A method for generating activity area summary based on aircraft track sequence processing
The aircraft track data is processed through the recurrent neural network model, divided into grid units and generated natural language descriptions, which solves the problem of converting track data into concise descriptions in the prior art, and improves efficiency and standardization.
Patent Information
- Application Number
- CN202210124194.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-02-10
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-02-10
AI Technical Summary
In the process of monitoring and statistics of aircraft navigation trajectory, the prior art requires a lot of manual statistics and calculations, which leads to high time costs and is difficult to automatically convert track data into concise text descriptions.
The recurrent neural network model is trained to generate a brief description generation model. By dividing the track data into grid cells and annotating it, a data set is formed, and a recurrent neural network is used to generate a concise description of natural language.
The automated track data is converted into concise description, which improves the standardization and efficiency of track description, reduces manual intervention and reduces time costs.
Smart Images

Figure CN114547973B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for generating activity area summary based on aircraft track path sequence processing. Background Art
[0002] Track data is widely present in the real world, such as the running tracks of football players in sports games, the driving routes of cars in road traffic, the routes of animal migration, etc. By analyzing track data and its associated background information, researchers can gain insights into athletes' usual running routes, car traffic conditions, and the areas where animals stay during migration.
[0003] In the fields of air traffic control and shipping, relevant staff need to monitor, observe, and count the flight paths of target aircraft, and then form a written summary report. The whole process requires professional knowledge, but also involves a large number of statistical and calculation processes, which is cumbersome and time-consuming. In this process, it is necessary to combine the geographic information near the track to convert the latitude and longitude data of the target's track over a period of time into a text description. According to the survey, only a few studies have been carried out in this field and business-specific products. Summary of the invention
[0004] Purpose of the invention: The purpose of the present invention is to provide a method for generating a brief description of an activity area based on the sequence processing of an aircraft track path, so as to automatically convert the track data into a concise summary description text. In the process of generating, the original time series distribution expression and restoration of the track are emphasized. The method of the present invention can generate a brief description of the activity area of the aircraft track in the area in combination with the landmarks in the predetermined area, and form a concise description text in natural language.
[0005] Technical solution: To achieve the above-mentioned invention object, the technical solution adopted by the present invention is as follows:
[0006] Determine the area where track generation can be performed. Tracks must be completely contained in the area to be generated. If there are tracks outside the area, the tracks outside the area cannot participate in the generation calculation, which may cause errors in the generated results.
[0007] The present invention comprises the following steps:
[0008] Step 1, training phase: using the recurrent neural network model to generate a summary generation model;
[0009] Step 2: Use phase: Apply the brief description generation model to perform computational processing on the input track to be processed and generate a text description of the track.
[0010] Step 1 includes the following steps:
[0011] Step 1-1: The user sets the global area where the track needs to be described in advance according to the usage requirements;
[0012] Step 1-2, dividing the global area established in step 1-1 into a regular grid array according to longitude and latitude;
[0013] Step 1-3, read each discrete point in the structured track of the aircraft, and calculate and project it to the grid unit;
[0014] Step 1-4: Arrange the corresponding relationship between the track points and grid cells obtained in step 1-3 according to the corresponding time sequence of the discrete points in the aircraft track to form a grid sequence corresponding to the track [ <c-Idx mn >], where c is arranged in ascending order from 0 to C-1 in chronological order;
[0015] Step 1-5: The user manually selects landmarks in the global area where the track needs to be described according to the description habits, and manually identifies the landmarks Mk s , where s∈[1, S], s∈N, s represents the subscript of the landmark. According to the track characteristics, the grid sequence corresponding to the structured track Tk of the aircraft is [ <c-Idx mn >] to mark and obtain the marking result R;
[0016] Step 1-6: Add the labeled raster sequence to the dataset;
[0017] Step 1-7, read the set of all aircraft tracks and determine whether all tracks have been processed. If not, proceed to step 1-8. If yes, proceed to step 1-9.
[0018] Step 1-8, select the next unprocessed track and return to step 1-3;
[0019] Step 1-9: Collect all the labeled raster sequences, align and fill them to form a data set. The data set is roughly as follows, where "pad" is recorded as a special filling value:
[0020] 2,3,4,5 Zhen, Pearl, Spring, Yi, North 9,15,<pad>,<pad> Zhen, Zhu, Quan, Yi, Xi …… ……
[0021] Step 1-10: Conduct training by building a recurrent neural network model, with the input being a grid sequence [ <c-Idx mn >], the output is the text R annotated by the annotator for the track, using the data set formed in steps 1-9 and conducting training;
[0022] Step 1-11: Form a trained summary generation model.
[0023] Step 1-1 includes: requiring the global area to cover all tracks that need to be described, assuming that the global area range is: lat∈[Lat min , Lat max ],lon∈[Lon min , Lon max ], and the side length l, where lat refers to latitude, Lat min Refers to setting the lower latitude limit of the global area, Lat max Refers to setting the upper limit of latitude for the global area, lon refers to longitude, Lon min Refers to setting the lower limit of longitude for the global area, Lon max Refers to setting the upper longitude limit of the global region.
[0024] In step 1-2, the longitude and latitude are divided into grid cells G according to the spacing of side length l. mn ;
[0025] Where m∈[0,M), m refers to the longitude subscript of the grid unit, M refers to the maximum value calculated by the grid cell longitude subscript;
[0026] n∈[0,N), n refers to the latitude subscript of the grid cell, N refers to the maximum value calculated by the grid cell latitude subscript;
[0027] Let the latitude of north be positive, the latitude of south be negative, the longitude of east be positive, and the longitude of west be negative;
[0028] Number each grid cell, number Idx mn For: Idx mn =m×N+n.
[0029] Steps 1-3 include: c Represents the cth discrete point, c∈[0,C), C represents the total number of points in the track, forming a corresponding relationship between track points and grid cells <c-Idx mm >, <c-Idx mn > indicates that the cth point in the track falls into grid cell G mn middle.
[0030] In steps 1-3, if the track point Tk c In grid cell G mn In the equation, F(Tk c ,G mn )=1, otherwise it is recorded as F(Tk c ,G mn )=0, the following formula will be used to calculate Tk c With grid cell G mn Projection relationship:
[0031]
[0032] Among them, Tk c,Lat and Tk c,Lon Represents the track point Tk c The latitude and longitude values in .
[0033] Step 2 includes the following steps:
[0034] Step 2-1: Read the structured track Tk′ of the aircraft to be generated c′ ;
[0035] Step 2-2, perform projection;
[0036] Step 2-3: Form the grid sequence corresponding to the track [ <c′-Idx′ m′n′ >];
[0037] Step 2-4: The grid sequence obtained in step 2-3 [ <c′-Idx′ m′n′ >] Input a brief description of the model to generate and get the output
[0038] Step 2-5: The user determines the generated result obtained in step 2-4 by observing the relative position of the original track and the landmark Is it reasonable? If yes, then end. If not, proceed to step 2-6.
[0039] Step 2-6: The user observes the relative position of the original track and the landmark and generates the result obtained in step 2-4. Modify it to the result you think is appropriate and will As the track Tk′ c′ The resulting grid sequence [ <c′-Idx′ m′n′ >] and added it to the dataset.
[0040] Step 2-2 includes: reusing the grid cells divided in steps 1-1 and 1-2, and converting each discrete point Tk′ in the structured track to be generated c′ Project to the grid unit to form the corresponding relationship between the track point and the grid unit <c′-Idx mn >, <c′-Idx mn > indicates that the c′th point in the track falls into grid cell G mn Where c′∈[0,C′), C′ represents the total number of points in the track.
[0041] In step 2-2, if the track point Tk′ c′ In the grid cell G′ with the subscript m′n′m′n′ In the equation, it is denoted as F(Tk′ c′ , G′ m′n′ )=1, otherwise it is recorded as F(Tk′ c′ , G′ m′n′ )=0, the following formula will be used to calculate Tk′ c′ With grid cell G′ m′n′ Projection relationship:
[0042]
[0043] Among them, Tk′ c′,Lat and Tk′ c′,Lon Represent the track points Tk′ c′ The latitude and longitude values in .
[0044] Step 2-3 includes: arranging the corresponding relationship between the track points and the grid cells obtained in step 2-2 according to the corresponding time sequence of the discrete points in the aircraft track to form a grid sequence corresponding to the track [ <c′-Idx′ m′n′ >], where c′ is arranged in ascending order from 0 to C′-1 in chronological order.
[0045] Beneficial effects: Compared with the prior art, the significant advantages of the present invention are: (1) This method adopts a recurrent neural network generation algorithm similar to "translation", so that similar tracks can obtain similar results, and the generated results are limited, which ensures the standardization and convergence of the track description results; (2) The factors considered are more focused, focusing only on the location information in the track, which is used to generate a brief description of the activity area where the track is located, eliminating the interference of other irrelevant factors. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.
[0047] Figure 1 is a flow chart of the method of the present invention;
[0048] Figure 2 is a grid area number map;
[0049] Figure 3 This is a schematic diagram of the calculation of the grid unit sequence of training data 1;
[0050] Figure 4 It is a schematic diagram of the calculation of the grid unit sequence of training data 2;
[0051] Figure 5 It is a schematic diagram of the calculation using the data grid cell sequence. DETAILED DESCRIPTION
[0052] Combination Figure 1 , introduces a method for generating activity area summary based on aircraft track path sequence processing of the present invention. In order to reduce the complexity of the introduction and facilitate understanding, the track is simplified to less than 10 points, and the area is also reduced accordingly, including the following steps:
[0053] (1) Training phase
[0054] Step 1: The user sets up a global area where the track needs to be described according to the usage requirements, and requires that the global area can cover all the tracks that need to be described. Assume that its range is: lat∈[10°,12°], lon∈[100°,103°];
[0055] Step 2: Divide the area selected in step 1 into a regular grid array according to longitude and latitude, and divide the longitude and latitude into grid units G with a side length of 0.5°. mn , where m∈[0,4), n∈[0,6), and each grid cell is numbered: Idx mn =m×N+n, for example, the number of the 2nd row and the 3rd column is 1×6+2=8, and the final numbering effect is as follows Figure 2 As shown;
[0056] Step 3: Assume that a structured track to be processed is as shown in Table 1, and read each discrete point Tk c , and projecting these values to these grid cells, the results are as follows:
[0057]
[0058] And form the grid sequence as follows:
[0059] [ <c-Idx mn >]=[<0,1>,<1,8>,<2,8>,<3,21>].
[0060] Table 1
[0061] Serial number Longitude (°) Latitude (°) time 0 100.8 11.6 09:00:00 1 101.4 11.4 09:06:54 2 101.3 11.1 09:10:08 3 101.7 10.3 09:18:04
[0062] Step 4: Through manual identification, relying on the landmark "Pearl Spring", the grid sequence corresponding to the structured track Tk of the aircraft is annotated according to the track characteristics. Figure 3 It can be seen that the tracks are mainly distributed on the west side of "Zhenzhuquan", so the annotation result is "west of Zhenzhuquan";
[0063] Step 5: Add the labeled matrix to the data set. The effect is shown in Table 2.
[0064] Table 2
[0065] 1 1,8,8,21 Zhenzhuquanyixi
[0066] Step 6: Read the set of all aircraft tracks and determine whether all tracks have been processed. If not, return to step 3 to continue processing until all tracks are averaged into a grid sequence and marked. Here we assume that there is another track, as shown in Table 3.
[0067] Table 3
[0068] Serial number Longitude (°) Latitude (°) time 0 102.8 11.6 09:00:00 1 102.9 11.4 09:06:54 2 102.8 11.1 09:10:08 3 102.7 10.3 09:18:04 4 102.6 10.7 09:22:22 5 102.8 10.9 09:27:14 6 102.9 10.1 09:31:01
[0069] Step 7: Select the next unprocessed track, return to step 3, and convert each discrete point Tk′ c′ , and projecting these values to these grid cells, the results are as follows:
[0070]
[0071] And form the grid sequence as follows:
[0072] [ <c′-Idx′ mn >] = [<0,5>,<1,11>,<2,11>,<3,23>,<4,17>,<5,23>];
[0073] Step 8: Through manual identification, relying on the landmark "Pearl Spring", the grid sequence corresponding to the structured track Tk' of the aircraft is annotated according to the track characteristics. Figure 4 It can be seen that the tracks are mainly distributed on the east side of "Zhenzhuquan", so the annotation result is "east of Zhenzhuquan";
[0074] Step 9. Add the labeled matrix to the data set. The effect is shown in Table 4. It can be seen that compared with Table 2, due to the different number of track points, the tracks with fewer points need to be aligned and filled, where "pad" is recorded as a special filling value;
[0075] Table 4
[0076] 1 1,8,8,21,pad,pad Zhenzhuquanyixi 2 5,11,11,23,17,23 Zhenzhuquanyidong
[0077] Step 10: Collect all labeled matrices to form a data set;
[0078] Step 11: Carry out training through a recurrent neural network model, with the input being the grid sequence corresponding to the track and the output being the description text of the track, and use the data set formed in step 9 to carry out training;
[0079] Step 12: A trained summary generation model is formed and is ready for use.
[0080] (2) Use phase
[0081] Step 1: Read the aircraft structured track Tk″ to be generated c″ , assuming that its structured track is as shown in Table 5;
[0082] Table 5
[0083] Serial number Longitude (°) Latitude (°) time 0 101.25 11.75 09:00:00 1 101.75 11.75 09:05:27 2 102.25 11.75 09:10:41 3 102.75 11.75 09:15:37
[0084]
[0085] And form the grid sequence as follows:
[0086] [ <c″-Idx″ mn >]=[<0,2>,<1,3>,<2,4>,<3,5>];
[0087] Step 3: Input the grid sequence obtained in step 2 into the brief generation model to generate the output It is "north of Pearl Spring";
[0088] Step 4: The user determines the generated result obtained in step 3 by observing the relative position of the original track and the landmark Is ("north of Pearl Spring") reasonable? Assuming that this is not reasonable, combined with Figure 5 , it is found that the track has a higher proportion in the northwest of "Zhenzhuquan", so it is assumed that the user changes to "Zhenzhuquan Northwest" and enters step 5;
[0089] Step 5: Modify the user's results (“Northwest of Pearl Spring”) as track Tk″ c″ The obtained raster sequence is annotated and added to the dataset, as shown in Table 6.
[0090] Table 6
[0091] 1 1,8,8,21,pad,pad Zhenzhuquanyixi 2 5,11,11,23,17,23 Zhenzhuquanyidong 3 2,3,4,5,pad,pad Zhen, Pearl, Spring, West, North
[0092] The present invention provides a method for generating an activity area summary based on aircraft track path sequence processing. There are many methods and approaches to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the protection scope of the present invention. All components not specified in this embodiment can be implemented by existing technologies.
Claims
1. A method for generating activity area profiles based on aircraft track sequence processing, characterized in that: The steps include: Step 1, training phase: using the recurrent neural network model to generate a summary generation model; Step 1 includes the following steps: Step 1-1, set the global area where the track needs to be described; Step 1-2, dividing the global area established in step 1-1 into a regular grid array according to longitude and latitude; Step 1-3, read each discrete point in the structured track of the aircraft, and calculate and project it to the grid unit; Step 1-4: Arrange the corresponding relationship between the track points and grid cells obtained in step 1-3 according to the corresponding time sequence of the discrete points in the aircraft track to form a grid sequence corresponding to the track [ <c-Idx mn >], where c is arranged in ascending order from 0 to C-1 in chronological order; Step 1-5: Select a landmark in the global area where the track needs to be described, and use the landmark Mk s , where s∈[1,S], s∈N, s represents the subscript of the landmark. According to the track characteristics, the grid sequence corresponding to the structured track Tk of the aircraft is [ <c-Idx mn >] to mark and obtain the marking result R; Step 1-6: Add the labeled raster sequence to the dataset; Step 1-7, read the set of all aircraft tracks and determine whether all tracks have been processed. If not, proceed to step 1-8. If yes, proceed to step 1-9. Step 1-8, select the next unprocessed track and return to step 1-3; Step 1-9: Collect all labeled raster sequences, align and fill them, and form a data set; Step 1-10: Train the recurrent neural network model with the input being a grid sequence [ <c-Idx mn >], the output is the text R annotated by the annotator for the track, and the training is carried out using the data set formed in steps 1-9; Step 1-11, forming a trained summary generation model; Step 2: Use phase: Apply the brief description generation model to calculate and process the input track to be processed, and generate a text description of the track; Step 2 includes the following steps: Step 2-1: Read the structured track Tk′ of the aircraft to be generated c′ ; Step 2-2, perform projection; Step 2-3: Form the grid sequence corresponding to the track [ <c′-Idx′ m′n′ >]; Step 2-4: The grid sequence obtained in step 2-3 [ <c′-Idx′ m′n′ >] Input a brief description of the model to generate and get the output Step 2-5: Determine the generated result obtained in step 2-4 by observing the relative position of the original track and the landmark Is it reasonable? If yes, then end. If not, proceed to step 2-6. Step 2-6: By observing the relative position of the original track and the landmark, the generated result obtained in step 2-4 is Modify it to the result you think is appropriate and will As the track Tk′ c′ The resulting grid sequence [ <c′-Idx′ m′n′ >] and added it to the dataset.
2. The method according to claim 1, characterized in that Step 1-1 includes: requiring the global area to cover all the tracks that need to be described, assuming that the global area range is: Lat∈[Lat min ,Lat max ],lon∈[Lon min ,Lon max ], and the side length l, where lat refers to latitude, Lat min Refers to setting the lower latitude limit of the global area, Lat max Refers to setting the upper limit of latitude for the global area, lon refers to longitude, Lon min Refers to setting the lower limit of longitude for the global area, Lon max Refers to setting the upper longitude limit of the global region.
3. The method according to claim 2, characterized in that In step 1-2, the longitude and latitude are divided into grid cells G according to the spacing of side length l. mn ; Where m∈[0,M), m refers to the longitude subscript of the grid unit, M refers to the maximum value calculated by the grid cell longitude subscript; n∈[0,N), n refers to the latitude subscript of the grid cell, N refers to the maximum value calculated by the grid cell latitude subscript; Let the latitude of north be positive, the latitude of south be negative, the longitude of east be positive, and the longitude of west be negative; Number each grid cell, number Idx mn For: Idx mn =m×N+n.
4. The method according to claim 3, characterized in that: Steps 1-3 include: c Represents the cth discrete point, c∈[0,C), C represents the total number of points in the track, forming a corresponding relationship between track points and grid cells <c-Idx mn >, <c-Idx mn > indicates that the cth point in the track falls into grid cell G mn middle.
5. The method according to claim 4, characterized in that In steps 1-3, if the track point Tk c In grid cell G mn In the equation, F(Tk c ,G mn )=1, otherwise it is recorded as F(Tk c ,G mn )=0, the following formula will be used to calculate Tk c With grid cell G mn Projection relationship: Among them, Tk c,Lat and Tk c,Lon Represents the track point Tk c The latitude and longitude values in .
6. The method according to claim 5, characterized in that Step 2-2 includes: reusing the grid cells divided in steps 1-1 and 1-2, and converting each discrete point Tk′ in the structured track to be generated c′ Project to the grid unit to form the corresponding relationship between the track point and the grid unit <c′-Idx mn >, <c′-Idx mn > indicates that the c′th point in the track falls into grid cell G mn Where c′∈[0,C′), C′ represents the total number of points in the track.
7. The method according to claim 6, characterized in that In step 2-2, if the track point Tk′ c′ In the grid cell G′ with the subscript m′n′ m′n′ In the equation, it is denoted as F(Tk′ c′ ,G′ m′n′ )=1, otherwise it is recorded as F(Tk′ c′ ,G′ m′n′ )=0, the following formula will be used to calculate Tk′ c′ With grid cell G′ m′n′ Projection relationship: Among them, Tk′ c′,Lat and Tk′ c′,Lon Represent the track points Tk′ c′ The latitude and longitude values in .
8. The method according to claim 7, characterized in that Step 2-3 includes: arranging the corresponding relationship between the track points and the grid cells obtained in step 2-2 according to the corresponding time sequence of the discrete points in the aircraft track to form a grid sequence corresponding to the track [ <c′-Idx′ m′n′ >], where c′ is arranged in ascending order from 0 to C′-1 in chronological order.
