CN112734111B - Horizontal transport task AGV dynamic time prediction method - Google Patents
Horizontal transport task AGV dynamic time prediction method Download PDFInfo
- Publication number
- CN112734111B CN112734111B CN202110027744.2A CN202110027744A CN112734111B CN 112734111 B CN112734111 B CN 112734111B CN 202110027744 A CN202110027744 A CN 202110027744A CN 112734111 B CN112734111 B CN 112734111B
- Authority
- CN
- China
- Prior art keywords
- data
- agv
- time
- machine learning
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000010801 machine learning Methods 0.000 claims abstract description 33
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 238000012545 processing Methods 0.000 claims abstract description 22
- 238000003062 neural network model Methods 0.000 claims abstract description 16
- 230000008569 process Effects 0.000 claims abstract description 9
- 230000003068 static effect Effects 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000012216 screening Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 5
- 238000003066 decision tree Methods 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000007637 random forest analysis Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 238000000844 transformation Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000011022 operating instruction Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06N3/047—Probabilistic or stochastic 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Game Theory and Decision Science (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a horizontal transport task AGV dynamic time estimation method in the technical field of dynamic time estimation, which comprises the following steps: step 1: selecting characteristics; step 2: data processing; step 3: machine learning; step 4: a neural network; step 5: verifying a dynamic time estimation model; according to the method for estimating the dynamic time of the AGV of the horizontal transport task, the machine learning model and the neural network model are used for dynamically predicting the task completion time of the AGV, and the influence of various factors on the AGV in the horizontal transport process of the wharf is described. Therefore, the time estimation result is improved, and more accurate basic data is provided for other application scenes.
Description
Technical Field
The invention relates to the technical field of dynamic time estimation, in particular to a horizontal transport task AGV dynamic time estimation method.
Background
AGV (Automated Guided Vehicles) is also called an unmanned carrier, an automatic navigation vehicle and a laser navigation vehicle, and is remarkably characterized by being unmanned. AGVs are used in large numbers in the current port industry for unmanned vehicle loading and unloading operations. In the AGV operation process, the AGV time prediction application scene is wide, the AGV running time is key basic data of a dispatching system, the AGV time prediction method can be applied to AGV task selection and arrangement, actual task AGV operation time estimation, port overall efficiency optimization, traffic conflict priority selection during AGV running, TOS task distribution of an automatic port, matching decision of vehicles and tasks, conflict optimization of vehicle path planning and the like, and has wide application scene and important research significance.
Static time matrix estimation is widely used in the current dock time estimation scene, and the static time estimation is to directly calculate the time required for reaching the end point from the starting point by simply considering the acceleration and deceleration principle when a path exists through the starting point and the end point of the AGV and by using a strategy without any collision and avoidance. The static time matrix estimation steps are as follows: 1. dividing the planned route into a plurality of different sections; 2. time calculation, namely calculating the time consumed by each AGV passing period respectively by utilizing the physical scene of an acceleration and deceleration principle and knowing the initial speed and the termination speed of each AGV passing period, and sequentially adding the time consumed by each AGV passing period to obtain final static time
In the prior art, the automatic wharf port operation always uses a static time matrix to estimate the AGV traffic time, and does not consider other vehicles on the wharf, directly uses a physical rule and calculates the speed planning. The deviation between the predicted result and the actual result exceeds 35%, and the excessively high time prediction deviation causes the phenomenon of resource idling or inefficiency in each link of port operation, and is not improved for a long time.
Disclosure of Invention
The invention aims to provide a horizontal transport task AGV dynamic time prediction method, which uses an machine learning model and a neural network model to dynamically predict the time for completing the task of the AGV and solves the technical problem that in the prior art, the deviation between a static time matrix prediction result and an actual result is large, so that each link of port operation generates the phenomenon of resource idling or low efficiency.
In order to achieve the technical purpose, the invention adopts the following technical scheme:
according to one aspect of the invention, a method for estimating the dynamic time of an AGV of a horizontal transport task is provided, which comprises the following steps:
step 1: feature selection: selecting and screening AGV operation command characteristics of a horizontal transport task by using a computer algorithm, and integrating characteristic data;
step 2: and (3) data processing: screening and processing the characteristic data in the step 1 by using a computer algorithm to form standardized data;
step 3: machine learning: adopting various machine learning models to conduct prediction on the standardized data in the step 2, and searching for optimal parameters;
step 4: neural network: constructing a full-connection layer neural network, and predicting the standardized data in the step 2 by using a neural network model;
step 5: verifying a dynamic time estimation model: comparing the predicted value of the machine learning model, the predicted value of the neural network model and the static time matrix, and evaluating the performance of the dynamic time prediction model;
according to the horizontal transport task AGV dynamic time estimation method of the above aspect of the invention, the feature selection in the step 1 comprises the following steps:
s11, extracting data from a processing order table; selecting a record about an AGV operation instruction from an AGV_ORDERS table;
s12, processing the extracted data in the track table; finding one or more indexes, screening out key points in the running process of the vehicle according to one Order, and dividing the Track corresponding to each Order data according to the key points;
s13, integrating data: and counting the number of the AGV path conflicts to generate a total feature.
According to the method for estimating the dynamic time of the AGV of the horizontal transport task, the step S12 comprises the following steps:
s121: according to the MoveFrequency field of the data table after feature extraction, counting and calculating the times of straight running, turning and inclined running of the AGV trolley;
s122: calculating the path length of the track according to the data in the step S121;
s123: processing the time of the start and end of the data in the above step S121;
s124: and calculating the running speed of the trolley according to the starting time, the ending time and the track distance.
According to the horizontal transport task AGV dynamic time estimation method of the above aspect of the present invention, the total characteristics in the step S13 include the vehicle driving distance, the number of straight runs, the number of turns, the number of diagonal runs, the number of conflicting vehicles, unloading, loading, turning, box exchange, box taking, PB section to PB section, PB section to WSTP section, WSTP section to PB section and WSTP section to WSTP section.
According to the horizontal transport task AGV dynamic time estimation method, the step 2 comprises the following steps:
s21: analyzing the characteristic data, eliminating abnormal values for a large amount of error data and invalid data which are obviously present;
s22: further screening abnormal values, and clustering two direct factors of time spent and total distance;
s23: after the clustering process is completed, selecting data meeting the conditions as final training data;
s24: the data were normalized, normalized and one-hot coded.
According to the horizontal transport task AGV dynamic time estimation method of the above aspect of the present invention, the various machine learning models in the step 3 include XGB, GBDT, randomForest, decisionTree, ridge and linearregprecision.
According to the horizontal transport task AGV dynamic time prediction method disclosed by the invention, the method for searching the optimal parameters in the various models by machine learning in the step 3 adopts a grid search method.
According to the method for estimating the dynamic time of the horizontal transport task AGV in the above aspect of the present invention, the first layer of the fully connected layer neural network in the step 4 is input, the last layer is output, and the hidden layers with 128×128 structures are shared.
According to the horizontal transport task AGV dynamic time prediction method of the above aspect of the present invention, the value range output by each layer of the fully connected layer neural network is compressed to (0, 1) by using the activation function, and the output value of each stage is used as the input value of the next stage.
According to the horizontal transport task AGV dynamic time prediction method of the above aspect of the present invention, in the step 5, the evaluation criteria for verifying the prediction of time by the machine learning model is root mean square error, the root mean square error between the prediction value, the static time matrix value, the average value and the true value of the computer learning model are respectively calculated, and the performance of the machine learning model is evaluated by comparing the calculation results.
By adopting the technical scheme, the invention has the following advantages:
the invention provides a horizontal transport task AGV dynamic time prediction method, which uses a machine learning model and a neural network model to dynamically predict the time for completing the task of an AGV, combines a model trained according to historical information with time information and site information by considering the path, characteristics, traffic conditions, congestion areas and the like of the AGV, greatly optimizes the model result of the dynamic time prediction, ensures that the deviation between a prediction result and an actual result is within 20%, solves the technical problem that factors such as AGV traffic conflict and time are difficult to directly describe, improves the phenomenon that resources are idle or low in efficiency in each link of port operation, improves the time prediction result and provides more accurate basic data for other application scenes.
Drawings
Other features, objects and advantages of the present invention will become more apparent upon reading of the detailed description of non-limiting embodiments, given with reference to the accompanying drawings.
FIG. 1 is a flow chart of the horizontal transport task AGV dynamic time estimation method of the present invention;
FIG. 2 is a data screenshot of an extract order table of the present invention;
FIG. 3 is a data screenshot of the extract track table of the present invention;
FIG. 4 is a data sheet screenshot after feature extraction of the present invention;
FIG. 5 is a screen shot of a time profile of a completed task prior to data processing in accordance with the present invention;
FIG. 6 is a screen shot of a time profile of a completed task after data processing in accordance with the present invention;
FIG. 7 is a screen shot of the machine learning various model prediction results of the present invention;
FIG. 8 is a screen shot of a segmented root mean square error table for machine learning of various model predictions in accordance with the present invention;
FIG. 9 is a screen shot of a segmented root mean square error table of the neural network prediction result of the present invention;
FIG. 10 is a screen shot of a static time matrix versus dynamic time prediction result comparison of the present invention.
Detailed Description
The technical scheme of the invention is specifically described below with reference to the attached drawings.
The detailed features and advantages of the present invention will be readily apparent to those skilled in the art from the following detailed description, claims, and drawings that follow.
Examples
FIG. 1 shows a flow chart of the method for estimating the dynamic time of the AGV of the horizontal transport task, which specifically comprises the following steps in combination with consideration of the path, characteristics, in-field traffic conditions and the like, as shown in FIG. 1:
step 1: feature selection: selecting and screening the operating instruction characteristics of the AGV of the horizontal transport task by using a computer program algorithm, and integrating characteristic data;
FIG. 2 shows a data screenshot of an extract process order table; FIG. 3 shows a data screenshot of an extraction processing track table; FIG. 4 illustrates a feature extracted data sheet screenshot;
wherein: the feature selection in step 1 includes the steps of:
s11, processing data in the order table is as shown in fig. 2, selecting records about AGV operation instructions from an AGV_ORDERS table (a sequence table of AGV trolleys), and only checking key points PB (sea side operation area) and WSTP (sea side exchange area) in the running process of the vehicle, and screening data according to whether the starting position (H_AGO_ST_location) and the DESTINATION position (H_AGO_PLANNED_DESTINATION) are PB, QC (quay bridge crane) and WSTP (container yard sea side exchange area).
S12, processing data in a Track table (a motion Track table of an AGV trolley) as shown in a figure 3, finding one or more indexes, measuring the route characteristics from the current starting point to the destination point of the AGV, screening out key points in the running process of the vehicle according to a sequence Order, such as QC to PB, PB to QC, PB to PB, PB to WSTP, WSTP to PB, and WSTP to WSTP, screening out corresponding Order data according to AGV_ID (information of the AGV trolley), and dividing the Track corresponding to each Order data according to the key points to obtain a characteristic extracted data table as shown in a figure 4.
The step S12 includes the steps of:
s121: according to the MoveFrequency field of the data, counting and calculating the number of times of straight going, turning and inclined going of the AGV;
s122: calculating the path length of the track according to the data;
s123: processing the starting and ending time of the data;
s124: and calculating the running speed of the trolley according to the starting time, the ending time and the track distance.
S13: integrating the data in FIG. 4, counting the number of path conflicts, and generating total features, wherein the total feature parameter description and the data sources are shown in the following table 1:
table 1: total feature parameter description and data sources
Field name | Field description | Data source mode |
Distance | Distance travelled by vehicle | From path computation |
DirectFrequency | Number of straight travel times | Reading from a path |
TurnFrequency | Number of turns | Reading from a path |
SlideFrequency | Number of diagonal runs | Reading from a path |
ConflictVehicle | Number of conflicting vehicles | Calculation using path keypoints |
AroundVehicle | Number of vehicles on the scene at the time of starting the mission | The system gives |
DSCH | Ship unloading | Reading from order |
LOAD | Shipping | Reading from order |
SHIFT | Transfer pile | Reading from order |
DELIVER | Cross box | Reading from order |
RECEIVE | Box taking | Reading from order |
REPOSITION | Redirecting | Reading from order |
PB2PB | From PB segment to PB segment | Filling in according to the setting |
PB2WSTP | From PB segment to WSTP segment | Filling in according to the setting |
WSTP2PB | From WSTP segment to PB segment | Filling in according to the setting |
WSTP2WSTP | From WSTP segment to WSTP segment | Filling in according to the setting |
earlymorning | 0 point to 4 points | Filling in according to the setting |
morning | 4 to 8 points | Filling in according to the setting |
Noon | 8 to 12 points | Filling in according to the setting |
Afternoon | 12 to 16 points | Filling in according to the setting |
Night | 16 to 20 points | Filling in according to the setting |
Latenight | 20 to 24 points | Filling in according to the setting |
Distance_no_path | Distance without path | Calculating the starting point and the end point |
FIG. 5 shows a time profile of a completion task prior to data processing; fig. 6 shows a time profile of completion of a task after data processing.
Step 2: and (3) data processing: screening and processing the characteristic data in the step 1 by using a computer algorithm to form standardized data;
the step 2 comprises the following steps:
s21: the task before data processing takes time as shown in fig. 5, for a significant amount of erroneous data and invalid data. Performing box line graph analysis on Distance data, and eliminating abnormal values;
s22: since the distribution of the screened data is still more scattered, further screening for outliers is required. Taking two direct factors of time and total distance into consideration for clustering;
s23: after clustering is completed, selecting the class meeting the line-up of the box line diagram as final training data;
s24: the data were normalized, and unithermally encoded, resulting in a task completion time as shown in fig. 7.
Step 3: machine learning: adopting various machine learning models to participate in the prediction of the standardized data in the step 2, and searching for optimal parameters;
FIG. 7 illustrates machine learning various model prediction outcome graph screenshots; FIG. 8 illustrates machine learning various model predictive result segmentation root mean square error table screen shots;
the various machine learning models in step 3 include XGB (a tree model), GBDT (gradient-lifted decision tree), random forest, decision tree, ridge, and linear regression.
Step 2 is involved in prediction by using the above multiple prediction models, the prediction results are shown in fig. 7, and it can be seen from fig. 7 that multiple prediction results XGB > GBDT > random forest > precision tree > ridge=linear regress
And searching for optimal parameters by using a grid search method. Setting the n_estimator (parameter tuning code) of XGB to 200, the result of XGB model prediction data is shown in fig. 9, and is calculated:
using the neural network model, the decision coefficient reaches 0.66585,
the segmentation look model does not take into account the special cases of 70-90 and 210-230, which are not representative and have small data size, in root mean square error of each segment, and the result is as follows:
the root mean square error of the neural network within 90-110s is about 16.72/100=16.7%
The root mean square error of the neural network within 110-130s is about 17.78/120=14.6%
The root mean square error of the neural network within 130-150s is about 20.21/140=14.5%
The root mean square error of the neural network within 150-170s is about 23.08/160=14.4%
The root mean square error of the neural network within 170-190s is about 27.96/180=15.6%
The root mean square error of the neural network within 190-210s is about 34.11/200=17.1%
The prediction error is about 10% -20%.
Step 4: neural network: setting up a fully-connected layer neural network for the multiple prediction models in the step 3, wherein the first layer is input, the last layer is output, and a hidden layer with a 128 x 128 structure is arranged, meanwhile, the value range of the output of each layer is compressed to (0, 1) by utilizing an activation function, the value of the output of each stage is used as the input of the next stage, and the condition that certain input is infinite, and other inputs are invalid is avoided. And finally, realizing accurate expression of the model on data by using a solution method of back propagation and gradient descent. The neural network computes a series of transformations that alter the similarity of the samples. The activity of each layer of neurons is a nonlinear function of the activity of the previous layer.
Predicting data using a neural network model; fig. 9 shows a screen shot of a section mean square error table of a neural network prediction result, a full-connection layer neural network is built, a function softmax is activated, a training algebra 300 is used, each generation of training number is 200, an evaluation index is MSE, an optimization (optimizer) is Adam (updating of the optimizer, a learning rate of each parameter is dynamically adjusted by using a first moment estimation and a second moment estimation of a gradient), a result of a neural network model prediction data is shown in fig. 9, and the result is calculated by:
using the neural network model, the decision coefficient reaches 0.814517,
the segmentation look model does not take into account the special cases of 70-90 and 210-230, which are not representative and have small data size, in root mean square error of each segment, and the result is as follows:
the root mean square error of the neural network within 90-110s is about 16.57/100=16.6%
The root mean square error of the neural network within 110-130s is about 17.42/120=14.5%
The root mean square error of the neural network within 130-150s is about 19.99/140=14.3%
The root mean square error of the neural network within 150-170s is about 23.17/160=14.5%
The root mean square error of the neural network within 170-190s is about 28.36/180=15.7%
The root mean square error of the neural network within 190-210s is about 34.69/200=17.3%
The prediction error is about 10% -20%.
Figure 10 shows a comparison of the static time matrix and the predicted result of the dynamic time estimation according to the present invention,
step 5: verifying a dynamic time estimation model: comparing the predicted value of the machine learning model predicted value neural network model with a static time matrix, and evaluating the performance of the dynamic time prediction model;
in order to verify the reliability of the machine learning model, the neural network model predictive value of the machine learning model predictive value is compared with a static time matrix, the prediction of time belongs to a regression problem, a typical regression performance evaluation standard is Mean Square Error (MSE), the root mean square error of the machine learning model predictive value, the static time matrix value and the neural network model predictive value are calculated respectively, and the performance of the machine learning model is evaluated through comparison of calculation results. As shown in fig. 10, it can be seen from fig. 10 that the root mean square error between the actual time consumption and the mean value and the root mean square error between the actual time consumption and the static time matrix are calculated, and the above comparison data can be seen that the root mean square error of the static time matrix and the mean value is far greater than the predicted value of the machine learning model, the root mean square error between the predicted value and the true value of the neural network model, and the machine learning model and the neural network model significantly improve the time prediction effect from the overall angle analysis. The model trained according to the historical information combines time information and site information, a model result of dynamic time estimation is greatly optimized, the deviation between the predicted result and the actual result can be within 20%, and the phenomenon that resources are idle or low in efficiency is generated in each link of port operation is improved.
Finally, it should be noted that while the present invention has been described with reference to the presently preferred embodiments, those skilled in the art will recognize that the above embodiments are for illustration purposes only and not for limitation, and that various equivalent changes or substitutions may be made without departing from the spirit of the invention, therefore, the invention as described in the foregoing embodiments will fall within the scope of the appended claims.
Claims (5)
1. The method for estimating the dynamic time of the AGV of the horizontal transport task for the automatic wharf port operation is characterized by comprising the following steps of:
step 1: feature selection: selecting and screening the AGV operation instruction characteristics of the horizontal transport task by using a computer algorithm, and integrating characteristic data, wherein the characteristic selection comprises the following steps:
s11, extracting data from a processing order table; selecting a record about an AGV operation instruction from an AGV_ORDERS table;
s12, processing the extracted data in the track table; finding one or more indexes, screening out key points in the running process of the vehicle according to one Order, and dividing the Track corresponding to each Order data according to the key points;
s13, integrating data: counting the number of the collision of the AGV paths, generating a total characteristic,
the total characteristics in the step S13 comprise a vehicle driving distance, a straight running frequency, a turning frequency, a diagonal running frequency, a conflict vehicle number, unloading, loading, piling, box exchange, box taking, PB section to PB section, PB section to WSTP section, WSTP section to PB section and WSTP section to WSTP section; PB is a sea-side operation area, and WSTP is a sea-side exchange area;
step 2: and (3) data processing: screening and processing the characteristic data in the step 1 by using a computer algorithm to form standardized data, wherein the method comprises the following steps of:
s21: analyzing the characteristic data, eliminating abnormal values for a large amount of error data and invalid data which are obviously present;
s22: further screening abnormal values, and clustering two direct factors of time spent and total distance;
s23: after the clustering process is completed, selecting data meeting the conditions as final training data;
s24: normalizing, normalizing and single-heat encoding the data;
step 3: machine learning: adopting various machine learning models to participate in prediction on the standardized data in the step 2, and searching for optimal parameters;
step 4: neural network: constructing a full-connection layer neural network, predicting the standardized data in the step 2 by using a neural network model,
the first layer of the fully-connected layer neural network is input, the last layer is output, a hidden layer with a 128 x 128 structure is shared, the value range of the output of each layer is compressed to (0, 1) by using an activation function, the value of the output of each stage is used as the input of the next stage, and the condition that certain input is infinite and other inputs are invalid is avoided; finally, a solution method of back propagation and gradient descent is utilized to realize accurate expression of the model on data; the neural network calculates a series of transformations that alter the similarity of the samples; the activity of each layer of neurons is a nonlinear function of the activity of the previous layer;
step 5: verifying a dynamic time estimation model: and comparing the predicted value of the machine learning model, the predicted value of the neural network model and the static time matrix, and evaluating the performance of the dynamic time prediction model.
2. The method for estimating the dynamic time of the horizontal transport task AGV for the automatic wharf port operation according to claim 1, wherein said step S12 comprises the steps of:
s121: according to the MoveFrequency field of the data table after feature extraction, counting and calculating the times of straight running, turning and inclined running of the AGV trolley;
s122: calculating the path length of the track according to the data in the step S121;
s123: processing the time of the start and end of the data in the above step S121;
s124: and calculating the running speed of the trolley according to the starting time, the ending time and the track distance.
3. The method for estimating the dynamic time of the horizontal transport task AGV for an automated dock-port job according to claim 1 wherein the machine learning models in step 3 include XGB, GBDT, randomForest, decisionTree, ridge and linearregprecision.
4. The method for estimating the dynamic time of the horizontal transport task AGV for the port operation of the automatic wharf according to claim 3, wherein the method for searching the optimal parameters in the various models by machine learning in the step 3 adopts a grid search method.
5. The method for estimating the dynamic time of the horizontal transport task AGV for the port operation of the automatic wharf according to claim 1, wherein the evaluation criterion for verifying the prediction of time by the machine learning model in the step 5 is root mean square error, the root mean square error between the predicted value, the static time matrix value, the average value and the true value of the machine learning model is calculated respectively, and the performance of the machine learning model is evaluated by comparing the calculation results.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110027744.2A CN112734111B (en) | 2021-01-11 | 2021-01-11 | Horizontal transport task AGV dynamic time prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110027744.2A CN112734111B (en) | 2021-01-11 | 2021-01-11 | Horizontal transport task AGV dynamic time prediction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112734111A CN112734111A (en) | 2021-04-30 |
CN112734111B true CN112734111B (en) | 2023-12-22 |
Family
ID=75589936
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110027744.2A Active CN112734111B (en) | 2021-01-11 | 2021-01-11 | Horizontal transport task AGV dynamic time prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112734111B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114137978A (en) * | 2021-11-29 | 2022-03-04 | 佛山市毕佳索智能科技有限公司 | Speed planning and control method for storage AGV |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106971545A (en) * | 2017-05-16 | 2017-07-21 | 青岛大学 | A kind of bus arrival time Forecasting Methodology |
KR101897407B1 (en) * | 2017-06-14 | 2018-10-18 | 국방과학연구소 | Method of Adaptive Dynamic Model-base]d Optimal Path Planning for Autonomous Navigation of Unmanned Ground Vehicle and Appratus thereof |
CN110570678A (en) * | 2019-10-23 | 2019-12-13 | 厦门大学 | Method and device for predicting total travel time of bus from starting point to end point |
CN112149888A (en) * | 2020-09-08 | 2020-12-29 | 北京踏歌智行科技有限公司 | Real-time travel time prediction method for unmanned mine card for strip mine |
-
2021
- 2021-01-11 CN CN202110027744.2A patent/CN112734111B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106971545A (en) * | 2017-05-16 | 2017-07-21 | 青岛大学 | A kind of bus arrival time Forecasting Methodology |
KR101897407B1 (en) * | 2017-06-14 | 2018-10-18 | 국방과학연구소 | Method of Adaptive Dynamic Model-base]d Optimal Path Planning for Autonomous Navigation of Unmanned Ground Vehicle and Appratus thereof |
CN110570678A (en) * | 2019-10-23 | 2019-12-13 | 厦门大学 | Method and device for predicting total travel time of bus from starting point to end point |
CN112149888A (en) * | 2020-09-08 | 2020-12-29 | 北京踏歌智行科技有限公司 | Real-time travel time prediction method for unmanned mine card for strip mine |
Non-Patent Citations (1)
Title |
---|
Learning heterogeneous traffic patterns for travel timeprediction of bus journeys;Peilan He et al;Information Sciences;第512卷;1394-1406 * |
Also Published As
Publication number | Publication date |
---|---|
CN112734111A (en) | 2021-04-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20140088865A1 (en) | Apparatus and method for predicting arrival times in a transportation network | |
CN114595607B (en) | Digital twin spinning barrel conveying method and system | |
CN111860989A (en) | Ant colony algorithm optimization-based LSTM neural network short-time traffic flow prediction method | |
CN114723125A (en) | Intercity vehicle order allocation method combining deep learning and multi-task optimization | |
CN112734111B (en) | Horizontal transport task AGV dynamic time prediction method | |
Seiler et al. | Flow-achieving online planning and dispatching for continuous transportation with autonomous vehicles | |
Gaidar et al. | Mathematical method for optimising the transport and logistics industry | |
Temkin et al. | Possible architecture and some neuro-fuzzy algorithms of an intelligent control system for open pit mines transport facilities | |
Arifin et al. | Development of ship allocation models using marine logistics data and its application to bulk carrier demand forecasting and basic planning support | |
Torabi et al. | Fuel‐Efficient Driving Strategies for Heavy‐Duty Vehicles: A Platooning Approach Based on Speed Profile Optimization | |
CN116629424A (en) | Novel automatic double-yard bridge intelligent scheduling method for container yard operation mode | |
CN113687651B (en) | Path planning method and device for dispatching vehicles on demand | |
CN114444809A (en) | Data-driven multi-target strip mine card path optimization method | |
Zhou et al. | Simulation‐Based Optimization for Yard Design at Mega Container Terminal under Uncertainty | |
Jun et al. | Evolutionary neural network for learning of scalable heuristics for pickup and delivery problems with time windows | |
CN116882673A (en) | Coal supply chain system and scheduling method | |
CN117094460A (en) | Multi-target travel business control method, system and medium based on OD data | |
Wang et al. | DSS approach for sustainable system design of shuttle-based storage and retrieval systems | |
CN115238789A (en) | Financial industry special data prediction method and system based on improved GRU | |
Gan et al. | Scheduling problems of automated guided vehicles in automated container terminals using a genetic algorithm | |
CN113222285A (en) | Strip mine charging pile site selection method based on adaptive disturbance goblet sea squirt algorithm | |
Ni et al. | Research on Port truck trajectory completion based on Long Short-Term Memory model and Speed distribution characteristics | |
Rezaeifar | Developing a framework to optimize the operations of an intermodal underground freight transportation terminal using simulation | |
CN113033893B (en) | Method for predicting running time of automatic guided vehicle of automatic container terminal | |
Mathias et al. | Concepts of Big Data Analysis of Container Terminals in the Digital Era |
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 |