CN113361190A - Troposphere delay correction method based on CMONOC-ZTD - Google Patents

Troposphere delay correction method based on CMONOC-ZTD Download PDF

Info

Publication number
CN113361190A
CN113361190A CN202110515252.8A CN202110515252A CN113361190A CN 113361190 A CN113361190 A CN 113361190A CN 202110515252 A CN202110515252 A CN 202110515252A CN 113361190 A CN113361190 A CN 113361190A
Authority
CN
China
Prior art keywords
model
network
weak
fnn
ztd
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
Application number
CN202110515252.8A
Other languages
Chinese (zh)
Inventor
朱明晨
房家伟
孙为
丁子扬
杨光
徐周
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tongling University
Original Assignee
Tongling University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tongling University filed Critical Tongling University
Priority to CN202110515252.8A priority Critical patent/CN113361190A/en
Publication of CN113361190A publication Critical patent/CN113361190A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)

Abstract

The invention discloses a troposphere delay correction method based on CMONOC-ZTD, which comprises the following steps: s1: obtaining modeling data and constructing a training sample; s2: constructing a plurality of 'weak' neural network models by using the samples obtained in the S1; s3: inputting sample training in a 'weak' neural network model; s4: determining a weight for each "weak" neural network model; s5: and (3) carrying out weighted average on the results of the weak neural network model, outputting a final model and verifying the precision of the final model. The invention is superior to the common CTRop model and SHATRop model, and the precision of the new model is improved by 9.6 percent and 3.3 percent compared with the two models; the modeling is carried out by adopting a GNSS actual measurement ZTD sequence, which is different from other models, or adopting meteorological reanalysis data or adopting sounding data, the self-consistency of the model in the GNSS application is realized to the maximum extent, and the application precision of navigation positioning in China areas can be ensured; the accuracy of the new model is higher than that of other common models in inland and high-altitude areas.

Description

Troposphere delay correction method based on CMONOC-ZTD
Technical Field
The invention relates to a global navigation satellite system, in particular to a troposphere delay correction method based on CMONOC-ZTD.
Background
The tropospheric delay of a radio signal is one of important error sources affecting the satellite navigation positioning accuracy, particularly the accuracy in the elevation direction, and the tropospheric zenith delay can reach 2m generally, and the delay is increased to 20m with the decrease of the elevation angle. Currently, the international common meteorological parameter troposphere zenith delay models mainly comprise models such as Hopfield, Saastamoinen and Black, and the application of the models depends on surface meteorological parameters, so that the application range of the models is greatly limited. Most of the common meteorological parameter-free models are established by adopting a numerical weather forecast model, and are troposphere delay models in a global scope established by analyzing global atmospheric average meteorological data and global climate. The accuracy of the model is poor, self-consistency in GNSS navigation positioning is difficult to realize, and the correction effect is limited particularly in areas with wide regions and complex environments. Currently, few ZTD models based on direct modeling of GNSS-ZTD data exist. The method aims at the problems that a traditional troposphere zenith delay model is low in precision and poor in stability in estimation of zenith delay and depends on meteorological parameters and the like, the troposphere zenith delay data are inverted based on GNSS provided by a Chinese regional terrestrial network, the defects of the parameterized model in the traditional ZTD model are avoided by utilizing the self-adaptive learning capability and the nonlinear approximation capability of an artificial neural network, the optimal approximation effect is achieved through automatic adjustment of weights, the effect of local refinement is achieved by establishing an accurate and reliable troposphere delay model, the troposphere delay correction precision in the Chinese region is improved, and the method has important practical significance for improving the precision and the reliability of GNSS navigation positioning.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a high-precision troposphere delay correction method based on CMONOC-ZTD.
The technical scheme is as follows: the invention relates to a troposphere delay correction method based on CMONOC-ZTD, which comprises the following steps:
s1: obtaining modeling data and constructing a training sample;
s2: constructing a plurality of 'weak' neural network models by using the samples obtained in the S1;
s3: inputting sample training in a 'weak' neural network model;
s4: determining a weight for each "weak" neural network model;
s5: and (3) carrying out weighted average on the results of the weak neural network model, outputting a final model and verifying the precision of the final model.
The modeling data in step S1 adopts a troposphere zenith delay (ZTD) product (CMONOC-ZTD for short) sequence provided by a survey station of the China continental environment monitoring network (CMONOC).
The "weak" neural network model in said step S2 assumes an FNN network.
The feed-Forward Neural Network (FNN) network consists of an input layer, a hidden layer and an output layer, wherein the number of nodes of the input layer is 5, the nodes of the input layer respectively represent longitude (°), latitude (°), elevation (m), Daypoint (DOY) and hour (h), the number of nodes of the hidden layer is 27, the number of nodes of the output layer is 1, and the nodes of the ZTD (mm) represent, the activating function selects a Tan-Sigmoid function, and the expression is as follows:
Figure BDA0003061686060000021
wherein x is the input signal value of the neuron, f (x) is the output signal value, the target error of FNN is 1mm, the maximum training frequency is 200 times, and the training is carried out by adopting a gradient descent method.
In step S4, the weight of each "weak" neural network model is determined according to the training error calculation, and the distribution weights of m training samples are initialized to:
D0(i)＝1/m,i＝1,2,3,…,m
the initial accumulated error for each FNN network is:
Error(t)＝0,t＝1,2,3,…,20
after the training of the tth FNN network is finished, if the FNN network is usedtWhen the prediction error of the ith sample is more than 50mm, the sample is considered as the sample needing reinforcement learning, and the accumulated error (t) of the network and the distribution error D of the sample are updated according to the following formulatOtherwise, the accumulated error of the network and the distribution weight of the sample are not changed;
Figure BDA0003061686060000022
calculating the weight of each FNN network, the weight p of the tth FNN networktCalculated and normalized by the following formula:
Figure BDA0003061686060000023
suppose that the tropospheric delay calculated by the tth FNN network is recorded as ZTDtThe final tropospheric delay is calculated from the following equation:
Figure BDA0003061686060000031
the verification precision in step S5 is to use ZTD0 of the land network measuring station not participating in step S1 as a true value, and use a mean deviation (Bias) and a Root mean square error (RMS) as evaluation criteria, where Bias represents accuracy, i.e., a deviation degree of the model from the true value, and RMS represents precision, and is used to measure reliability and stability of the model, and the calculation formula is:
Figure BDA0003061686060000032
has the advantages that: compared with the prior art, the invention has the following advantages:
1. the precision of the new model is improved by 9.6 percent and 3.3 percent compared with the two models by being superior to the common CTrop model and the SHATarop model;
2. the modeling is carried out by adopting a GNSS actual measurement ZTD sequence, which is different from other models, or adopting meteorological reanalysis data or adopting sounding data, the self-consistency of the model in the GNSS application is realized to the maximum extent, and the application precision of navigation positioning in China areas can be ensured;
3. the accuracy of the new model is higher than that of other common models in inland and high-sea wave regions.
Drawings
FIG. 1 is a flow chart of the steps of the present invention;
FIG. 2 is an information diagram of a land network modeling and verification survey station according to an embodiment of the present invention;
FIG. 3 is an algorithmic flow diagram according to an embodiment of the present invention;
fig. 4 is a Bias and RMS distribution diagram of each station of the model and the CTrop and SHAtrop model obtained by the method according to the present embodiment, where fig. 4a is Bias of CTrop, 4b is Bias of SHAtrop, 4c is Bias of BP-Adaboost, 4d is RMS of CTrop, 4e is RMS of SHAtrop, and 4f is RMS of BP-Adaboost;
fig. 5 is a comparison of RMS of the model obtained by the method according to the present embodiment with the CTrop and SHAtrop models in each season, where fig. 5a is RMS of each model in spring, 5b is RMS of each model in summer, 5c is RMS of each model in autumn, and 5d is RMS of each model in winter.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings.
As shown in fig. 1, the present embodiment discloses a ZTD correction method based on COMONC, which includes the following steps:
s1: and acquiring and screening a troposphere zenith delay product provided by a China continental structure environment monitoring network (land network) measuring station meeting the conditions. The method comprises the following specific steps:
in the specific embodiment, the CMONC stations with continuous data in the last 7 years are selected, observed values with errors larger than 5mm are removed, and 217 stations are obtained in total. And uniformly selecting data of 155 stations 2014-2018 for modeling, and verifying the data by using the data of the rest 62 stations (verification stations) and 141 stations (modeling stations) participating in modeling for 2019. 1883076 training samples were obtained, and 136958 and 271882 test samples were obtained.
S2: and (2) constructing a plurality of weak neural network models by using the samples obtained in the S1, determining that the input layer node is 5, and respectively representing longitude (°), latitude (°), elevation (m), Dayyfyer (DOY) and hour (h) of the survey station, the number of hidden layer nodes is 27, and the number of output layer nodes is 1, and represents an accurate ZTD (mm) predicted value.
S3: training establishes 20 FNN networks. Initializing the distribution weight of 1883076 training samples to
Figure BDA0003061686060000041
Each FNN network has three layers, which are respectively composed of an input layer, a hidden layer and an output layer. The number of nodes of the input layer is 5, which respectively represents: station longitude (°), latitude (°), elevation (m), yearly Days (DOY), and hours (h); the number of hidden layer nodes is 27; the number of output layer nodes is 1, representing ZTD (mm). The transfer function of the FNN model hidden layer selects a Tan-Sigmoid function, the expression of which is shown in formula (1), wherein x is an input value:
Figure BDA0003061686060000042
the target error of each FNN network is 1mm, the maximum iteration number is 200, and the training method adopts a standard steepest descent method.
S4: and (3) weight assignment of each FNN network, wherein 20 FNN networks are selected in the method, and the weight of the FNN networks is integrated into a strong predictor for accurately estimating the ZTD. The initial accumulated error for each network (error (t) 0, t 1,2,3, …, 20. The t-th FNN (FNN)t) After the network training is finished, if the network training is finished, the FNNtWhen the prediction error of the ith sample is more than 50mm, the sample is considered as the sample needing reinforcement learning, and the method is as followsEquation (2) updates the accumulated error (error) (t) of the network and the distributed error D of the sampletOtherwise, the accumulated error of the network and the distribution weight of the sample are not changed:
Figure BDA0003061686060000051
calculating the weight of each FNN network, the weight p of the tth FNN networktCan be calculated and normalized by equation (3) to yield:
Figure BDA0003061686060000052
s5: taking the weighted average of the 20 FNN network predictions as the final ZTD output, the formula is shown as (4), and determining the final improved weight and verifying the accuracy.
Figure BDA0003061686060000053
Therefore the tropospheric delays of the two sets of stations in step S1 are recorded as ZTD0As a true value, the mean deviation (Bias) and Root mean square error (RMS) are used as evaluation criteria. Wherein Bias represents accuracy, i.e., the degree of deviation of the model from the true value; the RMS represents the accuracy, which is used for measuring the reliability and stability of the model, and the calculation formula is shown as formula (5):
Figure BDA0003061686060000054
wherein N represents the number of samples, ZTDiThe calculated tropospheric delay for the model,
Figure BDA0003061686060000056
is an accurate value for tropospheric delay. To discuss the applicability of the new model, the CTrop model and SHAtrop model were compared in 2019 using ZTD of 141 modeling stations and 62 verification stations of CMONOC, respectivelyThe precision of the Chinese region, the specific results are shown in Table 1; for further analysis of the spatial applicability of each model, Bias and RMS distribution plots of each CMONC station and sounding station are plotted as shown in fig. 4;
table 1: bias and RMS (unit: mm) of the four models
Figure BDA0003061686060000055
Figure BDA0003061686060000061
Further analyzing the applicability of the four models at different altitudes, and respectively counting each measuring station according to three intervals of 0-2km, 2-4km and >4km according to the elevation of the measuring station, as shown in table 2:
table 2: bias and RMS (unit: mm) of two subregions
Figure BDA0003061686060000062
Further to verify the accuracy of the model in its applicability over different time periods, the RMS of the four models is calculated for each quarter, as shown in fig. 5.
Table 3: bias and RMS (unit: mm) for three elevation intervals
Figure BDA0003061686060000063
As can be seen from tables 1,2,3, 4 and 5:
(1) 155 stations participating in modeling of CMONOC in 2019 and 62 stations not participating in modeling are used for verifying that the total Bias of the BP-Adaboost model in the Chinese area is 0.62mm and minus 1.16mm, which is superior to the common CTRop model and SHATRop model, and shows that the model has no obvious system error; the RMS values were 25.30mm and 26.72mm, which are 9.6%, 2.8% and 3.3% better than the three models.
(3) The precision distribution of the three models shows the characteristics of high inland and low coastal areas in south east, and the advantages of the BP-Adaboost model in the inland area are more obvious. BP-Adaboost is also superior to the other three models in high altitude areas.
(4) The accuracy of the three models is low in summer and high in winter, and the verification results of the CMONOC two groups of data show that the accuracy of the BP-Adboost is improved by about 5% in spring and summer compared with that of a CTRop model and a SHATRop model due to other models.
(5) The ZTD sequence actually measured by the GNSS is adopted for modeling, and different from other models, the model is modeled by adopting weather reanalysis data or radio data, so that the model is self-consistent in GNSS application to the greatest extent, and the application precision of navigation positioning in a Chinese area can be ensured.
From the above conclusions, it can be seen that the reliability of the BP-Adaboost algorithm modeling accuracy based on the land state network of the chinese region, no matter the deviation or the root mean square error, is better than the model currently used in mainstream, and meanwhile, the model has more excellent performance in the inland region of china, so that the delay value of the chinese region can be calculated by using the method provided by the present invention.

Claims (6)

1. A troposphere delay correction method based on CMONOC-ZTD is characterized by comprising the following steps:
s1: obtaining modeling data and constructing a training sample;
s2: constructing a plurality of 'weak' neural network models by using the samples obtained in the S1;
s3: inputting sample training in a 'weak' neural network model;
s4: determining a weight for each "weak" neural network model;
s5: and (3) carrying out weighted average on the results of the weak neural network model, outputting a final model and verifying the precision of the final model.
2. The correction method according to claim 1, wherein the modeling data in step S1 employs troposphere zenith delay product sequences provided by survey stations of the china continental construction environment monitoring network.
3. The correction method according to claim 1, wherein said "weak" neural network model in step S2 adopts an FNN network.
4. The correction method according to claim 3, characterized in that said FNN network is composed of an input layer, a hidden layer and an output layer, the number of nodes of the input layer is 5, representing the longitude, latitude, elevation, the annual date and hour of the measuring station respectively, the number of nodes of the hidden layer is 27, the number of nodes of the output layer is 1, representing ZTD, the activation function is a Tan-Sigmoid function, the expression of which is:
Figure FDA0003061686050000011
wherein x is the input signal value of the neuron, f (x) is the output signal value, the target error of FNN is 1mm, the maximum training frequency is 200 times, and the training is carried out by adopting a gradient descent method.
5. The correction method according to claim 1, wherein the weight of each "weak" neural network model is determined according to the training error calculation in step S4, and the distribution weights of the m training samples are initialized as follows:
D0(i)＝1/m，i＝1，2，3，…，m
the initial accumulated error for each FNN network is:
Error(t)＝0，t＝1，2，3，…，20
after the training of the tth FNN network is finished, if the FNN network is usedtWhen the prediction error of the ith sample is more than 50mm, the sample is considered as the sample needing reinforcement learning, and the accumulated error (t) of the network and the distribution error D of the sample are updated according to the following formulatOtherwise, the accumulated error of the network and the distribution weight of the sample are not changed;
Figure FDA0003061686050000012
calculating the weight of each FNN network, the weight p of the tth FNN networktCalculated and normalized by the following formula:
Figure FDA0003061686050000021
suppose that the tropospheric delay calculated by the tth FNN network is recorded as ZTDtThe final tropospheric delay is calculated from the following equation:
Figure FDA0003061686050000022
6. the correction method according to claim 1, wherein the verification accuracy in step S5 is that ZTD of the land grid station that will not participate in step S10As a true value, the mean deviation and the root mean square error are used as evaluation criteria, and the calculation formula is as follows:
Figure FDA0003061686050000023
CN202110515252.8A 2021-05-12 2021-05-12 Troposphere delay correction method based on CMONOC-ZTD Pending CN113361190A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110515252.8A CN113361190A (en) 2021-05-12 2021-05-12 Troposphere delay correction method based on CMONOC-ZTD

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110515252.8A CN113361190A (en) 2021-05-12 2021-05-12 Troposphere delay correction method based on CMONOC-ZTD

Publications (1)

Publication Number Publication Date
CN113361190A true CN113361190A (en) 2021-09-07

Family

ID=77526164

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110515252.8A Pending CN113361190A (en) 2021-05-12 2021-05-12 Troposphere delay correction method based on CMONOC-ZTD

Country Status (1)

Country Link
CN (1) CN113361190A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117056732A (en) * 2023-10-11 2023-11-14 山东科技大学 Non-isotropic NARX troposphere delay grid prediction method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5479571A (en) * 1991-06-14 1995-12-26 The Texas A&M University System Neural node network and model, and method of teaching same
CN104777488A (en) * 2015-03-13 2015-07-15 中国科学院上海天文台 Modeling method and device for zenith tropospheric delay as well as measuring method and device
CN106022470A (en) * 2016-04-29 2016-10-12 东南大学 Troposphere delay correction method based on BP-EGNOS fusion model

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5479571A (en) * 1991-06-14 1995-12-26 The Texas A&M University System Neural node network and model, and method of teaching same
CN104777488A (en) * 2015-03-13 2015-07-15 中国科学院上海天文台 Modeling method and device for zenith tropospheric delay as well as measuring method and device
CN106022470A (en) * 2016-04-29 2016-10-12 东南大学 Troposphere delay correction method based on BP-EGNOS fusion model

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
王勇 等: "基于BP神经网络的对流层延迟预测研究", 大地测量与地球动力学, vol. 31, no. 3, pages 134 - 137 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117056732A (en) * 2023-10-11 2023-11-14 山东科技大学 Non-isotropic NARX troposphere delay grid prediction method
CN117056732B (en) * 2023-10-11 2023-12-15 山东科技大学 Non-isotropic NARX troposphere delay grid prediction method

Similar Documents

Publication Publication Date Title
CN109344865B (en) Data fusion method for multiple data sources
Arowolo et al. Comparison of spatial interpolation techniques to generate high‐resolution climate surfaces for Nigeria
US20220100997A1 (en) Farmland reference crop evapotranspiration prediction method considering uncertainty of meteorological factors
CN110990505B (en) Loran-C ASF correction method based on neural network
CN112163375B (en) Long-time sequence near-ground ozone inversion method based on neural network
CN111401602A (en) Assimilation method for satellite and ground rainfall measurement values based on neural network
CN105974495B (en) It will be averaged the method for cloud amount using classification fitting process anticipation target area future
CN112232554A (en) Construction method of local short-term rainfall forecast model based on BP neural network
CN111310386A (en) Near-surface ozone concentration estimation method
UA123863C2 (en) Computer-based representation of precipitation
CN114049570B (en) Satellite-borne remote sensing water vapor space inversion method and system based on neural network
CN111797131A (en) Extreme precipitation area frequency analysis method based on remote sensing precipitation product
CN109472283B (en) Dangerous weather prediction method and device based on multiple incremental regression tree model
CN113344149B (en) PM2.5 hourly prediction method based on neural network
CN106022470B (en) A kind of tropospheric delay correction method based on BP-EGNOS Fusion Model
CN108983324A (en) A kind of temperature forecast method and system based on Kalman filtering
Li et al. An improved model for detecting heavy precipitation using GNSS-derived zenith total delay measurements
CN113639893B (en) Near-earth weighted average temperature information acquisition method based on multiple meteorological factors
CN113361190A (en) Troposphere delay correction method based on CMONOC-ZTD
CN110968926B (en) Method for predicting atmospheric parameters based on improved background error covariance matrix
CN117172037B (en) Distributed hydrologic forecasting method, device, computer equipment and medium
Gusev et al. The application of the land surface model for calculating river runoff in high latitudes
CN112329334A (en) MWHTS and MWTS-II fusion inversion sea surface air pressure method based on simulated brightness temperature
Wright Rainfall information for global flood modeling
KR102002593B1 (en) Method and apparatus for analyzing harmful gas diffusion in a specific space

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