CN113393027A - Navigation mark drift intelligent prediction method based on deep learning - Google Patents
Navigation mark drift intelligent prediction method based on deep learning Download PDFInfo
- Publication number
- CN113393027A CN113393027A CN202110648282.6A CN202110648282A CN113393027A CN 113393027 A CN113393027 A CN 113393027A CN 202110648282 A CN202110648282 A CN 202110648282A CN 113393027 A CN113393027 A CN 113393027A
- Authority
- CN
- China
- Prior art keywords
- navigation mark
- data
- time sequence
- navigation
- drift
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000013135 deep learning Methods 0.000 title claims abstract description 22
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 38
- 230000007246 mechanism Effects 0.000 claims abstract description 17
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 230000033001 locomotion Effects 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000002159 abnormal effect Effects 0.000 claims description 8
- 238000010586 diagram Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 230000005856 abnormality Effects 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 125000004122 cyclic group Chemical group 0.000 description 3
- 238000006073 displacement reaction Methods 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 238000010219 correlation analysis Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000007667 floating Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 206010027175 memory impairment Diseases 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 239000003643 water by type Substances 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Game Theory and Decision Science (AREA)
- Molecular Biology (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Educational Administration (AREA)
- Remote Sensing (AREA)
Abstract
The invention provides an intelligent navigation mark drift prediction method based on deep learning, which relates to the technical field of navigation mark drift prediction and comprises the following steps: acquiring sample data and preprocessing the sample data; the sample data comprises navigation mark historical position data and water level historical data; constructing a time sequence based on the preprocessed sample data; establishing and training a navigation mark drift prediction model by using a navigation mark motion rule reflected by the time sequence; the navigation mark drift prediction model comprises a gating cycle unit for extracting time sequence characteristics in a time sequence and an attention mechanism for giving different weights to each time step of the time sequence; inputting preset moment data of the navigation mark, and predicting by using a trained navigation mark drift prediction model; and outputting the navigation mark predicted position. The time sequence characteristics are learned from a large amount of navigation marks and water level historical data based on a deep learning method, and the model prediction precision is further improved by considering the influence of water level factors and adding an attention mechanism.
Description
Technical Field
The invention relates to the technical field of navigation mark drift prediction, in particular to an intelligent navigation mark drift prediction method based on deep learning.
Background
A navigation mark is a facility or system that helps guide a vessel through, locate, and mark obstructions and artificial signs that indicate warnings, providing safety information for various water activities. Are often located in or near navigable waters to mark the location of fairway, anchor, beach and other obstructions. The buoy is a navigation mark floating on the water surface, has the largest quantity in the navigation mark and is widely applied. Buoys are usually arranged along the boundary of a channel, are tied to the water surface by tie rods and anchor chains, are affected by natural and artificial factors such as the rise and fall of damp water, the collision of passing ships, the self failure of a tying device and the like, can generate a deviation phenomenon, and directly affect the dimension of the channel. If the buoy is too large in position deviation and cannot be timely detected, wrong navigation information is sent to the passing ship, and navigation safety is threatened.
In recent years, deep learning techniques have been rapidly developed and widely used in the fields of numerical prediction, image recognition, and the like. In the field of numerical prediction, theories and methods based on a recurrent neural network are excellent in performance, strong in data adaptability, capable of capturing connection among multiple characteristics, good in short-term prediction and common in time series prediction problems. The digital channel telemetering and remote control system collects accumulated navigation mark position GPS observation data and water level data for a long time and is also typical time sequence data. A large amount of historical observation data provides data support for deeply analyzing factors influencing navigation mark drifting and mining the rule of navigation mark drifting motion.
If the latest deep learning method can be utilized, and the movement rule of the navigation mark drift is combined, a high-applicability navigation mark drift prediction method is constructed, the drift position and the drift distance of the navigation mark are accurately and quickly predicted, early warning is made in advance, and the method has important significance in the aspects of navigation ship route planning, anti-collision and the like.
At present, the conventional navigation mark drift prediction method includes:
journal articles: the study on the drift characteristics of inland navigation marks based on Kalman filtering and K-means + + algorithm (Zhou Yu Meng, first beautiful people, Jiang Zhong Lian, Zhong Cheng.
The paper proposes that Kalman filtering is used for preprocessing the navigation mark data, a navigation mark clustering center is obtained by combining a K-means + + algorithm, an inland navigation mark drifting-water level secondary fitting model is established by adopting a correlation analysis method, the navigation mark drifting distance can be estimated according to the water level value, and the model can better reflect the influence of the water level of the midstream river section of the Yangtze river on the navigation mark drifting.
This method has the following disadvantages: the kalman filter algorithm and the K-means + + algorithm (the flow chart is shown in fig. 1) can only process historical data, cannot perform real-time calculation, and the rationality of taking the clustering center as a reference position is yet to be verified; the constructed linear regression model is simple in structure, and large errors exist in prediction.
Journal articles: navigation mark drift calculation method research under the action of tidal current field (Zhou Chui, Zhao Jun Man, Gansu, xu shod people, Xucaiyun. navigation mark drift calculation method research under the action of tidal current field [ J ]. safety and environmental bulletin, 2021,21(01): 217-.
The paper proposes that navigation mark telemetering data is preprocessed through Layida mathematical criteria, the preprocessed data is clustered by combining a K-means + + algorithm and an ISODATA algorithm, a clustering center with higher accuracy is selected as a reference point for calculating navigation mark drift amount, and a navigation mark drift distance is calculated. The navigation mark drifting model under the action of the tidal flow field is constructed by adopting a Person correlation analysis method and a regression analysis method, can be used for reducing false alarm behaviors caused by navigation mark drifting in the tidal flow field, and can provide strong practical reference for maintenance and management of a remote control and remote measurement system of the marine navigation mark.
According to the method, the accuracy of the navigation mark drift prediction result is improved by analyzing the relation between the tidal current flow direction and the flow velocity and the navigation mark drift amount, but the model structure is still simpler, the time sequence characteristics of navigation mark data are ignored, and the prediction accuracy needs to be further improved.
Disclosure of Invention
In view of the above, the invention provides an intelligent navigation mark drift prediction method based on deep learning, and provides a navigation mark drift prediction model based on the combination of an attention mechanism and a cyclic neural network, which can predict a navigation mark drift position and a drift distance in a short period by navigation mark GPS observation data, thereby improving the prediction precision, and the prediction result can be referred by a channel department and a navigation ship, and is beneficial to improving the navigation safety level of the channel.
Therefore, the invention provides the following technical scheme:
the invention provides an intelligent navigation mark drift prediction method based on deep learning, which comprises the following steps:
s1, acquiring sample data and preprocessing the sample data; wherein the sample data comprises navigation mark historical position data and water level historical data;
s2, constructing a time sequence based on the preprocessed sample data;
s3, establishing and training a navigation mark drift prediction model by using the navigation mark motion rule reflected by the time sequence; the navigation mark drift prediction model comprises a gating cycle unit for extracting time sequence characteristics in a time sequence and an attention mechanism for giving different weights to each time step of the time sequence;
s4, inputting preset moment data of the navigation mark, and predicting by using the trained navigation mark drift prediction model;
and S5, outputting the predicted position of the navigation mark.
Further, the network structure of the navigation mark drift prediction model comprises: the device comprises an input layer, 2 GRU layers, an Attention layer and an output layer;
the network parameters of the navigation mark drift prediction model comprise:
the shape of the input vector of the input layer is [64,24,3], 64 is the data size of a primary input network, namely batch _ size, 24 is the time step, and 3 is the characteristic quantity, namely the longitude value, the latitude value and the water level value;
the number of neurons of the two GRU layers is 128 and 64 respectively, and the activation functions are both tanh functions.
Further, the Attention mechanism calculation formula is as follows:
hk′＝tanh(Wc[C；hk])；
wherein x is1,x2,...,xkRepresents an input sequence, h1,h2,...,hkRepresenting the state value of the hidden layer corresponding to the input sequence, a scoring function skjUsing dot product, alphakiAttention weight of hidden layer state of history input to current input, C is intermediate vector; h isk′Representing the finally output hidden layer state value at the current moment.
Further, preprocessing the sample data, including:
for data duplication, finding a duplicate record item and deleting the duplicate record item;
for data missing, linear interpolation is adopted for data filling;
and for data abnormality, detecting abnormal values by using a box diagram, and filling by using a data missing processing method after removing the abnormal values.
Further, the linear interpolation formula includes:
wherein, tiAnd xiRespectively a filling time point and a filling value, xi+1、xi-1Respectively adjacent time point observed values.
Further, the data preprocessing further comprises the following steps:
extracting navigation marks and water level data as a data set according to a preset time interval;
each column in the data set is normalized.
Further, the loss function of the navigation mark drift prediction model adopts the average absolute error.
The technical scheme has the following beneficial effects: the invention innovatively provides a navigation mark drift prediction model, learns time sequence characteristics from a large amount of navigation marks and water level historical data based on a deep learning method, further improves the model prediction precision by considering the influence of water level factors and adding an attention mechanism, and meets the actual application requirement of the prediction precision.
The method can be applied to the aspect of guiding the navigation of the ship, and the ship can plan the route by referring to the navigation mark drift prediction position, thereby improving the navigation efficiency of the channel. The method can also be applied to the aspect of navigation mark management, and a supervision department can better master navigation mark drift dynamic according to the predicted position of the navigation mark drift, carry out drift early warning and improve the navigation safety level of a navigation channel. The prediction of the drift position of the navigation mark has important significance in the aspect of anti-collision, and the ship can make collision prevention measures in advance according to the predicted position, so that the navigation safety is guaranteed, and the property loss is reduced.
In conclusion, the technical scheme of the invention realizes the navigation mark drift prediction, has important application in various aspects such as route planning, navigation mark management, anti-collision and the like, and has obvious economic and social benefits.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of the K-means + + algorithm in the prior art;
FIG. 2 is a flowchart of an intelligent navigation mark drift prediction method based on deep learning according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of an Attention-GRU-based navigation mark drift prediction model in an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a GRU unit structure in an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of an Attention unit structure according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The method aims to build time sequence characteristics among high-precision model learning data by using a deep learning method based on a large amount of navigation mark and water level historical data and predict the drift position and drift distance of the navigation mark in a short period.
The factors of float displacement are generally two: firstly, natural factors, namely displacement of the navigation mark under the influence of water flow, tide and wind; the other is human factor, namely navigation mark displacement caused by collision or dragging of passing ships. Under the influence of tidal current, river water makes reciprocating motion, so that the water level is periodically changed. The navigation mark as the suspension body floating on the water surface can also do reciprocating motion along with the navigation mark. Therefore, the water level change is an important factor influencing the navigation mark drift.
The method is based on navigation mark historical GPS position data, namely longitude values and latitude values, and takes water level characteristics into consideration to construct a time sequence, then utilizes a deep cycle neural network to extract time sequence characteristics of the sequence, further uses an attention mechanism to give different weights to each time step of an input sequence, highlights influences of more critical moments and helps a model to make more accurate judgment. And finally, predicting the longitude value and the latitude value of the navigation mark position in a short period by using the trained deep learning model.
Referring to fig. 2, a flowchart of an intelligent navigation mark drift prediction method based on deep learning according to an embodiment of the present invention is shown, where the method includes the following steps:
s1, acquiring sample data and preprocessing the sample data;
wherein the sample data comprises navigation mark historical position data and water level historical data;
due to the reasons of measurement accuracy, data transmission faults and the like, a small amount of navigation mark historical position data and water level historical data are poor in quality. For time series prediction, the quality of data is directly related to the usability and robustness of the model, and the prediction accuracy of the model is also greatly influenced. Therefore, it is important to perform preprocessing on the raw data.
The problems of data repetition, data loss, data abnormity and the like mainly exist in water level and navigation mark observation data.
And the data repetition means that a plurality of records with the same timestamp appear on the same navigation mark or water level station, and the recorded information is the same. For data duplication, only duplicate entries need to be found and deleted.
Data missing refers to the lack of data records at several points in time. The observation of the navigation mark and the original observation data of the water level shows that the data loss is less. Thus, linear interpolation is used for data padding, as shown in equation (1), tiAnd xiRespectively a filling time point and a filling value, xi+1、xi-1Respectively adjacent time point observed values.
Data anomalies (error values), which refer to extremely small or large values that have data values that are erroneous or that deviate significantly from normal levels over a time series. The method and the device for detecting the abnormal values detect the abnormal values by using the box diagram, and fill the abnormal values by using a data missing processing method after the abnormal values are removed.
After data preprocessing, navigation marks and water level data are extracted according to preset time intervals to serve as experimental data sets, and the final data set is shown in table 1, wherein the preset time intervals in table 1 are 2 hours.
TABLE 1
| Serial number | Time | Water level | Longitude (G) | Latitude |
| 0 | 2020/2/20 22：00 | 1.78 | 120.559559 | 32.009901 |
| 1 | 2020/2/21 0:00 | 2.83 | 120.559503 | 32.009912 |
| 2 | 2020/2/21 2:00 | 2.98 | 120.559422 | 32.009961 |
| 3 | 2020/2/21 4:00 | 2.32 | 120.559445 | 32.009959 |
| 4 | 2020/2/21 6:00 | 1.65 | 120.559524 | 32.009917 |
| 5 | 2020/2/21 8:00 | 1.24 | 120.559565 | 32.009914 |
| 6 | 2020/2/21 10:00 | 1.98 | 120.559559 | 32.009908 |
Because the difference of the water level, longitude and latitude data values is large, each column in the data set needs to be normalized, and the normalization calculation formula is shown as the formula (2):
where x is the value to be normalized, xmin、xmaxRespectively, when the minimum and maximum of the column, x' is normalizedThe latter value.
S2, constructing a time sequence based on the preprocessed sample data;
s3, establishing and training a navigation mark drift prediction model by using the navigation mark motion rule reflected by the time sequence; the navigation mark drift prediction model comprises a gating cycle unit for extracting time sequence characteristics in a time sequence and an attention mechanism for giving different weights to each time step of the time sequence;
and (3) constructing a navigation mark drift prediction model by utilizing the navigation mark motion rule, namely constructing the navigation mark drift prediction model by considering the water level characteristics, wherein the prediction model is fused with an attention mechanism. The navigation mark drift prediction model is based on an Attention mechanism (Attention) and a gating cycle unit (GRU), and the prediction model is constructed by an overall network structure and hyper-parameters, including parameters of an input layer, 2 GRU layers, an Attention layer and an output layer, and loss functions and an optimization method of the network. A GRU (Gate recovery Unit, gated cyclic Unit) is one of cyclic Neural networks (RNN), and is proposed to solve the problems of long-term memory and gradient in back propagation.
S4, inputting preset moment data of the navigation mark, and predicting by using the trained navigation mark drift prediction model;
in the embodiment of the invention, the preset time data is 24 time data.
And S5, outputting the predicted position of the navigation mark.
In the embodiment of the invention, the navigation marks of a plurality of navigation sections at the downstream of the Yangtze river are selected for comparison experiments, the RMSE value of the predicted drift distance value is used as an evaluation index, and the calculation result is shown in Table 2.
TABLE 2
The embodiment of the invention provides an Attention-GRU-based navigation mark drift prediction model, a deep learning-based method learns time sequence characteristics from a large amount of navigation marks and water level historical data, the model prediction precision is further improved by considering the influence of water level factors and adding an Attention mechanism, and the prediction precision meets the actual application requirement.
For the convenience of understanding, the following describes the building of the navigation mark drift prediction model in step S2 in the embodiment of the present invention in detail.
(1) Construction of models
The structure of the navigation mark drift prediction model based on the Attention and the GRU is shown in FIG. 3.
In this model, the core parameters are as follows:
1) and inputting the layer. The input vector shape is [64,24,3], 64 is the data size of one input network, namely, batch _ size, 24 is time step, 3 is the characteristic quantity, namely, longitude value, latitude value and water level value, wherein the batch _ size and the time step can be adjusted by self during subsequent test.
2) And a GRU layer. The number of neurons of the two GRU layers is 128 and 64 respectively, and the activation functions are both tanh functions. The GRU introduces a gating mechanism, makes full use of historical information, and can solve the long-term dependence problem existing in a long distance. The structure of the GRU unit is shown in FIG. 4.
The GRU only comprises an updating gate and a resetting gate, wherein the updating gate controls the degree of state information of the previous time to be brought into the current state, and the larger the value is, the more information is brought in the previous time. The reset gate controls the extent to which the current state is combined with the previous information, with a larger value indicating more information is retained. The specific calculation formula is as follows:
in the formula (3), rt、ztRespectively representing the values of reset gate, update gate, xtFor the input at the time t, the input is,
represents a candidate hidden state value, h, at time ttFor hiding layer state values at time t, Wr、Wz、WhFor training the parameter matrix, σ and tanh represent sigmoid activation function and hyperbolic tangent activation, respectivelyA function,. indicates multiplication of corresponding elements of the matrix. Reset gate rtFor controlling the state h hidden from the last time stept-1If r is the degree of forgetfulness oftClose to 0 means that the corresponding hidden state h is resett-1Is 0, the hidden state of the last time step is discarded if rtApproaching 1 indicates a greater preservation of the hidden state at the previous time step. Updating the door ztFor controlling the extent to which hidden states are updated by candidate hidden states containing current time step information, ztDetermines the hidden state h of the last time stept-1Flow into the current hidden state htDegree of (2), on the other hand, 1-ztDetermines candidate hidden states
For the current hidden state htThe degree of update of. Suppose the update gate is at time steps t 'to t (t'<t) always approximates to 1, i.e. the input information between time steps t' and t hardly flows into the hidden state h of time step ttCan be regarded as a hidden state h at an earlier momentt‘-1And storing and transmitting to the current time step t all the time.
3) An Attement layer. The Attention mechanism is a model simulating the Attention of the human brain, and the characteristics that the Attention of the human brain to things at a specific moment is concentrated to a specific place, and the Attention to other parts is reduced or even ignored are used for reference. In practical training, the GRU needs to capture sequence information step by step, and the performance on a long sequence gradually decays along with the increase of the step number, so that all useful information is difficult to keep. The Attention mechanism gives different weights to each time step of an input sequence, so that the influence of more critical moments is highlighted, the model is helped to make more accurate judgment, and the calculation and storage overhead of the model is not increased. The architecture of the Attention unit is shown in FIG. 5.
In FIG. 5, x1,x2,...,xkRepresents an input sequence, h1,h2,...,hkThe representation corresponds to the inputState values of hidden layers in sequence, scoring function skjUsing dot product, alphakiAttention weight to current input for hidden layer state of history input, C is the intermediate vector. h isk′Representing the finally output hidden layer state value at the current moment. The Attention mechanism calculation formula is as follows:
hk′＝tanh(Wc[C；hk]) (7)
(2) loss function
The loss calculation is performed according to the prediction result and the actual result of the forward propagation, so that the weight parameter is updated in the process of backward propagation. The loss function here adopts MAE (Mean Absolute Error), i.e., the average of the Absolute values of the errors between the observed value and the true value, and its calculation formula is shown in formula (8).
In the formula (8), m represents the total number of samples, yiAnd
respectively, the true position (latitude and longitude coordinates) and the predicted position (latitude and longitude coordinates) of the navigation mark.
The embodiment of the invention also discloses a computer-readable storage medium, wherein a computer instruction set is stored in the computer-readable storage medium, and when being executed by a processor, the computer instruction set realizes the intelligent navigation mark drift prediction method based on deep learning, which is provided by any one of the above embodiments.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (7)
1. A navigation mark drift intelligent prediction method based on deep learning is characterized by comprising the following steps:
s1, acquiring sample data and preprocessing the sample data; wherein the sample data comprises navigation mark historical position data and water level historical data;
s2, constructing a time sequence based on the preprocessed sample data;
s3, establishing and training a navigation mark drift prediction model by using the navigation mark motion rule reflected by the time sequence; the navigation mark drift prediction model comprises a gating cycle unit for extracting time sequence characteristics in a time sequence and an attention mechanism for giving different weights to each time step of the time sequence;
s4, inputting preset moment data of the navigation mark, and predicting by using the trained navigation mark drift prediction model;
and S5, outputting the predicted position of the navigation mark.
2. The intelligent prediction method for navigation mark drift based on deep learning of claim 1, wherein the network structure of the navigation mark drift prediction model comprises: the device comprises an input layer, 2 GRU layers, an Attention layer and an output layer;
the network parameters of the navigation mark drift prediction model comprise:
the shape of the input vector of the input layer is [64,24,3], 64 is the data size of a primary input network, namely batch _ size, 24 is the time step, and 3 is the characteristic quantity, namely the longitude value, the latitude value and the water level value;
the number of neurons of the two GRU layers is 128 and 64 respectively, and the activation functions are both tanh functions.
3. The intelligent prediction method for navigation mark drift based on deep learning of claim 1, wherein the Attention mechanism calculation formula is as follows:
hk′＝tanh(Wc[C；hk])；
wherein x is1,x2,...,xkRepresents an input sequence, h1,h2,...,hkRepresenting the state value of the hidden layer corresponding to the input sequence, a scoring function skjUsing dot product, alphakiAttention weight of hidden layer state of history input to current input, C is intermediate vector; h isk′Representing the finally output hidden layer state value at the current moment.
4. The intelligent navigation mark drift prediction method based on deep learning of claim 1, wherein the preprocessing of the sample data comprises:
for data duplication, finding a duplicate record item and deleting the duplicate record item;
for data missing, linear interpolation is adopted for data filling;
and for data abnormality, detecting abnormal values by using a box diagram, and filling by using a data missing processing method after removing the abnormal values.
5. The intelligent prediction method for navigation mark drift based on deep learning of claim 4, wherein the linear interpolation formula comprises:
wherein, tiAnd xiRespectively a filling time point and a filling value, xi+1、xi-1Respectively adjacent time point observed values.
6. The intelligent navigation mark drift prediction method based on deep learning of claim 4, wherein the data preprocessing further comprises:
extracting navigation marks and water level data as a data set according to a preset time interval;
normalizing each column in the data set;
correspondingly, constructing a time sequence based on the preprocessed sample data, comprising:
and constructing a time sequence based on the normalized data set.
7. The intelligent prediction method for navigation mark drift based on deep learning of claim 1, wherein the loss function of the navigation mark drift prediction model adopts average absolute error.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110648282.6A CN113393027B (en) | 2021-06-10 | 2021-06-10 | Navigation mark drift intelligent prediction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110648282.6A CN113393027B (en) | 2021-06-10 | 2021-06-10 | Navigation mark drift intelligent prediction method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113393027A true CN113393027A (en) | 2021-09-14 |
| CN113393027B CN113393027B (en) | 2023-11-07 |
Family
ID=77620220
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110648282.6A Active CN113393027B (en) | 2021-06-10 | 2021-06-10 | Navigation mark drift intelligent prediction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113393027B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113932817A (en) * | 2021-12-15 | 2022-01-14 | 思创数码科技股份有限公司 | Navigation mark offset evaluation method and system |
| WO2023216946A1 (en) * | 2022-05-09 | 2023-11-16 | 深圳大学 | Drift speed prediction method and apparatus, computing device, and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104913892A (en) * | 2015-05-08 | 2015-09-16 | 京杭运河江苏省交通运输厅苏北航务管理处 | Beacon collision detection system based on navigation positioning system and method |
| KR101778724B1 (en) * | 2016-12-05 | 2017-09-14 | 동국대학교 산학협력단 | Device and method for reconition of road markings |
| US20190034794A1 (en) * | 2017-07-27 | 2019-01-31 | Waymo Llc | Neural Networks for Vehicle Trajectory Planning |
| US20190265768A1 (en) * | 2018-02-24 | 2019-08-29 | Hefei University Of Technology | Method, system and storage medium for predicting power load probability density based on deep learning |
| CN111414442A (en) * | 2020-01-08 | 2020-07-14 | 重庆大学 | Navigation mark position checking method based on navigation channel topographic map and water level data |
| CN111488940A (en) * | 2020-04-15 | 2020-08-04 | 大连海事大学 | Intelligent navigation mark image classification method based on deep convolutional neural network |
| CN112288193A (en) * | 2020-11-23 | 2021-01-29 | 国家海洋信息中心 | Ocean station surface salinity prediction method based on GRU deep learning of attention mechanism |
| CN112364119A (en) * | 2020-12-01 | 2021-02-12 | 国家海洋信息中心 | Ocean buoy track prediction method based on LSTM coding and decoding model |
| US20210117796A1 (en) * | 2020-12-01 | 2021-04-22 | Harbin Engineering University | Ship Motion Prediction Method Based on Long Short-Term Memory Network and Gaussian Process Regression |
-
2021
- 2021-06-10 CN CN202110648282.6A patent/CN113393027B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104913892A (en) * | 2015-05-08 | 2015-09-16 | 京杭运河江苏省交通运输厅苏北航务管理处 | Beacon collision detection system based on navigation positioning system and method |
| KR101778724B1 (en) * | 2016-12-05 | 2017-09-14 | 동국대학교 산학협력단 | Device and method for reconition of road markings |
| US20190034794A1 (en) * | 2017-07-27 | 2019-01-31 | Waymo Llc | Neural Networks for Vehicle Trajectory Planning |
| US20190265768A1 (en) * | 2018-02-24 | 2019-08-29 | Hefei University Of Technology | Method, system and storage medium for predicting power load probability density based on deep learning |
| CN111414442A (en) * | 2020-01-08 | 2020-07-14 | 重庆大学 | Navigation mark position checking method based on navigation channel topographic map and water level data |
| CN111488940A (en) * | 2020-04-15 | 2020-08-04 | 大连海事大学 | Intelligent navigation mark image classification method based on deep convolutional neural network |
| CN112288193A (en) * | 2020-11-23 | 2021-01-29 | 国家海洋信息中心 | Ocean station surface salinity prediction method based on GRU deep learning of attention mechanism |
| CN112364119A (en) * | 2020-12-01 | 2021-02-12 | 国家海洋信息中心 | Ocean buoy track prediction method based on LSTM coding and decoding model |
| US20210117796A1 (en) * | 2020-12-01 | 2021-04-22 | Harbin Engineering University | Ship Motion Prediction Method Based on Long Short-Term Memory Network and Gaussian Process Regression |
Non-Patent Citations (3)
| Title |
|---|
| ILYA BELKIN 等: "Traffic Sign Recognition on Video Sequence Using Deep Neural Networks and Matching Algorithm", 2019 INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE: APPLICATIONS AND INNOVATIONS, pages 35 - 39 * |
| 权波 等: "基于LSTM的船舶航迹预测模型", 计算机科学, no. 2, pages 136 - 141 * |
| 潘明阳 等: "智能水位预测服务系统研究", 大连海事大学学报, vol. 46, no. 3, pages 31 - 37 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113932817A (en) * | 2021-12-15 | 2022-01-14 | 思创数码科技股份有限公司 | Navigation mark offset evaluation method and system |
| CN113932817B (en) * | 2021-12-15 | 2022-03-08 | 思创数码科技股份有限公司 | Navigation mark offset evaluation method and system |
| WO2023216946A1 (en) * | 2022-05-09 | 2023-11-16 | 深圳大学 | Drift speed prediction method and apparatus, computing device, and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113393027B (en) | 2023-11-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109359557B (en) | SAR remote sensing image ship detection method based on transfer learning | |
| CN109299262A (en) | A kind of text implication relation recognition methods for merging more granular informations | |
| CN109214107A (en) | A kind of ship's navigation behavior on-line prediction method | |
| CN110197215A (en) | A kind of ground perception point cloud semantic segmentation method of autonomous driving | |
| CN113393027A (en) | Navigation mark drift intelligent prediction method based on deep learning | |
| CN110082717A (en) | A kind of underwater wireless sensor node positioning method | |
| CN110414742A (en) | A kind of freshwater line intelligent Forecasting of more gaging station linkages | |
| CN111695299A (en) | Mesoscale vortex trajectory prediction method | |
| CN108008099A (en) | A kind of pollution sources localization method | |
| Tang et al. | Detection of abnormal vessel behaviour based on probabilistic directed graph model | |
| Rong et al. | Evaluation of near-collisions in the Tagus River Estuary using a marine traffic simulation model | |
| CN108960421A (en) | The unmanned surface vehicle speed of a ship or plane online forecasting method based on BP neural network of improvement | |
| CN110633892A (en) | Method for extracting long-line fishing state through AIS data | |
| CN112766308A (en) | Ocean vessel classification method | |
| CN115050214B (en) | AIS data-based ship collision risk prediction method | |
| CN117494871A (en) | Ship track prediction method considering ship interaction influence | |
| Yang et al. | An approach to ship behavior prediction based on AIS and RNN optimization model | |
| CN116608861A (en) | Ship track behavior abnormality detection method, system, device and storage medium | |
| CN114997296A (en) | Unsupervised track anomaly detection method and unsupervised track anomaly detection system based on GRU-VAE model | |
| CN115169527A (en) | AIS data-based ship abnormal state detection method | |
| CN117076893B (en) | Sound velocity distribution forecasting method based on long-term and short-term memory neural network | |
| CN117372854A (en) | Real-time detection method for hidden danger diseases of deep water structure of dam | |
| KR102561270B1 (en) | Method and apparatus for estimating the position of the sea drift object for search and rescue | |
| CN115761520A (en) | River runoff analysis method and system based on autonomous learning of river floaters | |
| CN115310559A (en) | Target identification method based on Bayes-Transformer neural network model |
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 |