CN111523641B - Sector delay prediction method based on ConvLSTM-SRU - Google Patents
Sector delay prediction method based on ConvLSTM-SRU Download PDFInfo
- Publication number
- CN111523641B CN111523641B CN202010277615.4A CN202010277615A CN111523641B CN 111523641 B CN111523641 B CN 111523641B CN 202010277615 A CN202010277615 A CN 202010277615A CN 111523641 B CN111523641 B CN 111523641B
- Authority
- CN
- China
- Prior art keywords
- time
- sector
- data
- network
- flight
- 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 37
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 23
- 230000004913 activation Effects 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 4
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 claims description 3
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000004519 manufacturing process Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 208000027744 congestion Diseases 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 10
- 238000013135 deep learning Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—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
- 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
- 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
- G06Q50/40—Business processes related to the transportation industry
-
- 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
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
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)
- Life Sciences & Earth Sciences (AREA)
- Strategic Management (AREA)
- Data Mining & Analysis (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a sector delay prediction method based on ConvLSTM-SRU, which comprises the following steps: (1) reading sector history flight ADS-B data; (2) processing the data; (3) dividing the index characteristics; (4) generating time-series data of each index; (5) Aiming at the space-time characteristic index, a ConvLSTM network space-time characteristic extraction model is established; (6) Establishing an SRU network time feature extraction model aiming at the time characteristic index; (7) establishing an index feature fully-connected network model; (8) outputting a predicted sector time of flight; (9) calculating the sector expected delay time. The prediction method can improve the prediction precision of the delay time of each route of the sector, provide prospective information for air traffic controllers, help the controllers to take control measures in advance, reduce the probability of increasing the congestion risk of the sector, and further improve the operation efficiency of the airport.
Description
Technical Field
The invention belongs to the technical field of air traffic flow management, and particularly relates to a sector delay prediction method based on ConvLSTM-SRU.
Background
The sector is taken as a control airspace unit, and the aircrafts in the range are concentrated on each air route, so that the air route is easy to be jammed in a state of large traffic, and the air route is spread to the whole sector and even other sectors. If the accuracy of predicting the delay time of each route of the sector in real time can be improved, important help is provided for relieving the congestion of the sector. The traditional methods for predicting the air traffic delay are mostly based on the sector flow, and although the prediction methods can judge which route the delay exists on, the prediction accuracy of the delay time is not enough, and the influence between the connected routes is not considered enough. In the process of sector operation, the route network structure of the sector is a key factor for determining the capacity of the sector, the flight time of the aircraft on the sector route (from the inlet to the outlet) is closely related to the operation state of the route, so that two aspects of route structure characteristics and flow rate must be considered in the process of predicting delay time.
Disclosure of Invention
The invention aims to: the invention aims to provide a sector delay time prediction method based on CNN-SRU, which effectively reduces the occurrence probability of sector congestion and improves the operation efficiency.
The technical scheme is as follows: the invention relates to a sector delay prediction method based on ConvLSTM-SRU, which comprises the following steps:
(1) Reading data: reading sector historical flight ADS-B data, counting to obtain flight flow data of each route point from a sector, flight flow data and flight time data of each route (from the sector inlet point to the sector outlet point) in the sector, and reading sector historical meteorological data in a corresponding time range;
(2) Processing data: arranging the data sets read in the step (1) according to time labels, filling the data missing on a time axis by using the average value of the time periods before and after, and replacing abnormal data on the time axis by using an interpolation method;
(3) Dividing index characteristics: the indexes are classified according to the index attribute, and the two indexes are classified into space-time characteristic indexes because the flying flow and the flying time of each route (from a sector inlet point to a sector outlet point) in the sector change along with time and space (a sector route network); the flying flow and the meteorological conditions of each route point entering and leaving the sector only change along with time in the same sector and are irrelevant to space, so the two indexes are divided into time characteristic indexes;
(4) Generating time series data of each index: on the basis of the step (3), selecting different time intervals (10 min,20min,30min and 60 min), and counting the data of each index to generate sequence data of the corresponding time intervals;
(5): aiming at the space-time characteristic index, a ConvLSTM network space-time characteristic extraction model is established;
(6): establishing an SRU network time feature extraction model aiming at the time characteristic index;
(7): establishing an index feature full-connection network model;
(8): outputting the predicted sector flight time;
(9): the sector estimated delay time and the prediction accuracy are calculated.
Further, the specific process of the step (5) is as follows:
(5.1) generating ConvLSTM network input set
Counting the number of the fan-in and fan-out waypoints to be m and n, and manufacturing a picture with the size of m multiplied by n, wherein the numerical value of each pixel point is the flight flow of the corresponding air route between the fan-in and fan-out points, generating an original picture of the flight flow at a desired corresponding time interval according to the time sequence data obtained in the step (4), and similarly, generating the original picture of the flight time at the corresponding time interval by taking the flight time of the air route between the fan-in and fan-out points as the pixel point;
(5.2) initializing ConvLSTM network parameters
(5.2.1) initializing input layer parameters
When constructing the input layer of the convLSTM network, the initial set value of the number of samples and the time step of each batch of training is given, and preferably, the initial number of samples of each batch of training is set to be 48; setting the initial value of the time step to be 6 time intervals;
(5.2.2) initializing network layer parameters
When constructing a network layer of the ConvLSTM network, setting initial set values of a convolution dimension, an input dimension, an output depth and a convolution kernel size to be 2; setting an initial setting value of an input dimension as [ m, n,1]; setting the initial setting value of the output depth to be 6; setting the initial size of the convolution kernel to be [3,3]; the activation function initially selects Relu; initializing the weight and bias of a network layer;
(5.3) setting input sample and output characteristics of ConvLSTM network
(5.3.1) constructing input samples
Dividing the original pictures of the flight flow and the flight time of the route obtained in the step (5.1) into a plurality of time sequences according to the initial time step set in the step (5.2.1);
(5.3.2) building output features
The network output characteristic takes the characteristic value of the last time step;
(5.3.3) data normalization
And normalizing the input sample by adopting min-max to generate a dimensionless training data set.
Further, the specific process of step (6) is as follows:
(6.1) initializing SRU network parameters;
(6.1.1) initializing input layer parameters
When an input layer of the SRU network is constructed, giving initial set values of the number of samples and time steps of each batch of training; preferably, the initial number of samples per batch of training is set to 48; setting the initial value of the time step to be 6 time intervals; (6.1.2) initializing network layer parameters
When constructing a network layer of an SRU network, giving initial setting values of the number of network layers and the number of neurons of each layer; preferably, the initial setting value of the network layer number is set to be 2; setting the initial value of each layer of neurons to be 50; the activation function initially selects Relu, and initializes the weight and bias of the network layer;
(6.2) setting input samples and output characteristics of SRU networks
(6.2.1) constructing the input samples
Dividing the flying flow and the meteorological data sequence of the entering and exiting fan waypoints obtained in the step (4) into a plurality of time sequences according to the initial time step set in the step (6.1.1);
(6.2.2) building output features
The network output characteristic takes the characteristic value of the last time step;
(6.2.3) data normalization
And normalizing the input sample by adopting min-max to generate a dimensionless training data set.
Further, the specific process of step (7) is as follows:
(7.1) initializing index feature fully connected network parameters
(7.1.1) initializing weights and biases of the input layers;
(7.1.2) initially selecting Relu by the output layer activation function;
(7.2) setting input and output samples of the index feature fully connected network
(7.2.1) constructing the input samples
Converting the space-time characteristic value output in the step (5.3.2) into a one-dimensional vector, and using the one-dimensional vector and the time characteristic extracted in the step (6.2.2) together as an input sample of a fully-connected network;
(7.2.2) constructing output samples
Taking the data of the last time step of the time sequence of flight obtained in the step (5.3.1) as an output sample of the index feature fully-connected network;
(7.2.3) data normalization
And normalizing the output sample by adopting min-max to generate a dimensionless training data set.
Further, the specific process of step (8) is as follows:
(8.1) setting training set and test set
Randomly extracting the data set obtained in the step (4) to obtain a training set, wherein k (0 < k < 1) is used as a test set, and k=0.8 is preferable;
(8.2) loss function setup
Selecting a square error loss function in a sector route flight time prediction neural network;
(8.3) optimizer settings
The optimizer selects Adam, adagrad, RMSprop, NAG, SGD to perform comparison experiments, preferably Adam;
(8.4) data Denormalization
And (3) carrying out min-max normalized inverse operation on the output value of the test set, wherein the obtained value is the predicted flight time.
Further, the specific process of step (9) is as follows:
(9.1) calculating a sector route time of flight reference
Sequencing the flight time data of each route of the sector obtained in the step (2) from small to large in sequence, and selecting a flight time reference value, wherein the flight time reference value is preferably a value corresponding to 20% of the quantiles;
(9.2) calculating sector expected delay time
Calculating the difference between the predicted flight time obtained in the step (8.4) and the flight time reference value obtained in the step (9.1) to obtain the predicted delay time of the sector;
(9.3) calculating the prediction precision
And selecting an average error duty ratio (MAPE) as an evaluation index, and calculating the prediction accuracy of the sector delay time.
Further, the method also comprises a step (10), and the specific process is as follows:
(10.1) calculating prediction accuracy of LSTM and GRU networks
Expanding the two-dimensional data generated in the step (5.1) to form one-dimensional data, combining the one-dimensional data with the data in the step (6), respectively inputting LSTM and GRU networks for prediction, and calculating prediction accuracy;
(10.2) prediction accuracy contrast
And (3) comparing the sector delay time prediction accuracy calculated in the step (9.3) with the prediction accuracy of the common deep learning method (LSTM and GRU) in the step (10.1).
The beneficial effects are that: the sector delay prediction method based on ConvLSTM-SRU can improve the prediction precision of sector delay time, provide prospective information for air traffic controllers, help the controllers to take control measures in advance, reduce the probability of increasing the sector congestion risk, and further improve the operation efficiency of airports. Compared with the traditional method for deducing delay time based on the flow, the method considers the factors of the flow and the own route network structure of the sector, has better adaptability and has practical engineering application value in the aspect of predicting the delay time of the sector.
Drawings
FIG. 1 is a flow chart of a ConvLSTM-SRU based sector delay prediction method of the present invention.
Detailed Description
For a further understanding of the present invention, the present invention will be described in detail with reference to the drawings and examples.
The sector delay prediction method based on ConvLSTM-SRU comprises the following steps:
(1) Reading data: reading sector historical flight ADS-B data, and counting to obtain flight flow data of each route point from a sector, flight flow data and flight time data of each route (from the sector inlet point to the sector outlet point) in the sector, reading sector historical meteorological data in a corresponding time range, wherein table 1 and table 2 are part of experimental data obtained by counting when the time interval is 10 min;
TABLE 1
TABLE 2
(2) Processing data: arranging the data sets read in the step (1) according to time labels, filling the data missing on a time axis by using the average value of the time periods before and after, and replacing abnormal data on the time axis by using an interpolation method;
(3) Dividing index characteristics: the indexes are classified according to the index attribute, and the two indexes are classified into space-time characteristic indexes because the flying flow and the flying time of each route (from a sector inlet point to a sector outlet point) in the sector change along with time and space (a sector route network); the flying flow and the meteorological conditions of each route point entering and leaving the sector only change along with time in the same sector and are irrelevant to space, so the two indexes are divided into time characteristic indexes;
(4) Generating time series data of each index: on the basis of the step (3), selecting different time intervals (10 min,20min,30min and 60 min), and counting the data of each index to generate sequence data of the corresponding time intervals;
(5): for space-time characteristic indexes, a ConvLSTM network space-time characteristic extraction model is established
(5.1) generating ConvLSTM network input set
Counting the number of the fan-in and fan-out waypoints to be 5 and 4, manufacturing pictures with the size of 5 multiplied by 4, wherein the numerical value of each pixel point is the flight flow of the corresponding inter-fan-in and fan-out waypoint route, generating the original pictures of the flight flow at the time interval according to the time sequence data obtained in the step (4), and similarly, generating the original pictures of the flight time at the corresponding time interval by taking the flight time of the inter-fan-in and fan-out waypoint route as the pixel point;
(5.2) initializing ConvLSTM network parameters
(5.2.1) initializing input layer parameters
When an input layer of a ConvLSTM network is constructed, setting the initial sample number of each batch of training to be 48; setting the initial value of the time step to be 6 time intervals;
(5.2.2) initializing network layer parameters
When a network layer of a ConvLSTM network is constructed, setting an initial setting value of a convolution dimension to be 2; setting an initial setting value of an input dimension to [5,4,1]; setting the initial setting value of the output depth to be 6; setting the initial size of the convolution kernel to be [3,3]; the activation function initially selects Relu; initializing the weight and bias of a network layer;
(5.3) setting input sample and output characteristics of ConvLSTM network
(5.3.1) constructing input samples
Dividing the original pictures of the flight flow and the flight time of the route obtained in the step (5.1) into a plurality of time sequences according to the initial time step set in the step (5.2.1);
(5.3.2) building output features
The network output characteristic takes the characteristic value of the last time step;
(5.3.3) data normalization
Normalizing an input sample by adopting min-max to generate a dimensionless training data set;
(6): establishing SRU network time feature extraction model aiming at time characteristic indexes
(6.1) initializing SRU network parameters;
(6.1.1) initializing input layer parameters
When an input layer of the SRU network is constructed, setting the initial sample number of each batch of training to be 48; setting the initial value of the time step to be 6 time intervals; the method comprises the steps of carrying out a first treatment on the surface of the
(6.1.2) initializing network layer parameters
When a network layer of the SRU network is constructed, setting an initial setting value of the network layer number to be 2; setting the initial value of each layer of neurons to be 50; the activation function initially selects Relu, and initializes the weight and bias of the network layer;
(6.2) setting input samples and output characteristics of SRU networks
(6.2.1) constructing the input samples
Dividing the flying flow and the meteorological data sequence of the entering and exiting fan waypoints obtained in the step (4) into a plurality of time sequences according to the initial time step set in the step (6.1.1);
(6.2.2) building output features
The network output characteristic takes the characteristic value of the last time step;
(6.2.3) data normalization
Normalizing an input sample by adopting min-max to generate a dimensionless training data set;
(7): establishing index feature full-connection network model
(7.1) initializing index feature fully connected network parameters
(7.1.1) initializing weights and biases of the input layers;
(7.1.2) initially selecting Relu by the output layer activation function;
(7.2) setting input and output samples of the index feature fully connected network
(7.2.1) constructing the input samples
Converting the space-time characteristic value output in the step (5.3.2) into a one-dimensional vector, and using the one-dimensional vector and the time characteristic extracted in the step (6.2.2) together as an input sample of a fully-connected network;
(7.2.2) constructing output samples
Taking the data of the last time step of the time sequence of flight obtained in the step (5.3.1) as an output sample of the index feature fully-connected network;
(7.2.3) data normalization
Normalizing the output sample by adopting min-max to generate a dimensionless training data set;
(8): outputting predicted sector time of flight
(8.1) setting training set and test set
Randomly extracting 80% of the data set obtained in the step (4) to serve as a training set, and the remaining 20% of the data set serves as a test sample;
(8.2) loss function setup
Selecting a square error loss function in a sector route flight time prediction neural network;
(8.3) optimizer settings
The optimizer selects Adam;
(8.4) data Denormalization
Performing min-max normalized inverse operation on the output value of the test set, wherein the obtained value is the predicted flight time;
(9): calculating sector expected delay time and prediction accuracy
(9.1) calculating a sector route time of flight reference
Sequencing the flight time data of each route of the sector obtained in the step (2) from small to large in sequence, and selecting a flight time reference value as a value corresponding to 20% of the quantiles;
(9.2) calculating sector expected delay time
Calculating the difference between the predicted flight time obtained in the step (8.4) and the flight time reference value obtained in the step (9.1) to obtain the predicted delay time of the sector;
(9.3) calculating the prediction precision
Selecting average error duty ratio (MAPE) as an evaluation index, and calculating the prediction accuracy of sector delay time;
(10): comparing the prediction precision;
(10.1) calculating prediction accuracy of LSTM and GRU networks
Expanding the two-dimensional data generated in the step (5.1) to form one-dimensional data, combining the one-dimensional data with the data in the step (6), respectively inputting LSTM and GRU networks for prediction, and calculating prediction accuracy;
(10.2) prediction accuracy contrast
Comparing the sector delay prediction result calculated in the step (9.3) and based on ConvLSTM-SRU with the prediction precision of the common deep learning method (LSTM, GRU) calculated in the step (10.1), as shown in the table 3, the method provided by the invention can improve the sector delay time prediction precision.
TABLE 3 Table 3
Claims (1)
1. The sector delay prediction method based on ConvLSTM-SRU is characterized by comprising the following steps:
(1) Reading data: reading sector historical flight ADS-B data, counting to obtain flight flow data of each route point from a sector, flight flow data and flight time data of each route in the sector, and reading sector historical meteorological data in a corresponding time range;
(2) Processing data: arranging the data sets read in the step (1) according to time labels, filling the data missing on a time axis by using the average value of the time periods before and after, and replacing abnormal data on the time axis by using an interpolation method;
(3) Dividing index characteristics: the method comprises the steps of performing type division on indexes according to index attributes, and dividing the flight flow and the flight time of each route in a sector into space-time characteristic indexes because the flight flow and the flight time of each route in the sector change along with time and space; the flying flow and meteorological conditions of each waypoint entering and leaving the sector only change along with time in the same sector and are irrelevant to space, so that the flying flow and meteorological conditions of each waypoint entering and leaving the sector are divided into time characteristic indexes;
(4) Generating time series data of each index: selecting different time intervals on the basis of the step (3), and counting the data of each index to generate sequence data of the corresponding time intervals;
(5): aiming at the space-time characteristic index, a ConvLSTM network space-time characteristic extraction model is established;
(6): establishing an SRU network time feature extraction model aiming at the time characteristic index;
(7): establishing an index feature full-connection network model;
(8): outputting the predicted sector flight time;
(9): calculating the estimated delay time and the estimated precision of the sector;
the specific process of the step (5) is as follows:
(5.1) generating ConvLSTM network input set
Counting the number of the fan-in and fan-out waypoints to be m and n, and manufacturing a picture with the size of m multiplied by n, wherein the numerical value of each pixel point is the flight flow of the corresponding inter-fan-in and fan-out waypoint route, taking the time sequence data obtained in the step (4) as the pixel point, generating an original picture of the flight flow at the corresponding time interval, and similarly, taking the flight time of the inter-fan-in and fan-out waypoint route as the pixel point, and generating the original picture of the flight time at the corresponding time interval;
(5.2) initializing ConvLSTM network parameters
(5.2.1) initializing input layer parameters
When an input layer of a ConvLSTM network is constructed, giving initial setting values of the number of samples and time steps of each batch of training;
(5.2.2) initializing network layer parameters
When constructing a network layer of a ConvLSTM network, giving initial set values of convolution dimension, input dimension, output depth and convolution kernel size; the activation function initially selects Relu, and initializes the weight and bias of the network layer;
(5.3) setting input sample and output characteristics of ConvLSTM network
(5.3.1) constructing input samples
Dividing the original pictures of the flight flow and the flight time of the route obtained in the step (5.1) into a plurality of time sequences according to the initial time step set in the step (5.2.1);
(5.3.2) building output features
The network output characteristic takes the characteristic value of the last time step;
(5.3.3) data normalization
Normalizing an input sample by adopting min-max to generate a dimensionless training data set;
the specific process of the step (6) is as follows:
(6.1) initializing SRU network parameters;
(6.1.1) initializing input layer parameters
When an input layer of the SRU network is constructed, giving initial set values of the number of samples and time steps of each batch of training; (6.1.2) initializing network layer parameters
When constructing a network layer of an SRU network, giving initial setting values of the number of network layers and the number of neurons of each layer; the activation function initially selects Relu, and initializes the weight and bias of the network layer;
(6.2) setting input samples and output characteristics of SRU networks
(6.2.1) constructing the input samples
Dividing the flying flow and the meteorological data sequence of the entering and exiting fan waypoints obtained in the step (4) into a plurality of time sequences according to the initial time step set in the step (6.1.1);
(6.2.2) building output features
The network output characteristic takes the characteristic value of the last time step;
(6.2.3) data normalization
Normalizing an input sample by adopting min-max to generate a dimensionless training data set;
the specific process of the step (7) is as follows:
(7.1) initializing index feature fully connected network parameters
(7.1.1) initializing weights and biases of the input layers;
(7.1.2) initially selecting Relu by the output layer activation function;
(7.2) setting input and output samples of the index feature fully connected network
(7.2.1) constructing the input samples
Converting the space-time characteristic value output in the step (5.3.2) into a one-dimensional vector, and using the one-dimensional vector and the time characteristic extracted in the step (6.2.2) together as an input sample of a fully-connected network;
(7.2.2) constructing output samples
Taking the data of the last time step of the time sequence of flight obtained in the step (5.3.1) as an output sample of the index feature fully-connected network;
(7.2.3) data normalization
Normalizing the output sample by adopting min-max to generate a dimensionless training data set;
the specific process of the step (8) is as follows:
(8.1) setting training set and test set
Randomly extracting the data set k obtained in the step (4) as a training set, wherein 0< k <1, and the rest is used as a test set;
(8.2) loss function setup
Selecting a square error loss function in a sector route flight time prediction neural network;
(8.3) optimizer settings
The optimizer selects Adam;
(8.4) data Denormalization
Performing min-max normalized inverse operation on the output value of the test set, wherein the obtained value is the predicted flight time;
the specific process of the step (9) is as follows:
(9.1) calculating a sector route time of flight reference
Sequencing the flight time data of each route of the sector obtained in the step (2) from small to large in sequence, and selecting a flight time reference value;
(9.2) calculating sector expected delay time
Calculating the difference between the predicted flight time obtained in the step (8.4) and the flight time reference value obtained in the step (9.1) to obtain the predicted delay time of the sector;
(9.3) calculating the prediction precision
And selecting MAPE as an evaluation index, and calculating the prediction accuracy of the sector delay time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010277615.4A CN111523641B (en) | 2020-04-10 | 2020-04-10 | Sector delay prediction method based on ConvLSTM-SRU |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010277615.4A CN111523641B (en) | 2020-04-10 | 2020-04-10 | Sector delay prediction method based on ConvLSTM-SRU |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111523641A CN111523641A (en) | 2020-08-11 |
| CN111523641B true CN111523641B (en) | 2023-05-30 |
Family
ID=71902708
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010277615.4A Active CN111523641B (en) | 2020-04-10 | 2020-04-10 | Sector delay prediction method based on ConvLSTM-SRU |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111523641B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105225007A (en) * | 2015-09-30 | 2016-01-06 | 中国民用航空总局第二研究所 | A kind of sector runnability method for comprehensive detection based on GABP neural network and system |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105679102B (en) * | 2016-03-03 | 2018-03-27 | 南京航空航天大学 | A kind of national flight flow spatial and temporal distributions prediction deduction system and method |
| CN106023655B (en) * | 2016-06-30 | 2019-04-05 | 南京航空航天大学 | Traffic congestion state monitoring method in a kind of sector null |
| CN109658741A (en) * | 2018-12-12 | 2019-04-19 | 中国船舶重工集团公司第七0九研究所 | A kind of sector short term traffic forecasting method and system |
-
2020
- 2020-04-10 CN CN202010277615.4A patent/CN111523641B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105225007A (en) * | 2015-09-30 | 2016-01-06 | 中国民用航空总局第二研究所 | A kind of sector runnability method for comprehensive detection based on GABP neural network and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111523641A (en) | 2020-08-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Prajwala | A comparative study on decision tree and random forest using R tool | |
| CN109472321B (en) | Time series type earth surface water quality big data oriented prediction and evaluation model construction method | |
| CN108985380B (en) | Point switch fault identification method based on cluster integration | |
| CN109492748B (en) | Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network | |
| CN112489497B (en) | Airspace operation complexity evaluation method based on deep convolutional neural network | |
| CN112966954A (en) | Flood control scheduling scheme optimization method based on time convolution network | |
| CN110164129B (en) | Single-intersection multi-lane traffic flow prediction method based on GERNN | |
| CN110442143B (en) | Unmanned aerial vehicle situation data clustering method based on combined multi-target pigeon swarm optimization | |
| CN109657709A (en) | A kind of equipment fault prediction technique based on particle group optimizing support vector regression | |
| CN111626366B (en) | Operation characteristic-based area sector scene similarity identification method | |
| CN110708318A (en) | Network abnormal flow prediction method based on improved radial basis function neural network algorithm | |
| CN112465199A (en) | Airspace situation evaluation system | |
| CN113657814B (en) | Aviation network risk prediction method and risk grade evaluation method | |
| CN109460872B (en) | Mobile communication user loss imbalance data prediction method | |
| CN111523641B (en) | Sector delay prediction method based on ConvLSTM-SRU | |
| Sagar et al. | Analysis of prediction techniques based on classification and regression | |
| CN116882584A (en) | Flight delay prediction method and system | |
| CN116578858A (en) | Air compressor fault prediction and health degree evaluation method and system based on graphic neural network | |
| CN115310999B (en) | Enterprise electricity behavior analysis method and system based on multi-layer perceptron and sequencing network | |
| CN113570120A (en) | Wind power prediction method and device based on improved particle swarm optimization | |
| CN111368408B (en) | Data center energy efficiency optimization continuous decision method based on seq2seq | |
| CN111652102A (en) | Power transmission channel target object identification method and system | |
| CN117422320B (en) | Method for extracting influence factors of weather on flight toughness operation | |
| CN111382761A (en) | CNN-based detector, image detection method and terminal | |
| CN113222229B (en) | Non-cooperative unmanned aerial vehicle track prediction method based on machine 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |