份有限公司 | Fault diagnosis method based on multi-period segmented sliding window standard deviation |
| CN112785091A (en) * | 2021-03-04 | 2021-05-11 | 湖北工业大学 | Method for performing fault prediction and health management on oil field electric submersible pump |
| CN113033663A (en) * | 2021-03-26 | 2021-06-25 | 同济大学 | Automatic container terminal equipment health prediction method based on machine learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015158198A1 (en) * | 2014-04-17 | 2015-10-22 | 北京泰乐德信息技术有限公司 | Fault recognition method and system based on neural network self-learning |
| CN110689075A (en) * | 2019-09-26 | 2020-01-14 | 北京工业大学 | Fault prediction method of self-adaptive threshold of refrigeration equipment based on multi-algorithm fusion |
| CN110763929A (en) * | 2019-08-08 | 2020-02-07 | 浙江大学 | Intelligent monitoring and early warning system and method for convertor station equipment |
-
2020
- 2020-04-01 CN CN202010252734.4A patent/CN111461551B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015158198A1 (en) * | 2014-04-17 | 2015-10-22 | 北京泰乐德信息技术有限公司 | Fault recognition method and system based on neural network self-learning |
| CN110763929A (en) * | 2019-08-08 | 2020-02-07 | 浙江大学 | Intelligent monitoring and early warning system and method for convertor station equipment |
| CN110689075A (en) * | 2019-09-26 | 2020-01-14 | 北京工业大学 | Fault prediction method of self-adaptive threshold of refrigeration equipment based on multi-algorithm fusion |
Non-Patent Citations (1)
| Title |
|---|
| 代杰杰等.采用LSTM 网络的电力变压器运行状态预测方法研究.《高电压技术》.2018,第44卷(第4期),1099-1106. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111461551A (en) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111461551B (en) | Deep learning and SPC criterion-based electric submersible pump fault early warning method | |
| CN109033450B (en) | Elevator equipment fault prediction method based on deep learning | |
| CN108197845B (en) | Transaction index abnormity monitoring method based on deep learning model LSTM | |
| CN112926273B (en) | Method for predicting residual life of multivariate degradation equipment | |
| CN110757510B (en) | Method and system for predicting remaining life of robot | |
| CN109471698B (en) | System and method for detecting abnormal behavior of virtual machine in cloud environment | |
| CN112434390B (en) | PCA-LSTM bearing residual life prediction method based on multi-layer grid search | |
| Skordilis et al. | A condition monitoring approach for real-time monitoring of degrading systems using Kalman filter and logistic regression | |
| US20240085274A1 (en) | Hybrid bearing fault prognosis with fault detection and multiple model fusion | |
| CN114841076A (en) | Power battery production process fluctuation abnormity detection method based on space-time diagram model | |
| CN114444582A (en) | Mechanical equipment fault diagnosis method based on convolutional neural network and Bayesian network | |
| Huang et al. | Bayesian neural network based method of remaining useful life prediction and uncertainty quantification for aircraft engine | |
| CN117273440A (en) | Engineering construction Internet of things monitoring and managing system and method based on deep learning | |
| CN115145899A (en) | Space-time data anomaly detection method based on data space of manufacturing enterprise | |
| CN117290726A (en) | CAE-BiLSTM-based fault early warning method for mobile equipment | |
| Lee et al. | In-line predictive monitoring framework | |
| KR102586845B1 (en) | Facility management system that enables preventive maintenance using deep learning | |
| KR102474332B1 (en) | Prediction method of machine health stability in smart factory system, and recording medium thereof | |
| CN113536671A (en) | Lithium battery life prediction method based on LSTM | |
| Qin et al. | Remaining useful life prediction using temporal deep degradation network for complex machinery with attention-based feature extraction | |
| CN117725548A (en) | Abnormality detection method and device, electronic equipment and storage medium | |
| CN117010683A (en) | Operation safety risk prediction method based on hybrid neural network and multiple agents | |
| Karagiorgou et al. | Unveiling trends and predictions in digital factories | |
| Dui et al. | Reliability Evaluation and Prediction Method with Small Samples. | |
| CN114486262B (en) | CNN-AT-LSTM-based bearing residual service life prediction method |
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 |