CN114252871A - Radar measurement accuracy compensation method based on machine learning - Google Patents
Radar measurement accuracy compensation method based on machine learning Download PDFInfo
- Publication number
- CN114252871A CN114252871A CN202111514264.5A CN202111514264A CN114252871A CN 114252871 A CN114252871 A CN 114252871A CN 202111514264 A CN202111514264 A CN 202111514264A CN 114252871 A CN114252871 A CN 114252871A
- Authority
- CN
- China
- Prior art keywords
- radar detection
- data
- radar
- target
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005259 measurement Methods 0.000 title claims abstract description 53
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000010801 machine learning Methods 0.000 title claims abstract description 33
- 238000001514 detection method Methods 0.000 claims abstract description 93
- 238000005516 engineering process Methods 0.000 claims abstract description 9
- 238000005070 sampling Methods 0.000 claims abstract description 9
- 230000002159 abnormal effect Effects 0.000 claims abstract description 7
- 230000008030 elimination Effects 0.000 claims abstract description 5
- 238000003379 elimination reaction Methods 0.000 claims abstract description 5
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 15
- 239000000126 substance Substances 0.000 claims description 5
- 238000003066 decision tree Methods 0.000 claims description 4
- 230000004907 flux Effects 0.000 claims description 4
- 238000012417 linear regression Methods 0.000 claims description 4
- 230000005855 radiation Effects 0.000 claims description 4
- 238000007637 random forest analysis Methods 0.000 claims description 4
- 238000012706 support-vector machine Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 abstract description 53
- 238000012360 testing method Methods 0.000 description 13
- 239000005433 ionosphere Substances 0.000 description 10
- 230000000694 effects Effects 0.000 description 9
- 230000008569 process Effects 0.000 description 5
- 238000010276 construction Methods 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 238000002790 cross-validation Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 244000061520 Angelica archangelica Species 0.000 description 1
- 235000001287 Guettarda speciosa Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000000523 sample Substances 0.000 description 1
- 239000005436 troposphere Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
- G01S7/497—Means for monitoring or calibrating
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/418—Theoretical aspects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a radar detection precision compensation method based on machine learning, which is characterized by collecting space environment characteristic parameters, space-time characteristic parameters and radar detection residual error data in a set duration to construct a data set, and preprocessing the obtained data by adopting an abnormal value elimination and data down-sampling technology; constructing a radar detection precision compensation model based on machine learning; resolving a target azimuth angle, a measurement time interval, a satellite point longitude and latitude and a target height from radar detection data; downloading and extracting radar detection time space environment parameters from the Internet; and inputting the extracted data into the constructed radar detection precision compensation model to obtain the radar detection data after precision compensation. On the basis of the electric wave environment refraction correction, the radar detection precision is further compensated, the loss of the detection precision caused by the space-time characteristic limitation of an electric wave refraction correction model, the radar system error and the like is compensated, and the radar detection precision is improved.
Description
Technical Field
The invention belongs to the field of space target detection, and relates to a radar measurement precision compensation method.
Background
When the radar detects a target, signals of the radar need to pass through an ionized layer to be transmitted, and are influenced by the nonuniformity of electron density of the ionized layer, and radar distance measurement and angle measurement can generate refraction errors. Generally, the refraction error caused by ionospheric disturbance can reach tens to hundreds of meters, and becomes a main error source influencing the detection accuracy of the radar. When a space environment event affecting the electron density of an ionized layer occurs, the radar measurement precision is further reduced, and the orbit determination precision of a space target is directly affected. Therefore, it is essential to correct radar measurement values for wave refraction.
Radar beam refraction correction has a considerable history of research, and Newcomb (1906) mentioned in his famous book "spherical astronomical summary (a complex of spherical astronomy) in the early twentieth century: "in practical astronomy, perhaps no branch is like the astronomical refraction problem, but the state is still unsatisfactory". In fact, as the requirement for the accuracy of the radar is higher and higher, the accuracy requirement for the refraction correction is higher and higher, more factors need to be considered in the process of constructing the model, and the difficulty of constructing the model is correspondingly higher.
The chinese radiowave propagation institute has started to develop radiowave refraction correction-related studies from the last 60 th century. In 2010, a first domestic radio wave refraction correction device (HZD-1) is successfully developed, and refraction errors caused by a troposphere and an ionosphere are effectively corrected. Aiming at the ionospheric environment characteristics in the middle and low latitude areas, after the algorithm model is improved and perfected, a new generation radio wave refraction corrector is applied to radar radio wave refraction correction in 2014. Although the accuracy of the detection data of the current radar is greatly improved after the radio wave refraction error correction, the residual error after the correction is still larger, and the residual error quantity is related to time and an elevation angle, which indicates that the refraction error caused by the ionosphere still has larger residual after the correction.
Through preliminary analysis, the reason for the larger error after radar refraction correction is as follows: at present, the radar usually adopts an ionosphere empirical model to correct the radio wave refraction error, and only can give the 'climatology' form of the ionosphere, which reflects the average state of the ionosphere. Although some radars adopt single-station measured data to drive an ionosphere empirical model, due to the limitation of the empirical model and insufficient coverage of the single-station data, the ionosphere model corrected by the measured data cannot accurately present the space-time characteristics of the ionosphere in the region, and cannot effectively reflect the disturbance characteristics of the ionosphere caused by solar explosive activity, so that the radar radio wave refraction correction effect is poor in certain time and space ranges or in the case of space environment disturbance.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a radar detection precision compensation method based on machine learning, which is used for extracting space-time and space environment influence factors of residual errors after radar refraction correction, constructing a functional relation between the factors and the residual errors based on the machine learning technology, further compensating radar detection precision, compensating loss of detection precision caused by space-time characteristic limitation of a radio wave refraction correction model, radar system errors and the like on the basis of radio wave environment refraction correction, and improving the radar detection precision.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
(1) collecting space environment characteristic parameters, space-time characteristic parameters and radar detection residue in set durationError data constructs a data set, noted as (x)1，x2，x3，x4，x5，x6，x7，x8，x9Error), wherein x1Representing a 10.7 cm solar radiation flux, x2Representing the geomagnetic index, x3Representing the radiation flux, X, of the solar X-ray band4Represents the total content, x, of vertical electrons in the ionized layer of the radar station5Indicates the hour, x, corresponding to the moment of measurement6Representing radar measurement azimuth, x7Representing the target latitude, x8Representing the target longitude, x9Representing the height of a target orbit, and error representing the radar detection residual error;
(2) preprocessing the data radar detection residual error data obtained in the step (1) by adopting an abnormal value elimination and data down-sampling technology;
(3) based on the data set preprocessed in the step (2), x is converted into1-x9As input, error is used as output, and a radar detection precision compensation model based on machine learning is constructed;
(4) resolving a target azimuth angle, a measurement time interval, a satellite point longitude and latitude and a target height from radar detection data; downloading and extracting radar detection time space environment parameters from the Internet; and inputting the extracted data into the constructed radar detection precision compensation model to obtain the radar detection data after precision compensation.
The data set in the step (1) comprises at least 1 year of space environment characteristic parameters, space-time characteristic parameters and radar detection residual error data.
The method comprises the following steps that (1) target positions detected by a radar and target precision rail data corresponding to a detection arc section are obtained; calculating the real position of the target at the radar detection time by adopting a 6-order Lagrange interpolation method; converting the real target position at the radar detection time from a rectangular earth coordinate system to a polar coordinate system of the survey station to obtain a spatial characteristic parameter x7-x9(ii) a And obtaining a radar detection residual error by interpolating radar detection data and the real position of the target.
The step (2) comprises the following steps: a. all arc sections with the duration time less than 30 seconds are removed, and the first arc section is taken as the current arc section from the rest arc sections; b. removing the first 30 seconds of measurement data of the current arc segment; c. judging whether the residual measurement value of the current measurement section is more than 50 points; d. if yes, randomly extracting 50 points from all the remaining measured values and adding the points into a training set; if not, adding all the residual measured values of the arc segment into a training set; e. and d, setting the next arc segment as the current arc segment, and repeating the steps b-d until all the arc segments are processed.
The step (3) adopts a cross validation mode to sequentially apply various machine learning regression models to solve the current problem, and selects the model with the best performance as a subsequent model f; the radar measurement precision compensation method comprises
Wherein the content of the first and second substances,
representing the radar measurement data and y representing the radar final output.
The machine learning regression model comprises a linear regression, a ridge regression, a Lasso regression, a minimum angle regression, a theilSen regression, a Huber regression, a decision tree, a random forest, an extreme random tree, a gradient lifting regression, an extreme gradient lifting, a light weight gradient lifting, a support vector machine, a K nearest neighbor and a multilayer perceptron model.
The invention has the beneficial effects that: a radar measurement precision compensation method based on machine learning is constructed by utilizing the relation between space environment and space-time parameters and radar measurement residual errors, the problem that the existing radio wave refraction correction model can only reflect the ionosphere average state so as to correct the refraction errors inaccurately is solved, the radar measurement errors are further reduced, and the detection precision of radar equipment is improved.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a schematic diagram of a radar detection accuracy compensation model construction;
FIG. 3 is a schematic diagram of compensating for radar measurement accuracy using a machine learning model.
Detailed Description
The radar detection precision compensation method based on machine learning makes up the defect of the existing experience model in radar radio wave refraction error correction, and effectively improves the radar measurement precision.
The invention adopts the following technical scheme:
(1) and (3) data set construction:
the data set comprises a certain time period (at least covering 1 year) of space environment characteristic parameters, space-time characteristic parameters and radar detection residual error data, which are marked as (x1, x2, x3, x4, x5, x6, x7, x8, x9, y). The data are shown in Table 1.
Table 1 data set data composition
Spatial environment characteristic parameter x1-x4(F10.7, Ap, X-ray and TEC) are acquired through the Internet open channel, and the time characteristic parameter X5(Hour) is the radar detection time rounding, and the radar measures the azimuth angle x6The output is detected for the radar. Spatial characteristic parameter x7-x9And the radar detection residual error is obtained by interpolating target fine track data (target real position information) and converting coordinates, and the steps are as follows:
(11) and acquiring the target position detected by the radar and target precision track data corresponding to the detection arc section. Recording target position data detected by the radar as (t, rho, a, e) in a polar coordinate system of the observation station, wherein t represents radar detection time, rho represents target distance detected by the radar, a represents azimuth detected by the radar, and e represents elevation detected by the radar; the target precision rail data of the radar detection arc segment is recorded as the ground-fixed rectangular coordinate system
As the "true value" of the target trajectory.
(12) And (4) interpolating to obtain the real position of the target at the radar detection time. Because the sampling time and the sampling frequency of the radar detection data and the fine track data are not consistent, the real position of the target at the radar detection time is calculated by adopting a 6-order Lagrange interpolation method
7 probe points are selected in the fine track data such that the designated interpolation point is located in the middle of the selected fine track data.
(13) And converting the real position of the target at the radar detection time into a polar coordinate system of the survey station by a rectangular coordinate system of the earth. Knowing the geodetic coordinates of the radar as (L, B, H), the position of the radar in the earth-fixed coordinate system
Wherein the content of the first and second substances,
REis the major semi-axis, R, of the equatorial ellipsoid of the earthpCalculating the position vector of the target in the rectangular coordinate system of the survey station for the polar radius of the earth
In the formula, ρx、ρy、ρzRespectively, x, y, z triaxial components of the position vector in a rectangular coordinate system, wherein,
then the real distance of the target under the polar coordinate system of the survey station
Azimuth angle
Elevation angle
Space characteristic parameter target latitude, longitude and altitude x7-x9The following coordinate transformation formula is used to obtain:
The radar detection residual error is obtained by interpolation of radar detection data and the real position of the target, i.e.
ρ-ρcAs distance error, e-ecIs the elevation error.
(2) Preprocessing a data set:
because the radar detection data has abnormal values and the number of data points of each arc section is different, the data obtained in the step (1) is directly utilized to carry out machine learning model training, the model precision is influenced, the data obtained in the step (1) is preprocessed by adopting an abnormal value elimination and data down-sampling technology, and the process is as follows: a. all arc sections with the duration time less than 30 seconds are removed, and the first arc section is taken as the current arc section from the rest arc sections; b. removing the first 30 seconds of measurement data of the current arc segment; c. judging whether the residual measurement value of the current measurement section is more than 50 points; d. if yes, randomly extracting 50 points from all the remaining measured values and adding the points into a training set; if not, adding all the residual measured values of the arc segment into a training set; e. and d, setting the next arc segment as the current arc segment, and repeating the steps b-d until all the arc segments are processed, namely executing the steps b-d on all the arc segments.
(3) Constructing a radar detection precision compensation model:
based on the data set preprocessed in the step (2), x is converted into1-x9The method comprises the following steps of taking error as an input and taking error as an output, and constructing a radar detection precision compensation model based on machine learning, wherein the method comprises the following steps:
(31) and (4) selecting a model. There are many regression models, and table 2 lists 15 models that are commonly used, and different models have their own advantages and disadvantages and are suitable for different problems. In general, model selection needs to rely on strong professional knowledge, and a good selection is difficult to make by ordinary non-professional personnel. To solve this problem, the present invention learns the model that best fits the current problem from the data in a traversal fashion. Because the problem data solved by the invention is structured data, the calculation complexity of the method shown in the table 2 is not very high, and the calculation capability is accelerated and improved at present, the traversal method is a feasible mode. Specifically, the models in table 2 are applied in sequence in a cross-validation manner to solve the current problem, and the model that performs best is selected as the subsequent model.
TABLE 2 machine learning regression model candidate set
| Serial number | Name (R) | Serial number | Name (R) |
| 1 | Linear regression | 9 | Extreme random tree |
| 2 | Ridge regression | 10 | Gradient boost regression |
| 3 | Lasso loopChinese angelica root-bark | 11 | Extreme gradient lift |
| 4 | Minimum angle regression | 12 | Light weight gradient lift |
| 5 | TheilSen regression | 13 | Support vector machine |
| 6 | Huber regression | 14 | Nearest neighbor of K |
| 7 | Decision tree | 15 | Multilayer perceptron |
| 8 | Random forest |
(32) And (5) training a model. The optimal model obtained by model selection is recorded as f, and the radar measurement accuracy compensation method is shown as the following formula:
wherein, the function f represents a machine learning correction model and needs to be learned from data; x is the number of1,...,xpThe parameters representing the spatial environment and the space-time are input parameters of the machine learning correction model, and the considered parameters include 9 (p is 9), and the detailed description is shown in table 1; error is the residual error predicted by the machine learning correction model;
the radar measurement data are expressed and are radar output values compensated by the embedded correction model; y represents the final output of the radar and is the output result after further compensation by the model constructed by the method.
(4) And (3) radar detection precision compensation:
(41) according to the step (1), resolving a target Azimuth angle, a measurement time interval, the Longitude and Latitude of a subsatellite point and a target Height from radar detection data, namely Azimuth, Hour, Latitude, Longitude and Height in model input variables;
(42) downloading and extracting radar detection time space environment parameters F10.7, Ap, X-ray and TEC from the Internet;
(43) and (4) inputting the data extracted in the steps (42) and (43) into the model constructed in the step (3) to obtain the radar detection data after the precision compensation.
The beneficial effects of the present invention are shown below quantitatively using the test results on the test data set. The test data set is from tracking data of a radar to five low-orbit satellites, and comprises 529 measurement arc sections. 70% of the arcs, i.e., 370 arcs, were randomly selected for training the model, and the remaining 159 arcs were the test set. The Mean Absolute Error (MAE) and the Root Mean Square Error (RMSE) are selected as evaluation criteria to quantitatively evaluate the radar measurement accuracy before and after the model correction, and the MAE and the RMSE are defined as follows:
wherein N is the number of the measuring points, ytrue(i) Is the true value (i.e. fine track data) of the ith measurement point, ymeasure(i) Is the measured value of the ith measuring point. To compare the degree of error reduction after correction and before correction, the percentage of correction is also used herein as an evaluation index, defined as:
this index can be interpreted as the validity of the modified model: when the value is less than or equal to 0, the model is invalid or counteractive; when the value is greater than 0, the model is effective, and the larger the value is, the better the model effect is; when the value is 1, all errors are corrected, which is the most ideal case.
The test results on the above test data set are shown in table 3, showing MAE, RMSE and percent correction for radar measured range and pitch before and after model correction according to the present invention. It can be seen that the average absolute error of the distance before correction is 24.4 meters, the value after correction is 10.3 meters, the correction percentage reaches 57 percent, and the effect is very obvious; for pitch angle, MAE before correction is 132.12 arc seconds, and MAE after correction is 42.48 arc seconds, and the correction percentage reaches 67.8%, which is more significant than the distance error correction. Tests show that the model has a very obvious effect of correcting radar measurement errors, greatly improves the radar measurement precision, and has a wide application prospect.
TABLE 3 MAE, RMSE and percent correction before and after model correction according to the invention on the test data set
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and examples. The specific embodiments described herein are merely illustrative of the invention and do not delimit the invention.
As shown in fig. 1, the present embodiment discloses a new method for compensating radar measurement accuracy, which includes the following steps:
(1) data set construction
The data set comprises space environment characteristic parameters, space-time characteristic parameters and certain radar detection residual error data in a part of time period from 2017 to 2018, and the data set is marked as (x)1，x2，x3，x4，x5，x6，x7，x8，x9Y). The data are shown in Table 1.
Spatial environment characteristic parameter x1-x4(F10.7, Ap, X-ray and TEC) are acquired through the Internet open channel, and the time characteristic parameter X5(Hour) is the radar detection time rounding, and the radar measures the azimuth angle x6The output is detected for the radar. Spatial characteristic parameter x7-x9And the radar detection residual error is obtained by interpolating target fine track data (target real position information) and converting coordinates, and the steps are as follows:
(11) and acquiring the target position detected by the radar and target precision track data corresponding to the detection arc section. Example experimental data from a radar tracking data on 5 targets, a total of 529 arc segments. Recording target position data detected by the radar as (t, rho, a, e) in a polar coordinate system of the observation station, wherein t represents radar detection time, rho represents target distance detected by the radar, a represents azimuth detected by the radar, and e represents elevation detected by the radar; the target precision rail data of the radar detection arc segment is recorded as the ground-fixed rectangular coordinate system
As the "true value" of the target trajectory.
(12) And (4) interpolating to obtain the real position of the target at the radar detection time. Because the sampling time and the sampling frequency of the radar detection data and the fine track data are not consistent, the real position of the target at the radar detection time is calculated by adopting a 6-order Lagrange interpolation method
Selecting 7 detection points in the fine track data to make the designated interpolation point be positioned in the middle of the selected fine track data, namely ti＜ti+1＜ti+2＜ti+3＜t＜ti+4＜ti+5＜ti+6。
(13) And converting the real position of the target at the radar detection time into a polar coordinate system of the survey station by a rectangular coordinate system of the earth. Knowing the geodetic coordinates of the radar as (L, B, H), the position of the radar in the earth-fixed coordinate system
Wherein the content of the first and second substances,
REis the major semi-axis, R, of the equatorial ellipsoid of the earthpCalculating the position vector of the target in the rectangular coordinate system of the survey station for the polar radius of the earth
Wherein the content of the first and second substances,
then the real distance of the target under the polar coordinate system of the survey station
Azimuth angle
Elevation angle
If ρxIf < 0, then ac＝ac+180。
Space characteristic parameter target latitude, longitude and altitude x7-x9The following coordinate transformation formula is used to obtain:
The radar detection residual error is obtained by interpolation of radar detection data and the real position of the target, i.e.
ρ-ρcAs distance error, e-ecIs the elevation error.
(2) Data set preprocessing
Because radar detection data has an abnormal value, and the number of data points of each arc section is different, the data obtained in the step (1) is directly utilized to carry out machine learning model training, the model precision is influenced, abnormal value elimination and data down-sampling technology is adopted to carry out pretreatment on the data radar detection residual error data obtained in the step (1), and the process is as follows: a. all arc sections with the duration time less than 30 seconds are removed, and the first arc section is taken as the current arc section from the rest arc sections; b. removing the first 30 seconds of measurement data of the current arc segment; c. judging whether the residual measurement value of the current measurement section is more than 50 points; d. if yes, randomly extracting 50 points from all the remaining measured values and adding the points into a training set; if not, adding all the residual measured values of the arc segment into a training set; e. and d, setting the next arc segment as the current arc segment, and repeating the steps b-d until all the arc segments are processed, namely executing the steps b-d on all the arc segments.
(3) Radar detection precision compensation model construction
The model was trained with 70% of the 529 measurement arcs (370 arcs) randomly selected. As shown in fig. 2, based on the data set preprocessed in step (2), x is divided1-x9The method comprises the following steps of taking error as an input and taking error as an output, and constructing a radar detection precision compensation model based on machine learning, wherein the method comprises the following steps:
(31) and (4) selecting a model. The existing data is used for guiding model learning, a method of traversing candidate models including linear regression, support vector machines, random forests, artificial neural networks and the like is adopted, and a cross validation technology is used for selecting the model with the optimal performance. The process is a public technology and is not explained too much, and the detailed description refers to a machine learning open source tool pycaret to explain the compare _ models function in the document.
(32) And (5) training a model. On the constructed test data set, through model selection, the obtained optimal model f is light GBM (light Gradient Boosting machine), and the radar measurement accuracy compensation model is shown as the following formula:
wherein, the function f represents a machine learning correction model and needs to be learned from data; x is the number of1,...,xpThe parameters representing the spatial environment and the space-time are input parameters of the machine learning correction model, and the considered parameters include 9 (p is 9), and the detailed description is shown in table 1; error is the residual error predicted by the machine learning correction model;
the radar measurement data are expressed and are radar output values compensated by the embedded correction model; y represents the final output of the radar and is the output result after further compensation by the model constructed by the method.
LightGBM is a rapid, distributed and high-performance decision tree-based gradient lifting method for Microsoft open source, and has the advantages of high accuracy, high convergence rate and reasonable memory occupation. The training of the model is an open technology, a plurality of open source software can realize the process, and a user only needs to send a training data set to call a corresponding interface. Without undue elaboration herein, details may be found in the Python toolkit lightgbm specification document.
(4) Radar detection accuracy compensation
Using the remaining 159 arc segment measurements as a test set, as shown in fig. 3, the procedure is as follows:
(41) according to the step (1), resolving a target Azimuth angle, a measurement time interval, the Longitude and Latitude of a subsatellite point and a target Height from radar detection data, namely Azimuth, Hour, Latitude, Longitude and Height in model input variables;
(42) downloading and extracting radar detection time space environment parameters F10.7, Ap, X-ray and TEC from the Internet;
(43) and (4) inputting the data extracted in the steps (42) and (43) into the model constructed in the step (3) to obtain the radar detection data after the precision compensation.
(5) Compensation effect evaluation
Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) were chosen as criteria for quantitative assessment of accuracy of radar measurements before and after correction by the model of the invention, and MAE and RMSE were defined as follows:
wherein N is the number of the measuring points, ytrue(i) Is the true value (i.e. fine track data) of the ith measurement point, ymeasure(i) Is the measured value of the ith measuring point. To compare the degree of error reduction after correction and before correction, the percentage of correction is also used herein as an evaluation index, defined as:
this index can be interpreted as the validity of the modified model: when the value is less than or equal to 0, the model is invalid or counteractive; when the value is greater than 0, the model is effective, and the larger the value is, the better the model effect is; when the value is 1, all errors are corrected, which is the most ideal case.
The model was tested on the test data set by the above procedure and the results are shown in table 4.
TABLE 4 MAE, RMSE and percent correction before and after model correction according to the invention on the test data set
It can be seen that the average absolute error of the distance before correction is 24.4 meters, the value after correction is 10.3 meters, the correction percentage reaches 57 percent, and the effect is very obvious; for pitch angle, MAE before correction is 132.12 arc seconds, and MAE after correction is 42.48 arc seconds, and the correction percentage reaches 67.8%, which is more significant than the distance error correction. Tests show that the model has a very obvious effect of correcting radar measurement errors, can greatly improve the radar measurement accuracy, and has a wide application prospect.
Claims (6)
1. A radar detection precision compensation method based on machine learning is characterized by comprising the following steps:
(1) collecting space environment characteristic parameters, space-time characteristic parameters and radar detection residual error data in set duration to construct a data set, and recording as (x)1，x2，x3，x4，x5，x6，x7，x8，x9Error), wherein x1Representing a 10.7 cm solar radiation flux, x2Representing the geomagnetic index, x3Representing the radiation flux, X, of the solar X-ray band4Represents the total content, x, of vertical electrons in the ionized layer of the radar station5Indicates the hour, x, corresponding to the moment of measurement6Representing radar measurement azimuth, x7Representing the target latitude, x8Representing the target longitude, x9Representing the height of a target orbit, and error representing the radar detection residual error;
(2) preprocessing the data obtained in the step (1) by adopting an abnormal value elimination and data down-sampling technology;
(3) based on the data set preprocessed in the step (2), x is converted into1-x9As input, error is used as output, and a radar detection precision compensation model based on machine learning is constructed;
(4) resolving a target azimuth angle, a measurement time interval, a satellite point longitude and latitude and a target height from radar detection data; downloading and extracting radar detection time space environment parameters from the Internet; and inputting the extracted data into the constructed radar detection precision compensation model to obtain the radar detection data after precision compensation.
2. The machine-learning-based radar detection accuracy compensation method according to claim 1, wherein the data set in step (1) comprises at least 1 year of spatial environment characteristic parameters, spatiotemporal characteristic parameters and radar detection residual error data.
3. The machine learning-based radar detection accuracy compensation method according to claim 1, wherein the step (1) obtains target positions detected by the radar and target precision track data corresponding to detection arc sections; calculating the real position of the target at the radar detection time by adopting a 6-order Lagrange interpolation method; converting the real target position at the radar detection time from a rectangular earth coordinate system to a polar coordinate system of the survey station to obtain a spatial characteristic parameter x7-x9(ii) a And obtaining a radar detection residual error by interpolating radar detection data and the real position of the target.
4. The machine-learning-based radar detection accuracy compensation method according to claim 1, wherein the step (2) comprises the steps of: a. all arc sections with the duration time less than 30 seconds are removed, and the first arc section is taken as the current arc section from the rest arc sections; b. removing the first 30 seconds of measurement data of the current arc segment; c. judging whether the residual measurement value of the current measurement section is more than 50 points; d. if yes, randomly extracting 50 points from all the remaining measured values and adding the points into a training set; if not, adding all the residual measured values of the arc segment into a training set; e. and d, setting the next arc segment as the current arc segment, and repeating the steps b-d until all the arc segments are processed.
5. The machine-learning-based radar detection accuracy compensation method according to claim 1, wherein the step (3) employs cross validationThe method sequentially applies each machine learning regression model to solve the current problem, and selects the model with the best performance as a subsequent model f; the radar measurement precision compensation method comprises
Wherein the content of the first and second substances,
representing the radar measurement data and y representing the radar final output.
6. The machine-learning based radar detection accuracy compensation method of claim 5, wherein the machine-learning regression model comprises linear regression, ridge regression, Lasso regression, least angle regression, TheilSen regression, Huber regression, decision tree, random forest, extreme random tree, gradient elevation regression, extreme gradient elevation, light-weight gradient elevation, support vector machine, K-nearest neighbor, and multi-layered perceptron model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111514264.5A CN114252871A (en) | 2021-12-13 | 2021-12-13 | Radar measurement accuracy compensation method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111514264.5A CN114252871A (en) | 2021-12-13 | 2021-12-13 | Radar measurement accuracy compensation method based on machine learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114252871A true CN114252871A (en) | 2022-03-29 |
Family
ID=80791917
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111514264.5A Pending CN114252871A (en) | 2021-12-13 | 2021-12-13 | Radar measurement accuracy compensation method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114252871A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116343554A (en) * | 2023-05-26 | 2023-06-27 | 山东建筑大学 | Intelligent information processing radar calibration simulation method based on machine learning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109839620A (en) * | 2019-03-11 | 2019-06-04 | 深圳大学 | A kind of least square method for estimating radar system error for combining ADS-B |
| US20190235067A1 (en) * | 2018-01-31 | 2019-08-01 | Wave3D Co.,Ltd. | Moving object detection method in real-time using fmcw radar |
| CN110346844A (en) * | 2019-07-15 | 2019-10-18 | 南京恩瑞特实业有限公司 | Quantitative Precipitation estimating and measuring method of the NRIET based on cloud classification and machine learning |
| CN111414891A (en) * | 2020-04-07 | 2020-07-14 | 云南电网有限责任公司昆明供电局 | Power transmission line channel tree height inversion method based on laser radar and optical remote sensing |
| CN111695440A (en) * | 2020-05-21 | 2020-09-22 | 河海大学 | GA-SVR lake level measurement and prediction method based on radar altimeter |
| CN112613532A (en) * | 2020-11-26 | 2021-04-06 | 西安电子科技大学 | Moving target tracking method based on radar and recurrent neural network complete infrared fusion |
| CN113777627A (en) * | 2021-11-15 | 2021-12-10 | 浙江大学 | Laser radar ratio region transmission method based on novel foundation laser radar network |
-
2021
- 2021-12-13 CN CN202111514264.5A patent/CN114252871A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190235067A1 (en) * | 2018-01-31 | 2019-08-01 | Wave3D Co.,Ltd. | Moving object detection method in real-time using fmcw radar |
| CN109839620A (en) * | 2019-03-11 | 2019-06-04 | 深圳大学 | A kind of least square method for estimating radar system error for combining ADS-B |
| CN110346844A (en) * | 2019-07-15 | 2019-10-18 | 南京恩瑞特实业有限公司 | Quantitative Precipitation estimating and measuring method of the NRIET based on cloud classification and machine learning |
| CN111414891A (en) * | 2020-04-07 | 2020-07-14 | 云南电网有限责任公司昆明供电局 | Power transmission line channel tree height inversion method based on laser radar and optical remote sensing |
| CN111695440A (en) * | 2020-05-21 | 2020-09-22 | 河海大学 | GA-SVR lake level measurement and prediction method based on radar altimeter |
| CN112613532A (en) * | 2020-11-26 | 2021-04-06 | 西安电子科技大学 | Moving target tracking method based on radar and recurrent neural network complete infrared fusion |
| CN113777627A (en) * | 2021-11-15 | 2021-12-10 | 浙江大学 | Laser radar ratio region transmission method based on novel foundation laser radar network |
Non-Patent Citations (2)
| Title |
|---|
| 张瑜;赤娜;侯佳;师帅涛;: "大气折射引起的雷达定位误差模型", 电光与控制, no. 07, 15 July 2009 (2009-07-15) * |
| 李川川;刘琨;朱庆林;朱杉;孔韬;: "S波段雷达探测中的高精度电波折射误差修正方法", 空军预警学院学报, no. 06, 15 December 2019 (2019-12-15) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116343554A (en) * | 2023-05-26 | 2023-06-27 | 山东建筑大学 | Intelligent information processing radar calibration simulation method based on machine learning |
| CN116343554B (en) * | 2023-05-26 | 2023-08-22 | 山东建筑大学 | Intelligent information processing radar calibration simulation method based on machine learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6929927B2 (en) | Methods and equipment for predicting strong convective weather | |
| Huang et al. | Ionospheric single-station TEC short-term forecast using RBF neural network | |
| CN111145227B (en) | Iterative integral registration method for space multi-view point cloud of underground tunnel | |
| CN111680870B (en) | Comprehensive evaluation method for quality of target motion trail | |
| CN112965146A (en) | Quantitative precipitation estimation method combining meteorological radar and rainfall barrel observation data | |
| CN105974495B (en) | It will be averaged the method for cloud amount using classification fitting process anticipation target area future | |
| CN111401602A (en) | Assimilation method for satellite and ground rainfall measurement values based on neural network | |
| CN110297218B (en) | Method for detecting unknown modulation mode of radar signal based on generation countermeasure network | |
| CN114252871A (en) | Radar measurement accuracy compensation method based on machine learning | |
| CN115494007A (en) | Random forest based high-precision rapid detection method and device for soil organic matters | |
| CN113592143A (en) | Rainfall forecast-based medium and long term runoff ensemble forecasting method and system | |
| CN114723188A (en) | Water quality prediction method, device, computer equipment and storage medium | |
| RU2713188C1 (en) | Method for single-position determination of coordinates of sources of high-frequency radio waves during ionospheric propagation | |
| CN117148360B (en) | Lightning approach prediction method and device, electronic equipment and computer storage medium | |
| CN116719062A (en) | Satellite signal quality evaluation method and device, equipment and storage medium | |
| CN115062526B (en) | Three-dimensional ionosphere electron concentration distribution model training method based on deep learning | |
| CN111239682B (en) | Electromagnetic emission source positioning system and method | |
| CN110321528A (en) | A kind of Hyperspectral imaging heavy metal-polluted soil concentration evaluation method based on semi-supervised geographical space regression analysis | |
| RU2411468C1 (en) | Method of evaluating quantitative characteristics of probed earth's surface | |
| KR20230072793A (en) | Service providing apparatus and method for prediction of fine dust and ultrafine dust | |
| CN113177702B (en) | Meteorological input data matching method and system | |
| CN116258023B (en) | Wind speed prediction method of wind power plant and terminal equipment | |
| US20240094434A1 (en) | Information processing apparatus, radar apparatus, method, and storage medium | |
| CN115511011B (en) | Radar data correction method and system based on countermeasure generation network model | |
| CN116909320B (en) | Electric power collaborative inspection strategy analysis method based on ant colony algorithm |
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 |