A BDS-3 real-time clock difference prediction modeling method and system based on grey model
By establishing the homogeneous and non-homogeneous ratio characteristics of clock difference sequences and constructing a first-order univariate non-homogeneous white index law gray model, the GM(1,1) model has a large error in satellite clock difference forecast. The problem of this paper improves the accuracy and stability of forecasting, and enhances the positioning and timing service quality of navigation system.
Patent Information
- Application Number
- CN202410315243.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-19
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-03-19
AI Technical Summary
The existing GM(1,1) and its optimization models often have a few satellite forecasts with large errors in satellite clock difference forecasts, and there is a lack of analysis and research on the mechanism of generating large errors, insufficient adaptability, and no non-construction New model of GM(1,1) and selection strategies for multiple gray models.
The characteristics of the clock difference sequence are correlated with the response equation of the gray model by using homogeneous and non-homogeneous ratios to construct the first-order univariate non-homogeneous white exponential law gray model (NEGM(1,1) and adaptive gray model (AGM), the prediction model is selected according to the clock difference sequence of different characteristics, and the feasibility and effectiveness of the model are verified through multiple prediction experiments.
It improves the accuracy and stability of satellite clock difference forecasting, enhances the adaptability and flexibility of the model, can more accurately capture the internal laws of clock difference changes, reduces time and space errors, and enhances the positioning and timing service quality of the navigation system.
Smart Images

Figure CN118259320B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of satellite clock error modeling and prediction, and in particular relates to a BDS-3 real-time clock error prediction modeling method and system based on a grey model. Background Art
[0002] The Global Navigation Satellite System (GNSS) is essentially a high-precision time and frequency measurement system. The satellite clock, as the on-board reference time, directly affects the accuracy of time and frequency transmission and navigation positioning. It is well known that under the influence of the external environment, the on-board time maintained by the satellite clock cannot be completely synchronized with the standard system time. Therefore, the modeling and prediction of satellite clock errors are essential, especially for ensuring real-time dynamic precise single-point positioning and optimizing navigation message clock error parameters. Generally speaking, the improvement of GNSS real-time performance can be achieved mainly by improving the stability of satellite clocks, updating model parameters more frequently, or improving the accuracy of the model.
[0003] The BDS system has been widely studied since it was put into operation in 2020. However, the quality of BDS satellite clock products has not been well resolved compared with orbit products. There are currently three BDS real-time clock difference products available. Among them, the broadcast ephemeris clock difference product is achieved by the system broadcasting a set of satellite clock parameters every hour, and the user performs quadratic polynomial fitting prediction based on the parameters, with an accuracy of about 10ns. The ultra-fast clock difference product updated every 6 hours uses the observation data of the 24 hours before release (observation accuracy is about 0.5ns) to predict the clock difference of the next 24 hours (prediction accuracy is about 2ns~6ns). Although its accuracy is better than the broadcast ephemeris, it still cannot meet the requirements of real-time positioning and high-precision timing with decimeter-level accuracy. The satellite clock difference real-time correction product uses the real-time observation traffic of local or wide-area BDS stations to estimate the satellite clock difference in real time. The estimated clock difference interval is short and the accuracy is within 0.5ns. However, this product relies on high-performance processing equipment, and the data stream often loses information or is interrupted, and it is difficult for ordinary users to obtain.
[0004] In order to improve the accuracy of real-time clock difference products, a variety of forecast models including mathematical models and neural network types have been proposed, among which the main mathematical models suitable for real-time forecasting include polynomial model (QP, quadratic polynomial), spectral analysis model (SA, spectral analysis), grey model (GM(1,1), time series model (ARIMA, auto-regressive integrated moving average), Kalman filter (KF, Kalman filter) and neural network. Polynomial model and spectral analysis model are widely used because of their clear physical meaning and strong practicality. For example, the ultra-fast clock product of IGS obtains the predicted value through SA, and iGMAS integrates the estimated values obtained by each analysis center through QP or SA to form a clock product. GM(1,1) is very suitable for real-time clock difference forecasting because of its many advantages such as small data modeling and high short-term forecast accuracy. Related work can be found in: Li Chenglong et al. used an improved particle swarm algorithm to enhance the generalization ability of the grey model, and the 6-hour prediction error of the four GPS satellite clocks did not exceed 1.60ns. Tan et al. proposed an improved grey model based on the new information theory, and the average prediction performance was improved compared with the basic GM; many literatures have used the grey model and polynomial together as comparison models.
[0005] The above work optimizes the grey model by optimizing initial conditions or intelligent algorithms, which improves the accuracy of clock error prediction to a certain extent. However, when GM(1,1) and its optimized model predict satellite clock errors, a few satellites often have large prediction errors, and so far there is no relevant literature to analyze and study the mechanism of large errors, the adaptability of GM(1,1), the construction of new models that are not optimized by GM(1,1), and the selection strategy of multiple grey models.
[0006] Through the above analysis, the problems and defects of the prior art are as follows:
[0007] When GM(1,1) and its optimized model predict satellite clock errors, a few satellites often have large prediction errors. So far, there is no relevant literature analyzing and studying the mechanism of large errors, the adaptability of GM(1,1), the construction of new models that are not optimized by GM(1,1), and the selection strategy of multiple grey models. Summary of the invention
[0008] In view of the problems existing in the prior art, the present invention provides a BDS-3 real-time clock difference prediction modeling method and system based on a grey model.
[0009] The present invention is implemented in this way: a BDS-3 real-time clock difference prediction modeling method based on a grey model, the BDS-3 real-time clock difference prediction modeling method based on a grey model comprises:
[0010] Step 1: Use homogeneous ratios and non-homogeneous ratios to associate the characteristics of the clock difference series with the response equation of the grey model, clarify the modeling basis of GM (1,1) and the root cause of large forecast errors;
[0011] Step 2: construct a first-order univariate nonhomogeneous white exponential law grey model (NEGM (1,1));
[0012] Step 3: Select different forecast models based on clock error series with different characteristics, and determine the characteristic indicators and selection criteria;
[0013] Step 4: Conduct multiple forecast tests using 18 BDS-3 satellites.
[0014] Furthermore, in step 1, the homogeneous ratio and non-homogeneous ratio sequence of the sequence are used to characterize the variation characteristics of the clock error data. Let x be the n-data sequence, and x = (x(1), x(2), ..., x(n)), called
[0015]
[0016] and
[0017]
[0018] are the homogeneous class ratio (HCR) and non-homogeneous class ratio (NHCR) of x, respectively, where k = 2, 3, …, n, σ ξ (k) where k = 3, 4, ..., n. Furthermore, the homogeneous order ratio σ corresponding to the GM (1,1) fitting forecast model GM (k) The analytical expression is as follows:
[0019]
[0020] Estimated value of the development coefficient is a constant, so σ GM (k) is also a constant. GM(1,1) performs fitting and forecasting based on the approximation of the homogeneous order ratio of the modeling sequence as a constant. The error between the approximate value and the true value determines the accuracy of the forecast.
[0021] Furthermore, the specific modeling steps of step 2, NEGM (1,1) are as follows:
[0022] Step 1: Generate 1-IAGO and background value sequences from the original signal, and establish the definition model and whitening model of NEGM (1,1);
[0023] Step 2: Solve the whitened differential equation to obtain the continuous response function of NEGM(1,1), find the initial conditions and discretize the time variable to obtain the discrete response function with unknown parameters;
[0024] Step 3: Reconstruct the background value using the linear combination of the adjacent values, and calculate the weight coefficient of the background value by the equation system consisting of the original background value, the definition model of NEGM (1,1) and the response function;
[0025] Step 4: Use the least squares method to estimate the development coefficient and gray action, and obtain the fitting and prediction model of NEGM (1,1).
[0026] Furthermore, in step three, the selection strategy of the grey model is to independently select a suitable grey forecasting model according to the size of the homogeneous order ratio standard deviation (or the non-homogeneous order ratio standard deviation).
[0027] Furthermore, in step four, the ultra-fast observation clock error product of the 927th Beidou week (October 7 to October 13, 2023) of the BDS-3 system provided by iGMAS is used to conduct experiments to verify the feasibility and effectiveness of the satellite clock error adaptive prediction strategy and its superiority over GM (1,1), NEGM (1,1) and ISU-P. In the specific forecasting process, the model constructed using the clock error data of the previous day is used to predict the clock errors of 1h, 6h, 12h and 24h on the next day, and the root mean square error (RMS) between the predicted clock error and the corresponding ultra-fast observation clock error in the forecast time period is used to evaluate the accuracy of the clock error forecast, and the confidence interval width of the error mean is used to evaluate the stability of the clock error forecast.
[0028] Another object of the present invention is to provide a BDS-3 real-time clock difference prediction modeling system based on a grey model, the BDS-3 real-time clock difference prediction modeling system based on a grey model comprises:
[0029] Analysis module, used to analyze the error generation mechanism of grey model;
[0030] Modeling module, used to construct the first-order univariate non-homogeneous white exponential law grey model;
[0031] Strategy selection module, used to select the strategy of the forecast model;
[0032] The test module is used to conduct forecasting tests.
[0033] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the BDS-3 real-time clock difference prediction modeling method based on the gray model.
[0034] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the BDS-3 real-time clock difference prediction modeling method based on the gray model.
[0035] Another object of the present invention is to provide an information data processing terminal, which is used to implement the BDS-3 real-time clock difference prediction modeling system based on the grey model.
[0036] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:
[0037] First, the present invention studies the construction method of the grey prediction model suitable for the approximate non-homogeneous order ratio white exponential law clock difference sequence and the real-time / near real-time satellite clock difference adaptive high-precision prediction strategy. First, through the analysis of the mechanism of the large error generated by the GM (1,1) clock difference prediction, the concepts of homogeneous order ratio and non-homogeneous order ratio are introduced, and the grey model NEGM (1,1) with a non-homogeneous order ratio of 1 is proposed for predicting the approximate non-homogeneous white exponential law clock difference sequence. Then, through a large number of experiments, the characteristic parameters and their discrimination criteria for measuring the fluctuation amplitude of the homogeneous order ratio and non-homogeneous order ratio of the clock difference sequence are clarified, and an adaptive clock difference prediction strategy AGM is proposed that autonomously selects the best prediction model based on the discrimination criteria. Finally, the feasibility and effectiveness of the proposed model NEGM (1,1) and method AGM are verified through single-day and multi-day clock difference prediction experiments, and the following conclusions are obtained.
[0038] (1) The introduction of the NEGM(1,1) model provides a new perspective for enriching the navigation satellite clock error prediction model and the model's adaptability research. Compared with the GM(1,1) model, which is suitable for predicting clock error sequences with homogeneous order ratios approximately constant, the NEGM(1,1) model is more suitable for predicting clock error sequences with non-homogeneous order ratios approximately constant.
[0039] (2) The adaptive forecasting strategy AGM provides a new idea for high-precision real-time / near-real-time clock difference forecasting. AGM combines the advantages of the GM(1,1) and NEGM(1,1) models, and can autonomously select the best forecasting model based on the characteristic parameters of the clock difference sequence and its discrimination criteria. Compared with GM(1,1) and NEGM(1,1), AGM is not only applicable to satellite clock differences with various characteristics, but also has greatly improved forecasting accuracy and stability. In addition, the predicted clock difference of AGM is superior to that of ISU-P products in all aspects, especially in terms of real-time / near-real-time forecast accuracy. This will better serve high-precision applications such as precise point positioning, and provide new methods and strategy references for iGMAS ultra-fast product forecasting.
[0040] Second, (1) NEGM (1,1) model is a grey forecasting model mainly constructed for a large number of approximately non-homogeneous exponential clock difference series existing in real life. The model has a simple structure and is easy to implement. Compared with the traditional GM (1,1) model, it has better fitting and forecasting accuracy.
[0041] (2) AGM combines the advantages of both GM (1,1) and NEGM (1,1) models. It can solve the characteristic parameters of the sequence based on the input data and automatically select the appropriate prediction model based on the discrimination criteria. It takes into account the characteristics of homogeneous exponential and approximately non-homogeneous exponential clock difference sequences and improves the average prediction accuracy of all satellites.
[0042] Third, the technical solution of the present invention solves a technical problem that people have been eager to solve but have never been successful: the traditional GM (1,1) model is used for clock error prediction, and the prediction accuracy is unstable, which is mainly manifested as follows: (1) for the clock error prediction of different satellites in the same time period, some satellites have high prediction accuracy, while some satellites have very low prediction accuracy; (2) for the clock error prediction of the same satellite in different time periods, some time periods have high prediction accuracy, while some time periods have low prediction accuracy. After in-depth research, it is found that GM (1,1) is based on the approximation of the homogeneous order ratio of the modeling sequence as a constant for fitting prediction processing, and the error between the approximate value and the true value determines the prediction accuracy. It is well known that the time-frequency characteristics of satellite clocks are relatively complex and easily affected by the external environment, resulting in unstable clock signals and nonlinear characteristics. Therefore, the homogeneous order ratio of the BDS satellite clock error time series is generally not a constant. In addition, the present invention also finds that most clock error sequences with large homogeneous basic fluctuation amplitudes have very small fluctuation amplitudes of their non-homogeneous order ratios, which can be approximated as a constant. The NEGM(1,1) model is a grey forecasting model constructed for clock difference sequences with non-homogeneous order ratios approximately equal to constants, thus avoiding the large forecasting errors caused by GM(1,1) in such clock difference processing. The AGM forecasting strategy combines the advantages of the GM(1,1) and NEGM(1,1) models, and can solve the characteristic parameters of the sequence according to the input data, and automatically select the appropriate model to forecast the clock difference according to the discrimination criteria, ultimately solving the problem of unstable forecasting accuracy when using the grey model for clock difference forecasting.
[0043] Fourth, the BDS-3 real-time clock error prediction modeling method based on the grey model proposed in the present invention represents a significant technical advancement in the field of satellite navigation system clock error prediction. This advancement is mainly reflected in the following aspects:
[0044] 1. Improve forecast accuracy: By using homogeneous and non-homogeneous ratios to analyze clock error sequences, the present invention can more accurately capture the inherent laws of clock error changes compared to traditional forecasting methods, thereby significantly improving forecast accuracy. This method can effectively reduce the temporal and spatial errors of the navigation system and enhance the overall positioning, navigation and timing (PNT) service quality.
[0045] 2. Enhance model adaptability: The constructed first-order univariate non-homogeneous white exponential gray model (NGBM) can adapt to the nonlinear characteristics and dynamic changes of the clock difference series, making the model more adaptable and flexible. This improvement in adaptability enables the present invention to cope with changing environments and complex clock behaviors, ensuring the stability and reliability of the forecast results.
[0046] 3. Optimize the selection of forecast models: This invention analyzes the characteristics of the clock difference series, adopts different forecast model strategies, and selects the most suitable model for different types of clock difference series. The application of this method not only improves the forecast accuracy of the model, but also optimizes the model selection process, making the forecast more targeted and efficient.
[0047] 4. Empirical verification and model optimization: Through multiple forecast tests on 18 BDS-3 satellites, this paper not only verifies the effectiveness of the proposed model, but also provides an empirical basis for further optimization and improvement of the model. These test results provide an important reference for accurately predicting satellite clock errors, and help guide model selection and parameter adjustment in practical applications.
[0048] The significant technical progress of the present invention provides new methodological support for improving the accuracy and reliability of real-time clock difference prediction in satellite navigation systems, and provides a solid technical guarantee for the high-precision service quality of satellite navigation systems such as BDS-3, which has important theoretical significance and application value. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 It is a flow chart of a BDS-3 real-time clock difference prediction modeling method based on a grey model provided in an embodiment of the present invention;
[0050] Figure 2 is a modeling process diagram of NEGM(1,1) provided in an embodiment of the present invention;
[0051] Figure 3 It is a module diagram of a BDS-3 real-time clock difference prediction modeling system based on a grey model provided by an embodiment of the present invention;
[0052] Figure 4 It is a forecast accuracy map of the satellite clock error GM(1,1) on October 9, 2023 provided by an embodiment of the present invention;
[0053] Figure 5 is the homogeneous order ratio of the satellite clock error provided in the embodiment of the present invention;
[0054] Figure 6 is a non-homogeneous ratio of the satellite clock error provided by an embodiment of the present invention;
[0055] Figure 7 is the non-homogeneous ratio of C19, C20, C21 and C24 provided in the embodiment of the present invention;
[0056] Figure 8 is the standard deviation of the homogeneous ratio and the inhomogeneous ratio of the satellite clock from 2021 to 2023 provided by the embodiment of the present invention;
[0057] Fig. 9is a homogeneous ratio and a non-homogeneous ratio of clock errors corresponding to some abnormal prediction values provided by an embodiment of the present invention;
[0058] Fig.10 are the homogeneous ratios, non-homogeneous ratios and original clock error sequences of the C30 satellite clock error on October 2, 2022, provided by an embodiment of the present invention;
[0059] Fig.11 It is a grey model adaptive clock error prediction strategy provided by an embodiment of the present invention;
[0060] Fig.12 is a homogeneous order ratio variation curve of C20 and C24 satellites provided in an embodiment of the present invention;
[0061] Fig.13 It is the average prediction accuracy of each satellite for 6 days provided by the embodiment of the present invention. DETAILED DESCRIPTION
[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0063] like Figure 1 As shown, the BDS-3 real-time clock difference prediction modeling method based on the grey model provided in the embodiment of the present invention comprises the following steps:
[0064] S101, using homogeneous ratios and non-homogeneous ratios to relate the characteristics of the clock difference series to the response equation of the grey model (GM(1,1)), clarifying the modeling basis of GM(1,1) and the source of large forecast errors;
[0065] S102, constructing a first-order univariate nonhomogeneous white exponential law grey model (NEGM(1,1));
[0066] S103, selecting different forecast models according to clock difference sequences with different characteristics, and determining characteristic indicators and selection criteria;
[0067] S104, multiple forecast tests through 18 BDS-3 satellites.
[0068] The modeling of GM(1,1) is to generate the internal law of the mining sequence by accumulation, and use the whitening equation to describe the sequence characteristics to obtain the response function. Let the original sequence be x (0) , and x (0) =(x (0) (1),x (0) (2),…,x (0) (n)), n is the number of data in the sequence; x (1) For x(0) The first-order accumulated generating operation (1-AGO) sequence, x (1) =(x (1) (1),x (1) (2),…,x (1) (n)), and z (1) For x (1) The immediate neighboring mean background value is x (0) (k)+az (1) (k) = b is the definition form of GM(1,1), dx (1) / dt+ax (1) =b is the whitening model of GM(1,1).
[0069] use Find the initial conditions of the response function and get the fitting prediction model of GM(1,1) as
[0070]
[0071] in, and is the estimated value of the development coefficient a and the gray action b, k=1,2,…,n,n+1,…,n+m, n is the number of data in the original sequence, and m is the number of data in the forecast. When 1≤k≤n, the obtained is the fitting value of the original sequence; when n+1≤k≤m, the obtained is the predicted value of the original series.
[0072] In order to better explain the fundamental cause of the error in the grey model prediction of satellite clock error, the homogeneous order ratio and non-homogeneous order ratio sequence are used to characterize the variation characteristics of clock error data. Let x be an n-dimensional data sequence, and x = (x(1), x(2), ..., x(n)), called
[0073]
[0074] and
[0075]
[0076] are the homogeneous class ratio (HCR) and non-homogeneous class ratio (NHCR) of x, respectively, where k = 2, 3, …, n, σ ξ(k) where k = 3, 4, ..., n. Furthermore, the homogeneous order ratio σ corresponding to the GM (1,1) fitting forecast model GM (k) The analytical expression is as follows:
[0077]
[0078] Estimated value of the development coefficient is a constant, so σ GM (k) is also a constant. That is, GM(1,1) is based on the approximation of the homogeneous order ratio of the modeling sequence as a constant for fitting and forecasting. The error between the approximate value and the true value determines the accuracy of the forecast. It is well known that the time-frequency characteristics of satellite clocks are relatively complex and are easily affected by the external environment, resulting in unstable clock signals and nonlinear characteristics.
[0079] The non-homogeneous white exponential grey model (NEGM (1,1)) provided in the embodiment of the present invention is constructed, including:
[0080] 2.1NEGM(1,1) definition
[0081] In the non-homogeneous white exponential law grey model, the grey difference equation and grey differential equation are established by using the original sequence and its first-order cumulative reduction sequence, and then the response equation of the model is obtained. (0) is the original sequence, and x (0) =(x (0) (1),x (0) (2),…,x (0) (n)), x (-1) For x (0) A first-order decreasing sequence, x (-1) =(x (-1) (1),x (-1) (2),…,x (-1) (n)), and x (-1) (k) = x (0) (k)-x (0) (k-1), called
[0082] x (-1) (k)+az (0) (k) = b (6)
[0083] is the definition of the non-homogeneous white exponential law grey model (NEGM), referred to as NEGM (1,1).
[0084]
[0085] It is called the whitening model of NEGM (1,1). Among them, a and b are the development coefficient and gray action, respectively, and z (0) (k) is x (0) The background value of
[0086]
[0087] To obtain the NEGM (1,1) forecast model, it is also necessary to solve the grey model response equation x (0) Solution of (k), background value z (0) (k) and the estimation of the development coefficient a and the gray action b.
[0088] 2.2 Solution of NEGM(1,1) response equation
[0089] According to formula (7), the approximate solution of the whitened differential equation is
[0090]
[0091] Where c is the initial condition. If we take x (0) The first component of is used as the basis for calculating the initial conditions, that is, The initial condition is obtained from formula (9):
[0092]
[0093] Substituting the value of c into equation (9), we can obtain the continuous response equation of the NEGM (1,1) model:
[0094]
[0095] After discretizing the time variable t in equation (10), the discrete response equation of the NEGM (1,1) model is obtained as follows:
[0096]
[0097] In the equation, the value of k is: k=1,2,…,n,n+1,…,n+m, n is the number of data in the original sequence, and m is the number of forecast data. When 1≤k≤n, the obtained is the fitting value of the original sequence; when n+1≤k≤m, the obtained is the predicted value of the original series.
[0098] Assume that the nonhomogeneous order ratio of the response equation of the NEGM (1,1) model is σ ξ (k), according to formula (3) we can get
[0099]
[0100] Obviously, NEGM(1,1) is a grey model established by approximating the non-homogeneous order ratio of the original sequence to 1.
[0101] 2.3 Background value construction
[0102] Since the background value construction method close to the mean adopted by the traditional grey model often produces lag errors for high growth index series, this paper uses the background value z (0) (k) is set as a linear combination of a pair of adjacent sequences, that is,
[0103] z (0) (k) = ωx (0) (k-1)+(1-ω)x (0) (k) (12)
[0104] Among them, the coefficient to be determined is ω∈[0,1].
[0105] For the convenience of description,
[0106]
[0107] but and The expressions are simplified to
[0108]
[0109] Use x (0) The true value replaces the fitted value, that is Substituting equation (13) into equation (8) we can obtain
[0110]
[0111] Substituting equation (14) into equation (12), we can obtain
[0112] z (0) (k) = ω[Ee -a(k-2) +F]+(1-ω)[Ee -a(k-1) +F] (16)
[0113] Combining equations (15) and (16), we get
[0114]
[0115] After arranging and simplifying formula (17), we can obtain
[0116]
[0117] It can be seen from this that if the development coefficient a can be obtained, the background value can be obtained.
[0118] 2.4 Estimation of development coefficient and grey action parameters
[0119] The following equations are obtained by defining NEGM(1,1) and constructing the background value:
[0120]
[0121] Solving the system of equations yields
[0122]
[0123] make
[0124]
[0125] Formula (18) can be simplified as
[0126] x (0) (k) = λ1x (0) (k-1)+λ2 (20)
[0127] Where k = 2, 3, ..., n, let
[0128]
[0129] Using the least squares parameter method to estimate the parameters in equation (20), we get
[0130] The estimated values of the parameters and Substituting into equation (19), we get the estimated values of parameters a and b:
[0131]
[0132] In summary, if Figure 2 As shown, the specific modeling steps of NEGM(1,1) are as follows:
[0133] Step 1: Generate 1-IAGO and background value sequences from the original signal, and establish the definition model and whitening model of NEGM (1,1);
[0134] Step 2: Solve the whitened differential equation to obtain the continuous response function of NEGM(1,1), find the initial conditions and discretize the time variable to obtain the discrete response function with unknown parameters;
[0135] Step 3: Reconstruct the background value using the linear combination of the adjacent values, and calculate the weight coefficient of the background value by the equation system consisting of the original background value, the definition model of NEGM (1,1) and the response function;
[0136] Step 4: Use the least squares method to estimate the development coefficient and gray action, and obtain the fitting and prediction model of NEGM (1,1).
[0137] The satellite clock difference sequences with different fluctuation amplitudes of aligned and non-homogeneous ratios have different corresponding GM(1,1) and NEGM(1,1) prediction accuracies. In addition, standard deviation is often used to measure the dispersion or fluctuation amplitude of a signal. Generally speaking, the larger the value of the standard deviation, the larger the amplitude of signal fluctuation. Therefore, the selection strategy of the grey model proposed in the embodiment of the present invention is based on the standard deviation of the homogeneous ratio (abbreviated as STD HCR ) or non-homogeneous ratio standard deviation (STD NHCR ) size and select the appropriate grey forecast model (AGM for short).
[0138] The embodiment of the present invention uses the ultra-fast observation clock difference product of the 927th Beidou week (October 7 to October 13, 2023) of the BDS-3 system provided by iGMAS to conduct experiments to verify the feasibility and effectiveness of the satellite clock difference adaptive prediction strategy and its superiority over GM (1,1), NEGM (1,1) and ISU-P. In the specific forecasting process, the model constructed using the clock difference data of the previous day predicts the clock differences of 1h, 6h, 12h and 24h on the next day respectively. The root mean square error (RMS) between the predicted clock difference and the corresponding ultra-fast observation clock difference in the forecast time period is used to evaluate the accuracy of the clock difference forecast, and the confidence interval width of the error mean is used to evaluate the stability of the clock difference forecast. The smaller the RMS value, the higher the accuracy; the narrower the confidence interval width, the higher the stability.
[0139] like Figure 3 As shown, the BDS-3 real-time clock difference prediction modeling system based on the grey model provided by the embodiment of the present invention includes:
[0140] Analysis module, used to analyze the error generation mechanism of grey model;
[0141] Modeling module, used to construct the first-order univariate non-homogeneous white exponential law grey model;
[0142] Strategy selection module, used to select the strategy of the forecast model;
[0143] The test module is used to conduct forecasting tests.
[0144] The BDS-3 real-time clock difference prediction modeling method based on the grey model provided in the application embodiment of the present invention is applied to a computer device, wherein the computer device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the BDS-3 real-time clock difference prediction modeling method based on the grey model.
[0145] The BDS-3 real-time clock difference prediction modeling method based on the grey model provided in the application embodiment of the present invention is applied to an information data processing terminal, and the information data processing terminal is used to implement the BDS-3 real-time clock difference prediction modeling system based on the grey model.
[0146] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.
[0147] The embodiments of the present invention have achieved some positive effects during the development or use process, and indeed have great advantages over the prior art. The following content is described in conjunction with data, charts, etc. of the test process.
[0148] 1 Analysis of the error generation mechanism of grey model
[0149] The homogeneous order ratio of the BDS satellite clock error time series is generally not a constant. The clock error data on October 9, 2023 are randomly selected for prediction. The prediction accuracy (RMS) of each satellite clock GM(1,1) is as follows: Figure 4 shown.
[0150] It can be seen that the prediction accuracy of the gray model for most BDS satellite clocks is in the nanosecond level, while the prediction error of a few satellite clock errors is dozens of times higher than that of other satellite clocks, such as C20 and C24 satellite clocks. In order to more intuitively show the impact of homogeneous ratios on prediction errors, Figure 5 The homogeneous ratio sequence of all satellite clock errors of the ultra-rapid observation clock error of the BDS-3 system on October 9, 2023 provided by iGMAS is shown. Figure 5 The homogeneous ratio series of C28, which has the smallest prediction error, and C20 and C24, which have larger errors, are shown separately.
[0151] from Figure 5It can be seen that (1) except for a few satellite clock error sequences whose homogeneous order ratios fluctuate significantly, the homogeneous order ratio variation curves of most satellite clock errors can be approximated as a straight line parallel to the horizontal axis. (2) For the C28 satellite with the smallest RMS value of the grey model prediction accuracy, the homogeneous order ratio variation curve only fluctuates smoothly within a small range. (3) For the C20 and C24 satellites with larger RMS values of the grey model prediction accuracy, the homogeneous order ratio variation curves have an obvious upward trend. It can be seen that there is a mathematical mapping relationship between the GM(1,1) model and the homogeneous order ratios of the modeled sequence, so it is suitable for predicting clock error sequences whose homogeneous order ratios are approximately straight lines. But in fact, the homogeneous order ratios of almost all satellite clock errors are just variables that fluctuate within a certain range and are not constants, even for the satellite with the smallest RMS value.
[0152] For the convenience of description, the following definitions are quoted:
[0153] (1) If σ(k)∈[μ1,μ2], then x is said to have a grey exponential law, or x is called a grey exponential law sequence;
[0154] (2) If σ(k) is a constant, that is, σ(k) = const, then x is said to have a homogeneous white exponential law, or x is called a homogeneous white exponential law sequence;
[0155] (3) If σ ξ (k) is a constant, that is, σ ξ (k) = const, then x is said to have a non-homogeneous White exponential law, or x is called a non-homogeneous White exponential law sequence.
[0156] Under the above definition, the use of the GM (1,1) model for clock error prediction requires the gray exponential law sequence to be approximated as a homogeneous white exponential law sequence. Obviously, the approximation error is the root cause of the prediction error of the GM (1,1) model. The smaller the approximation error, the higher the prediction accuracy. If the homogeneous order ratio of the satellite clock error sequence has a significant upward or downward trend, a large approximation error will be generated, which is reflected in the low prediction accuracy.
[0157] The above phenomenon inspired the present invention to think deeply about the characteristics of the satellite clock error sequence itself. A very natural idea is, what are the characteristics of the non-homogeneous order ratio of the clock error sequence? To this end, the present invention analyzes Figure 5 The corresponding non-homogeneous order ratio change curve is as follows: Figure 6 As shown, (a) is the non-homogeneous ratio of all satellite clock errors, (b), (c) and (d) are the non-homogeneous ratios of C28, C20 and C24 satellites respectively.
[0158] Depend on Figure 6It can be seen that the non-homogeneous magnitude ratios of the 18 satellites basically fluctuate around 1, and the fluctuation range of most non-homogeneous magnitude ratios is relatively small. Even the C20 and C24 satellites whose homogeneous magnitude ratios have obvious upward or downward trends have the same fluctuation pattern. Figure 6 (c) and (d). Therefore, in theory, the non-homogeneous order ratio of the original clock difference sequence can be approximated as a constant, thereby constructing a grey model suitable for the non-homogeneous white-law clock difference sequence. At this time, seeking a function model with a constant non-homogeneous order ratio is the key to modeling.
[0159] 2 Construction of non-homogeneous white exponential grey model (NEGM(1,1))
[0160] The ultra-rapid observed clock errors of all BDS-3 system satellites on October 9, 2023 are used to perform NEGM(1,1) and GM(1,1) modeling, and the clock errors on October 10 are predicted. The results are shown in Table 1.
[0161] Table 1 GM(1,1) and NEGM(1,1) satellite clock error prediction accuracy (RMS / ns)
[0162]
[0163] The data in Table 1 show that, except for C19 and C21, the NEGM(1,1) prediction accuracy of most satellites is high and higher than that of GM(1,1). For the same satellite, the accuracy of NEGM(1,1) is high while the accuracy of GM(1,1) is relatively low, and vice versa, especially for C19, C20, C21 and C24. Since the modeling basis of NEGM(1,1) is to approximate the non-homogeneous order ratio of the clock error sequence as a constant, satellites with lower prediction accuracy of GM(1,1) and NEGM(1,1) are selected for comparative analysis. The non-homogeneous order ratios of the four satellites C19, C20, C21 and C24 are as follows: Figure 7 shown.
[0164] observe Figure 7 The results show that the non-homogeneous order ratio change curves of the four satellites all fluctuate smoothly around 1, and there is no obvious upward or downward trend, but the fluctuation amplitudes of C19 and C21 are significantly greater than those of C20 and C24. Referring to the data in Table 2, it is also found that the NEGM (1,1) model is not suitable for predicting satellites with large NHCR fluctuations, such as C19 and C21, but is suitable for predicting C20 and C24 satellites, which are just not suitable for the GM (1,1) model. It can be seen that the clock error prediction results of the GM (1,1) and NEGM (1,1) models are complementary.
[0165] The above phenomenon shows that a single model cannot solve all satellite clock error prediction problems, and this conclusion is widely recognized by existing researchers. It also shows that there is a corresponding relationship between the satellite clock error signal and the GM (1,1) or NEGM (1,1) model. If the clock error sequence can be linked to the prediction model through relevant parameters, the appropriate model can be selected to predict the clock error in order to obtain high-precision prediction results. Based on the previous analysis and conclusions, this paper intends to solve the problem of model selection by exploring the characteristic parameters and their discrimination criteria that characterize the essential properties of the original clock error sequence.
[0166] 3 Selection strategy of grey model
[0167] Table 2 gives the STD of all BDS-3 satellite clock errors on October 9, 2023. HCR and STD NHCR , and the RMS prediction accuracy of the satellite clock error on October 10 using GM(1,1) and NEGM(1,1) to model the clock error on October 9 GM and RMS NEGM .
[0168] Table 2 Standard deviation (STD) of magnitude ratio and accuracy of clock error prediction (RMS / ns)
[0169]
[0170]
[0171] From the data given in Table 2, we can see that STD HCR Large and STD NHCR Hours, RMS GM Large values, such as C20, C24 and C36 satellites; STD HCR Small and STD NHCR When large, RMS NEGM Large, such as C19, C21 and C32 satellites; STD NHCR and STD HCR When both are large, RMS NEGM Less than RMS GM ;STD NHCR and STD HCR When moderate, RMS NEGM and RMS GM are relatively small, and the RMS NEGM Less than RMS GM Obviously, the premise of selecting the best model is to determine the threshold of the standard deviation. Below, the BDS-3 satellite clock error of one day is randomly selected from each month from 2021 to 2023 and its STD is calculated. HCR and STD NHCR, in order to obtain a stronger universal threshold, the results are as follows Figure 8 The two black straight lines parallel to the horizontal axis are STD HCR =10 -5 and STD NHCR =5×10 -2 .
[0172] from Figure 8 It can be seen that the STD of most satellite clocks HCR <1E-5 and STD NHCR <5E-2, STD of a few satellite clocks HCR >1E-5 or STD NHCR >5E-2, STD of individual satellite clocks only HCR >10E-5 and STD NHCR >5E-2, such as C30. Table 3 gives some STD HCR >1E-5 or STD NHCR >5E-2 and when both are larger, the corresponding satellite related information. Fig. 9 It is a curve of homogeneous order ratio or non-homogeneous order ratio sequence change of a part of corresponding satellites. Fig.10 The homogeneous ratio, non-homogeneous ratio and original clock error series of the C30 satellite clock error on September 6, 2022 are shown.
[0173] Table 3 Some forecast anomalies and related information of satellite clocks
[0174]
[0175] From Table 3 and Figures 9-10 It can be seen that STD HCR >1E-5 and STD NHCR When <5E-2, the homogeneous order ratio change curve shows an obvious upward trend. At this time, the NEGM (1,1) model should be used to predict the clock error; STD NHCR >5E-2 and STD HCR When the non-homogeneous ratio is less than 1E-5, the fluctuation range of the non-homogeneous ratio is very large, and it is advisable to use the GM(1,1) model to predict the clock error; the original clock error sequence of the C30 satellite fluctuates frequently, resulting in large fluctuations in both the homogeneous ratio and the non-homogeneous ratio, so that STD HCR >1E-5 and STD NHCR >5E-2, it is advisable to use NEGM(1,1) to predict the clock error. HCR and STD NHCR The thresholds are set to 1E-5 and 5E-2 respectively. The selection process of the prediction model is as follows: Fig.11 shown.
[0176] 4. Satellite clock error adaptive prediction example and analysis
[0177] The ultra-rapid observation clock error products of the 927th Beidou week (October 7 to October 13, 2023) of the BDS-3 system provided by iGMAS were used to conduct experiments to verify the feasibility and effectiveness of the satellite clock error adaptive prediction strategy and its superiority over GM (1,1), NEGM (1,1) and ISU-P. In the specific forecasting process, the model constructed using the clock error data of the previous day predicted the clock errors of 1h, 6h, 12h and 24h on the next day respectively. The root mean square error (RMS) between the predicted clock error and the corresponding ultra-rapid observation clock error in the forecast period was used to evaluate the accuracy of the clock error forecast, and the confidence interval width of the error mean was used to evaluate the stability of the clock error forecast. The smaller the RMS value, the higher the accuracy; the narrower the confidence interval width, the higher the stability.
[0178] 4.1 Single-day forecast results and analysis
[0179] The model constructed based on the clock error data on October 7, 2023 is used to predict the satellite clock error on October 8. The prediction results are shown in Tables 5-6 and Fig.11 As shown in Table 5, due to the large amount of data, only the RMS of the 6-h and 24-h forecast durations of each satellite are listed. Since the forecast errors of a few satellites in GM(1,1) and NEGM(1,1) are very large, Fig.11 Only the 24h forecast error curves of ISU-P and AGM are given. Table 5 shows the confidence intervals and their widths of the clock error forecasts of the 18 satellites in each forecast period.
[0180] Table 4 Prediction accuracy (RMS / ns)
[0181]
[0182]
[0183] Table 5927 Confidence interval and width of the first day of Beidou week
[0184]
[0185] From the results shown in Tables 4 and 5, we can see that:
[0186] (1) Unlike C19 and C20 in GM(1,1) and C21 and C24 in NEGM(1,1), the RMS values of all satellites in AGM are lower. The average RMS of AGM forecast for 6h and 24h are 0.182ns and 0.855ns, respectively. Compared with IOGM(1,1), NEGM(1,1) and ISU-P, its 6h forecast accuracy is improved by 69.2%, 89.02% and 84.55%, respectively, and its 24h forecast accuracy is improved by 72.22%, 74.76% and 49.53%, respectively.
[0187] (2) The variation curves of the homogeneous ratio series of the C20 and C24 satellite clock errors show a clear upward trend (e.g. Fig.12 As shown in Figure 2, it is not suitable to use the GM(1,1) model to predict the clock error. The standard deviations of the non-homogeneous magnitude ratios of C19 and C21 satellites are 0.136 and 0.361, respectively, which are much larger than the threshold of 5×10^(-2). Therefore, it is not suitable to use the NEGM(1,1) model to predict the clock error. It can be seen that the experimental results are consistent with the previous theoretical analysis.
[0188] (3) When the AGM clock error is used, the prediction errors of all satellites are concentrated around 0 when the prediction time is short (6h). As the prediction time increases, the prediction errors of most satellites can still maintain a stable change, while the error curves of a few satellites are seriously emitted. The width of the confidence interval given in Table 5 also shows that the stability of the AGM clock error prediction is better than other methods.
[0189] 4.26-day statistical results and analysis
[0190] The model constructed using the clock difference of the day before October 7 to October 13, 2023 is used to predict the clock difference of the next day. The statistical results for the six days are shown in Tables 6-7 and Fig.13 shown.
[0191] Table 6 Average prediction accuracy of 18 satellites in 6 days and its improvement rate
[0192]
[0193] Fig.11 Average forecast accuracy for each satellite over 6 days
[0194] Table 7 Average confidence intervals and their widths for each forecast duration for 6 days
[0195]
[0196] Tables 6 to 7 and Fig.13 The statistical forecast results for multiple days show.
[0197] (1) Compared with IOGM(1,1), NEGM(1,1) and ISU-P, the forecast accuracy of AGM can be improved by up to 89.02%, and the stability can be improved by up to 93.68%.
[0198] (2) From Fig.12 From the statistical results of each satellite for 6 days, it can be seen that in the 927th Beidou week, the C20, C24 and C36 satellites are not suitable for using the IOGM (1,1) model to predict the clock error, and the C19, C21 and C32 satellites are not suitable for using the NEGM (1,1) model to predict the clock error.
[0199] In summary, the AGM proposed in this paper can autonomously select a forecast model suitable for the modeling sequence characteristics according to the discrimination criteria, thereby avoiding the large forecast error caused by the mismatch between the essential characteristics of the clock difference series and the adaptability of the model. Both the single-day forecast and multi-day forecast results show that the forecast accuracy and stability of AGM are better than IOGM (1,1), NEGM (1,1) and ISU-P.
[0200] This paper proposes a BDS-3 real-time clock error prediction modeling method based on a grey model, aiming to improve the accuracy of satellite clock error prediction in the Beidou satellite navigation system. The working principle of this method can be divided into the following steps for detailed explanation:
[0201] Step 1: Establish the basis of grey model
[0202] In this step, the characteristics of the satellite clock error series are linked to the response equation of the grey model by using homogeneous and non-homogeneous ratios. This method clarifies the basis for the GM (1,1) model (a commonly used grey prediction model) and points out the possible sources of forecast errors. By analyzing the characteristics of the clock error series, we can better understand its changing laws and internal mechanisms, providing a scientific basis for the construction of subsequent models.
[0203] Step 2: Constructing the white exponential law gray model
[0204] Based on the analysis results of step 1, a first-order univariate non-homogeneous white exponential gray model (NGBM) is constructed. This model can more accurately reflect the nonlinear characteristics of the clock difference series and the dynamic changes of the time series, thereby improving the accuracy of the forecast. By introducing white noise terms, the NGBM model increases the model's adaptability to actual data changes, making the forecast results more reliable.
[0205] Step 3: Choose an appropriate forecast model
[0206] Different forecast model strategies are adopted according to the different characteristics of the clock difference series. By determining the characteristic indicators and selection criteria, the most appropriate forecast model is selected for each clock difference series. The key to this step is to analyze the statistical characteristics of the clock difference data, such as the trend of change, periodicity, etc., to select the model that best captures its characteristics, thereby improving the accuracy and reliability of the forecast.
[0207] Step 4: Conduct forecasting tests
[0208] The validity and accuracy of the selected model were verified by conducting multiple forecast tests on 18 BDS-3 satellites. These tests not only considered the performance of the model under different conditions, but also covered the model's ability to predict future clock error changes. Through repeated tests and data analysis, the model parameters were further optimized to improve the overall performance of the forecast method.
[0209] Embodiment 1
[0210] In the implementation process of the BDS-3 real-time clock difference prediction modeling method based on the grey model, the real-time clock difference data of the BDS-3 satellite is first obtained. For the obtained clock difference sequence, the homogeneous order ratio and non-homogeneous order ratio are calculated to analyze the change trend and stability of the clock difference sequence, so as to determine whether it is suitable to use the GM (1,1) model for modeling.
[0211] Then, according to the characteristics of the clock difference sequence, a first-order univariate non-homogeneous white exponential law grey model is constructed. When constructing the model, the nonlinear variation characteristics and influencing factors of the clock difference sequence are fully considered to ensure the accuracy and reliability of the model.
[0212] After the model is built, the appropriate forecast model is selected based on the pre-set characteristic indicators and selection criteria. These characteristic indicators may include the rate of change, stability, and periodicity of the clock error series, while the selection criteria are based on factors such as the model's prediction accuracy, stability, and computational complexity.
[0213] Finally, one or more of the 18 BDS-3 satellites were selected for multiple forecast tests. In each test, the model was applied to the forecast of actual clock error data, and the accuracy and reliability of the forecast results were evaluated. Through multiple tests and comparative analysis, the effectiveness and superiority of the modeling method can be verified.
[0214] Embodiment 2
[0215] When implementing the BDS-3 real-time clock difference prediction modeling method based on the grey model, the present invention first preprocesses the clock difference data of the BDS-3 satellite, including data cleaning and outlier processing, to ensure the accuracy and reliability of the data.
[0216] Then, the present invention uses the homogeneous ratio and non-homogeneous ratio methods to conduct an in-depth analysis of the clock error series. Through analysis, the present invention finds that the clock error series of some satellites show a relatively stable trend, which is suitable for modeling using the GM (1,1) model; while the clock error series of other satellites show strong nonlinear variation characteristics, which require the construction of a more complex grey model for description.
[0217] Based on the above analysis, the present invention constructs corresponding grey models for clock difference sequences with different characteristics. For stable clock difference sequences, the present invention adopts the standard GM (1,1) model for modeling; for clock difference sequences with strong nonlinear changes, the present invention constructs a first-order univariate non-homogeneous white exponential law grey model to better capture its changing law.
[0218] After the model is built, the present invention formulates a detailed forecast strategy, including the parameter setting of the model, the determination of the forecast period, and the evaluation method of the forecast results. Then, the present invention selects a number of representative BDS-3 satellites to conduct multiple forecast tests. By comparing with the actual clock error data, the present invention verifies the superiority of the modeling method in terms of forecast accuracy and stability.
[0219] At the same time, the present invention also conducts an in-depth analysis and discussion of the forecast results, explores various factors that affect the forecast accuracy, and proposes corresponding improvement measures. These improvement measures include optimizing model parameters, improving forecast strategies, and improving data processing accuracy, etc., aiming to further improve the performance and application effect of the BDS-3 real-time clock difference forecast modeling method based on the grey model.
[0220] The BDS-3 real-time clock difference prediction modeling method based on the grey model of the present invention significantly improves the accuracy and reliability of clock difference prediction through refined steps and scientific model selection. The method provides an efficient and feasible technical solution for clock difference management and control in the BDS-3 satellite navigation system.
[0221] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.
Claims
1. A BDS-3 real-time clock difference prediction modeling method based on a grey model, characterized in that: include: Step 1: Use homogeneous ratios and non-homogeneous ratios to associate the characteristics of the clock difference series with the response equation of the grey model, and clarify the modeling basis of GM (1,1) and the source of forecast errors; Step 2: construct the first-order univariate non-homogeneous white exponential law grey model NEGM (1,1); Step 3: Select different forecast models based on clock error series with different characteristics, and determine the characteristic indicators and selection criteria; Step 4: multiple forecast tests using 18 BDS-3 satellites; In the first step, the homogeneous ratio and non-homogeneous ratio sequences of the sequence are used to characterize the variation characteristics of the clock error data. Let x be the n-data sequence, and x = (x(1), x(2), ..., x(n)), called and are the homogeneous ratio and non-homogeneous ratio of x, respectively, where k = 2, 3, ..., n, σ ξ (k) where k = 3, 4, ..., n. Furthermore, the homogeneous order ratio σ corresponding to the GM (1,1) fitting forecast model GM (k) The analytical expression is as follows: Estimated value of the development coefficient is a constant, so σ GM (k) is also a constant. GM(1,1) performs fitting and forecasting based on the approximation of the homogeneous order ratio of the modeling sequence as a constant. The error between the approximate value and the true value determines the accuracy of the forecast. The specific modeling steps of NEGM(1,1) are as follows: Step 1: Generate a first-order cumulative reduction sequence and a background value sequence from the original signal, and establish the definition model and whitening model of NEGM (1,1); Let x (0) is the original sequence, and x (0) =(x (0) (1),x (0) (2),…,x (0) (n)), x (-1) For x (0) A first-order decreasing sequence, x (-1) =(x (-1) (1),x (-1) (2),…,x (-1) (n)), and x (-1) (k) = x (0) (k)-x (0) (k-1), called x (-1) (k)+az (0) (k)=b (6) For the non-homogeneous white exponential law grey model, the first-order differential equation It is called the whitening model of NEGM (1,1); where a and b are the development coefficient and the gray action, respectively, and z (0) (k) is x (0) The background value of Step 2: Solve the whitened differential equation to obtain the continuous response function of NEGM(1,1), find the initial conditions and discretize the time variable to obtain the discrete response function with unknown parameters; The discrete response function of the NEGM(1,1) model is Where k = 1, 2, ..., n, n + 1, ..., n + m, n is the number of data in the original sequence, and m is the number of forecast data; when 1 ≤ k ≤ n, the obtained is the fitting value of the original sequence; when n+1≤k≤m, the obtained is the predicted value of the original series; Step 3: Reconstruct the background value using the linear combination of the adjacent values, and calculate the weight coefficient of the background value by the equation system consisting of the original background value, the definition model of NEGM (1,1) and the response function; The background value z (0) (k) is set as a linear combination of a pair of adjacent sequences, that is, z (0) (k)=ωx (0) (k-1)+(1-ω)x (0) (k) (12) Among them, the coefficient to be determined is ω∈[0,1]; make but and The expressions are simplified to Use x (0) The true value replaces the fitted value, that is Substituting equation (13) into equation (8) we can obtain Substituting equation (14) into equation (12), we can obtain z (0) (k)=ω[Ee -a(k-2) +F]+(1-ω)[Ee -a(k-1) +F] (16) Combining equations (15) and (16), we get After arranging and simplifying formula (17), we can obtain By finding the development coefficient a, we can get the background value; Step 4: Use the least square method to estimate the development coefficient and gray action, and obtain the fitting and prediction model of NEGM (1,1); The following equations are obtained by defining NEGM(1,1) and constructing the background value: Solving the system of equations yields make Formula (18) can be simplified as x (0) (k)=λ1x (0) (k-1)+λ2 (20) Where k = 2, 3, ..., n, let Using the least squares parameter method to estimate the parameters in equation (20), we get The estimated values of the parameters and Substituting into equation (19), we get the estimated values of parameters a and b:
2. The BDS-3 real-time clock difference prediction modeling method based on the grey model according to claim 1 is characterized in that: In the step three, the selection strategy of the grey model is to independently select a suitable grey forecasting model according to the size of the homogeneous order ratio standard deviation or the non-homogeneous order ratio standard deviation.
3. The BDS-3 real-time clock difference prediction modeling method based on the grey model according to claim 1 is characterized in that: In the step 4, the ultra-fast observation clock difference product of the 927th Beidou cycle of the BDS-3 system provided by iGMAS is used to conduct experiments to verify the feasibility and effectiveness of the satellite clock difference adaptive prediction strategy and its superiority compared with GM (1,1), NEGM (1,1) and ISU-P. In the specific forecasting process, the model constructed using the clock difference data of the previous day is used to predict the clock differences of 1h, 6h, 12h and 24h on the next day respectively. The root mean square error (RMS) between the predicted clock difference and the corresponding ultra-fast observation clock difference in the forecast time period is used to evaluate the accuracy of the clock difference forecast, and the confidence interval width of the error mean is used to evaluate the stability of the clock difference forecast.
4. A BDS-3 real-time clock difference prediction modeling system based on a grey model that implements the BDS-3 real-time clock difference prediction modeling method based on a grey model as claimed in any one of claims 1 to 3, characterized in that: The BDS-3 real-time clock difference prediction modeling system based on the grey model includes: Analysis module, used to analyze the error generation mechanism of grey model; Modeling module, used to construct the first-order univariate non-homogeneous white exponential law grey model; Strategy selection module, used to select the strategy of the forecast model; The test module is used to conduct forecasting tests.
5. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the BDS-3 real-time clock difference prediction modeling method based on the grey model as described in any one of claims 1 to 3.
6. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the BDS-3 real-time clock difference prediction modeling method based on the grey model as described in any one of claims 1 to 3.
