Method, system and device for generating a flexible time scale oriented to changes in a group of atomic clocks

Abstract

The application relates to an elastic time scale generation method, system and device for atomic clock group change. The method comprises the following steps: constructing an elastic factor graph model for time scale generation, generating two consistent time scales of a master and a backup, accumulating the time difference change of the master and the backup time scales by using atomic clock group change, compensating the time difference and the frequency deviation jump value of the time scale under the clock group change, generating an elastic time scale, ensuring the continuity and stability of the time scale, and improving the monitoring performance of the atomic clock. Meanwhile, by using the plug-and-play architecture of the factor graph, factor nodes can be dynamically added or removed for model updating when the clock group changes, and the overall elastic factor graph model does not need to be reconstructed, so that the generation of the elastic time scale has the advantages of flexibility and scalability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of time frequency and signal processing, in particular to an elastic time scale generation method, system and device for atomic clock group change. BACKGROUND

[0002] Time scale algorithm plays an important role in time keeping laboratories and Global Navigation Satellite System (GNSS). The purpose of time scale algorithm is to integrate all atomic clocks in a clock group to generate a new free paper time with higher reliability and frequency stability. Both a single clock and a paper time can be regarded as a time scale; a time scale without being governed is called a free time scale. Therefore, the purpose of time scale algorithm is to “generate” a new paper time scale, which is generally represented by the time difference between the new paper time scale and a single clock. The time scale algorithm usually does not consider the frequency accuracy of the time scale, and leaves the task of improving the frequency accuracy to the time scale governing algorithm. For a time keeping laboratory and time frequency system (hereinafter referred to as time frequency system), the free paper time is usually denoted as TA(k) (where k is the laboratory code), which can be used to monitor the performance of atomic clocks.

[0003] The existing time scale algorithm can be divided into two categories: 1) one category is weighted average algorithm, mainly including: the weighted average algorithm adopted by Bureau International des Poids et Measures (BIPM) (referred to as: ALGOS algorithm), the weighted average algorithm adopted by National Institute of Standards and Technology (NIST) (referred to as: AT1 algorithm); 2) one category is Kalman filter algorithm, mainly including: the standard Jones-Tyron type Kalman filter algorithm used by NIST in the 1980s, and a series of Kalman filter algorithms improved on this basis, especially the single state variable Kalman filter time scale generation algorithm. Since the time scales generated by AT1 algorithm and Kalman filter algorithm are real-time, they are more suitable for GNSS.

[0004] Compared with other algorithms, Kalman filter algorithm can better filter out atomic clock noise, and perform clock state estimation and fault detection, and performs better in systems such as GNSS that provide real-time time service. However, it still has certain limitations: the number of atomic clocks in the clock group changes, which leads to the time scale that must be recalculated, the time difference and frequency difference of the time scale jump, the generated time scale is unstable and discontinuous, which leads to the deterioration or failure of the monitoring performance of the atomic clock. SUMMARY

[0005] Therefore, it is necessary to provide a flexible time scale generation method, system and device for atomic clock group changes in view of the above technical problems.

[0006] A flexible time scale generation method for atomic clock group changes, the method comprising:

[0007] Step 1, time scale parameter initialization: constructing an elastic factor graph model for time scale generation and initializing a main time scale and a backup time scale with consistent states;

[0008] Step 2, clock group operation: automatically or manually performing clock group operation;

[0009] Step 3, clock group time difference update: updating the clock group time difference based on the clock group change after the clock group operation;

[0010] Step 4, time scale model update: if there is no change in the clock group, updating the elastic factor graph model according to the current clock group time difference and clock group configuration; if there is a change in the clock group, updating the clock group configuration of the backup time scale according to the clock group operation result, calculating and accumulating the time difference between the main time scale and the backup time scale, and determining whether the clock group change duration meets the preset accumulated duration; if not, updating the elastic factor graph model according to the current time difference between the main time scale and the backup time scale and the clock group configuration; if so, first estimating and compensating the time difference and frequency deviation of the time scale based on the accumulated time difference result, then updating the clock group configuration of the main time scale according to the clock group operation result, initializing the backup time scale using the main time scale updated by the clock group configuration, maintaining the consistency of the main time scale and the backup time scale, obtaining the optimized adjustment time difference result between the main time scale and the backup time scale, and updating the elastic factor graph model in combination with the current clock group configuration;

[0011] Step 5, time scale calculation: generating the updated main time scale and backup time scale by solving the updated elastic factor graph model, calculating the time difference of the updated main time scale and backup time scale, and outputting the updated main time scale as the flexible time scale for atomic clock group changes.

[0012] In one embodiment, constructing an elastic factor graph model for time scale generation and initializing a main time scale and a backup time scale with consistent states comprises:

[0013] The state equation and the observation equation of the single-state variable Kalman filter time scale algorithm are constructed and represented as:

[0014] ;

[0015] wherein, is the time difference of the clock group at time t, the superscript T represents transposition, k T is the time difference of the clock group at time t, the superscript T represents transposition,​ Is the first in the Zhong group i The time difference of the atomic clocks and , is the number of atomic clocks; is the state transition matrix; is the process noise of the clock group, Is the first in the Zhong group i Process noise of atomic clocks; It's the Zhong group k The observed time difference between moments; is the observation matrix; is the observation noise, It is i The observation noise of each group of time difference is independent of each other and ;

[0016] State transition matrix , process noise covariance matrix , observation matrix and the observation noise covariance matrix They are represented as follows:

[0017] ;

[0018] in, is the unit matrix; i Process state noise variance of an atomic clock , Indicates the i The Allan variance of an atomic clock, is the virtual Kalman sampling interval; for The noise covariance of

[0019] By converting the state equation and observation equation of the single-state variable Kalman filter time scaling algorithm into state factors and observation factors in the factor graph, converting the state variables into variable nodes, and converting the covariance of the estimation error into a priori factors, a time scale generation elastic factor graph model is constructed, which can be expressed as:

[0020] ;

[0021] The elastic factor graph model is composed of variable nodes, state factors, observation factors and prior factors, the variable nodes correspond to state variables, the state factors correspond to state transition constraints connecting adjacent state variables, the observation factors correspond to Kalman sensor observation constraints connecting state variables and observation values, and the prior factors correspond to covariance of estimation errors in the Kalman filter; when the clock group changes, the elastic factor graph model is updated by adding or deleting factors corresponding to the atomic clock;

[0022] The time scale initial parameters are set, including initializing the atomic clock group configuration, setting the initial phase and accuracy of the time scale, the initial time difference of each atomic clock, the covariance of the estimation error and setting the cumulative duration of the clock group change;

[0023] Based on the elastic factor graph model and the time scale initial parameters, the state-consistent main time scale and backup time scale are initialized and generated.

[0024] In one embodiment, the clock group time difference is updated based on the clock group change after the clock group operation, including:

[0025] Based on the clock group operation result, the clock group change is judged, if there is no change in the clock group, step 4 is directly entered; if the atomic clock is added, the time difference of the added atomic clock is added in the clock group time difference, and then step 4 is entered; if the atomic clock is removed, it is judged whether the clock group change duration meets the preset cumulative duration, if not, it is considered that the removal operation is not completed, the time difference of the atomic clock to be removed is predicted based on the state estimation result of the atomic clock, and then the predicted time difference is added in the clock group time difference, and step 4 is entered; otherwise, it is considered that the removal operation is completed and step 4 is entered.

[0026] In one embodiment, the cumulative duration is determined by the accuracy of the atomic clock time difference prediction and the accuracy of the frequency deviation estimated by the least square estimation based on the main and backup time scales, and is expressed as:

[0027] ;

[0028] Wherein, is the weighted result, is the weighted coefficient of the cumulative duration, is the weight of the atomic clock time difference prediction accuracy, is the weight of the frequency deviation estimation accuracy; , , satisfy the following constraints respectively:

[0029] ;

[0030] Wherein, is the cumulative duration, is a constant threshold of 1.345, is the prediction accuracy of the atomic clock time difference, is a normalization factor of is the accuracy of the frequency bias estimation, is a normalization factor of is the minimum, at this time the corresponding is a preset accumulation time length.

[0031] In one embodiment, the clock group configuration of the backup time scale is updated according to the clock group operation result, including:

[0032] The state variable and the estimation error covariance matrix of the backup time scale before the clock group change are obtained;

[0033] If the atomic clock i is added, the time difference of the atomic clock i predicted based on the state estimation result of the atomic clock is added in the state variable, a new initial state variable is constructed, and the mean of the diagonal elements of the estimation error covariance matrix is added in the estimation error covariance matrix as a new initial estimation error covariance matrix;

[0034] If the atomic clock i is removed, the value of the atomic clock i is removed in the state variable, a new initial state variable is constructed, and the value corresponding to the atomic clock i is removed in the estimation error covariance matrix, and a new initial estimation error covariance matrix is constructed.

[0035] In one embodiment, the time difference and the frequency bias of the time scale are estimated and compensated based on the accumulated time difference result, including:

[0036] The least square estimation is performed based on the accumulated time difference result, the jump values of the time difference and the frequency bias between the primary and backup time scales are obtained, and the jump values are compensated to the initial phase and the accuracy of the time scale, so as to realize the compensation of the time difference and the frequency bias of the time scale.

[0037] In one embodiment, the clock group configuration of the primary time scale is updated according to the clock group operation result, including:

[0038] The state variable and the estimation error covariance matrix of the primary time scale before the clock group change are obtained;

[0039] If the atomic clock i is added, the atomic clock estimated by the Kalman filter through ithe time difference of the atomic clock, a new initial state variable is constructed, and the mean of the diagonal elements of the estimated error covariance matrix is added to the estimated error covariance matrix as a new initial estimated error covariance matrix;

[0040] If the atomic clock is i eliminated, the value of the atomic clock in the state variable is eliminated, a new initial state variable is constructed, and the value of the atomic clock corresponding to the atomic clock i is eliminated from the estimated error covariance matrix, a new initial estimated error covariance matrix is constructed; i

[0041] wherein the Kalman filter estimates the time difference of the atomic clock through a group of observations, i the time difference of the atomic clock, is expressed as , and the iterative calculation step is expressed as:

[0042] ;

[0043] wherein, is the estimated value of the time difference of the clock group at time k -1, is the predicted value of the time difference of the clock group at time k -1 to time k , and is the state transition matrix; is the estimated error matrix of the clock group time difference at time k -1, is the predicted error matrix of the clock group time difference at time k ; is the process noise covariance matrix; is the Kalman filter gain at time k , is the observation matrix, is the observation noise covariance matrix, is the observed time difference of the clock group at time , is the identity matrix, and the superscript k represents transposition. T In one embodiment, the above method further comprises:

[0044] Step 6, atomic clock state estimation: using a 2-state Kalman filter or a 3-state Kalman filter to estimate the state of the atomic clock, and predicting the time difference of the atomic clock based on the state estimation result of the atomic clock.

[0045] Step 6, atomic clock state estimation: using a 2-state Kalman filter or a 3-state Kalman filter to estimate the state of the atomic clock, and predicting the time difference of the atomic clock based on the state estimation result of the atomic clock.

[0046] A flexible time scale generation system for changes in a group of atomic clocks, the system comprising:

[0047] ​​A time scale parameter initialization module is configured to construct an elastic factor graph model generated by a time scale and initialize a primary time scale and a backup time scale in a consistent state.

[0048] A clock group operation module is configured to automatically or manually perform clock group operation.

[0049] A clock group time difference updating module is configured to update the clock group time difference based on the change of the clock group after the clock group operation.

[0050] A time scale model updating module is configured to, if there is no change in the clock group, update the elastic factor graph model according to the current clock group time difference and clock group configuration; if there is a change in the clock group, update the clock group configuration of the backup time scale according to the clock group operation result, calculate and accumulate the time difference between the primary time scale and the backup time scale, and determine whether the duration of the change in the clock group meets a preset accumulated duration; if not, update the elastic factor graph model according to the current time difference between the primary time scale and the backup time scale and the clock group configuration; if yes, first estimate and compensate the time difference and frequency deviation of the time scale based on the accumulated time difference result, then update the clock group configuration of the primary time scale according to the clock group operation result, initialize the backup time scale based on the primary time scale updated by the clock group configuration, maintain the consistency of the primary time scale and the backup time scale, obtain the optimized time difference result between the primary time scale and the backup time scale, and update the elastic factor graph model in combination with the current clock group configuration.

[0051] A time scale calculation module is configured to generate the updated primary time scale and backup time scale and calculate the time difference between the updated primary time scale and backup time scale by solving the updated elastic factor graph model, and output the updated primary time scale as the elastic time scale facing the change of the atomic clock group.

[0052] A computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0053] Step 1, time scale parameter initialization: constructing an elastic factor graph model generated by a time scale and initializing a primary time scale and a backup time scale in a consistent state.

[0054] Step 2, clock group operation: automatically or manually performing clock group operation.

[0055] Step 3, clock group time difference updating: updating the clock group time difference based on the change of the clock group after the clock group operation.

[0056] Step 4, time scale model updating: if the clock group does not change, the elastic factor graph model is updated according to the current clock group time difference and clock group configuration; if the clock group changes, the clock group configuration of the backup time scale is updated according to the clock group operation result, and the time difference between the main time scale and the backup time scale is calculated and accumulated, and it is judged whether the clock group change duration meets the preset accumulated duration; if not, the elastic factor graph model is updated according to the current time difference between the main time scale and the backup time scale and the clock group configuration; if yes, the time difference and the frequency deviation of the time scale are estimated and compensated based on the accumulated time difference result, then the clock group configuration of the main time scale is updated according to the clock group operation result, and the initialization of the backup time scale is performed in the main time scale updated by the clock group configuration, the consistency of the main time scale and the backup time scale is maintained, and the optimized adjustment time difference result between the main time scale and the backup time scale is obtained, and the elastic factor graph model is updated by combining the current clock group configuration;

[0057] Step 5, time scale calculation: the updated main time scale and backup time scale are generated by solving the updated elastic factor graph model, and the time difference between the updated main time scale and backup time scale is calculated, and the updated main time scale is output as the elastic time scale facing the change of the atomic clock group.

[0058] The above-mentioned elastic time scale generation method, system and device facing the change of the atomic clock group are based on the single-state variable Kalman filter time scale algorithm, construct the elastic factor graph model of time scale generation, generate two consistent main time scale and backup time scale, and compensate the time difference and frequency deviation jump value of the time scale under the clock group change by using the time difference change accumulation of the main time scale and the backup time scale, generate the elastic time scale, ensure the continuity and stability of the time scale, and improve the monitoring performance of the atomic clock. At the same time, by using the plug-and-play architecture of the factor graph, the factor node can be dynamically added or removed for model updating when the clock group changes, without the need to reconstruct the overall elastic factor graph model, so that the generation of the elastic time scale has the advantages of flexibility and scalability. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 It is a flowchart of the elastic time scale generation method facing the change of the atomic clock group in an embodiment;

[0060] Figure 2 It is a factor graph model diagram of the Kalman filter in an embodiment;

[0061] Figure 3 It is a schematic diagram of the elastic factor graph model of time scale generation in an embodiment;

[0062] Figure 4 It is a schematic diagram of the frequency stability of the atomic clock in the atomic clock group in an embodiment;

[0063] Figure 5 Figure 2 is a schematic diagram of the time scale variation under the clock removal experiment in one embodiment; wherein, Figure 5 (a) is the overall change of the time difference; Figure 5 (b) is the local change of the time difference; Figure 5 (c) is the overall change of the frequency deviation; Figure 5 (d) is the local change of the frequency deviation;

[0064] Figure 6 Figure 3 is a schematic diagram of the time difference results of the main time scale and the backup time scale under the clock removal experiment in one embodiment; wherein, Figure 6 (a) is the overall time difference result; Figure 6 (b) is the local time difference result before and after the clock group change;

[0065] Figure 7 Figure 4 is a schematic diagram of the frequency stability of the reference time scale, the main time scale and the backup time scale under the clock removal experiment in one embodiment;

[0066] Figure 8 Figure 5 is a schematic diagram of the time scale variation under the clock addition experiment in one embodiment; wherein, Figure 8 (a) is the overall change of the time difference; Figure 8 (b) is the local change of the time difference; Figure 8 (c) is the overall change of the frequency deviation; Figure 8 (d) is the local change of the frequency deviation;

[0067] Figure 9 Figure 6 is a schematic diagram of the time difference results of the main time scale and the backup time scale under the clock addition experiment in one embodiment; wherein, Figure 9 (a) is the overall time difference result; Figure 9 (b) is the local time difference result before and after the clock group change;

[0068] Figure 10 Figure 7 is a schematic diagram of the frequency stability of the reference time scale, the main time scale and the backup time scale under the clock addition experiment in one embodiment;

[0069] Figure 11 Figure 8 is a schematic diagram of the time scale variation under the mixed clock operation experiment; wherein, Figure 11 (a) is the overall change of the time difference; Figure 11 (b) is the local change of the time difference; Figure 11 (c) is the overall change of the frequency deviation; Figure 11 (d) is the local change of the frequency deviation;

[0070] Figure 12 Figure 9 is a schematic diagram of the time difference results of the main time scale and the backup time scale under the mixed clock operation experiment; wherein, Figure 12 (a) is the overall time difference result; Figure 12(b) is the local time difference result before and after the clock group change;

[0071] Figure 13 The frequency stability diagram of the reference time scale, the main time scale and the backup time scale for the mixed clock operation experiment;

[0072] Figure 14 The internal structure diagram of the computer device in an embodiment. DETAILED DESCRIPTION

[0073] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0074] In one embodiment, as shown in Figure 1 A flexible time scale generation method for atomic clock group change is provided, comprising the following steps:

[0075] Step 1, time scale parameter initialization: constructing a flexible factor graph model for time scale generation and initializing the generated main time scale and backup time scale to be consistent.

[0076] Step 2, clock group operation: automatically or manually performing clock group operation through the time-frequency system.

[0077] Step 3, clock group time difference update: updating the clock group time difference based on the clock group change after the clock group operation. Specifically, based on the clock group operation result, the clock group change is determined. If there is no change in the clock group, go directly to step 4; if an atomic clock is added, add the time difference of the new atomic clock in the clock group time difference, and then go to step 4; if an atomic clock is removed, determine whether the clock group change duration meets the preset cumulative duration. If not, it is considered that the removal operation is not completed, the time difference of the atomic clock to be removed is predicted based on the state estimation result, and the predicted time difference is added in the clock group time difference, and then go to step 4; otherwise, it is considered that the removal operation is completed and go to step 4.

[0078] Step 4, time scale model update: If the clock group has not changed, the elastic factor graph model is updated according to the current clock group time difference and clock group configuration; if the clock group has changed, the clock group configuration of the backup time scale is updated according to the clock group operation result, and the time difference between the main and backup time scales is calculated and accumulated, and it is determined whether the clock group change duration meets the preset cumulative duration; if not, the elastic factor graph model is updated according to the current time difference between the main and backup time scales and the clock group configuration; if satisfied, the time difference and frequency deviation of the time scale are first estimated and compensated based on the accumulated time difference results, and then the clock group configuration of the main time scale is updated according to the clock group operation results, and the backup time scale is initialized on the main time scale after the clock group configuration is updated. After maintaining the consistency of the main and backup time scales, the optimized time difference result between the main and backup time scales is obtained and combined with the current clock group configuration to update the elastic factor graph model.

[0079] Step 5, time scale calculation: By solving the updated elastic factor graph model, the updated primary time scale and backup time scale are generated and the time difference between the updated primary and backup time scales is calculated. The updated primary time scale is output as the elastic time scale for the changes in the atomic clock group.

[0080] In one embodiment, step 1 includes:

[0081] The state equation and observation equation for constructing the single state variable Kalman filter time scaling algorithm are expressed as:

[0082] ;

[0083] in, It's the Zhong group k Time difference, superscript T represents transpose, Is the first in the Zhong group i The time difference of the atomic clocks and , is the number of atomic clocks; is the state transition matrix; is the process noise of the clock group, Is the first in the Zhong group i Process noise of atomic clocks; It's the Zhong group k The observed time difference between moments; is the observation matrix; is the observation noise, It is i The observation noise of each group of time difference is independent of each other and ;State transition matrix , process noise covariance matrix , observation matrix and observation noise covariance matrix are represented as follows, respectively:

[0084]

[0085] wherein, is an identity matrix; the first i process state noise variance of the mth atomic clock , denotes the Allan variance of the mth atomic clock, i is a virtual Kalman sampling interval; is the noise covariance of

[0086] As shown in Figure 2 , there is an equivalent conversion relationship between the factor graph and the Kalman filter. The state equation and the observation equation of the Kalman filter can be converted into the state factor and the observation factor in the factor graph. The covariance of the estimation error can be converted into the prior factor. The elastic factor graph model generated in the time scale is constructed as shown in Figure 3 , and is represented as:

[0087]

[0088] wherein, the elastic factor graph model is composed of variable nodes, state factors, observation factors and prior factors. The variable nodes correspond to state variables. The state factors correspond to the state transition constraints connecting adjacent state variables. The observation factors correspond to the Kalman sensor observation constraints connecting the state variables and the observation values. The prior factors correspond to the covariance of the estimation error in the Kalman filter. When the clock group changes, the elastic factor graph model is updated by adding or deleting the factors corresponding to the atomic clocks.

[0089] Further, the factor graph model of the Kalman filter shown in Figure 2 and the elastic factor graph model shown in Figure 3 can be solved. In the factor graph model of the Kalman filter, the error variables are constructed as follows:

[0090]

[0091] and the least square objective function is constructed as: . Wherein, , state error and observation error, respectively.

[0092] In the elastic factor graph model, the error variables are constructed as follows:

[0093] ​​​​​ ;

[0094] And the least square objective function is constructed as: . Wherein, , respectively represent the state error and observation error of the first atomic clock. i

[0095] By solving the above two least square objective functions, it is known that the solution is consistent, that is, it can be proved that the factor graph model of Kalman filter shown in Figure 2 and the elastic factor graph model shown in Figure 3 are equivalent.

[0096] It should be understood that the factor graph is a kind of binary graph model, which decomposes a complex global function into the product of local functions. The factor graph contains variable nodes (random variables) and factor nodes (local functions), and the edges represent the dependence between variables and factors. The factor graph can be considered as a generalized form of Kalman filter. The recursive estimation process (prediction + update) of Kalman filter can be realized by Gaussian message passing of factor graph. When the process noise and observation noise are both zero mean, the state mean estimation of factor graph is consistent with the posteriori estimation result of Kalman filter. Kalman filter usually assumes a linear Gaussian system, while factor graph can naturally represent a nonlinear, non-Gaussian system and improve robustness and adaptability by iterative optimization of complex model. Factor graph also has flexibility and scalability, which uses plug-and-play architecture to dynamically add or remove factor nodes without reconstructing the overall model.

[0097] Therefore, based on the single state variable Kalman filter time scale algorithm, when the time scale generated elastic factor graph model constructed is updated, only the clock group time difference information and clock group configuration information input in step 3 are used to obtain the atomic clock information and its time difference, and the observation factor (time difference) and state factor of different clocks in the elastic factor graph model shown in Figure 3 are used to update the prior factor in the graph. In short, the clock group configuration and time difference information corresponding to the operation results of different clock groups are different. For model updating, only the input changes, which has no effect on the model updating operation. It is a new idea to improve the flexibility of adding or removing atomic clocks.

[0098] Further, based on the elastic factor graph model and the preset time scale initial parameters, the main time scale and the backup time scale with consistent states are initialized, which can be denoted as TA (main) and TA (backup). Wherein, the time scale initial parameters include initializing the atomic clock group configuration, setting the initial phase and accuracy of the time scale, the initial time difference of each atomic clock, the covariance of the estimation error, and setting the cumulative length of the clock group change. ​

[0099] In one embodiment, the preset accumulated time length depends on two aspects according to the method provided in the present application: one aspect is the accuracy of the atomic clock time difference prediction, and the other aspect is the accuracy of the frequency bias estimated by the least square method of the TA (main) and TA (backup) time difference sequence. For the atomic clock time difference prediction, the error is the deviation of the predicted atomic clock time difference from the actual measurement result, and the prediction accuracy is the root mean square (RMS) of the error sequence. For the accuracy of the frequency bias estimation, it is the difference between the estimation result of the previous epoch and the estimation result of the next epoch. The smaller the difference is, the closer to the theoretical true value it is, and the higher the accuracy is. Therefore, the accumulated time length is determined by the accuracy of the atomic clock time difference prediction and the accuracy of the frequency bias estimated by the least square method of the main and backup time scales, and is represented as:

[0100] ;

[0101] wherein, is the weighted result, is the weighted coefficient of the accumulated time length, is the weight of the atomic clock time difference prediction accuracy, is the weight of the frequency bias estimation accuracy, each weight can be set by different users according to the importance of different parameters, and , , satisfy the following constraints, respectively:

[0102] ;

[0103] wherein, is the accumulated time length, is a constant threshold of 1.345, is the prediction accuracy of the atomic clock time difference, is a normalization factor of , generally 0.1 ns; is the accuracy of the frequency bias estimation, is a normalization factor of , generally 1x10 −16 s / s.

[0104] It can be understood that the weighted coefficient is proportional to the accumulated time length , in addition, the starting value of the weighted coefficient cannot be too small, and cannot differ from other values by an order of magnitude. Therefore, the standard S curve is modified to obtain the weighted coefficient of the accumulated time length . and is modified from the inverse function of Huber weight function. and is set by user according to the state of time-frequency system and empirical statistical data. In summary, R value is calculated according to different accumulated time length , when R is the minimum, the corresponding is the most reasonable value calculated, that is, the preset accumulated time length.

[0105] In one embodiment, the clock group configuration of the backup time scale is updated according to the clock group operation result, including:

[0106] The state variable and the estimated error covariance matrix of the backup time scale before the clock group change are obtained;

[0107] If it is an atomic clock i , the time difference of the atomic clock i predicted based on the state estimation result of the atomic clock is added in the state variable, and the mean value of the diagonal elements of the estimated error covariance matrix is added in the estimated error covariance matrix as the new initial estimated error covariance matrix, so as to construct a new initial state variable.

[0108] If it is an atomic clock i , the value of the atomic clock i is removed in the state variable, and the corresponding value of the atomic clock i is removed in the estimated error covariance matrix, so as to construct a new initial estimated error covariance matrix.

[0109] In one embodiment, the time difference and the frequency deviation of the time scale are estimated and compensated based on the accumulated time difference result, including:

[0110] The least square estimation is performed based on the accumulated time difference result, the jump value of the time difference and the frequency deviation between the primary time scale and the backup time scale is obtained, and the jump value is compensated to the initial phase and the accuracy of the time scale, so as to realize the compensation of the time difference and the frequency deviation of the time scale.

[0111] In one embodiment, the clock group configuration of the primary time scale is updated according to the clock group operation result, including:

[0112] The state variable and the estimated error covariance matrix of the primary time scale before the clock group change are obtained;

[0113] If it is an atomic clock i , the atomic clock estimated by the Kalman filter through ithe time difference of the atomic clock, the new initial state variable is constructed, and the mean of the diagonal elements of the estimated error covariance matrix is added to the estimated error covariance matrix as a new initial estimated error covariance matrix;

[0114] If the atomic clock is i eliminated, the value of the atomic clock in the state variable is eliminated, the new initial state variable is constructed, and the value of the atomic clock corresponding to the time difference in the estimated error covariance matrix is eliminated, the new initial estimated error covariance matrix is constructed; i i

[0115] wherein the Kalman filter estimates the time difference of the atomic clock through the time difference of the atomic clock i is expressed as The iterative calculation step is expressed as:

[0116] ;

[0117] wherein, is the estimated value of the time difference of the clock group at time k -1, is the predicted value of the time difference of the clock group at time k -1 to time k , and is a state transition matrix; is the estimated error matrix of the clock group time difference at time k -1, is the predicted error matrix of the clock group time difference at time k ; is a process noise covariance matrix; is the Kalman filter gain at time k , and is an observation matrix, is an observation noise covariance matrix, is the observed time difference of the clock group at time k , and is an identity matrix, and the superscript T represents transposition.

[0118] Further, after updating the clock group configuration of the main time scale, the parameters of the backup time scale can be updated using the updated main time scale calculation result of the clock group configuration, and the backup time scale is initialized to ensure the consistency of TA (main) and TA (backup).

[0119] In one of the embodiments, the above method further comprises:

[0120] ​​Step 6, atomic clock state estimation. It should be understood that the state estimation of the atomic clock using the Kalman filter and the time difference prediction are existing mature algorithms, and typically, a 2-state Kalman filter or a 3-state Kalman filter is used for atomic clock state estimation, and the atomic clock time difference is predicted based on the state estimation result of the atomic clock.

[0121] wherein the state transition matrix of the 2-state Kalman filter , the process noise covariance matrix , the observation matrix and the observation noise covariance matrix are respectively represented as follows:

[0122] .

[0123] The state transition matrix of the 3-state Kalman filter , the process noise covariance matrix , the observation matrix and the observation noise covariance matrix are respectively represented as follows:

[0124] ;

[0125] wherein is the time sampling interval; is the observation noise standard deviation of the atomic clock i . , and are noise parameters, and the specific values of the noise parameters are generally obtained by the method of Allan variance inversion.

[0126] In summary, the above-mentioned elastic time scale generation method facing atomic clock group changes is based on the single-state variable Kalman filter time scale algorithm, constructs an elastic factor graph model for time scale generation, generates two consistent time scales of the main and backup, and uses the time difference change accumulation results of the main and backup time scales under the atomic clock group change to compensate for the time difference and frequency deviation jump values of the time scale under the clock group change, generate an elastic time scale, ensure the continuity and stability of the time scale, and improve the monitoring performance of the atomic clock. At the same time, using the plug-and-play architecture of the factor graph, the factor nodes can be dynamically added or removed for model updating when the clock group changes, without the need to reconstruct the overall elastic factor graph model, so that the generation of the elastic time scale has the advantages of flexibility and scalability.

[0127] Further, the performance of the method is verified by experiments. Specifically, relevant experimental data are obtained from a stable time-frequency system, which has 6 hydrogen clocks H01-H06 and 3 cesium clocks Cs01-Cs03 in the atomic clock group. The time difference of each atomic clock in the atomic clock group is measured by a high-precision time interval measurement device, with a sampling interval of 1 s, a data length of 45 days, and a time difference between the real-time physical signal and coordinated universal time UTC(k) obtained by fiber bidirectional time transfer. The frequency stability performance of each clock is obtained by the triangular cap method, as shown in Table 1. Figure 4 Figure 4 Among them, H01-H03 and H06 have similar performance, H04 and H05 have similar performance, and are slightly worse than other hydrogen clocks; Cs01-Cs03 have similar performance.

[0128] In order to show the effectiveness of the method, 6 clocks H01-H04, Cs01 and Cs02 are selected to construct the original atomic clock group in the subsequent experiment, and the experiment shown in Table 1 is carried out on the clock group configuration. At the same time, the frequency stability of the generated time scale is evaluated and analyzed by using the obtained time difference of UTC(k).

[0129] Table 1 Experimental setting

[0130]

[0131] 1. Cumulative time length selection experiment: under the parameter setting of a typical time scale calculation interval of 300 s and a virtual sampling interval of 86400 s in the time-frequency system, the results of different cumulative time lengths within 24 hours are calculated as shown in Table 2. It can be found that when the cumulative time length increases, the prediction accuracy of the time difference decreases, and the frequency deviation estimated by TA (main) and TA (backup) decreases. At the same time, when R is the minimum, the cumulative time length is 7 hours. Therefore, the cumulative time length selected in the subsequent experiment is 7 hours. t

[0132] Table 2 Cumulative time length calculation results

[0133]

[0134] In the subsequent experiment, the experimental parameters are shown in Table 3. The cumulative time length in the method is determined according to the cumulative time length selection experiment. The virtual sampling interval and the time scale calculation interval are set according to the construction index and the actual running situation of the time-frequency system.

[0135] Table 3 Experimental parameters

[0136]

[0137] ​​2. Elimination of clock experiment: In this experiment, according to the type of atomic clock and its weight ratio, 6 clock groups are constructed as shown in Table 4. The time scale of the clock group change is taken as the reference time scale, denoted as TA (reference).

[0138] Table 4 Elimination of clock experiment scene setting

[0139]

[0140] As Figure 5 shown, before the clock group changes, the generated elastic time scale is consistent with the reference time scale, and after the clock group changes, the time difference changes. As Figure 5 (a) and Figure 5 (c) show that the trends of scenarios Del 1 and Del 5 are basically the same; scenario Del 4 has a part consistent with scenarios Del 2 and Del 6, and then separates. The frequency deviation of the clock group after the change is better than 1x10 −15 s / s. As Figure 5 (b) and Figure 5 (d) show that the time difference jump in scenario Del 3 can be ignored; scenarios Del 2 and Del 6 have basically the same trend, with small time difference jumps; scenarios Del 2, Del 4 and Del 6 have the same trend, with large time difference jumps; the time difference jump value before and after the clock group change is better than 20 ps, and the frequency deviation jump value is better than 1x10 −17 s / s. According to the experiment set in Table 4, scenario Del 3 only eliminates Cs02, scenarios Del 2 and Del 6 eliminate H04, scenarios Del 1 and Del 5 eliminate H03, and scenario Del 4 eliminates H03 and H04. At the same time, according to the theoretical and experimental data analysis and the set experimental parameters, the weight of the hydrogen clock in the generated time scale is much higher than that of the cesium clock, and the weight of H03 is higher than that of H04. The weight of the hydrogen clock in the generated time scale is much higher than that of the cesium clock, and the weight of H03 is higher than that of H04. Therefore, when the clock group eliminates the atomic clock with low weight, the time scale changes little; when the clock group eliminates the atomic clock with high weight, the time scale changes greatly. When multiple clocks are eliminated, the time difference change depends on the weight ratio of the atomic clock, that is, the performance of the atomic clock.

[0141] The time difference results of TA (main) and TA (backup) under different elimination clock experiment scenarios are shown in Figure 6 . As Figure 6 shown, before the clock group changes, TA (main) and TA (backup) are consistent, and after the clock group changes, the time difference changes. When the clock group changes, the time difference change value of TA (main) and TA (backup) is less than 0.4 ns. When the cumulative time is met, the time difference of TA (main) and TA (backup) is quickly adjusted and maintained at a level better than 0.1 ns.

[0142] The frequency stability results of TA (reference), TA (master) and TA (backup) under the cut-off clock experiment are shown in Figure 7 . Figure 7 In the table, Ref, Master and Backup represent TA (reference), TA (master) and TA (backup) respectively, and their frequency stabilities are basically consistent. Although TA (backup) appeared time difference adjustment operation in the clock group change process, since its adjustment amount is less than 0.3 ns and the adjustment time is short, 7 hours, the influence on long time data (45 days) is small, so the stability performance is shown in Figure 7 .

[0143] Table 5. Add clock experiment scene setting

[0144]

[0145] 3. Add clock experiment: in this experiment, according to the type of atomic clock and its weight ratio, six clock group adding clock scenes shown in Table 5 are constructed. The time scale without clock group change is taken as the reference time scale, denoted as TA (reference).

[0146] For the above experimental scene, the method proposed in the application is used to generate the elastic time scale. The time scale changes generated under different adding clock experimental scenes are shown in Figure 8 . As shown in Figure 8 , before the clock group changes, the generated elastic time scale is consistent with the reference time scale, and after the clock group changes, the time difference and frequency deviation change. As shown in Figure 8 (a) and Figure 8 (c), the trends of scenes Add 2 and Add 6 are basically consistent; the trends of scenes Add 1 and Add 5 are basically consistent; scene Add 4 has a part consistent with the trends of scenes Add 2 and Add 6, and then separates. The time scale change under scene Add 4 is the largest. The frequency deviation after the clock group change under all scenes is better than 3×10 −16 s / s. As shown in Figure 8 (b) and Figure 8 (d), the time difference jump value after the clock group change under all scenes is better than 10 ps, and the frequency deviation jump value is better than 4×10 −18s / s. According to the experiment set in Table 5, the scene Add 3 only adds Cs02, the scenes Add 2 and Add 6 both add H06, the scenes Add 1 and Add 5 both add H05, and the scene Add 4 adds both H05 and H06. At the same time, according to the theoretical and experimental data analysis and the set experimental parameters, the weight of the hydrogen clock in the time scale is much higher than that of the cesium clock, and the weight of H06 is higher than that of H05. Therefore, when the clock group adds an atomic clock with a lower weight, the time scale time difference changes less; when the clock group adds an atomic clock with a higher weight, the time scale changes more. When multiple clocks are added, the size of the time scale change depends on the proportion of the atomic clock weight, that is, the performance of the atomic clock.

[0147] The time difference results of TA (main) and TA (backup) under the clock adding experiment are shown in Figure 9 . Figure 9 In the table, before the clock group changes, TA (main) and TA (backup) are consistent, and after the clock group changes, the time difference changes. After the clock group changes, the time difference of TA (main) and TA (backup) changes, and the change trend of each scene is different, but the change value is less than 0.3 ns. When the cumulative time is met, the time difference of TA (main) and TA (backup) is quickly adjusted and maintained at a level of better than 0.1 ns.

[0148] The frequency stability results of TA (reference), TA (main) and TA (backup) under the clock adding experiment are shown in Figure 10 . Figure 10 In the table, Ref, Master and Backup represent TA (reference), TA (main) and TA (backup) respectively, and their frequency stabilities are basically consistent. Although TA (backup) appears time difference adjustment during the clock group change process, since the adjustment amount is less than 0.3 ns and the adjustment time is short, the adjustment time is 7 hours, the influence is small in the long time data (45 days), and therefore the stability is shown in Figure 10 .

[0149] 4. Mixed clock operation experiment: In this experiment, according to the type and weight proportion of the atomic clock, six clock group adding / eliminating clock mixed operation scenes shown in Table 6 are constructed. The time scale without clock group change is taken as the reference time scale, denoted as TA (reference).

[0150] Table 6 Mixed clock operation experiment scene setting

[0151]

[0152] For the above experimental scenes, the method proposed in the present application is used to generate the elastic time scale. The time scale changes generated under different mixed clock operation experimental scenes are shown in Figure 11 . Figure 11In the middle, the generated elastic time scale is consistent with the reference time scale before the clock group changes, and the time difference and frequency deviation change after the clock group changes. As shown in Figure 11 (a) and Figure 11 (c), the trends of scenarios Mix 1 and Mix 5 are basically consistent; scenario Mix 4 has a part consistent with the trends of scenarios Mix 1 and Mix 5, and then separates. The time difference and frequency deviation change of scenario Mix 6 are the largest. The frequency deviation of the clock group after the change is better than 1.5×10 −15 s / s under all scenarios. As shown in Figure 11 (b) and Figure 11 (d), the time difference jump value after the change of the clock group is better than 20ps, and the frequency deviation jump value is better than 1×10 −17 s / s under all scenarios. According to the experiment set in Table 6, scenario Mix 1 removes and adds a hydrogen clock with equivalent performance, and scenario Mix 2 removes and adds a cesium clock with equivalent performance. Scenarios Mix 1, Mix 3 and Mix 4 remove clock H04 and add atomic clocks with different weights respectively. Scenarios Mix 5 and Mix 6 increase the number of removed and added atomic clocks. At the same time, according to the theoretical and experimental data analysis and the set experimental parameters, the weight of the hydrogen clock in the generated time scale is much higher than that of the cesium clock, and the weights of H03 and H06 are higher than those of H04 and H05. Therefore, when the clock group adds or removes an atomic clock with a low weight, the time scale time difference changes little; when the clock group adds or removes an atomic clock with a high weight, the time scale time difference changes a lot. When multiple clocks are added or removed, the size of the time scale change depends on the weight proportion of the atomic clock, that is, the performance of the atomic clock.

[0153] The time difference results of TA (main) and TA (backup) under the experimental scenarios of different mixed clock operations are shown in Figure 12 . Figure 12 In the middle, TA (main) and TA (backup) are consistent before the clock group changes, and the time difference changes after the clock group changes. When the clock group changes, the time difference of TA (main) and TA (backup) changes, and the change trend of each scenario is different, but the change value is less than 0.3ns. When the cumulative time length is met, the time difference of TA (main) and TA (backup) is quickly adjusted and maintained at a level better than 0.1ns.

[0154] The frequency stability results of TA (reference), TA (main) and TA (backup) under the mixed clock operation experiment are shown in Figure 13 . Figure 13In the experiment, Ref, Master and Backup represent TA (reference), TA (main) and TA (backup) respectively, and their frequency stabilities are basically consistent. Although TA (backup) appears time difference adjustment operation in the clock group change process, since its adjustment amount is less than 0.3 ns and the adjustment time is short (7 hours), the influence on long-time data (45 days) is small, so the stability is as shown in Figure 13

[0155] In summary, the elastic time scale generated by the application changes the time difference and frequency deviation under the clock group change, which includes removing and adding atomic clocks. The performance of the removed and added atomic clocks is proportional to the influence on the scale of time: the better the performance of the atomic clock, the greater the change of the time scale, the worse the performance, and the smaller the change of the time scale. The jump value of the elastic time scale time difference is better than 20 ps, and the jump value of the frequency deviation is better than 1×10 −17 s / s, and the frequency deviation is better than 1.5×10 −15 s / s. At the same time, the time difference between the generated TA (backup) and TA (main) in the elastic scale generation method is better than 0.3 ns, and the time difference between the generated TA (main) and TA (backup) is maintained at a level better than 0.1 ns after the clock group change operation is completed, and their frequency stabilities are consistent without deterioration.

[0156] In one embodiment, an elastic time scale generation system for atomic clock group change is provided, comprising:

[0157] A time scale parameter initialization module for constructing an elastic factor graph model for time scale generation and initializing a main time scale and a backup time scale with consistent generation states;

[0158] A clock group operation module for automatically or manually performing clock group operation;

[0159] A clock group time difference updating module for updating the clock group time difference based on the clock group change after the clock group operation;

[0160] ​The time scale model updating module is configured to update the elastic factor graph model according to the current time difference between the clock groups and the clock group configuration if the clock groups do not change, and to update the clock group configuration of the backup time scale according to the clock group operation result if the clock groups change, while calculating and accumulating the time difference between the primary time scale and the backup time scale, and judging whether the duration of the clock group change meets a preset accumulated duration; if not, updating the elastic factor graph model according to the current time difference between the primary time scale and the backup time scale and the clock group configuration; if yes, first estimating and compensating the time difference and the frequency deviation of the time scale based on the accumulated time difference result, then updating the clock group configuration of the primary time scale according to the clock group operation result, and initializing the backup time scale by using the primary time scale updated by the clock group configuration, maintaining the consistency of the primary time scale and the backup time scale, obtaining the optimized adjustment result of the time difference between the primary time scale and the backup time scale, and updating the elastic factor graph model in combination with the current clock group configuration.

[0161] The time scale calculation module is configured to generate the updated primary time scale and the backup time scale by solving the updated elastic factor graph model, to calculate the time difference between the updated primary time scale and the backup time scale, and to output the updated primary time scale as the elastic time scale facing the change of the atomic clock group.

[0162] The specific limitation of the elastic time scale generation system facing the change of the atomic clock group can refer to the limitation of the elastic time scale generation method facing the change of the atomic clock group in the foregoing, and will not be repeated here. Each module in the elastic time scale generation system facing the change of the atomic clock group can be realized by software, hardware, or a combination thereof, in whole or in part. Each module can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operation corresponding to each module.

[0163] In one embodiment, a computer device is provided, which can be a terminal, and the internal structure diagram thereof can be as shown in Figure 14As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to implement an elastic time scale generation method for atomic clock group changes. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0164] Those skilled in the art can understand that, Figure 14 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0165] In one embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the following steps:

[0166] Step 1, time scale parameter initialization: constructing an elastic factor graph model of time scale generation and initializing the main time scale and the backup time scale consistent with the generation state;

[0167] Step 2, clock group operation: automatically or manually performing clock group operation;

[0168] Step 3, clock group time difference update: updating the clock group time difference based on the clock group change after the clock group operation;

[0169] Step 4, time scale model updating: if the clock group does not change, the elastic factor graph model is updated according to the current clock group time difference and clock group configuration; if the clock group changes, the clock group configuration of the backup time scale is updated according to the clock group operation result, and the time difference between the primary time scale and the backup time scale is calculated and accumulated, and it is judged whether the clock group change duration meets the preset accumulated duration; if not, the elastic factor graph model is updated according to the current time difference between the primary time scale and the backup time scale and the clock group configuration; if yes, the time difference and the frequency deviation of the time scale are estimated and compensated based on the accumulated time difference result, then the clock group configuration of the primary time scale is updated according to the clock group operation result, and the initialization of the backup time scale is performed in the primary time scale after the clock group configuration is updated, the consistency of the primary time scale and the backup time scale is maintained, the optimized adjustment time difference result between the primary time scale and the backup time scale is obtained, and the elastic factor graph model is updated in combination with the current clock group configuration;

[0170] Step 5, time scale calculation: the updated primary time scale and backup time scale are generated by solving the updated elastic factor graph model, the time difference between the updated primary time scale and backup time scale is calculated, and the updated primary time scale is output as the elastic time scale facing the change of the atomic clock group.

[0171] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0172] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, it should be understood that the application encompasses all possible combinations of the technical features described above unless such a combination is not technically possible.

[0173] The above embodiments only express several implementation manners of the application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the application. It should be pointed out that for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the application, and these all belong to the protection scope of the application.

Claims

1. A method for generating a flexible time scale oriented to changes in a group of atomic clocks, characterized in that, The elastic time scale generation method comprises: Step 1, time scale parameter initialization: constructing an elastic factor graph model for time scale generation and initializing a main time scale and a backup time scale in a consistent state; Step 2, automatically or manually performing clock group operation; Step 3, updating clock group time difference based on clock group change after clock group operation; Step 4, time scale model updating: if there is no change in the clock group, updating the elastic factor graph model according to the current clock group time difference and clock group configuration; if there is a change in the clock group, updating the clock group configuration of the backup time scale according to the clock group operation result, and calculating and accumulating the time difference between the main time scale and the backup time scale, and judging whether the clock group change duration meets the preset accumulated duration; if not, updating the elastic factor graph model according to the time difference between the current main time scale and the backup time scale and the clock group configuration; if yes, first estimating and compensating the time difference and frequency deviation of the time scale based on the accumulated time difference result, then updating the clock group configuration of the main time scale according to the clock group operation result, and initializing the backup time scale based on the main time scale updated by the clock group configuration, maintaining the consistency of the main time scale and the backup time scale, obtaining the optimized adjustment time difference result between the main time scale and the backup time scale, and updating the elastic factor graph model in combination with the current clock group configuration; Step 5, time scale calculation: generating the updated main time scale and backup time scale by solving the updated elastic factor graph model, and calculating the time difference of the updated main time scale and backup time scale, and outputting the updated main time scale as the elastic time scale facing the change of the atomic clock group.

2. The method of claim 1, wherein the method is characterized by, The elastic factor graph model for time scale generation is constructed and the main time scale and the backup time scale in a consistent state are initialized, comprising: The state equation and the observation equation of the single-state variable Kalman filter time scale algorithm are constructed, respectively represented as: ； wherein is the time difference of the clock group at the time instant k , the superscript T denotes the transpose, is the time difference of the i th atomic clock in the clock group and , is the number of atomic clocks; is the state transition matrix; is the process noise of the clock group, is the process noise of the i th atomic clock in the clock group; is the observed time difference of the clock group at the time instant k ; is the observation matrix; is the observation noise, is the observation noise of the i th group of time differences, the observation noises of the groups of time differences are mutually independent and ; state transition matrix , process noise covariance matrix , observation matrix and observation noise covariance matrix are represented as follows, respectively: ； wherein is the identity matrix; the i process state noise variance of the first atomic clock , denotes the i Allen variance of the first atomic clock is the virtual Kalman sampling interval; is the noise covariance of The state equation and the observation equation of the single-state variable Kalman filter time scale algorithm are converted into state factors and observation factors in the factor graph, respectively, the state variable is converted into a variable node, and the covariance of the estimation error is converted into a prior factor, to construct the elastic factor graph model for time scale generation, represented as: ； The elastic factor graph model is composed of variable nodes, state factors, observation factors and prior factors, the variable nodes correspond to state variables, the state factors correspond to state transition constraints connecting adjacent state variables, the observation factors correspond to Kalman sensor observation constraints connecting state variables and observation values, and the prior factors correspond to the covariance of the estimation error in the Kalman filter; when the clock group changes, the elastic factor graph model is updated by adding or deleting factors corresponding to the atomic clock; The time scale initial parameters are set, including initializing the atomic clock group configuration, setting the initial phase and accuracy of the time scale, the initial time difference of each atomic clock, the covariance of the estimation error, and setting the accumulated duration of the clock group change; Based on the elastic factor graph model and the time scale initial parameters, the main time scale and the backup time scale in a consistent state are initialized.

3. The method of claim 1, wherein the method is characterized by, The clock group time difference is updated based on the clock group change after the clock group operation, including: The clock group change is judged based on the clock group operation result. If there is no change in the clock group, step 4 is directly entered. If an atomic clock is added, the time difference of the added atomic clock is added in the clock group time difference, and then step 4 is entered. If an atomic clock is removed, it is judged whether the clock group change duration meets the preset accumulated duration. If not, it is considered that the removal operation is not completed, the time difference of the atomic clock to be removed is predicted based on the state estimation result of the atomic clock, the predicted time difference is added in the clock group time difference, and then step 4 is entered. Otherwise, it is considered that the removal operation is completed and step 4 is entered.

4. The method of claim 3, wherein the method is characterized by, The accumulated duration is determined by the accuracy of the atomic clock time difference prediction and the accuracy of the frequency deviation obtained by the least square estimation based on the main time scale and the backup time scale, and is expressed as: ； wherein, is a weighted result, is a weighting coefficient of the accumulated time length, is a weight of the atomic clock time difference prediction accuracy, is a weight of the frequency bias estimation accuracy; , , respectively satisfy the following constraints: ； wherein, is the accumulated time length, is a constant threshold of 1.345, is the prediction accuracy of the atomic clock time difference, is is a normalization factor of is the accuracy of the frequency bias estimation, is a normalization factor of when is the minimum, at which time the corresponding is the preset accumulated time length.

5. The method of claim 1, wherein the method is further characterized by: The clock group configuration of the backup time scale is updated according to the clock group operation result, including: The state variable and the estimation error covariance matrix of the backup time scale before the clock group change are obtained; If it is an atomic clock i In addition, an atomic clock i based on the state estimation result of the atomic clock is added in the state variable, the time difference of the atomic clock is added, a new initial state variable is constructed, and the mean of the diagonal elements of the estimation error covariance matrix is added as a new initial estimation error covariance matrix. if the atomic clock is rejected i the value of the atomic clock in the state variable, a new initial state variable is constructed and the value of the atomic clock i is rejected in the estimation error covariance matrix, a new initial estimation error covariance matrix is constructed. The value of the atomic clock i is rejected in the estimation error covariance matrix, a new initial estimation error covariance matrix is constructed.

6. The method of claim 1, wherein, The time scale time difference and the frequency deviation are estimated and compensated based on the accumulated time difference result, including: The least square estimation is performed based on the accumulated time difference result, the jump value of the time difference and the frequency deviation between the main time scale and the backup time scale is obtained, and the jump value is compensated to the initial phase and the accuracy of the time scale, so as to realize the compensation of the time scale time difference and the frequency deviation.

7. The method of claim 1, wherein the method is further characterized by: The clock group configuration of the main time scale is updated according to the clock group operation result, including: The state variable and the estimation error covariance matrix of the main time scale before the clock group change are obtained; If it is an atomic clock i In addition, a Kalman filter is added in the state variable to estimate the time difference of the atomic clock The time difference of the atomic clock estimated by the group observation i The new initial state variable is constructed by the time difference of the atomic clock, and the mean of the diagonal elements of the estimation error covariance matrix is added in the estimation error covariance matrix as the new initial estimation error covariance matrix. if the atomic clock is rejected i the value of the atomic clock in the state variable, a new initial state variable is constructed and the value of the atomic clock i is rejected in the estimation error covariance matrix, a new initial estimation error covariance matrix is constructed; and i the value of the atomic clock is rejected in the estimation error covariance matrix, a new initial estimation error covariance matrix is constructed; and wherein the Kalman filter estimates the time difference between the atomic clocks by the group of observed time difference estimates of the atomic clocks i is denoted as the iterative calculation step is denoted as ； in, For the Zhong Group k -1 time difference, For the Zhong Group k -1 moment k The predicted value of the time difference between the time and the hour, is the state transfer matrix; for k The estimated error matrix of the time difference of the clock group at -1 hour, for k The prediction error matrix of the time difference of the clock group; is the process noise covariance matrix; for k The Kalman filter gain at time , is the observation matrix, is the observation noise covariance matrix, It's the Zhong group k The observed time difference, is the identity matrix, the superscript T Indicates transpose.

8. The method of claim 1 or 5, wherein, The elastic time scale generation method further includes: Step 6, atomic clock state estimation: the state of the atomic clock is estimated by using a 2-state Kalman filter or a 3-state Kalman filter, and the atomic clock time difference is predicted based on the state estimation result of the atomic clock; where the state transition matrix of the 2-state Kalman filter , the process noise covariance matrix , the observation matrix , and the observation noise covariance matrix are given by ； State transition matrix of a 3-state Kalman filter Process noise covariance matrix Observation matrix and observation noise covariance matrix are represented as follows, respectively: ； wherein is the time sampling interval; is the atomic clock i observed noise standard deviation, , and are noise parameters, the specific values of which are generally obtained using the Allan variance inversion method.

9. A resilient time scale generation system oriented towards ensemble of atomic clocks variations, characterized in that, The system includes: A time scale parameter initialization module is configured to construct an elastic factor graph model for generating a time scale and initialize the main time scale and the backup time scale with consistent states. A clock group operation module is configured to automatically or manually perform clock group operation. A clock group time difference updating module is configured to update the clock group time difference based on the clock group change after the clock group operation. A time scale model updating module is configured to update the elastic factor graph model according to the current clock group time difference and the clock group configuration if there is no change in the clock group, and update the clock group configuration of the backup time scale according to the clock group operation result if there is a change in the clock group. Meanwhile, the time difference between the main time scale and the backup time scale is calculated and accumulated, and it is judged whether the clock group change duration meets the preset accumulated duration. If not, the elastic factor graph model is updated according to the time difference between the current main time scale and the backup time scale and the clock group configuration. If yes, the time difference and the frequency deviation of the time scale are estimated and compensated based on the accumulated time difference result, the clock group configuration of the main time scale is updated according to the clock group operation result, the backup time scale is initialized by using the main time scale updated by the clock group configuration, the consistency of the main time scale and the backup time scale is maintained, the time difference result between the main time scale and the backup time scale is optimized and adjusted, and the elastic factor graph model is updated by combining the current clock group configuration. A time scale calculation module is configured to generate an updated primary time scale and a backup time scale by solving the updated elastic factor graph model, calculate a time difference between the updated primary time scale and the backup time scale, and output the updated primary time scale as an elastic time scale facing changes of the atomic clock group. 10.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-9. The processor implements the steps of the elastic time scale generation method in any one of claims 1 to 8 when executing the computer program.

Images