CN117493837B - Machine learning-based shield tunneling machine attitude item prediction method - Google Patents

Machine learning-based shield tunneling machine attitude item prediction method Download PDF

Info

Publication number
CN117493837B
CN117493837B CN202410004533.0A CN202410004533A CN117493837B CN 117493837 B CN117493837 B CN 117493837B CN 202410004533 A CN202410004533 A CN 202410004533A CN 117493837 B CN117493837 B CN 117493837B
Authority
CN
China
Prior art keywords
parameter data
long
trend
attitude
memory network
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
Application number
CN202410004533.0A
Other languages
Chinese (zh)
Other versions
CN117493837A (en
Inventor
杨剑
崔凯
赵志坚
李佩业
乔亚飞
袁海林
王皓宇
王秋实
刘宏伟
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.)
Tongji University
China Railway South Investment Group Co Ltd
Original Assignee
Tongji University
China Railway South Investment Group Co Ltd
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 Tongji University, China Railway South Investment Group Co Ltd filed Critical Tongji University
Priority to CN202410004533.0A priority Critical patent/CN117493837B/en
Publication of CN117493837A publication Critical patent/CN117493837A/en
Application granted granted Critical
Publication of CN117493837B publication Critical patent/CN117493837B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/211—Selection of the most significant subset of features
    • G06F18/2113—Selection of the most significant subset of features by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/08—Learning methods
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • E—FIXED CONSTRUCTIONS
    • E21—EARTH OR ROCK DRILLING; MINING
    • E21D—SHAFTS; TUNNELS; GALLERIES; LARGE UNDERGROUND CHAMBERS
    • E21D9/00—Tunnels or galleries, with or without linings; Methods or apparatus for making thereof; Layout of tunnels or galleries
    • E21D9/06—Making by using a driving shield, i.e. advanced by pushing means bearing against the already placed lining
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a machine learning-based method for predicting the attitude of a shield tunneling machine in terms of terms, and relates to the field of attitude prediction of the shield tunneling machine. The method comprises the following steps: trend items and fluctuation items of attitude parameter data and trend items and fluctuation items of construction parameter data are obtained; determining strong correlation parameters, and determining trend items and fluctuation items of the strong correlation parameter data; constructing and training a first long-time memory network model and a second long-time memory network model; and performing super-parameter optimization and second training on the trained first long-short-time memory network model and second long-short-time memory network model to obtain the gesture separate prediction result of the shield tunneling machine. The method considers the influence of the construction parameters of the shield tunneling machine on the posture of the shield tunneling machine, solves the problem of complex association factors of the time sequence parameters of the shield tunneling machine, and further can realize accurate prediction of the posture of the shield tunneling machine.

Description

Machine learning-based shield tunneling machine attitude item prediction method
Technical Field
The invention relates to the field of attitude prediction of shield tunneling machines, in particular to a machine learning-based attitude item prediction method of a shield tunneling machine.
Background
The attitude of the shield tunneling machine is an important judging index for judging the construction safety of the shield tunneling machine, and comprises a shield head horizontal deviation, a shield head vertical deviation, a shield tail horizontal deviation, a shield tail vertical deviation, a pitch angle and a roll angle. Therefore, the attitude prediction of the shield tunneling machine has important engineering significance, the construction risk of the shield tunneling machine can be predicted in advance, and valuable reference information is provided for the adjustment of the attitude of the shield tunneling machine. Especially in shield sections with complex geological conditions such as upper soft and lower hard, the attitude prediction of the shield tunneling machine plays an important role in ensuring the safety and stability of engineering.
At present, most of existing shield tunneling machine attitude prediction methods are based on long-range autocorrelation of the shield tunneling machine attitude, the shield tunneling machine attitude is predicted, influence of shield tunneling machine construction parameters on the shield tunneling machine attitude is ignored, and the shield tunneling machine attitude is closely related to the shield tunneling machine construction parameters, so that the precision and the interpretability of the existing shield tunneling machine attitude prediction methods are low. The existing shield tunneling machine prediction method based on machine learning predicts the complex shield tunneling machine attitude change process without any pretreatment, and because the shield tunneling machine time sequence parameter is generally a noisy non-stationary sequence, the correlation factor is complex, so that the existing method is difficult to accurately predict.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a machine learning-based method for predicting the attitude of the shield tunneling machine in terms of the terms, considers the influence of the construction parameters of the shield tunneling machine on the attitude of the shield tunneling machine, solves the problem of complex association factors of the time sequence parameters of the shield tunneling machine, and can further realize accurate prediction of the attitude of the shield tunneling machine.
In order to achieve the aim of the invention, the invention adopts the following technical scheme:
a method for predicting the attitude of a shield tunneling machine based on machine learning comprises the following steps:
s1, acquiring attitude parameter data and construction parameter data of a shield tunneling machine, and acquiring trend items and fluctuation items of the attitude parameter data and trend items and fluctuation items of the construction parameter data by adopting a variation modal decomposition method;
s2, calculating a de-trend correlation index of trend items of the attitude parameter data and trend items of the construction parameter data to determine strong correlation parameters, and determining trend items and fluctuation items of the strong correlation parameter data;
s3, constructing a first long-short-time memory network model and a second long-short-time memory network model, training the first long-short-time memory network model by using trend items of gesture parameter data and trend items of strong related parameter data, and training the second long-short-time memory network model by using fluctuation items of gesture parameter data and fluctuation items of strong related parameter data;
and S4, performing super-parameter optimization and second training on the trained first long-short-time memory network model and second long-time memory network model to obtain a final first long-short-time memory network model and a final second long-time memory network model, and obtaining a shield tunneling machine attitude item prediction result by using the final first long-time memory network model and second long-time memory network model.
Further, step S2 includes the following sub-steps:
s21, respectively calculating centralized accumulated signals of trend items of the attitude parameter data and trend items of the construction parameter data;
s22, respectively determining a least square linear fitting signal of trend items of the attitude parameter data and trend items of the construction parameter data by adopting a least square method;
s23, determining the fluctuation index of the trend item of the attitude parameter data according to the centralized accumulated signal and the least square straight line fitting signal of the trend item of the attitude parameter data, and determining the fluctuation index of the trend item of the construction parameter data according to the centralized accumulated signal and the least square straight line fitting signal of the trend item of the construction parameter data;
s24, calculating a cross fluctuation index according to the centralized accumulated signal of the trend item of the attitude parameter data, the least square straight line fitting signal, the centralized accumulated signal of the trend item of the construction parameter data and the least square straight line fitting signal;
s25, calculating a de-trend correlation index of the attitude parameter data and the construction parameter data according to the fluctuation index of the trend item of the attitude parameter data, the fluctuation index and the cross fluctuation index of the trend item of the construction parameter data;
s26, determining strong correlation parameters according to the de-trend correlation indexes of the attitude parameter data and the construction parameter data, and determining trend items and fluctuation items of the strong correlation parameter data.
Further, in the substep S25, a detrend correlation index of the attitude parameter data and the construction parameter data is calculated, expressed as:
wherein:for calculating the detrending correlation index of the attitude parameter data and the construction parameter data +.>Sequence number of trend item for gesture parameter data, +.>Sequence number of trend item for construction parameter data, +.>Cross fluctuation index, ++, for trend term of attitude parameter data and trend term of construction parameter data>Fluctuation index of trend term for posture parameter data, +.>And the fluctuation index of trend items of construction parameter data.
Further, in step S3, the first long-short-time memory network model and the second long-short-time memory network model that are constructed each include an input layer, a long-short-time memory network block, and an output layer;
the input layer is used for inputting normalized vectors of the gesture parameters and the strong correlation parameters in a set time range and transmitting the normalized vectors of the gesture parameters and the strong correlation parameters in the set time range to the long-short-time memory network block;
the long-time and short-time memory network block is used for receiving normalized vectors of the attitude parameters and the strong correlation parameters in the set time range, calculating one-dimensional feature vectors of the attitude of the shield tunneling machine in the next time range according to the normalized vectors of the attitude parameters and the strong correlation parameters in the set time range, and transmitting the one-dimensional feature vectors of the attitude of the shield tunneling machine in the next time range to the output layer;
the output layer is used for receiving the one-dimensional feature vector of the shield tunneling machine gesture in the next time range, calculating the weight of each element in the one-dimensional feature vector of the shield tunneling machine gesture in the next time range to the shield tunneling machine gesture sub-term prediction result, calculating the weighted sum of each element, and outputting the weighted sum result of each element as the output to obtain the shield tunneling machine gesture sub-term prediction result.
Further, the long-short time memory network block comprises a long-short time memory network layer, a regularization layer and a selective inhibition layer which are connected in sequence; the long-short time memory network layer comprises a plurality of long-short time memory network units.
Further, the long-short-time memory network unit comprises a forgetting gate substructure, an input gate substructure and an output gate substructure.
Further, the calculation formula of the forgetting door substructure is expressed as:
wherein:is->Output of forgetting gate substructure of individual cell, < >>To activate the function +.>Trainable weights for forgetting door substructure, +.>For a two-vector splice symbol, < >>Is->History parameter of individual cell output,/->Is the firstNormalized vector of pose parameter and strongly correlated parameter of individual cell,/->Trainable bias for forgetting gate substructure.
Further, the computational formula of the input gate substructure is expressed as:
，
wherein:is->Memory parameters of choice of the input gate substructure of the individual cells,/->To activate the function +.>Trainable weights for selecting memory parameters, +.>For a two-vector splice symbol, < >>Is->The history parameter output by the individual units,is->Normalized vector of pose parameter and strongly correlated parameter of individual cell,/->Trainable deviations for the selection of memory parameters +.>Is->Candidate memory parameters of the input gate substructure of the individual cells,/->As an arctangent function, +.>Trainable weights for candidate memory parameters, +.>Trainable deviations for candidate memory parameters.
Further, the calculation formula of the output gate mechanism is expressed as:
wherein:is->The output of the output gate mechanism of the individual units, is->To activate the function +.>To output trainable weights of the gate mechanism, < +.>For a two-vector splice symbol, < >>Is->History parameter of individual cell output,/->Is the firstNormalization of pose parameters and strongly correlated parameters of individual unitsVector (S)>To output the trainable bias of the gate mechanism.
Further, in step S4, performing the super-parameter optimization on the trained first long-short-time memory network model and the trained second long-time memory network model includes the following steps:
a1, initializing the number of long-short-time memory network blocks and the number of long-short-time memory network units of the trained model;
a2, training the model with the initialized long-short-time memory network block number and the initialized long-short-time memory network unit number by using the normalized vector of the gesture parameter and the strong correlation parameter in the set time range, and determining the model with the minimum loss after parameter initialization;
a3, carrying out parameter variation on the model with the minimum loss after parameter initialization, obtaining models with different parameters, training the models with different parameters by using normalized vectors of attitude parameters and strong related parameters within a set time range, and determining the model with the minimum loss after parameter variation.
The invention has the following beneficial effects:
(1) According to the invention, through calculating the trending correlation index of the trending item of the attitude parameter data and the trending item of the construction parameter data, strong correlation parameters are screened, the parameter range of machine learning data is reduced, the convergence of network model fitting is facilitated, the calculation resources are saved, and the prediction precision is improved;
(2) According to the method, the shield timing sequence parameters are decomposed into trend items and fluctuation items while the shield timing sequence parameters are reduced in a variation modal decomposition method, wherein the trend items describe parameter variation trend, the fluctuation items correspond to fluctuation changes of parameters, and the trend items and the fluctuation items are predicted according to the fluctuation changes of the parameters, so that the reliability and the precision of model results are improved, the model convergence is accelerated, meanwhile, the engineering significance of the trend items and the fluctuation items is clear, and the method has high engineering practical application value;
(3) According to the invention, the super parameters of the long-short-time memory network model are optimized, so that the scale of the trainable parameters of the long-short-time memory network model is matched with the scale of the data set and the prediction complexity, the prediction precision of the long-short-time memory network model is improved, the calculation resources are saved, the scale of the long-short-time memory network model is reduced, and the application and deployment of the long-short-time memory network model are facilitated.
Drawings
FIG. 1 is a schematic flow diagram of a machine learning-based method for predicting the attitude of a shield tunneling machine in terms of terms;
fig. 2 is a schematic diagram of an initializing long-short-term memory network model for the gesture item prediction of the shield tunneling machine.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and all the inventions which make use of the inventive concept are protected by the spirit and scope of the present invention as defined and defined in the appended claims to those skilled in the art.
As shown in FIG. 1, the method for predicting the attitude of the shield tunneling machine based on machine learning comprises the following steps S1-S4:
s1, acquiring attitude parameter data and construction parameter data of a shield tunneling machine, and acquiring trend items and fluctuation items of the attitude parameter data and trend items and fluctuation items of the construction parameter data by adopting a variation modal decomposition method.
In an optional embodiment of the invention, attitude parameter data and construction parameter data of the shield tunneling machine are obtained, and trend items and fluctuation items of the attitude parameter data and trend items and fluctuation items of the construction parameter data are obtained by adopting a variation modal decomposition method. The attitude parameter data of the shield tunneling machine comprise pitch angle, rolling angle, forward horizontal deviation, forward vertical deviation, backward horizontal deviation and forward vertical deviation. The construction parameter data of the shield tunneling machine comprise cutter head rotating speed, pushing pressure, tunneling speed, cutter head torque, ring number, jack pushing pressure, soil bin pressure, grouting pressure, jack pushing displacement and total pushing force.
Specifically, the invention collects the time sequence parameter data of the shield tunneling machine from the industrial personal computer of the shield tunneling machine so as to obtain the attitude parameter data and the construction parameter data of the shield tunneling machine. The attitude parameter data and the construction parameter data of the shield tunneling machine obtained by the invention are time sequence parameter data of the shield tunneling machine. The time sequence parameter data of the shield tunneling machine are the shield tunneling machine parameters which are arranged in time sequence and are continuously collected and stored at fixed frequency.
Specifically, in a continuous tunneling section, the invention adopts a variation modal decomposition method to obtain trend items and fluctuation items of attitude parameter data and trend items and fluctuation items of construction parameter data. The continuous tunneling section is time sequence parameter data of the shield tunneling machine within a certain time range, and in the time range, the cutter head rotating speed, the pushing pressure, the tunneling speed and the cutter head torque are all not 0.
The invention adopts a variation modal decomposition method to obtain trend items and fluctuation items of attitude parameter data and trend items and fluctuation items of construction parameter data, and comprises the following steps of B1-B3:
b1, constructing a constraint variation model of attitude parameter data and construction parameter data, wherein the constraint variation model is expressed as follows:
wherein:to take->And->Is the minimum value of the variable, +.>Is the +.>Decomposing the modal function->Also denoted as->，/>Is->Individual decomposition modality function->Center frequency of>For decomposing the sequence number of the mode->For time of day->Deviation-inducing and->Time of time sequence data corresponding to attitude parameter data and construction parameter data, +.>For +.>Is a Croneck function of->Is of circumference rate>Is imaginary unit, ++>Is natural and normalCount (n)/(l)>Is the square of the spectrum norm, +.>Is attitude parameter data and construction parameter data.
The decomposition modal function of the attitude parameter data and the construction parameter data is expressed as:
wherein:is->Envelope magnitude of each decomposition mode, and +.>，/>Is->Instantaneous phase of the individual decomposition modes, and +.>，/>Is->For->And (5) deriving results.
And B2, acquiring an augmented Lagrange function of the time sequence parameter data of the shield tunneling machine by using a secondary punishment item and a Lagrange multiplier method according to the constraint variation model of the shield attitude parameter data and the construction parameter data, wherein the augmented Lagrange function is expressed as follows:
wherein:to->、/>And->An augmented Lagrangian function for parameters, +.>Is Lagrangian multiplier +.>Also denoted as->，/>To be defined as symbol>For punishment parameters->For the inner product of the matrix.
And B3, acquiring decomposition mode data of the attitude parameter data and the construction parameter data by using an alternate direction multiplier method according to the extended Lagrangian function of the attitude parameter data and the construction parameter data.
The invention obtains the decomposition mode data of the time sequence parameter data of the shield tunneling machine by using the alternate direction multiplier method, namely, firstly fixingAnd->Two variables, update only +.>The solution expression is:
wherein:is->Post-iteration->，/>Is->Post-iteration->，/>Is->Post-iteration->。
The invention solves the solving expression in the frequency domain by utilizing the Pasteur theorem, and can obtain:
，
wherein:is->Fourier transform results of->For Fourier transform, ++>Is the center frequency, here the variable, +.>Is->Fourier transform results of->Is->Fourier transform results of->For the number of decomposition modality function at the time of accumulated summation, +.>Is->Fourier transform results of->Is imaginary unit, ++>Namely +.>。
Then updateThe solution expression is:
wherein:is->Post-iteration->。
Finally updateThe solution expression is:
wherein:is->Post-iteration->，/>Is the noise figure.
The update termination conditions are:meets the precision requirement, or reaches the maximum update times, and the termination condition expression is as follows:
or->
Wherein:for the precision condition value +.>Is the maximum number of updates.
When the invention utilizes the process to carry out variation modal decomposition on the attitude parameter data and the construction parameter data, firstly, the method takesInitializing->The method comprises the steps of carrying out a first treatment on the surface of the Then constantly take +.>Update +.>Output +.>。
Output ofAfter that, continuously take +.>Iterative calculation of the output +.>Finally obtainInstant postureThe invention determines a first item in the decomposition mode set of the attitude parameter data and the construction parameter data as a trend item of the attitude parameter data and the construction parameter data, and the rest items as fluctuation items of the attitude parameter data and the construction parameter data.
S2, calculating a de-trend correlation index of trend items of the attitude parameter data and trend items of the construction parameter data to determine strong correlation parameters, and determining trend items and fluctuation items of the strong correlation parameter data.
In an alternative embodiment of the invention, the invention calculates a de-trend correlation index of trend terms of the attitude parameter data and trend terms of the construction parameter data to determine strongly correlated parameters, and determines trend terms and fluctuation terms of the strongly correlated parameter data.
Step S2 comprises the following sub-steps:
s21, calculating centralized accumulated signals of trend items of the attitude parameter data and trend items of the construction parameter data respectively.
The invention calculates the centralized accumulated signal of the data, expressed as:
wherein:for the centralized accumulation of data, +.>For the moment of->Time of time sequence data corresponding to attitude parameter data and construction parameter data, +.>Is->Time attitude parameter numberAccording to the value of->Is the mean value of the attitude parameter data.
S22, a least square method is adopted to respectively determine a least square straight line fitting signal of trend items of the attitude parameter data and trend items of the construction parameter data.
Specifically, a signal window is set, and a least square method is adopted to conduct linear fitting on centralized accumulated signals of trend items of attitude parameter data and trend items of construction parameter data in the signal window, so that least square linear fitting signals of the trend items of the attitude parameter data and the trend items of the construction parameter data are obtained.
The invention sets a signal window specifically as follows: at least 5 signal window lengths are set to form a signal window of an arithmetic series. The method comprises the steps of setting signal windows with different window lengths, and carrying out linear fitting on centralized accumulated signals of trend items of attitude parameter data and trend items of construction parameter data in the signal windows with different window lengths by adopting a least square method to obtain least square linear fitting signals of the trend items of the attitude parameter data and the trend items of the construction parameter data with different window lengths.
S23, determining the fluctuation index of the trend item of the attitude parameter data according to the centralized accumulated signal and the least square straight line fitting signal of the trend item of the attitude parameter data, and determining the fluctuation index of the trend item of the construction parameter data according to the centralized accumulated signal and the least square straight line fitting signal of the trend item of the construction parameter data.
Specifically, the root mean square error corresponding to the two data is respectively determined according to the centralized accumulated signal and the least square straight line fitting signal of the two data, and is expressed as:
wherein:is the first/>Root mean square error of data within the individual signal windows, < >>Time of time sequence data corresponding to attitude parameter data and construction parameter data, +.>For a signal window length of +.>Window number, & gt>For the length of the signal window,for the centralized accumulation of data, +.>For data in->The least squares straight line within each signal window fits the signal.
According to the root mean square error of the data, the fluctuation function of the data is calculated, and is expressed as:
wherein:for a signal window length of +.>Fluctuation function of data of->For a signal window length of +.>Is a function of the total number of signal windows of the mobile terminal.
The invention changes window sizenAt least 5 of the units are formed into an equal ratio arraynIs marked as，/>Linear fitting by least squares method>Relation between them, fluctuation index of trend item of the obtained attitude parameter data +.>And fluctuation index of trend term of construction parameter data +.>。
S24, calculating a cross fluctuation index according to the centralized accumulated signal of the trend item of the attitude parameter data, the least square straight line fitting signal, the centralized accumulated signal of the trend item of the construction parameter data and the least square straight line fitting signal.
Specifically, the invention integrates signals according to trend items of attitude parameter dataAnd least squares straight line fit signal +.>And trend item accumulation signal of construction parameter data +.>And least squares straight line fit signal +.>Calculating trend items of attitude parameter data and trend items of construction parameter dataAnd calculating a cross fluctuation function according to the root mean square error of the trend item of the attitude parameter data and the trend item of the construction parameter data.
The invention calculates the root mean square error of the trend item of the attitude parameter data and the trend item of the construction parameter data, which is expressed as follows:
wherein:is->Root mean square error of trend term of attitude parameter data and trend term of construction parameter data within individual signal windows, +.>Sequence number of trend item for gesture parameter data, +.>Sequence number of trend item for construction parameter data, +.>Time of time sequence data corresponding to attitude parameter data and construction parameter data, +.>For the signal window length, +.>Is->Centralized accumulated signal of individual gesture parameter data trend item,/->Is->The trend item of the personal posture parameter data is in the +.>Least squares straight line fit signal within each signal window,/->Is->Centralized accumulated signal of individual construction parameter data trend items,/->Is->The data trend item of the individual construction parameters is +.>The least squares straight line within each signal window fits the signal.
According to the method, a cross fluctuation function is calculated according to the root mean square error of the trend item of the attitude parameter data and the trend item of the construction parameter data, and the cross fluctuation function is expressed as follows:
wherein:for a signal window length of +.>Cross fluctuation function of trend item of attitude parameter data and trend item of construction parameter data, +.>Sequence number of trend item for gesture parameter data, +.>For applyingSequence number of trend item of engineering parameter data, +.>For the signal window length, +.>For a signal window length of +.>Signal window total number,/->For a signal window length of +.>Is a window sequence number of (c).
The invention changes window sizenAt least 5 of the units are formed into an equal ratio arraynIs marked as，/>Linear fitting by least squares method>Relation between them, calculated cross fluctuation index +.>。
S25, calculating a de-trend correlation index of the attitude parameter data and the construction parameter data according to the fluctuation index of the trend item of the attitude parameter data, the fluctuation index and the cross fluctuation index of the trend item of the construction parameter data.
The invention calculates the de-trend correlation index of the attitude parameter data and the construction parameter data, which is expressed as:
wherein:for calculating the detrending correlation index of the attitude parameter data and the construction parameter data +.>Sequence number of trend item for gesture parameter data, +.>Sequence number of trend item for construction parameter data, +.>Cross fluctuation index, ++, for trend term of attitude parameter data and trend term of construction parameter data>Fluctuation index of trend term for posture parameter data, +.>And the fluctuation index of trend items of construction parameter data.
S26, determining strong correlation parameters according to the de-trend correlation indexes of the attitude parameter data and the construction parameter data, and determining trend items and fluctuation items of the strong correlation parameter data.
In particular, the present invention determines a decorrelation index thresholdAnd determining the construction parameters corresponding to the decorrelation indexes greater than or equal to the decorrelation index threshold as strong correlation parameters.
And S3, constructing a first long-short-time memory network model and a second long-short-time memory network model, training the first long-short-time memory network model by using trend items of gesture parameter data and trend items of strong related parameter data, and training the second long-short-time memory network model by using fluctuation items of gesture parameter data and fluctuation items of strong related parameter data.
In an alternative embodiment of the invention, the invention constructs a first long-short-time memory network model and a second long-short-time memory network model, trains the first long-short-time memory network model by using trend items of gesture parameter data and trend items of strong related parameter data, trains the second long-short-time memory network model by using fluctuation items of gesture parameter data and fluctuation items of strong related parameter data.
As shown in fig. 2, a first long-short-term memory network model constructed according to the present invention is the trend item prediction network in fig. 2, and a second long-short-term memory network model constructed according to the present invention is the fluctuation item prediction network in fig. 2. The first long-short time memory network model and the second long-short time memory network model constructed by the method comprise an input layer, a long-short time memory network block and an output layer.
Specifically, the present invention sets the block numbers of the long-short-time memory network blocks of the first long-short-time memory network model and the second long-short-time memory network model to 2. The 2 long-short-time memory network blocks are a first long-short-time memory network block and a second long-short-time memory network block respectively. The invention sets the long-short time memory network units in the first long-short time memory network block of the first long-short time memory network model to 64 units. The invention sets the long-short-time memory network units in the second long-short-time memory network block of the first long-short-time memory network model to 32 units. The invention sets 128 long-short time memory network units in a first long-short time memory network block of a second long-short time memory network model. The invention sets the long-short-time memory network units in the second long-short-time memory network block of the second long-short-time memory network model to 64 units.
The input layer is used for inputting a set time rangeT S ) Normalized vector of internal attitude parameter and strong correlation parameterNa+Nc) And transmitting the normalized vectors of the attitude parameters and the strong correlation parameters within the set time range to the long-short-time memory network block.
The long-time and short-time memory network block is used for receiving normalized vectors of the attitude parameters and the strong correlation parameters in the set time range, calculating one-dimensional feature vectors of the attitude of the shield tunneling machine in the next time range according to the normalized vectors of the attitude parameters and the strong correlation parameters in the set time range, and transmitting the one-dimensional feature vectors of the attitude of the shield tunneling machine in the next time range to the output layer.
Specifically, the method updates the weight value in the long-short-term memory network block, calculates the correlation between the normalized vector of the posture parameter and the strong correlation parameter in the current time range and the posture parameter in the next time range by using the updated weight value, and further calculates the one-dimensional feature vector of the posture of the shield tunneling machine in the next time range.
The long-short time memory network block comprises a long-short time memory network layer, a regularization layer and a selective inhibition layer which are connected in sequence; the long-short time memory network layer comprises a plurality of long-short time memory network units. The regularization layer is used for regularizing the output vector of the upper layer, and the selective inhibition layer is used for randomly inhibiting the connection number of the upper layer and the lower layer. In the present invention, the inhibition ratio of the selective inhibition layer was set to 0.2.
The long-and-short-term memory network unit comprises a forgetting gate substructure, an input gate substructure and an output gate substructure.
Specifically, the present invention relates to a method for manufacturing a semiconductor device. The first of the long-short-term memory network unitsThe unit input is->Historical parameters of individual cell outputsMemory state->First->Normalized vector of attitude parameters and strongly correlated parameters within the set time range of the individual units +.>. The first part of the long-time memory network unit>The output of the individual units is the history parameter +.>And memory state->。
The invention calculates the memory state, expressed as:
wherein:is->Memory state of individual cells->Is->Output of individual cell forgetting gate substructure, +.>For the Hadamard product of the matrix, +.>Is->Memory state of individual cells->Is->The individual cells are entered into the memory parameters of the gate structure, for example>Is->The individual cells input candidate memory parameters for the gate structure.
The calculation history parameters of the invention are expressed as follows:
wherein:is->History parameters of individual units->Is->The output of the individual cell output gate structure, < >>For the Hadamard product of the matrix, +.>As an arctangent function.
The invention is the firstThe individual unit calculates the output of the forgetting gate substructure as a function of the input parameters +.>Input of the memory parameter of choice of the gate substructure +.>And candidate memory parameter->Output of output gate structure +.>。
The calculation formula of the forgetting door substructure is expressed as:
wherein:is->The output of the forgetting gate substructure of the individual cells, and +.>，/>Representing historical parameters->Degree of forgetting, ->Indicating history parameter->Is completely reserved (I)>Indicating history status->Is left behind in the complete process of being forgotten,to activate the function +.>Trainable weights for forgetting door substructure, +.>For a two-vector splice symbol, < >>Is the firstHistory parameter of individual cell output,/->Is->Normalized vector of pose parameter and strongly correlated parameter of individual cell,/->Trainable bias for forgetting gate substructure.
The computational formula of the input gate substructure is expressed as:
，
wherein:is->Memory parameters of the input gate structure of the individual cells and +.>，/>Representing candidate memory parameters +.>The degree of being retained, ++>Indicating candidate memory parameters->Is completely reserved (I)>Indicating candidate memory parameters->Is forgotten completely, is added with>To activate the function +.>Trainable weights for selecting memory parameters, +.>For a two-vector splice symbol, < >>Is->History parameter of individual cell output,/->Is->Normalized vector of pose parameter and strongly correlated parameter of individual cell,/->Trainable deviations for the selection of memory parameters +.>Is->Candidate memory parameters of the input gate substructure of the individual cells,/->As an arctangent function, +.>Trainable weights for candidate memory parameters, +.>Trainable deviations for candidate memory parameters.
The output gate mechanism is calculated as:
wherein:is->The output of the output gate mechanism of the unit for inputting +.>Normalized vector and history parameters of pose parameters and strongly correlated parameters of individual units +.>Fusion (S)>To activate the function +.>To output trainable weights of the gate mechanism, < +.>For a two-vector splice symbol, < >>Is->History parameter of individual cell output,/->Is->Normalized vector of pose parameter and strongly correlated parameter of individual cell,/->To output the trainable bias of the gate mechanism.
The output layer is used for receiving the one-dimensional feature vector of the shield tunneling machine gesture in the next time range, calculating the weight of each element in the one-dimensional feature vector of the shield tunneling machine gesture in the next time range to the shield tunneling machine gesture sub-term prediction result, calculating the weighted sum of each element, and outputting the weighted sum result of each element as the output to obtain the shield tunneling machine gesture sub-term prediction result.
The method comprises the steps of updating weight values in an output layer, calculating the weight of each element in a one-dimensional feature vector of the posture of the shield tunneling machine in the next time range to a posture item prediction result of the shield tunneling machine according to the updated weight values, calculating the weighted sum of each element, and taking the weighted sum result of each element as output to obtain the posture item prediction result of the shield tunneling machine.
And S4, performing super-parameter optimization and second training on the trained first long-short-time memory network model and second long-time memory network model to obtain a final first long-short-time memory network model and a final second long-time memory network model, and obtaining a shield tunneling machine attitude item prediction result by using the final first long-time memory network model and second long-time memory network model.
In an optional embodiment of the invention, the super-parameter optimization and the second round of training are performed on the trained first long-short-time memory network model and the trained second long-time memory network model so as to obtain a final first long-time memory network model and a final second long-time memory network model, and the final first long-time memory network model and the final second long-time memory network model are utilized to obtain the gesture item prediction result of the shield tunneling machine.
The invention optimizes the super parameters of the trained first long-short-time memory network model and second long-short-time memory network model, which comprises the following steps:
a1, initializing the number of long-short-time memory network blocks and the number of long-short-time memory network units of the trained model.
A2, training the model with the initialized long-short time memory network block number and the initialized long-short time memory network unit number by using the normalized vector of the gesture parameter and the strong related parameter in the set time range, and determining the model with the minimum loss after parameter initialization.
A3, carrying out parameter variation on the model with the minimum loss after parameter initialization, obtaining models with different parameters, training the models with different parameters by using normalized vectors of attitude parameters and strong related parameters within a set time range, and determining the model with the minimum loss after parameter variation.
The method comprises the steps of performing a second training on the model with the minimum loss after parameter variation until a loss function is not in gradient descent, so as to obtain a final first long-short-time memory network model and a final second long-short-time memory network model, and obtaining a shield tunneling machine attitude item prediction result by using the final first long-short-time memory network model and the final second long-time memory network model.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The principles and embodiments of the present invention have been described in detail with reference to specific examples, which are provided to facilitate understanding of the method and core ideas of the present invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present invention, the present description should not be construed as limiting the present invention in view of the above.
Those of ordinary skill in the art will recognize that the embodiments described herein are for the purpose of aiding the reader in understanding the principles of the present invention and should be understood that the scope of the invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific modifications and combinations from the teachings of the present disclosure without departing from the spirit thereof, and such modifications and combinations remain within the scope of the present disclosure.

Claims (7)

1. The machine learning-based method for predicting the attitude of the shield tunneling machine is characterized by comprising the following steps of:
s1, acquiring attitude parameter data and construction parameter data of a shield tunneling machine, and acquiring trend items and fluctuation items of the attitude parameter data and trend items and fluctuation items of the construction parameter data by adopting a variation modal decomposition method;
s2, calculating a de-trend correlation index of trend items of the attitude parameter data and trend items of the construction parameter data to determine strong correlation parameters, and determining trend items and fluctuation items of the strong correlation parameter data;
step S2 comprises the following sub-steps:
s21, respectively calculating centralized accumulated signals of trend items of the attitude parameter data and trend items of the construction parameter data;
s22, respectively determining a least square linear fitting signal of trend items of the attitude parameter data and trend items of the construction parameter data by adopting a least square method;
s23, determining the fluctuation index of the trend item of the attitude parameter data according to the centralized accumulated signal and the least square straight line fitting signal of the trend item of the attitude parameter data, and determining the fluctuation index of the trend item of the construction parameter data according to the centralized accumulated signal and the least square straight line fitting signal of the trend item of the construction parameter data;
s24, calculating a cross fluctuation index according to the centralized accumulated signal of the trend item of the attitude parameter data, the least square straight line fitting signal, the centralized accumulated signal of the trend item of the construction parameter data and the least square straight line fitting signal;
s25, calculating a de-trend correlation index of the attitude parameter data and the construction parameter data according to the fluctuation index of the trend item of the attitude parameter data, the fluctuation index and the cross fluctuation index of the trend item of the construction parameter data, wherein the de-trend correlation index is expressed as follows:
wherein:for calculating the detrending correlation index of the attitude parameter data and the construction parameter data +.>Sequence number of trend item for gesture parameter data, +.>Sequence number of trend item for construction parameter data, +.>Cross fluctuation index, ++, for trend term of attitude parameter data and trend term of construction parameter data>Fluctuation index of trend term for posture parameter data, +.>A fluctuation index which is a trend term of construction parameter data;
s26, determining strong correlation parameters according to the de-trend correlation indexes of the attitude parameter data and the construction parameter data, and determining trend items and fluctuation items of the strong correlation parameter data;
s3, constructing a first long-short-time memory network model and a second long-short-time memory network model, training the first long-short-time memory network model by using trend items of gesture parameter data and trend items of strong related parameter data, and training the second long-short-time memory network model by using fluctuation items of gesture parameter data and fluctuation items of strong related parameter data;
the built first long-short-time memory network model and the built second long-short-time memory network model comprise an input layer, a long-short-time memory network block and an output layer;
the input layer is used for inputting normalized vectors of the gesture parameters and the strong correlation parameters in a set time range and transmitting the normalized vectors of the gesture parameters and the strong correlation parameters in the set time range to the long-short-time memory network block;
the long-time and short-time memory network block is used for receiving normalized vectors of the attitude parameters and the strong correlation parameters in the set time range, calculating one-dimensional feature vectors of the attitude of the shield tunneling machine in the next time range according to the normalized vectors of the attitude parameters and the strong correlation parameters in the set time range, and transmitting the one-dimensional feature vectors of the attitude of the shield tunneling machine in the next time range to the output layer;
the output layer is used for receiving one-dimensional feature vectors of the shield tunneling machine gesture in the next time range, calculating weights of all elements in the one-dimensional feature vectors of the shield tunneling machine gesture in the next time range on the shield tunneling machine gesture sub-term prediction result, calculating weighted sums of all elements, and outputting the weighted sum result of all elements as an output to obtain the shield tunneling machine gesture sub-term prediction result;
and S4, performing super-parameter optimization and second training on the trained first long-short-time memory network model and second long-time memory network model to obtain a final first long-short-time memory network model and a final second long-time memory network model, and obtaining a shield tunneling machine attitude item prediction result by using the final first long-time memory network model and second long-time memory network model.
2. The machine learning-based shield tunneling machine attitude subitem prediction method according to claim 1, wherein the long-short-time memory network block comprises a long-short-time memory network layer, a regularization layer and a selective inhibition layer which are sequentially connected; the long-short time memory network layer comprises a plurality of long-short time memory network units.
3. The machine learning-based method for predicting the attitude and the subentry of a shield tunneling machine according to claim 2, wherein the long-short-term memory network unit comprises a forgetting gate substructure, an input gate substructure and an output gate substructure.
4. A machine learning based method for predicting attitude of a shield tunneling machine according to claim 3, wherein the calculation formula of the forgetting gate substructure is expressed as:
wherein:is->Output of forgetting gate substructure of individual cell, < >>To activate the function +.>Trainable weights for forgetting door substructure, +.>For a two-vector splice symbol, < >>Is->History parameter of individual cell output,/->Is->Normalized vector of pose parameter and strongly correlated parameter of individual cell,/->Trainable bias for forgetting gate substructure.
5. A machine learning based method for predicting attitude of a shield tunneling machine according to claim 3, wherein the input gate substructure is calculated by the formula:
，
wherein:is->Memory parameters of choice of the input gate substructure of the individual cells,/->To activate the function +.>Trainable weights for selecting memory parameters, +.>For a two-vector splice symbol, < >>Is->History parameter of individual cell output,/->Is->Normalized vector of pose parameter and strongly correlated parameter of individual cell,/->Trainable deviations for the selection of memory parameters +.>Is->Candidate memory parameters of the input gate substructure of the individual cells,/->As an arctangent function, +.>Trainable weights for candidate memory parameters, +.>Trainable deviations for candidate memory parameters.
6. A machine learning based method for predicting attitude of a shield tunneling machine according to claim 3, wherein the output gate mechanism is calculated by the formula:
wherein:is->The output of the output gate mechanism of the individual units, is->To activate the function +.>To output trainable weights of the gate mechanism, < +.>Is in two directionsSpliced sign of quantity->Is->History parameter of individual cell output,/->Is->Normalized vector of pose parameter and strongly correlated parameter of individual cell,/->To output the trainable bias of the gate mechanism.
7. The machine learning-based method for predicting the attitude of a shield tunneling machine according to claim 1, wherein in step S4, performing super-parametric optimization on the trained first long-short-time memory network model and second long-short-time memory network model comprises the following steps:
a1, initializing the number of long-short-time memory network blocks and the number of long-short-time memory network units of the trained model;
a2, training the model with the initialized long-short-time memory network block number and the initialized long-short-time memory network unit number by using the normalized vector of the gesture parameter and the strong correlation parameter in the set time range, and determining the model with the minimum loss after parameter initialization;
a3, carrying out parameter variation on the model with the minimum loss after parameter initialization, obtaining models with different parameters, training the models with different parameters by using normalized vectors of attitude parameters and strong related parameters within a set time range, and determining the model with the minimum loss after parameter variation.
CN202410004533.0A 2024-01-03 2024-01-03 Machine learning-based shield tunneling machine attitude item prediction method Active CN117493837B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410004533.0A CN117493837B (en) 2024-01-03 2024-01-03 Machine learning-based shield tunneling machine attitude item prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410004533.0A CN117493837B (en) 2024-01-03 2024-01-03 Machine learning-based shield tunneling machine attitude item prediction method

Publications (2)

Publication Number Publication Date
CN117493837A CN117493837A (en) 2024-02-02
CN117493837B true CN117493837B (en) 2024-03-19

Family

ID=89674861

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410004533.0A Active CN117493837B (en) 2024-01-03 2024-01-03 Machine learning-based shield tunneling machine attitude item prediction method

Country Status (1)

Country Link
CN (1) CN117493837B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112100841A (en) * 2020-09-09 2020-12-18 中铁二十局集团有限公司 Shield tunneling machine attitude prediction method and device, terminal equipment and storage medium
CN114117599A (en) * 2021-11-22 2022-03-01 中铁高新工业股份有限公司 Shield attitude position deviation prediction method
CN114329810A (en) * 2021-11-16 2022-04-12 中国水利水电科学研究院 Real-time shield tunneling machine working attitude prediction method based on big data
CN114810100A (en) * 2022-06-28 2022-07-29 中铁工程服务有限公司 Shield tunneling attitude prediction method based on deep neural network
CN117272139A (en) * 2023-09-18 2023-12-22 湖北工业大学 Shield tunneling machine attitude prediction method combining characteristic correlation and time correlation

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112100841A (en) * 2020-09-09 2020-12-18 中铁二十局集团有限公司 Shield tunneling machine attitude prediction method and device, terminal equipment and storage medium
CN114329810A (en) * 2021-11-16 2022-04-12 中国水利水电科学研究院 Real-time shield tunneling machine working attitude prediction method based on big data
CN114117599A (en) * 2021-11-22 2022-03-01 中铁高新工业股份有限公司 Shield attitude position deviation prediction method
CN114810100A (en) * 2022-06-28 2022-07-29 中铁工程服务有限公司 Shield tunneling attitude prediction method based on deep neural network
CN117272139A (en) * 2023-09-18 2023-12-22 湖北工业大学 Shield tunneling machine attitude prediction method combining characteristic correlation and time correlation

Also Published As

Publication number Publication date
CN117493837A (en) 2024-02-02

Similar Documents

Publication Publication Date Title
Sehovac et al. Forecasting building energy consumption with deep learning: A sequence to sequence approach
CN111027686B (en) Landslide displacement prediction method, device and equipment
CN109902801A (en) A kind of flood DATA PROCESSING IN ENSEMBLE PREDICTION SYSTEM method based on variation reasoning Bayesian neural network
CN106709640A (en) Method for predicting ship energy consumption based on deep learning and support vector regression
CN111815033A (en) Offshore wind power prediction method based on RCNN and meteorological time sequence characteristics
CN108879732B (en) Transient stability evaluation method and device for power system
CN111030889B (en) Network traffic prediction method based on GRU model
Wu et al. Development of convolutional neural network based Gaussian process regression to construct a novel probabilistic virtual metrology in multi-stage semiconductor processes
CN109471049B (en) Satellite power supply system anomaly detection method based on improved stacked self-encoder
Liu et al. Correntropy long short term memory soft sensor for quality prediction in industrial polyethylene process
CN112879024A (en) Dynamic prediction method, system and equipment for shield attitude
CN112766603A (en) Traffic flow prediction method, system, computer device and storage medium
CN115018193A (en) Time series wind energy data prediction method based on LSTM-GA model
Wei et al. An ensemble multi-step forecasting model for ship roll motion under different external conditions: A case study on the South China Sea
CN112241802A (en) Interval prediction method for wind power
CN117493837B (en) Machine learning-based shield tunneling machine attitude item prediction method
CN113807951A (en) Transaction data trend prediction method and system based on deep learning
CN113393034A (en) Electric quantity prediction method of online self-adaptive OSELM-GARCH model
CN112149896A (en) Attention mechanism-based mechanical equipment multi-working-condition fault prediction method
CN116975775A (en) Deep learning method for coal mine gas outburst prediction based on multi-mode data fusion
CN115794805A (en) Medium-low voltage distribution network measurement data supplementing method
CN115630582A (en) Multi-sliding-window model fused soft rock tunnel surrounding rock deformation prediction method and equipment
Ma et al. A data-driven fault detection approach for unknown large-scale systems based on GA-SVM
CN115719040B (en) Penicillin fermentation process key variable soft measurement method and system
CN116756531B (en) Method and device for monitoring hot continuous rolling process of strip steel for model interaction linkage

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