CN114662782A - Method for predicting instantaneous yield of trailing suction hopper dredger based on LSTM neural network - Google Patents
Method for predicting instantaneous yield of trailing suction hopper dredger based on LSTM neural network Download PDFInfo
- Publication number
- CN114662782A CN114662782A CN202210369799.6A CN202210369799A CN114662782A CN 114662782 A CN114662782 A CN 114662782A CN 202210369799 A CN202210369799 A CN 202210369799A CN 114662782 A CN114662782 A CN 114662782A
- Authority
- CN
- China
- Prior art keywords
- neural network
- lstm neural
- network model
- dredging
- instantaneous yield
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 25
- 238000003062 neural network model Methods 0.000 claims abstract description 104
- 238000012360 testing method Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000005457 optimization Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 22
- 210000004027 cell Anatomy 0.000 claims description 18
- 230000002159 abnormal effect Effects 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 15
- 230000000694 effects Effects 0.000 claims description 12
- 238000011010 flushing procedure Methods 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 10
- 238000001914 filtration Methods 0.000 claims description 9
- 238000011156 evaluation Methods 0.000 claims description 7
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 abstract 1
- 238000010276 construction Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000011217 control strategy Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000002068 genetic effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 125000004432 carbon atom Chemical group C* 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000011112 process operation Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—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
- G06N3/02—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- 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)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Strategic Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Primary Health Care (AREA)
- Manufacturing & Machinery (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a method for predicting the instantaneous yield of a trailing suction hopper dredger based on an LSTM neural network, which comprises the following steps: collecting operating parameters of a trailing suction hopper dredger during dredging and loading; integrating, cleaning and converting data to form a dredging and cabin loading data training and testing sample; initializing an LSTM neural network model; based on a Bayesian adjustment optimization neural network model, obtaining an optimization neural network prediction model through sample training and testing; and inputting the control parameters related to the instantaneous yield of the trailing suction hopper dredger during dredging and loading into the cabin into a prediction model, and predicting to obtain the instantaneous yield of the trailing suction hopper dredger during dredging and loading into the cabin. The method fully considers the influence factors of dredging and loading, optimizes the neural network structure, the storage space, the iteration speed and the stability, and provides reliable parameter basis for the prediction of the instantaneous yield of the trailing suction hopper dredger.
Description
Technical Field
The invention belongs to the technical field of dredging, and particularly relates to a method for predicting instantaneous yield of a trailing suction hopper dredger based on an LSTM neural network model.
Background
The drag suction dredger can independently complete the whole dredging process operations such as digging, loading, transporting, unloading, hydraulic filling and the like, can change the site, is widely applied and dominates the international dredging market. When the trailing suction dredger works, the operation usually depends on the experience of technicians, but in the construction process, the dredging efficiency is not high due to the complicated and changeable working conditions. The dredging project is short in construction time, heavy in task and high in construction period pressure, so that a dredging yield prediction model of the dredger is built, and the loading efficiency is improved at present. The dredger realizes efficient dredging and intelligent dredging through the historical construction data of the trailing suction dredger, and has very important significance for promoting the development of the dredging industry.
In recent years, more and more scholars have conducted intensive research on trailing suction dredge through neural network models. In a paper & lt & gt prediction research on yield of trailing suction dredger based on genetic BP neural network, published by Qian Guangting 2018 in the electronic engineering journal 38 of warship, vol.08, 141 and 145, a model prediction control strategy is used, and optimization control is performed by combining a mode search algorithm; on the basis of establishment of a dredging model, an academic paper entitled "research on-line optimization control strategy for dredging performance of a trailing suction dredger", published by Yangting 2016, in the value engineering journal 35, No. 01, No. 107 and No. 109, model parameters are adjusted by comprehensively applying intelligent algorithms such as a genetic algorithm, a particle filter and the like; in a paper, "drag suction dredger dredging process cabin optimization research", published in 2012 by Sun Jian in the scientific and technical and engineering journal 12, vol.05, 1065 and 1068, extreme learning machines and BP neural network algorithms were used to predict dredging data. However, most of the models have simple structure, poor prediction accuracy, poor fault tolerance and low convergence rate, and cannot effectively process massive data and intuitively reflect the state of instantaneous yield.
Most of traditional neural network models are simple in structure, massive data, especially time series data cannot be effectively processed, and prediction accuracy is often low.
Disclosure of Invention
The invention aims to provide a method for predicting the instantaneous yield of a trailing suction hopper dredger based on an LSTM neural network, aiming at the problems and the defects of the method for predicting the instantaneous yield in the prior art.
The method selects a long-time memory network (LSTM) to model the instantaneous yield of the dredger, trains the network through historical data, and predicts the instantaneous yield of the dredger in a short time. The LSTM is a modified recurrent neural network, has a memory function in a longer time range, and can eliminate the problem of gradient disappearance. In addition, the Bayesian optimization LSTM neural network model is utilized to search the hyper-parameter combination which minimizes the objective function, so that the neural network model can be predicted more accurately. The dredging personnel can predict the instantaneous yield result through the neural network model, and the smooth and efficient dredging is ensured.
In order to achieve the purpose, the invention adopts the following technical scheme to realize.
A method for predicting the instantaneous yield of a trailing suction hopper dredger based on an LSTM neural network comprises the following specific steps:
step one, data acquisition: collecting parameters related to the yield of the dredging and loading of the trailing suction hopper dredger through a sensor on the trailing suction hopper dredger comprises the following steps: the mud pump rotating speed, the high-pressure flushing pump rotating speed, the rake head angle, the navigational speed, the height of an overflow cylinder, the pressure of a wave compensator, the suction vacuum, the cross-sectional area of a pipeline, the mud pump density and the mud pump flow speed parameters are drawn into a mud dredging and loading data set of the drag suction dredger;
step two, data preprocessing: removing abnormal values in the collected parameter data set by adopting a triple standard deviation method, directly deleting missing values, carrying out min-max normalization and haar wavelet filtering processing; the data set is divided into 8: 2, forming a dredging cabin loading data training set and a testing set by division;
step three, building an LSTM neural network model, building the LSTM neural network model by adopting Python language, setting a model structure and parameter initial values and substituting the data set in the step two into the LSTM neural network model, wherein the LSTM neural network model comprises an input layer, an output layer and an intermediate layer;
optimizing, training and testing an LSTM neural network model prediction model, optimizing the LSTM neural network model in the third step through a Bayes algorithm, and training and testing the model by using a dredging and cabin-loading data set to obtain a Bayes-LSTM neural network prediction model;
and fifthly, evaluating the LSTM neural network model in the third step and the Bayes-LSTM neural network model in the fourth step according to the fitness evaluation index in the prediction to obtain a neural network model with a good prediction effect.
Further, the specific content, method and steps of the step one are as follows:
s11, collecting parameters related to instantaneous yield of dredging and loading of the trailing suction hopper dredger through a sensor of the trailing suction hopper dredger, namely collecting the rotating speed of a dredge pump and the rotating speed of a high-pressure flushing pump through a rotating speed sensor, collecting the angle of a rake head through an angle sensor, collecting the navigational speed through a marine log, collecting the height of an overflow cylinder through an overflow weir position sensor, collecting the pressure of a wave compensator through a pressure sensor, collecting the suction vacuum through a vacuum pressure sensor, collecting the density of a left dredge pump and the density of a right dredge pump through a densimeter, collecting the flow rate of the left dredge pump and the flow rate of the right dredge pump through a flowmeter, and counting the cross-sectional area of a pipeline through drawing data;
s12, calculating the instantaneous yield of the trailing suction hopper dredger during dredging and loading into the cabin according to the following formula:
R＝A·(ρ1·ν1+ρ2·ν2)
wherein A represents the cross-sectional area of the pipe (m)2) Is a constant of 1; rho1,ρ2(kg/m3) Respectively shows the density of the left mud pumpRight dredge pump density; v. of1,v2(m/s) respectively representing the flow rate of the left dredge pump and the flow rate of the right dredge pump, wherein the optimal dredging efficiency corresponds to the maximum instantaneous dredging yield R (kg/s);
and S13, drawing the acquired parameters and the instantaneous dredging output data of the drag suction dredger during dredging and loading into the cabin into a data set of the drag suction dredger during dredging and loading into the cabin.
Further, the specific content, method and steps of step two are as follows:
s21, removing abnormal values: observing the integral step-by-step condition of data in a data set through a box diagram, obtaining abnormal data by calculating an upper boundary, a lower boundary, a 25% quantile and a 75% quantile, wherein the abnormal data are data outside the upper boundary and the lower boundary, and removing by adopting a triple standard deviation method, wherein the formula is as follows:
s22, missing value processing: deleting the missing values in the data set after the abnormal values are removed;
s23, min-max normalization processing: in order to eliminate the influence of dimension, the data set after the treatment is subjected to normalization treatment by adopting min-max normalization. Wherein, the normalized calculation formula is as follows:
wherein, XmaxIs the maximum value in the data set, XminIs the minimum value of the data set, XnormIs the result of the normalization;
s24, haar wavelet filtering: filtering the mud pump rotating speed, the high-pressure flushing pump rotating speed, the drag head angle, the navigational speed, the height of an overflow cylinder, the pressure of a wave compensator and the suction vacuum in the processed data set by using haar wavelets;
s25, dividing the processed data set into a training set and a test set, wherein the division ratio is 8: 2.
further, the specific content, method and steps of step three are as follows:
s31, taking the rotating speed of a mud pump, the rotating speed of a high-pressure flushing pump, the angle of a drag head, the navigational speed, the height of an overflow cylinder, the pressure of a wave compensator and the suction vacuum as input layer neurons;
s32, building an LSTM neural network model by Python language, setting the number of nodes of the LSTM neural network model to be 50, the learning rate to be 0.01 and setting the number of model iterations epoch to be 100;
first, the cell output h at the previous time (t-1)t-1Last cell state Ct-1And input x at this time (t)tInput to the forgetting gate according to ht-1And xtSelectively extracting the ground information in the memory unit, forgetting the output function f of the gatetThe expression is as follows:
ft＝σ(W[ht-1,xt])+b
wherein, σ is sigmoid function, W is weight of forgetting gate, b is bias weight of forgetting gate, and [ ] represents splicing two vectors.
The data processed by the forgetting gate comes to the updating gate, and the updating gate is according to ht-1And xtAdding new information into the memory unit to obtain information i to be memorizedtAnd candidate memory cell for refresh memory cell
The expression is as follows:
wherein, WiFor information to be memorized, WcIs the weight of the memory cell, biFor information to be memorized, bcIs the bias of the memory cell.
Output C after memory cell refreshtThe expression is as follows:
the data processed by the update gate comes to the output gate, the output function o of whichtAnd then htThe expression is as follows:
ot＝σ(Wo[ht-1,xt])+bo
ht＝ot·tanhCt
at this time, the predicted instantaneous yield of the LSTM neural network model is output, namely the instantaneous yield is used as the neuron of the output layer.
Further, the specific content, method and steps of step four are as follows:
s41, the basic theorem of the Bayesian algorithm is as follows:
where f is the unknown, p (D | f) is the likelihood distribution of f, p (f) is the prior probability of f, and p (D) is the marginal likelihood distribution of f. D { (x)1,y1),(x2,y2),…(xn,yn) Is a candidate set of hyper-parameters and observations, xiFor observed hyperparameters, yi＝f(xi) + xi is an observed value, xi is a trade-off scalar, and the objective function fi＝f(xi) The objective function is the absolute error in the LSTM neural network model, and the idea of Bayesian optimization is to use the objective function f (x)i) The posterior probability is selected so that f (x)i) The number of the carbon atoms is reduced to the minimum,
wherein x isi *Expressing the parameter value for minimizing the objective function, wherein X is equal to X;
and S42, substituting the optimal parameter combination obtained in S41 into a Bayes-LSTM neural network model, inputting a training set in a data set to train the Bayes-LSTM neural network model, inputting a test set in the data set into the Bayes-LSTM neural network model to obtain a result, and outputting the result as the instantaneous yield predicted by the Bayes-LSTM neural network model.
Further, the evaluation in step five includes two methods, one, by determining the coefficient R2And (4) judging: modeling R of LSTM neural network2R with Bayes-LSTM neural network model2By comparison, when R is2The neural network model with a large numerical value is a neural network model with a good prediction effect; and secondly, visually evaluating the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model and the instantaneous yield value R of the dredging and binning of the trailing suction hopper dredger calculated in the step one S12 through drawing pictures, wherein the prediction effect of the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model is better when the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model are more consistent with the instantaneous yield value R of the dredging and binning of the trailing suction hopper dredger calculated in the step one S12.
Compared with the prior art, the method has the advantages that the instantaneous yield prediction model is established through the Bayes-LSTM neural network model through the historical construction data of the trailing suction hopper dredger, the characteristic of poor prediction precision of the LSTM neural network model is solved, the development of the dredging industry has very important significance, and the method has the following advantages and beneficial effects:
1) the operation experience of constructors selects control parameters and output quantity, and the mechanism model and mathematical drive are combined, so that the consideration is more comprehensive;
2) the LSTM neural network model solves the problem of dependence between data, and meanwhile, the Bayesian optimization algorithm is adopted for optimization, so that the prediction precision is greatly improved.
3) The method efficiently utilizes the data resources of the operation of the trailing suction hopper dredger and provides reliable basis for the prediction of the instantaneous yield. According to the method, a Bayes-LSTM neural network model is established, historical data of the trailing suction hopper dredger are used for testing, instantaneous yield is used as a prediction variable, the importance degree of influencing the instantaneous yield is reflected through correlation analysis, goodness of fit is used as an evaluation standard, and the prediction of the instantaneous yield of the trailing suction hopper dredger at a certain moment in the future is finally realized.
Drawings
FIG. 1 is a flow chart of a method for predicting the instantaneous yield of a mud excavation binning based on an optimized neural network of the present invention;
FIG. 2 is a block diagram of a data set formed by data collected by a sensor according to the steps of the present invention;
FIG. 3 is a flow chart of a Bayesian optimization LSTM neural network prediction model of the present invention.
FIG. 4 is a graph of the prediction results of the LSTM, Bayes-LSTM neural network prediction models of the present invention.
Detailed Description
In order to make the technical solutions of the present invention more clearly understood by those skilled in the art, the present invention is fully described below with reference to the following embodiments. Those skilled in the art can make modifications to the invention based on the embodiments disclosed in the present application without departing from the principle of the invention, and such modifications are also protected by the following claims.
The invention aims to predict the instantaneous yield of a specific time point according to parameters related to the instantaneous yield in the historical data of the operation of the trailing suction hopper dredger by establishing a Bayes-LSTM neural network model, thereby providing related technical support for dredging engineering.
As shown in FIG. 1, the method for predicting the instantaneous yield of the trailing suction hopper dredger based on the LSTM neural network comprises the following specific steps:
the method comprises the following steps: data acquisition: the method comprises the following steps of acquiring control parameters related to the yield of the dredging cabin of the trailing suction hopper dredger through a sensor on the trailing suction hopper dredger, wherein the control parameters comprise: dredge pump speed, high-pressure flushing pump speed, drag head angle, navigational speed, overflow cylinder height, wave compensator pressure, suction vacuum. And the parameters of the cross section area of the pipeline, the density of the dredge pump and the flow speed of the dredge pump are collected. The specific method and steps are as follows:
s11, the method adopts operation data of Xinhaihu at Yangtze river from 2018, 8, 23 days and 2018, 8, 25 days, collects control parameters related to instantaneous yield of dredging and cabin loading of a trailing suction hopper dredger from a sensor of the trailing suction hopper dredger according to actual requirements, collects dredge pump rotation speed and high-pressure flushing pump rotation speed through a rotation speed sensor, collects rake head angles through an angle sensor, collects navigation speed through a marine log, collects overflow cylinder height through an overflow weir position sensor, collects pressure of a wave compensator through a pressure sensor, collects suction vacuum through a vacuum pressure sensor, collects left dredge pump density and right dredge pump density through a densimeter, collects left dredge pump flow speed and right dredge pump flow speed through a flowmeter, and counts the cross-sectional area of a pipeline through a graph of the Xinhaihu.
S12, calculating the instantaneous yield of the trailing suction hopper dredger during dredging and loading into the cabin according to the following formula:
R＝A·(ρ1·ν1+ρ2·ν2)
wherein A represents the cross-sectional area of the pipe (m)2) Is a constant of 1; rho1,ρ2(kg/m3) Respectively representing the density of a left mud pump and the density of a right mud pump; v. of1,v2(m/s) represents the left and right dredge pump flow rates, respectively, with the optimum dredging efficiency corresponding to the maximum instantaneous dredging yield R (kg/s).
And S13, drawing the acquired parameters and the instantaneous dredging output data of the drag suction dredger during dredging and loading into the cabin into a drag suction dredger dredging and loading data set (as shown in figure 2).
Step two, data preprocessing: removing abnormal values in a data set drawn by the collected data by adopting a triple standard deviation method, directly deleting missing values, and carrying out min-max normalization and haar wavelet filtering processing; the data set is divided into 8: 2, forming a dredging cabin loading data training set and a testing set by division; the specific method and steps are as follows:
s21, removing abnormal values: observing the integral step-by-step condition of data in a data set through a box diagram, obtaining abnormal data by calculating an upper boundary, a lower boundary, a 25% quantile and a 75% quantile, wherein the abnormal data are data outside the upper boundary and the lower boundary, and removing by adopting a three-time standard difference method, wherein the formula is as follows:
s22, missing value processing: deleting missing values in the data set after the abnormal values are removed;
s23, min-max normalization processing: in order to eliminate the influence of dimension, the data set after the treatment is subjected to normalization treatment by adopting min-max normalization. Wherein, the normalized calculation formula is as follows:
wherein, XmaxIs the maximum value in the data set, XminIs the minimum value of the data set, XnormIs the result of the normalization.
S24, haar wavelet filtering: and filtering the mud pump rotating speed, the high-pressure flushing pump rotating speed, the drag head angle, the navigational speed, the height of the overflow cylinder, the pressure of the wave compensator and the suction vacuum in the processed data set by using haar wavelets.
S25, dividing the processed data set into a training set and a test set, wherein the division ratio is 8: 2.
and step three, building an LSTM neural network model, and building the LSTM neural network model by adopting Python language. The LSTM neural network model includes an input layer, an output layer, and an intermediate layer. Setting the model structure and parameter initial values and substituting the data set in the step two into the LSTM neural network model. The specific method and steps are as follows:
s31, taking the rotating speed of a mud pump, the rotating speed of a high-pressure flushing pump, the angle of a drag head, the navigational speed, the height of an overflow cylinder, the pressure of a wave compensator and the suction vacuum as input layer neurons;
s32, building an LSTM neural network model by Python language, setting the number of nodes of the LSTM neural network model to be 50, the learning rate to be 0.01, and setting the number of model iterations epoch to be 100.
First, the cell output h at the previous time (t-1)t-1Last cell state Ct-1And input x at this time (t)tInput to the forgetting gate according to ht-1And xtSelectively fetching information in a memory cellOutput function f of forgetting gatetThe expression is as follows:
ft＝σ(W[ht-1,xt])+b
wherein, σ is sigmoid function, W is weight of forgetting gate, b is bias weight of forgetting gate, and [ ] represents splicing two vectors.
The data processed by the forgetting gate comes to the updating gate, and the updating gate is based on ht-1And xtAdding new information into the memory unit to obtain information i to be memorizedtAnd candidate memory cell for refresh memory cell
The expression is as follows:
wherein, WiFor information to be memorized, WcIs the weight of the memory cell, biFor information to be memorized, bcIs the bias of the memory cell.
Output C after memory cell refreshtThe expression is as follows:
the data processed by the update gate comes to the output gate, the output function o of whichtAnd then htThe expression is as follows:
ot＝σ(Wo[ht-1,xt])+bo
ht＝ot·tanh Ct
at this time, the predicted instantaneous yield of the LSTM neural network model is output, namely the instantaneous yield is used as the neuron of the output layer.
And step four, optimizing, training and testing the prediction model of the LSTM neural network model. Optimizing the LSTM neural network model in the third step by a Bayesian algorithm, and training and testing the model by using a dredging and cabin-loading data set to obtain a Bayes-LSTM neural network prediction model, wherein the specific method comprises the following steps:
s41, the basic theorem of the Bayesian algorithm is as follows:
where f is the unknown, p (D | f) is the likelihood distribution of f, p (f) is the prior probability of f, and p (D) is the marginal likelihood distribution of f. D { (x)1,y1),(x2,y2),…(xn,yn) Is a candidate set of hyper-parameters and observations, xiFor observed hyperparameters, yi＝f(xi) + xi is an observed value, xi is a trade-off scalar, and the objective function fi＝f(xi) The objective function is the absolute error in the LSTM neural network model. The idea of Bayesian optimization is to utilize an objective function f (x)i) The posterior probabilities are selected so that f (x)i) To a minimum.
Wherein x isi *Expressed as the parameter value that minimizes the objective function, where X ∈ X.
And S42, substituting the optimal parameter combination obtained in S41 into a Bayes-LSTM neural network model, inputting a training set in a data set to train the Bayes-LSTM neural network model, and inputting a test set in the data set into the Bayes-LSTM neural network model to obtain a result. The output at this time is the predicted instantaneous yield of the Bayes-LSTM neural network model.
And fifthly, evaluating the LSTM neural network model in the third step and the Bayes-LSTM neural network model in the fourth step according to the fitness evaluation index in the prediction to obtain a neural network model with a good prediction effect. Includes two methods, method 1, by determining the coefficient R2And (4) judging: r of LSTM neural network model2R with Bayes-LSTM neural network model2The comparison is carried out in such a way that,when R is2The neural network model with a large numerical value is a neural network model with a good prediction effect; the method 2 is that the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model and the instantaneous yield value R of the drag suction dredger for loading into the tank calculated in the step one S12 are visually evaluated through picture drawing, and the prediction effect of the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model is better when the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model are more consistent with the instantaneous yield value R of the drag suction dredger for loading into the tank calculated in the step one S12. The specific method and steps are as follows:
s51, determining the coefficient R in the method 12The calculation formula of (c) is as follows:
wherein SST is the sum of the total squares, SSR is the sum of the regression squares, SSE is the sum of the residual squares, and the specific expression is as follows:
wherein, let y be the true value, i.e. the instantaneous yield value R calculated by step 1,
is the average value of the values of y,
the predicted value is the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model.
S52, writing method 1 according to the above formula by python language to obtain R of LSTM neural network model2And R of Bayes-LSTM neural network model2. The results are shown in Table 1, and from Table 1, the R of the LSTM neural network model can be seen295.33% R of Bayes-LSTM neural network model297.75%, 97.75％>95.33 percent, and the Bayes-LSTM neural network model has better prediction effect.
TABLE 1 evaluation index
| Algorithm | Goodness of fit |
| LSTM | 95.33％ |
| Bayes-LSTM | 97.75％ |
S53, drawing by adopting python language, and drawing the instantaneous yield value predicted by the instantaneous yield value R, LSTM neural network model calculated in the step 1 and the instantaneous yield value predicted by the Bayes-LSTM neural network model. The generated prediction graph is shown in FIG. 4, where the abscissa is the sample point and the ordinate is the instantaneous yield. Through visual evaluation, the instantaneous yield value calculated in the step I S12 of the instantaneous yield predicted by the Bayes-LSTM neural network model is more consistent, and the prediction effect of the Bayes-LSTM neural network model is better.
According to the invention, the Bayes-LSTM neural network model has more excellent overall prediction effect and relatively stable performance, and has better application value in predicting the instantaneous yield of the trailing suction hopper dredger during dredging and loading.
Claims (6)
1. A method for predicting the instantaneous yield of a trailing suction hopper dredger based on an LSTM neural network is characterized by comprising the following steps: the method comprises the following specific steps:
step one, data acquisition: collecting parameters related to the yield of the dredging and loading of the trailing suction hopper dredger through a sensor on the trailing suction hopper dredger comprises the following steps: the method comprises the following steps of (1) drawing a dredging and cabin loading data set of a drag suction dredger according to the parameters of dredge pump rotating speed, high-pressure flushing pump rotating speed, drag head angle, navigational speed, overflow cylinder height, wave compensator pressure, suction vacuum, pipeline cross-sectional area, dredge pump density and dredge pump flow speed;
step two, data preprocessing: removing abnormal values in the collected parameter data set by adopting a triple standard deviation method, directly deleting missing values, carrying out min-max normalization and haar wavelet filtering processing; the data set is divided into 8: 2, forming a dredging cabin loading data training set and a testing set by division;
step three, building an LSTM neural network model, building the LSTM neural network model by adopting Python language, setting a model structure and parameter initial values and substituting the data set in the step two into the LSTM neural network model, wherein the LSTM neural network model comprises an input layer, an output layer and an intermediate layer;
optimizing, training and testing an LSTM neural network model prediction model, optimizing the LSTM neural network model in the third step through a Bayes algorithm, and training and testing the model by using a dredging and cabin-loading data set to obtain a Bayes-LSTM neural network prediction model;
and fifthly, evaluating the LSTM neural network model in the third step and the Bayes-LSTM neural network model in the fourth step according to the fitness evaluation index in the prediction to obtain a neural network model with a good prediction effect.
2. The method for predicting the instantaneous yield of the trailing suction hopper dredger based on the LSTM neural network as claimed in claim 1, wherein: the specific content, method and steps of the first step are as follows:
s11, collecting parameters related to instantaneous yield of dredging and loading of the trailing suction hopper dredger through a sensor of the trailing suction hopper dredger, namely collecting the rotating speed of a dredge pump and the rotating speed of a high-pressure flushing pump through a rotating speed sensor, collecting the angle of a rake head through an angle sensor, collecting the navigational speed through a marine log, collecting the height of an overflow cylinder through an overflow weir position sensor, collecting the pressure of a wave compensator through a pressure sensor, collecting the suction vacuum through a vacuum pressure sensor, collecting the density of a left dredge pump and the density of a right dredge pump through a densimeter, collecting the flow rate of the left dredge pump and the flow rate of the right dredge pump through a flowmeter, and counting the cross-sectional area of a pipeline through drawing data;
s12, calculating the instantaneous yield of the drag suction dredger during dredging and loading the cabin according to the following formula:
R＝A·(ρ1·ν1+ρ2·ν2)
wherein A represents the cross-sectional area of the pipe (m)2) Is a constant of 1; rho1,ρ2(kg/m3) Respectively representing the density of a left mud pump and the density of a right mud pump; v. of1,v2(m/s) respectively representing the flow rate of the left dredge pump and the flow rate of the right dredge pump, wherein the optimal dredging efficiency corresponds to the maximum instantaneous dredging yield R (kg/s);
and S13, drawing the acquired parameters and the instantaneous dredging output data of the drag suction dredger during dredging and loading into the cabin into a data set of the drag suction dredger during dredging and loading into the cabin.
3. The method for predicting the instantaneous yield of the trailing suction hopper dredger based on the LSTM neural network as claimed in claim 1, wherein: the specific content, method and steps of the second step are as follows:
s21, removing abnormal values: observing the integral step-by-step condition of data in a data set through a box diagram, obtaining abnormal data by calculating an upper boundary, a lower boundary, a 25% quantile and a 75% quantile, wherein the abnormal data are data outside the upper boundary and the lower boundary, and removing by adopting a three-time standard difference method, wherein the formula is as follows:
s22, missing value processing: deleting the missing values in the data set after the abnormal values are removed;
s23, min-max normalization processing: in order to eliminate the influence of dimension, the data set after the processing is normalized by adopting min-max normalization, wherein the calculation formula of the normalization is as follows:
wherein, XmaxIs the maximum value in the data set, XminIs the minimum value of the data set, XnormIs the result of the normalization;
s24, haar wavelet filtering: filtering the mud pump rotating speed, the high-pressure flushing pump rotating speed, the drag head angle, the navigational speed, the height of an overflow cylinder, the pressure of a wave compensator and the suction vacuum in the processed data set by using haar wavelets;
s25, dividing the processed data set into a training set and a test set, wherein the division ratio is 8: 2.
4. the LSTM neural network-based instantaneous yield prediction method for the trailing suction hopper dredger according to claim 1, characterized in that: the concrete content, method and steps of the third step are as follows:
s31, taking the rotating speed of a mud pump, the rotating speed of a high-pressure flushing pump, the angle of a drag head, the navigational speed, the height of an overflow cylinder, the pressure of a wave compensator and the suction vacuum as input layer neurons;
s32, building an LSTM neural network model by Python language, setting the number of nodes of the LSTM neural network model to be 50, the learning rate to be 0.01 and setting the number of model iterations epoch to be 100;
first, the cell output h at the previous time (t-1)t-1Last cell state Ct-1And input x at this time (t)tInput to the forgetting gate according to ht-1And xtSelectively extracting the ground information in the memory unit, forgetting the output function f of the gatetThe expression is as follows:
ft＝σ(W[ht-1,xt])+b
wherein sigma is a sigmoid function, W is the weight of a forgetting gate, b is the bias weight of the forgetting gate, and [ ] represents that two vectors are spliced;
the data processed by the forgetting gate comes to the updating gate, and the updating gate is according to ht-1And xtNew to memory cellAdding information to obtain information i needing to be memorizedtAnd candidate memory cell for refresh memory cell
The expression is as follows:
wherein, WiFor information to be memorized, WcIs the weight of the memory cell, biFor information to be memorized, bcIs the bias of the memory cell;
output C after memory cell refreshtThe expression is as follows:
the data processed by the update gate comes to the output gate, the output function o of whichtAnd then htThe expression is as follows:
ot＝σ(Wo[ht-1,xt])+bo
ht＝ot·tanhCt
at this time, the predicted instantaneous yield of the LSTM neural network model is output, namely the instantaneous yield is used as the neuron of the output layer.
5. The method for predicting the instantaneous yield of the trailing suction hopper dredger based on the LSTM neural network as claimed in claim 1, wherein: the concrete content, method and step of step four are as follows:
s41, the basic theorem of the Bayesian algorithm is as follows:
where f is the unknown, p (Df) is the likelihood distribution of f p (f) is the prior probability of f,p (D) is the marginal likelihood distribution of f, D { (x)1,y1),(x2,y2),…(xn,yn) Is a candidate set of hyper-parameters and observations, xiFor observed hyperparameters, yi＝f(xi) + xi is an observed value, xi is a trade-off scalar, and the objective function fi＝f(xi) The objective function is the absolute error in the LSTM neural network model, and the idea of Bayesian optimization is to use the objective function f (x)i) The posterior probability is selected so that f (x)i) The number of the grooves is reduced to the minimum,
wherein x isi *Expressing the parameter values for minimizing the objective function, wherein X ∈ X;
and S42, substituting the optimal parameter combination obtained in S41 into a Bayes-LSTM neural network model, inputting a training set in a data set to train the Bayes-LSTM neural network model, inputting a test set in the data set into the Bayes-LSTM neural network model to obtain a result, and outputting the result as the instantaneous yield predicted by the Bayes-LSTM neural network model.
6. The LSTM neural network-based instantaneous yield prediction method for the trailing suction hopper dredger according to claim 1, characterized in that: the method for evaluating in the step five comprises the following two methods,
method one, by determining the coefficient R2And (4) judging: modeling R of LSTM neural network2Compared with R2 of Bayes-LSTM neural network model, when R2 is a large-value neural network model, the neural network model is good in prediction effect;
and secondly, visually evaluating the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model and the instantaneous yield value R of the dredging and binning of the trailing suction hopper dredger calculated in the step one S12 through drawing pictures, wherein the prediction effect of the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model is better when the instantaneous yield value predicted by the LSTM neural network model and the instantaneous yield value predicted by the Bayes-LSTM neural network model are more consistent with the instantaneous yield value R of the dredging and binning of the trailing suction hopper dredger calculated in the step one S12.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210369799.6A CN114662782A (en) | 2022-04-08 | 2022-04-08 | Method for predicting instantaneous yield of trailing suction hopper dredger based on LSTM neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210369799.6A CN114662782A (en) | 2022-04-08 | 2022-04-08 | Method for predicting instantaneous yield of trailing suction hopper dredger based on LSTM neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114662782A true CN114662782A (en) | 2022-06-24 |
Family
ID=82035006
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210369799.6A Pending CN114662782A (en) | 2022-04-08 | 2022-04-08 | Method for predicting instantaneous yield of trailing suction hopper dredger based on LSTM neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114662782A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115600746A (en) * | 2022-10-24 | 2023-01-13 | 哈尔滨工程大学（Cn） | Convolutional neural network-based drag suction ship energy efficiency prediction method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113642801A (en) * | 2021-08-20 | 2021-11-12 | 江苏科技大学 | Cutter suction dredger yield prediction method based on LSTM |
| CN114065639A (en) * | 2021-11-19 | 2022-02-18 | 江苏科技大学 | Closed-loop real-time inversion method for construction parameters of dredger |
| CN114091339A (en) * | 2021-11-24 | 2022-02-25 | 江苏科技大学 | Method for predicting bow blowing instantaneous yield of drag suction dredger based on GRU |
-
2022
- 2022-04-08 CN CN202210369799.6A patent/CN114662782A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113642801A (en) * | 2021-08-20 | 2021-11-12 | 江苏科技大学 | Cutter suction dredger yield prediction method based on LSTM |
| CN114065639A (en) * | 2021-11-19 | 2022-02-18 | 江苏科技大学 | Closed-loop real-time inversion method for construction parameters of dredger |
| CN114091339A (en) * | 2021-11-24 | 2022-02-25 | 江苏科技大学 | Method for predicting bow blowing instantaneous yield of drag suction dredger based on GRU |
Non-Patent Citations (2)
| Title |
|---|
| 石怀涛: "基于贝叶斯优化的SWDAE-LSTM滚动轴承早期故障预测方法研究", 《振动与冲击》, vol. 40, no. 18, 30 September 2021 (2021-09-30), pages 1 - 12 * |
| 雷松博: "面向电网负荷频率控制特性的监控分析系统设计与实现", 《硕士电子期刊》, 31 January 2021 (2021-01-31), pages 26 - 31 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115600746A (en) * | 2022-10-24 | 2023-01-13 | 哈尔滨工程大学（Cn） | Convolutional neural network-based drag suction ship energy efficiency prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110195592B (en) | Intelligent shield tunneling pose prediction method and system based on hybrid deep learning | |
| Bai et al. | Data mining approach to construction productivity prediction for cutter suction dredgers | |
| CN112183993B (en) | TBM (tunnel boring machine) tunneling control parameter intelligent prediction and optimization decision method | |
| CN110826790A (en) | Intelligent prediction method for construction productivity of cutter suction dredger | |
| CN113006188B (en) | Excavator staged power matching method based on LSTM neural network | |
| CN114662782A (en) | Method for predicting instantaneous yield of trailing suction hopper dredger based on LSTM neural network | |
| CN114810100B (en) | Shield tunneling attitude prediction method based on deep neural network | |
| CN115310532A (en) | Basin multipoint prediction early warning method based on space-time association mixed deep learning model | |
| CN112131794A (en) | Hydraulic structure multi-effect optimization prediction and visualization method based on LSTM network | |
| CN114661006A (en) | Optimized control system and method for cabin pumping and bank blowing process of trailing suction hopper dredger | |
| CN115983488A (en) | Method and device for predicting water inflow of mine | |
| CN115563863A (en) | Method and terminal for determining abrasion loss of slurry shield hobbing cutter in real time based on deep learning | |
| CN114091339A (en) | Method for predicting bow blowing instantaneous yield of drag suction dredger based on GRU | |
| CN113642801A (en) | Cutter suction dredger yield prediction method based on LSTM | |
| CN113156492B (en) | Real-time intelligent early warning method applied to TBM tunnel rockburst disasters | |
| Liu et al. | Predicting variation of multipoint earth pressure in sealed chambers of shield tunneling machines based on hybrid deep learning | |
| Fu et al. | Productivity estimation of cutter suction dredger operation through data mining and learning from real-time big data | |
| CN117521511A (en) | Granary temperature prediction method based on improved wolf algorithm for optimizing LSTM | |
| CN117422165A (en) | Urban water delivery system water quantity prediction method and system based on low carbon emission | |
| CN111485896A (en) | Multi-variable fusion control method and system for slurry air pressure balance shield | |
| CN115952590A (en) | Shield tunnel optimization design method and device based on BO-RF-MOMPA hybrid algorithm | |
| CN115563475A (en) | Pressure soft sensor of excavator hydraulic system | |
| Huang et al. | Prediction and feature importance of earth pressure in shields using machine learning algorithms | |
| CN113762635B (en) | EPB tunnel construction earth surface settlement full-period self-adaptive fuzzy reasoning prediction method | |
| CN117669008B (en) | Foundation settlement prediction method and system based on deep learning |
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 |