Implementation method of troposphere zenith statics delay correction model

Abstract

The invention discloses an implementation method of a troposphere zenith static delay correction model, which comprises the following steps of: calculating approximate truth values of zenith static delay, zenith wet delay and zenith total delay based on high-altitude meteorological observation data of a radiosonde, and respectively recording the approximate truth values as ZHD, ZWD and ZTD; calculating zenith static delay and zenith wet delay by adopting a GTrop model, and respectively recording the zenith static delay and the zenith wet delay as ZHDGTrop and ZWDGTrop; constructing an NNSZHD model by taking the ZTD, the ZHDGTrop and the ZWDGtrop as input variables of a neural network input layer and taking the ZHD as an output variable of a neural network output layer; and taking the obtained ZTD, ZHD, ZHDGTrop and ZWDGTrop as a sample set, training an NNSZHD model, and storing an optimal model for forecasting. The obtained ZHD empirical model has high precision and good applicability.

Description

Technical Field

[0001] The invention belongs to the field of global navigation system and GNSS meteorology, and in particular relates to a method for realizing a tropospheric zenith statics delay correction model. Background Art

[0002] The satellite signals of the Global Navigation Satellite System (GNSS) are affected by tropospheric delay when passing through the earth's atmosphere. As one of the error sources affecting the GNSS positioning accuracy, tropospheric delay needs to be weakened or eliminated through certain methods (such as tropospheric delay empirical model). Zenith Tropospheric Delay (ZTD) is also called zenith total delay. ZTD modeling is an important part of the tropospheric delay research. There are many advantages for GNSS users to use ZTD empirical models, such as universality, convenience, cheapness and real-time performance. At present, ZTD empirical models without measured meteorological parameters have been widely studied. This type of model does not require any measured meteorological parameters (such as measured temperature, measured humidity and measured atmospheric pressure) to be input when calculating ZTD. GNSS users can use this type of model when they cannot obtain measured meteorological data from the station. The input parameters of this type of model usually include: location parameters (longitude, latitude and altitude) and time parameters (year, annual cumulative day and time). The output parameters are usually: 1) ZHD (ZenithHydrostatic Delay) and ZWD (Zenith Wet Delay), 2) meteorological parameters, and then ZHD and ZWD are calculated based on these empirical meteorological parameters. Some early models (such as the UNB series models and the EGNOS model) belong to this type of model. The GPT3 model established by Professor is the most famous, and its accuracy is better than the earlier UNB series models and EGNOS models. At present, the GTrop model of Wuhan University is a ZTD model without measured meteorological parameters applicable to the entire troposphere of the world, and its ZHD and ZWD accuracy are slightly better than the GPT3 model.

[0003] In fact, ZTD and ZHD are strongly correlated. Based on this, when only the ZTD model without measured meteorological parameters is available, ZHD can be obtained in two ways: 1) through the ZHD model without measured meteorological parameters (such as ZHD GTrop ) to obtain ZHD; 2) to obtain the ZWD model without measured meteorological parameters by subtracting ZTD (such as by subtracting ZWD from ZTD GTrop ).

[0004] However, currently, when GNSS users use the ZTD model without measured meteorological parameters to separate the ZTD products measured by GNSS, they only consider the ZHD model without measured meteorological parameters as the reference value of ZHD. In this case, GNSS users do not consider the importance of ZTD itself in the modeling process, nor do they consider that ZHD can also be obtained by using ZTD minus the ZWD model without measured meteorological parameters. Therefore, how to consider the ZHD obtained by the two methods and combine them, and then introduce them into ZHD modeling is an important issue. Summary of the invention

[0005] Purpose of the invention: To provide a method for implementing a tropospheric zenith statics delay correction model, wherein the zenith statics delay correction model is used to separate the ZHD and ZWD in the ZTD measured by GNSS, and obtain a ZHD empirical model with high accuracy and good applicability.

[0006] Technical solution: A method for implementing a tropospheric zenith statics delay correction model, comprising the following steps:

[0007] S1, based on the high-altitude meteorological observation data of radiosonde, calculate the approximate true values ​​of zenith static delay, zenith wet delay and zenith total delay, which are denoted as ZHD, ZWD and ZTD respectively;

[0008] S2, zenith static delay and zenith wet delay are calculated using the GTrop model and are denoted as ZHD GTrop and ZWD GTrop ;

[0009] S3, ZTD, ZHD GTrop and ZWD Gtrop The NNSZHD model is constructed by taking ZHD as the input variable of the neural network input layer and ZHD as the output variable of the neural network output layer;

[0010] S4, using the ZTD, ZHD, and ZHD obtained in steps S1 and S2 GTrop and ZWD GTrop As a sample set, the NNSZHD model is trained and the best model is saved for prediction.

[0011] Furthermore, in the step of calculating ZWD, an exponential interpolation method is used to process the wet refractive index between two adjacent layer heights.

[0012] Furthermore, in the NNSZHD model, the hidden layer is set to a single layer with 5 neurons, and the activation function uses a hyperbolic tangent function as a transfer function.

[0013] Furthermore, in step S4, the conditions for the end of model training are as follows: the expected training error is set to 0, and if the minimum training error of the neural network does not change for 6 consecutive iterations, the training is stopped.

[0014] Compared with the prior art, the present invention has the following significant effects:

[0015] The present invention relies on two ways to obtain ZHD, namely, 1) ZHD obtained by ZHD model without measured meteorological parameters and 2) ZHD obtained by subtracting ZWD of model without measured meteorological parameters from ZTD; and combines the ZHD obtained by these two ways to introduce ZHD modeling. GTrop 、ZWD GTrop The approximate true value of ZTD is used as the modeling parameter, and an empirical ZHD model is established based on artificial neural network technology as a modeling tool. Compared with the two existing mature models (GTrop-ZHD model and GZHD model), the empirical ZHD model has higher accuracy and good applicability in different application areas around the world. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the present invention;

[0017] Figure 2 This is the structure diagram of the NNSZHD neural network model;

[0018] Figure 3 This is a schematic diagram of the global training sites selected by the NNSZHD neural network model;

[0019] Figure 4 This is a schematic diagram of the global test sites selected by the NNSZHD neural network model;

[0020] Figure 5 This is a schematic diagram of the comparison results of the NNSZHD model, GTrop-ZHD model and GZHD model at different latitudes. DETAILED DESCRIPTION

[0021] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0022] The present invention provides a method for implementing a tropospheric zenith statics delay correction model, the flow chart of which is as follows: Figure 1 As shown, the steps include:

[0023] Step 1: Based on the high-altitude meteorological observation data from the radiosonde, the approximate true values ​​of the zenith static delay ZHD and the zenith wet delay ZWD are calculated.

[0024] Based on the upper-air meteorological data from radiosondes, the approximate true value of the zenith static delay is calculated using the following formula:

[0025]

[0026] Where P S is the atmospheric pressure at the measuring station, is the latitude of the station, H S is the orthometric height of the measuring station.

[0027] The approximate true value of the zenith wet delay is calculated based on the upper air meteorological data from radiosondes using the following formula:

[0028]

[0029] Among them, H S is the orthometric height of the station, T is the temperature, e is the water vapor pressure, k'2 and k3 are the constant coefficients of atmospheric refractive index, which are k'2 = 16.52K / mbar and k3 = 377600000K respectively. 2 / mbar.

[0030] The ZWD is calculated using radiosonde upper air parameters (temperature, relative humidity and geopotential height). In the process of calculating ZWD using radiosonde meteorological data, the exponential interpolation method is used when dealing with the wet refractivity between two adjacent layers.

[0031] According to formulas (1) and (2), the approximate true values ​​of ZHD and ZWD are obtained, and the approximate true value of ZTD can be obtained according to ZTD=ZHD+ZWD.

[0032] The high-altitude meteorological observation data of 775 sounding stations around the world are selected to calculate the approximate true values ​​of ZHD, ZWD and ZTD. In this embodiment, some of the calculated data are listed in Table 1.

[0033] Step 2: Use GTrop (global tropospheric model) to calculate ZHD and ZWD, respectively, denoted as ZHD GTrop and ZWD GTrop .

[0034] Used in GTrop model to calculate ZHD GTrop and ZWD GTrop The formulas are:

[0035] ZHD GTrop =ZHD G @[1-α ZHD (H S -H G )] 5.225 (3)

[0036] ZWD GTrop =ZWD G ·[1-α ZWD (H S -H G )] 5.225 (4)

[0037] In the formula, ZHD G 、ZWD G are ZHD and ZWD on the global grid points, α ZHD , α ZWD are the height adjustment coefficients for ZHD and ZWD respectively.

[0038] In the GTrop model, ZHD G 、ZWD G , α ZHD and α ZWD These four parameters all take into account the trend, annual change and semi-annual change characteristics and are calculated according to the following formula:

[0039]

[0040] Among them, y is the year, which is used to describe the changing trend of these parameters; DOY is the number of days corresponding to year y. A1~A6 are ZHD G The time series model coefficients of ZHD The time series model coefficients; C1~C6 are ZWD G The time series model coefficients; D1~D6 are α ZWD Model coefficients for the time series model.

[0041] The ZHD was calculated by using high-altitude meteorological observation data from 775 sounding stations around the world. GTrop and ZWD GTrop Due to space limitations, some of the calculation data are listed in Table 1.

[0042] Table 1 ZHD, ZWD, ZTD, ZHD GTrop and ZWD GTrop Part of the calculation data

[0043]

[0044] Step three, establish a neural network model of ZHD, that is, establish the NNSZHD model.

[0045] NNSZHD model is a neural network model based on multilayer perceptron. The neural network structure of NNSZHD model is designed as follows: the parameter to be predicted is usually set as the output variable of the multilayer perceptron model, so the output variable of NNSZHD model is ZHD. Therefore, there is only one neuron in the output layer, and a linear transfer function (i.e., f(x)=x) is used in the neuron. The design of the input layer should take into account the parameters that affect ZHD modeling.

[0046] In this embodiment, ZHD modeling is mainly carried out from two aspects: on the one hand, a ZHD model based on unmeasured meteorological parameters (such as ZHD GTrop ); On the other hand, the ZTD minus the ZWD model without measured meteorological parameters (i.e., ZTD minus ZWD GTrop ). So the input variables are: ZTD, ZHD GTrop and ZWD GTrop The hidden layer of NNSZHD is set to a single layer with 5 neurons, and the hyperbolic tangent function is used in all 5 neurons. as a transfer function.

[0047] The neural network structure of the NNSZHD model is as follows Figure 2 , the main parameters are shown in Table 2.

[0048] Table 2 Main parameters of the neural network of the NNSZHD model

[0049]

[0050] Step 4: Train the NNSZHD neural network model to obtain the best model for prediction.

[0051] According to step 1 and step 2, calculate the ZTD, ZHD, and ZHD of the whole sounding station. GTrop and ZWD GTrop Sample set, based on which the NNSZHD model is trained. Figure 3 The data of the 452 radiosonde stations shown (i.e., ZHD GTrop 、ZWD GTrop , ZTD and ZHD data sets) are used to train the NNSZHD model. The training process of the NNSZHD model is completed in the MATLAB software system. It is necessary to use the neural network toolkit commonly used in MATLAB software, use the newff function to build a neural network model, and use the train function to train the neural network model. The conditions for the end of model training are as follows: set the expected training error to 0, and stop training when the minimum training error of the neural network does not change for 6 consecutive iterations.

[0052] After the NNSZHD model training step is completed, the use and prediction of the NNSZHD model can be realized, that is, it can be used to separate ZHD and ZWD of ZTD globally.

[0053] In order to verify the application effect and accuracy of the method of the present invention (i.e., NNSZHD model), 323 sounding stations around the world were selected as test stations, with the following geographical distribution: Figure 4 At the same time, two ZHD models with relatively high accuracy (GTrop-ZHD model and GZHD model) were selected as comparison objects. GTrop-ZHD model and GZHD model are existing mature technologies, and both are ZHD empirical models that do not require actual meteorological parameters. Figure 5 As shown in the figure, the accuracy results in the global range (90°S~90°N) show that the average bias (Bias) and root mean square error (RMSE) of the NNSZHD model are -0.002mm and 7.51mm, respectively. In terms of the RMSE comparison results, the NNSZHD model is 6.74mm and 9.53mm less than the GTrop-ZHD model and the GZHD model, respectively. Because the average values ​​of ZHD, ZWD and ZTD of stations around the world vary greatly with latitude. Although the average ZHD and model errors have different characteristics in different latitude bands, the NNSZHD model performs better among the three models. In actual use, the ZTD data measured by GNSS can be substituted into the ZTD in the input variable of the model.

[0054] 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 regarded as the scope of protection of the present invention.

Claims

1. A method for implementing a tropospheric zenith statics delay correction model, characterized in that: The steps include: S1, based on the high-altitude meteorological observation data of radiosonde, calculate the approximate true values ​​of zenith static delay, zenith wet delay and zenith total delay, which are denoted as ZHD, ZWD and ZTD respectively; S2, zenith static delay and zenith wet delay are calculated using the GTrop model and are denoted as ZHD GTrop and ZWD GTrop ; S3, ZTD, ZHD GTrop and ZWD Gtrop The NNSZHD model is constructed by taking ZHD as the input variable of the neural network input layer and ZHD as the output variable of the neural network output layer; S4, using the ZTD, ZHD, and ZHD obtained in steps S1 and S2 GTrop and ZWD GTrop As a sample set, the NNSZHD model is trained and the best model is saved for prediction.

2. The method for implementing the tropospheric zenith statics delay correction model according to claim 1, characterized in that: In the step of calculating ZWD, the exponential interpolation method is used to deal with the wet refractive index between two adjacent layer heights.

3. The method for implementing the tropospheric zenith statics delay correction model according to claim 1, characterized in that: In the NNSZHD model, the hidden layer is set to a single layer with 5 neurons, and the activation function uses the hyperbolic tangent function as the transfer function.

4. The method for implementing the tropospheric zenith statics delay correction model according to claim 1, characterized in that: In step S4, the conditions for the end of model training are as follows: the expected training error is set to 0, and the training is stopped when the minimum training error of the neural network does not change for 6 consecutive iterations.

Images