CN112149896A - Attention mechanism-based mechanical equipment multi-working-condition fault prediction method - Google Patents

Attention mechanism-based mechanical equipment multi-working-condition fault prediction method Download PDF

Info

Publication number
CN112149896A
CN112149896A CN202010986026.3A CN202010986026A CN112149896A CN 112149896 A CN112149896 A CN 112149896A CN 202010986026 A CN202010986026 A CN 202010986026A CN 112149896 A CN112149896 A CN 112149896A
Authority
CN
China
Prior art keywords
data
vector
layer
network
attention
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.)
Withdrawn
Application number
CN202010986026.3A
Other languages
Chinese (zh)
Inventor
孙雁飞
张及棠
亓晋
许斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202010986026.3A priority Critical patent/CN112149896A/en
Publication of CN112149896A publication Critical patent/CN112149896A/en
Withdrawn legal-status Critical Current

Links

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
    • G06Q10/04—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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/045—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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a mechanical equipment multi-working-condition fault prediction method based on an attention mechanism, which comprises the following steps of: step 1) preprocessing data, vectorizing and normalizing original data, and dividing by using a sliding time window; step 2) embedding the preprocessed data into a long-term and short-term memory network layer, obtaining a data long-range dependency relationship through training, and calculating a hidden vector; step 3) inputting the hidden vector in the step 2) into the attention layer; step 4), in the network training process, optimizing network parameters by using a particle swarm optimization technology; and 5) inputting the result obtained by network training into a full-link layer of the neural network, and performing linear regression calculation by using the feature representation of a higher layer learned by the full-link layer to obtain the RUL corresponding to the period of the machine for manually adjusting the operation parameters or the operation conditions of the machine.

Description

Attention mechanism-based mechanical equipment multi-working-condition fault prediction method
Technical Field
The invention relates to a fault prediction method, in particular to a multi-working-condition fault prediction method for mechanical equipment, and belongs to the technical field of mechanical equipment.
Background
With the rapid development of sensor technology and industrial systems, industrial equipment sensor data is rapidly increasing. However, in the industrial manufacturing process, the mechanical equipment inevitably suffers performance degradation and even failure due to complicated physical and chemical changes in the manufacturing process. Since a fault occurring in a production process may cause a serious industrial production problem and economic loss, it is important to implement fault Prediction and Health Management (PHM) for industrial production.
Nowadays, in order to realize fault prediction and health management in industrial production, researchers have proposed the concept of Remaining Useful Life (RUL) of mechanical equipment components or systems. The RUL is a life cycle calculation of a device component or system from a current working time to a device failure time, and the occurrence of fading or failure of an industrial device can be effectively avoided by predicting the RUL, which is extremely important for industrial production.
In general, methods of predicting an estimated RUL are classified into three, a mathematical physical model-based method, a data-driven method, and a hybrid method thereof. The method based on the mathematical physical model simulates the change of physical characteristics of the mechanical production process and the change of system faults by establishing an accurate mathematical model for the machine per se, and then predicts the RUL of the mechanical equipment according to the state change of the mechanical equipment. However, the complex physical and chemical changes occurring in the production process, and the variable operating environment and environmental noise, it is difficult for the model-based method to accurately mathematically model the mechanical production process. On the other hand, a large amount of sensor data is generated in an industrial process, and thus a data driving method relying on a large amount of data is attracting attention of researchers and industries. The data-driven method can well extract and utilize a large amount of potential information in multi-dimensional sensor data, and the traditional data-driven method comprises a support vector machine, a correlation vector machine and a support vector regression, but the output characteristics of the methods are all made manually, so that good prediction performance is difficult to realize under different industrial conditions. Because a large amount of sensor data generated in industrial production engineering are time series data, in recent years, a recurrent neural network, particularly a long-short term memory neural network, is proposed to model the sensor time series data, so that potential information in the sensor data can be well extracted, and the RUL prediction performance is good.
However, these methods all assume that all sensor data is obtained from the same operating conditions. However, in a real industrial production environment, the mechanical operation conditions are often different, which means that different operation conditions can affect the prediction of the RUL and even lead to wrong prediction results. Different production conditions therefore present challenges to the failure prediction and health management of mechanical equipment.
The prior art discloses a method and a system for calculating the remaining service life of mechanical equipment under multiple working conditions, wherein the application numbers are as follows: 201710137261.1, filing date: 2017-07-14, which comprises six modules, a data acquisition module, a construction module, a prediction module, a judgment module, a calculation module and an optimization module. The data acquisition module is used for acquiring historical data and current data of the mechanical equipment to form an original training data set of the Gaussian process regression model; the construction module constructs a Gaussian process regression model according to the original data; the prediction module predicts the characteristic value according to the obtained regression model to obtain a predicted value corresponding to the residual service life; the judgment module judges whether the predicted value exceeds a threshold value; the calculation module calculates the residual service life corresponding to the predicted value; and when the predicted value does not exceed the threshold value, the optimization module inputs the predicted value into the model as new training data to optimize the training of the model. The disadvantages are that: the method for acquiring data by the technology is a manual manufacturing characteristic, the process is extremely complex and difficult, and time and labor are consumed, and secondly, the method cannot fully utilize the potential time dependence relationship in the sensor data and cannot find the operation condition which has larger influence on the mechanical production process, so that the method is convenient for manual adjustment of the mechanical production process.
Disclosure of Invention
The invention aims to provide a mechanical equipment multi-working-condition fault prediction method based on an attention mechanism, which combines a natural language processing technology and the attention mechanism, obtains potential information in sensor data by using a long-term and short-term memory network in the natural language processing technology, realizes more abstract feature representation in mechanical sensor data, trains by using the abstract feature representation and accurately predicts mechanical fault data; by introducing an attention mechanism and adding an attention weight value to the operation condition, the operation condition with large influence on fault prediction is found, and the prediction accuracy of the network is improved, so that the fault prediction under multiple working conditions is realized.
The purpose of the invention is realized as follows: a multi-condition fault prediction method for mechanical equipment based on an attention mechanism comprises the following steps:
step 1) preprocessing data, vectorizing sensor data in an original mechanical database, then normalizing vectors according to a normalization principle, and finally dividing the normalized data by using a sliding time window;
step 2) embedding the preprocessed data obtained in the step 1) into a long-short term memory network layer as input, and obtaining a long-range dependency relationship of the data through training so as to calculate a hidden vector of the operation condition in the mechanical input vector;
step 3) inputting hidden vectors of the operating conditions in the mechanical data obtained in the step 2) into an attention layer;
step 4), in the network training process, optimizing network parameters by using a particle swarm optimization technology;
and 5) inputting the result obtained by network training into a full connection layer of the neural network, acquiring the input higher layer feature representation by the full connection layer, and performing linear regression calculation by using the higher layer feature representation learned by the full connection layer to obtain the RUL corresponding to the period of the machine for manually adjusting the operation parameters or the operation conditions of the machine.
As an improvement of the present invention, step 1) specifically includes:
1-1) vectorizing the raw data to obtain sensor vector data, i.e.
Figure BDA0002689297810000031
Figure BDA0002689297810000032
Wherein T iscC ∈ {1,2, …, n } for the failure period of the c-th component; each component s in the vectoriIs a k +1 dimensional vector { x1,x2,…,xk,yiDenotes the k +1 dimensional sensor data input features and the corresponding remaining life cycle, i.e. xiFor the i-th sensor data, yiIs the corresponding remaining life cycle;
1-2) normalization of the input vector, normalizing each term in S, i.e.
Figure BDA0002689297810000033
Figure BDA0002689297810000041
Expressing the normalization value of jth sensor data in the ith period to obtain a value S' after normalization processing;
1-3) dividing input data in a time domain by using a sliding time window to obtain data O ═ { O ═ Oi|i＞＝0,i＜Tc-L }, wherein oi＝si' Window Length is L and sliding step is T.
As an improvement of the present invention, step 2) specifically comprises:
2-1) receiving the vector O ═ s in step 1)1′,s2′,…,sL' }, denote the vector component as xtThe sensor vector after the normalization processing is carried out; the vectors are then input into the long-short term memory network, which computes, for each time t, the use of the input xtAnd h of the previous momentt-1To calculate htWherein h istHidden vectors representing data sequences, specific calculation methodsThe method is as follows,
it＝σ(Wixt+Uiht-1+bi) (1)
ft＝σ(Wfxt+Ufht-1+bf) (2)
Figure BDA0002689297810000042
Figure BDA0002689297810000043
ot＝σ(Woxt+Uoht-1+bo) (5)
ht＝ot⊙tanh(ct) (6)
Hi＝[h1,h2,…,hL] (7)
wherein h, i, f and o are respectively a hidden vector, an input gate, a forgetting gate and an output gate; w, U and b corresponding to the subscripts are respectively calculated connection weights during the training of the long-term and short-term memory network, sigma is a sigmoid function, and sigma is a dot product; the network updating algorithm is a back propagation algorithm (BP);
2-2) inputting the obtained hidden vector to the attention layer as an input to obtain an attention weight value of each operation condition hidden vector.
As a development of the invention, step 3) the attention tier is for each input htThe attention weight is given, the attention weight is calculated as follows,
eti＝tanh(ht) (8)
Figure BDA0002689297810000044
wherein a isiThe attention weight value of each hidden vector is a minimum value, and the denominator is avoided being zero; e.g. of the typetiTo countCalculating intermediate variables with the meaning of htThe attention layer adds an attention weight to each hidden vector, and the output result is Hi′＝[h1′,h2′,…,hL′]Wherein h'i＝aihi。
As an improvement of the present invention, step 4) specifically includes:
4-1) calculating a solution space dimension D of the optimization problem according to parameters needing to be optimized; the parameter to be optimized related to the patent is the length L of a sliding time window and the step length T, namely D is 1+ 1;
4-2) initializing a particle swarm, wherein the particle swarm is assumed to form a D-dimensional solution space by N particles, and each particle vector is Xi＝(xi1,xi2,…,xiD) Velocity of each particle is Vi＝(vi1,vi2,…,viD)，i∈{1,2,…,n}；
4-3) establishing an optimization target, wherein in order to make the network training convergence faster and the accuracy higher, the particle swarm optimization target equation is a network loss function, namely
Figure BDA0002689297810000051
Wherein W and b are respectively the weight and the deviation of the LSTM network, and parameters are optimized to ensure that a loss function is minimum;
4-4) updating the position and the speed of the particles, and the calculation formula method is as follows,
Vt+1＝w·vt+c1rand()(pbestt-pt)+c2rand()(gbestt-pt)
(10)
pt+1＝pt+vt+1 (11)
wherein, ViIs the particle velocity, piIs the particle position, w is the velocity inertia value, c1,c2The degrees of influence of receiving the locally optimal solution and the globally optimal solution respectively.
As an improvement of the present invention, the calculation of the RUL prediction in step 5) is as follows,
oi＝f(H′i；θdense)＝f(WH′i+b) (12)
Figure BDA0002689297810000052
wherein o isiIs the output vector of the full-link layer, W, b respectively represent the weight and deviation of the full-link layer, f is the activation function of the full-link layer,
Figure BDA0002689297810000053
for the prediction of RUL, θdenseAbbreviated as full connectivity layer network weight, i.e., W, b above.
Compared with the prior art, the invention adopting the technical scheme has the following technical effects: the method combines the time series modeling capability of a long-short term memory network (LSTM) and the selection capability of an Attention (Attention) mechanism, and improves the convergence speed and accuracy of the model by utilizing an optimization algorithm; the neural network provided by the invention mainly comprises three parts, namely an LSTM (long short term memory network) layer, an Attention layer and a full connection layer; the method comprises the steps of extracting potential information of input data by using an LSTM, embedding the potential information into an Attention layer as input, giving a corresponding Attention weight value to each operation condition, obtaining a corresponding weight value by each operation condition through network training, indicating that the larger the weight value is, the larger the influence of the operation condition on a prediction result is, finally inputting the result of the Attention layer into a full-connection layer, and performing linear regression on the output result of the full-connection layer to obtain a final RUL value. The system fully considers the time dependence of sensor data and the different influences of different environments on the prediction result, so that a fault prediction model with higher accuracy and better generalization performance is obtained, and more convenience is brought to fault prediction and health management in the mechanical production process.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a schematic diagram of a neural network according to the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the attached drawings:
as shown in fig. 1-2, the present embodiment provides a method for predicting a multi-condition fault of a mechanical device based on an attention mechanism, including the following steps:
step 1: taking out original data from a database received and stored on a sensor, and preprocessing the original data; first, the raw data is received and vectorized, denoted as
Figure BDA0002689297810000061
TcC ∈ {1,2, …, n } for the failure period of the c-th component; each of which is a K +1 dimensional vector { x +1,x2,…,xk,yiRepresents k sensor input characteristic values and corresponding RUL values, respectively; then, the data is normalized, and the calculation formula is
Figure BDA0002689297810000062
I.e. the corresponding normalized value of each term in S is
Figure BDA0002689297810000063
Finally, the data is divided in time to form sliding time window data, namely O ═ Oi|i＞＝0,i＜Tc-L }, wherein oi＝si' Window Length is L and sliding step is T.
Step 2: inputting the processed data into the LSTM layer, and calculating a hidden vector h corresponding to each time periodtThus, a hidden vector corresponding to the whole time window, i.e. H, is obtainedi＝[h1,h2,…,hL]The hidden vector preserves time-dependent information in the sensor data.
And step 3: inputting the hidden vectors obtained by calculation in the LSTM layer into the corresponding attention layer, calculating the attention weight value of each hidden vector, wherein the greater the influence of different running conditions on the RUL prediction, the greater the attention weight value, and the attention weight value can be obtained by the formulas (8) and (9), namelyThe attention weight to each operating condition is aiAnd taking the product of the attention weight value and the hidden vector as an input to be input into the full-connection layer.
And 4, step 4: optimizing parameters designed by the network model, and calculating a solution space dimension of the particle swarm, wherein D is 2; initializing group velocity and position vt,ptEstablishing an optimization equation
Figure BDA0002689297810000071
Finally, the velocity and position v of the particle group are updatedt+1,pt+1。
And 5: obtaining the output of the attention layer and calculating the formula oi＝f(H′i；θdense)＝f(WH′i+ b) to get a higher level representation of the feature, and for oiDirectly performing linear regression to obtain corresponding RUL value,
Figure BDA0002689297810000072
and comparing the predicted RUL value with a preset fault threshold value of the system, and if the predicted RUL value is lower than the preset fault threshold value, feeding back the machine to be in a pre-fault state and prompting to perform artificial adjustment.
The above description is only an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can understand that the modifications or substitutions within the technical scope of the present invention are included in the scope of the present invention, and therefore, the scope of the present invention should be subject to the protection scope of the claims.

Claims (6)

1. A multi-condition fault prediction method for mechanical equipment based on an attention mechanism is characterized by comprising the following steps:
step 1) preprocessing data, vectorizing sensor data in an original mechanical database, then normalizing vectors according to a normalization principle, and finally dividing the normalized data by using a sliding time window;
step 2) embedding the preprocessed data obtained in the step 1) into a long-short term memory network layer as input, and obtaining a long-range dependency relationship of the data through training so as to calculate a hidden vector of the operation condition in the mechanical input vector;
step 3) inputting hidden vectors of the operating conditions in the mechanical data obtained in the step 2) into an attention layer;
step 4), in the network training process, optimizing network parameters by using a particle swarm optimization technology;
and 5) inputting the result obtained by network training into a full connection layer of the neural network, acquiring the input higher layer feature representation by the full connection layer, and performing linear regression calculation by using the higher layer feature representation learned by the full connection layer to obtain the RUL corresponding to the period of the machine for manually adjusting the operation parameters or the operation conditions of the machine.
2. The method for predicting the multi-condition fault of the mechanical equipment based on the attention mechanism as claimed in claim 1, wherein the step 1) specifically comprises:
1-1) vectorizing the raw data to obtain sensor vector data, i.e.
Figure FDA0002689297800000011
Figure FDA0002689297800000012
Wherein T iscC ∈ {1, 2.., n }, which is a failure period of the c-th component; each component s in the vectoriIs a k +1 dimensional vector { x1，x2，...，xk，yiDenotes the k +1 dimensional sensor data input features and the corresponding remaining life cycle, i.e. xiFor the i-th sensor data, yiIs the corresponding remaining life cycle;
1-2) normalization of the input vector, normalizing each term in S, i.e.
Figure FDA0002689297800000013
Figure FDA0002689297800000014
Figure FDA0002689297800000015
Expressing the normalization value of jth sensor data in the ith period to obtain a value S' after normalization processing;
1-3) dividing input data in a time domain by using a sliding time window to obtain data O ═ { O ═ Oi|i＞＝0，i＜Tc-L }, wherein oi＝si' Window Length is L and sliding step is T.
3. The method for predicting the multi-condition fault of the mechanical equipment based on the attention mechanism as claimed in claim 2, wherein the step 2) specifically comprises the following steps:
2-1) receiving the vector O ═ s in step 1)1′，s2′，...，sL' }, denote the vector component as xtThe sensor vector after the normalization processing is carried out; the vectors are then input into the long-short term memory network, which computes, for each time t, the use of the input xtAnd h of the previous momentt-1To calculate htWherein h istThe hidden vector representing the data sequence is calculated as follows,
it＝σ(Wixt+Uiht-1+bi) (1)
ft＝σ(Wfxt+Ufht-1+bf) (2)
Figure FDA0002689297800000021
Figure FDA0002689297800000022
ot＝σ(Woxt+Uoht-1+bo) (5)
ht＝ot⊙tanh(ct) (6)
Hi＝[h1，h2，...，hL] (7)
wherein h, i, f and o are respectively a hidden vector, an input gate, a forgetting gate and an output gate; w, U and b corresponding to the subscripts are respectively calculated connection weights during the training of the long-term and short-term memory network, sigma is a sigmoid function, and sigma is a dot product; the network updating algorithm is a back propagation algorithm (BP);
2-2) inputting the obtained hidden vector to the attention layer as an input to obtain an attention weight value of each operation condition hidden vector.
4. The method for predicting the multi-condition fault of the mechanical equipment based on the attention mechanism as claimed in claim 3, wherein the attention layer in the step 3) is used for each input htThe attention weight is given, the attention weight is calculated as follows,
eti＝tanh(ht) (8)
Figure FDA0002689297800000023
wherein a isiThe attention weight value of each hidden vector is a minimum value, and the denominator is avoided being zero; e.g. of the typetiFor calculation of intermediate variables, the meaning is htThe attention layer adds an attention weight to each hidden vector, and the output result is Hi′＝[h1′，h2′，...，hL′]Wherein h'i＝aihi。
5. The method for predicting the multi-condition fault of the mechanical equipment based on the attention mechanism is characterized in that the step 4) specifically comprises the following steps:
4-1) calculating a solution space dimension D of the optimization problem according to parameters needing to be optimized; the parameter to be optimized related to the patent is the length L of a sliding time window and the step length T, namely D is 1+ 1;
4-2) initializing a particle swarm, wherein the particle swarm is assumed to form a D-dimensional solution space by N particles, and each particle vector is Xi＝(xi1，xi2，...，xiD) Velocity of each particle is Vi＝(vi1，vi2，...，viD)，i∈{1，2，...，n}；
4-3) establishing an optimization target, wherein in order to make the network training convergence faster and the accuracy higher, the particle swarm optimization target equation is a network loss function, namely
Figure FDA0002689297800000031
Wherein W and b are respectively the weight and the deviation of the LSTM network, and parameters are optimized to ensure that a loss function is minimum;
4-4) updating the position and the speed of the particles, and the calculation formula method is as follows,
Vt+1＝w·vt+c1rand()(pbestt-pt)+c2rand()(gbestt-pt)
(10)
pt+1＝pt+vt+1 (11)
wherein, ViIs the particle velocity, piIs the particle position, w is the velocity inertia value, c1，c2The degrees of influence of receiving the locally optimal solution and the globally optimal solution respectively.
6. The method of claim 5, wherein the RUL prediction in step 5) is calculated as follows,
oi＝f(H′i；θdense)＝f(WH′i+b) (12)
Figure FDA0002689297800000032
wherein o isiIs the output vector of the full-link layer, W, b respectively represent the weight and deviation of the full-link layer, f is the activation function of the full-link layer,
Figure FDA0002689297800000033
for the prediction of RUL, θdenseAbbreviated as full connectivity layer network weight, i.e., W, b above.
CN202010986026.3A 2020-09-18 2020-09-18 Attention mechanism-based mechanical equipment multi-working-condition fault prediction method Withdrawn CN112149896A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010986026.3A CN112149896A (en) 2020-09-18 2020-09-18 Attention mechanism-based mechanical equipment multi-working-condition fault prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010986026.3A CN112149896A (en) 2020-09-18 2020-09-18 Attention mechanism-based mechanical equipment multi-working-condition fault prediction method

Publications (1)

Publication Number Publication Date
CN112149896A true CN112149896A (en) 2020-12-29

Family

ID=73893244

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010986026.3A Withdrawn CN112149896A (en) 2020-09-18 2020-09-18 Attention mechanism-based mechanical equipment multi-working-condition fault prediction method

Country Status (1)

Country Link
CN (1) CN112149896A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113705922A (en) * 2021-09-06 2021-11-26 内蒙古科技大学 Improved ultra-short-term wind power prediction algorithm and model establishment method
CN114386693A (en) * 2022-01-11 2022-04-22 福建中锐网络股份有限公司 Reservoir water level prediction early warning method based on PSO-LSTM-Attention model
CN116050665A (en) * 2023-03-14 2023-05-02 淄博热力有限公司 Heat supply equipment fault prediction method

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113705922A (en) * 2021-09-06 2021-11-26 内蒙古科技大学 Improved ultra-short-term wind power prediction algorithm and model establishment method
CN113705922B (en) * 2021-09-06 2023-09-12 内蒙古科技大学 Improved ultra-short-term wind power prediction algorithm and model building method
CN114386693A (en) * 2022-01-11 2022-04-22 福建中锐网络股份有限公司 Reservoir water level prediction early warning method based on PSO-LSTM-Attention model
CN116050665A (en) * 2023-03-14 2023-05-02 淄博热力有限公司 Heat supply equipment fault prediction method
CN116050665B (en) * 2023-03-14 2024-04-02 淄博热力有限公司 Heat supply equipment fault prediction method

Similar Documents

Publication Publication Date Title
CN110245801A (en) A kind of Methods of electric load forecasting and system based on combination mining model
CN109800875A (en) Chemical industry fault detection method based on particle group optimizing and noise reduction sparse coding machine
CN108596327B (en) Seismic velocity spectrum artificial intelligence picking method based on deep learning
CN110837915B (en) Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning
CN112329990A (en) User power load prediction method based on LSTM-BP neural network
CN111079343A (en) Wind turbine generator effective wind speed estimation method based on width learning
CN116303786B (en) Block chain financial big data management system based on multidimensional data fusion algorithm
CN112149896A (en) Attention mechanism-based mechanical equipment multi-working-condition fault prediction method
CN112766603A (en) Traffic flow prediction method, system, computer device and storage medium
CN113850320A (en) Transformer fault detection method based on improved support vector machine regression algorithm
CN117313795A (en) Intelligent building energy consumption prediction method based on improved DBO-LSTM
CN104634265A (en) Soft measurement method for thickness of mineral floating foam layer based on multivariate image feature fusion
CN115619028A (en) Clustering algorithm fusion-based power load accurate prediction method
CN116822652A (en) Subway fault prediction method, subway fault prediction device, electronic equipment, subway fault prediction system and storage medium
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
CN113449919B (en) Power consumption prediction method and system based on feature and trend perception
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
CN117477544A (en) LSTM ultra-short-term photovoltaic power prediction method and system integrating time mode features
CN117113086A (en) Energy storage unit load prediction method, system, electronic equipment and medium
CN112014757A (en) Battery SOH estimation method integrating capacity increment analysis and genetic wavelet neural network
CN116646929A (en) PSO-CNN-BILSTM-based short-term wind power prediction method
CN115794805A (en) Medium-low voltage distribution network measurement data supplementing method
CN115759343A (en) E-LSTM-based user electric quantity prediction method and device
Zhang et al. Short-term electrical load forecasting based on Attention-GRU networks
CN113971489A (en) Method and system for predicting remaining service life based on hybrid neural network

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20201229