Multi-scale set adjustment Kalman filtering method suitable for small samples
Patent Information
- Application Number
- CN202410886011.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-03
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Under the constraints of computational resources, existing technologies, such as ensemble Kalman filtering, cannot effectively maintain the multi-scale representativeness of the background error covariance, resulting in insufficient initialization accuracy of atmospheric and oceanic ensemble forecasts.
A multi-scale ensemble adjustment Kalman filter method with small samples is adopted. By designing an interface during the mode operation, static and dynamic ensembles are constructed. An efficient multi-scale assimilation algorithm is used to calculate the observation increment and perform ensemble Kalman filter update to maintain the flow dependence characteristics and the representativeness of low-frequency signals.
It improves the initialization accuracy and state estimation accuracy of atmospheric and oceanic ensemble forecasts, and can still obtain good state estimation results, especially in the case of small samples.
Smart Images

Figure CN118939992A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data assimilation and earth system coupled data assimilation, and in particular to a multi-scale ensemble adjusted Kalman filtering method suitable for small samples. Background Art
[0002] Data assimilation is based on Bayesian conditional probability theory, which integrates discrete and discontinuous observation data in time and space into mathematical models to obtain the optimal estimate of the system state. For ocean and atmospheric research, data assimilation technology can not only reconstruct the state evolution of the atmosphere and ocean in the past history, but also provide the initial conditions required for the initialization of model forecasts and predictions. It plays an important role in the scientific research and forecast application practice of the ocean atmosphere. At present, data assimilation algorithms are mainly divided into two categories: sequential data assimilation algorithms and continuous data assimilation algorithms. Ensemble Kalman filtering is a typical representative of sequential data assimilation, and is widely used for its good robustness and lower computational cost than other algorithms.
[0003] However, due to the limitation of computing resources, the ensemble Kalman filter can only use a limited number of ( ) ensemble and finite time integration length are used to calculate the error covariance matrix, which makes the calculated error covariance matrix lack representation of the low-frequency signal of the background flow. Although some studies have solved this problem well by using a multi-scale statistics of historical data without model ensemble integration, it seriously weakens the instantaneous flow dependence of the original ensemble filter and cannot meet the needs of initialization of atmospheric and ocean ensemble forecasts. Summary of the invention
[0004] In order to make up for the shortcomings of the prior art, the present invention provides a multi-scale ensemble adjustment Kalman filter method suitable for small samples. The present invention aims to solve the problem of how to maintain the characteristics of ensemble filter flow dependence when the background error covariance has multi-scale representativeness in the case of a small number of ensemble members, design a new algorithm to realize the initialization of atmosphere and ocean ensemble forecast, and further improve the forecast accuracy of the atmosphere and ocean.
[0005] The present invention is implemented by the following technical scheme: a multi-scale set adjustment Kalman filtering method suitable for small samples, based on a numerical model, designs an interface during the operation of the model, and specifically includes the following steps: Step S1: Use the interface to read the mode state variables that need to be modified into the assimilation module; in the assimilation module, for each dynamic ensemble member (Dynamical Ensemble, DE), a queue structure is used to record the integral value of each dynamic ensemble member mode in the recent period of time to construct Static Ensemble (SE) of time scales, each of which has The static set is the Members are the distance from the current integration step to the previous The average of the step integration results, where s is the time scale of the τth static set; Step S2: Calculate the observation increment at each scale using an efficient multi-scale assimilation algorithm: in, It is The mode value of the dynamic set members at the observation point (obtained by interpolation), represents the observed value at the observation point, It is The first member of the dynamic set The standard deviation calculated from the static set of groups, is the standard deviation of the observation, which represents the uncertainty of the observation; in the calculation process of the observation increment of the i-th dynamic set member, the static set is only used to calculate the uncertainty and error covariance, and the model background field uses the model value of the single dynamic set member. After the observation increment is calculated, it is projected into the state space of the model to obtain the adjusted value of the model state variable of each dynamic set: It is the weight coefficient corresponding to the observation increment of each scale, which is determined according to the actual problem; Is to use the The dynamic ensemble members construct the static ensemble members of the τ time scale, and the calculated mode grid points ( ) and the observation grid ( ) is the error covariance between them.
[0006] Step S3: Use each updated dynamic set member to perform set Kalman filtering: Step S3-1, calculate the observation increment based on the multi-scale algorithm: in, is the standard deviation of the dynamic set after being updated by formula (2), is the mean of the dynamic set; Step S3-2, project the observation increment into the pattern space: As a preferred solution, the method for updating each static set in step S1 is: when the model is integrated forward, each step will be recorded first, and when the set is full of s steps, the mean of s time scales is calculated, and then the 2nd to Ne members of the static set member array replace the 1st to Ne-1 data in turn, and then the Ne member is replaced with the value just calculated. The whole process complies with the principle of "first in, first out" to complete an online update of a time scale static set.
[0007] Due to the adoption of the above technical solution, the present invention has the following beneficial effects compared with the prior art: 1. Compared with traditional assimilation methods (EAKF / MSHea-EnKF, etc.), this assimilation method not only maintains the characteristics of ensemble filter flow dependence, but also is representative of the low-frequency signals of the background flow, thus providing a more accurate estimation of the system state.
[0007] 2. Since each ensemble member uses a multi-scale assimilation algorithm to perform a one-step filtering before executing the ensemble adjusted Kalman filter, this method can achieve better state estimation results with a small number of ensemble members.
[0008] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 is the assimilation result of sea surface pressure. Figure a is the time series of the root mean square error of different experimental methods. The black line CTL is the control experiment, and the solid line is the patent method ( ), the dotted line is the multi-scale assimilation method implemented for each dynamic ensemble member ( ) results. The red line is the coupled data assimilation (CDA), the blue line is the assimilation of only ocean observation data (ODA), and the green line is the assimilation of only atmospheric observation data (ADA). Figures b to d show the root mean square error reduction rate compared to the control experiment, defined as , Figures e to g are compared The root mean square error reduction rate.
[0010] Figure 2 Assimilation results of ocean temperature and salinity under coupled data assimilation. Figures a and c show the root mean square error reduction rate of ocean temperature and salinity from 0 to 2000 m compared with the control experiment, and Figures b and d show the results from 2000 m to the bottom of the ocean. Figures e, f, and gh are the same as Figures a, b, c, and d, but the results are the same as In comparison. DETAILED DESCRIPTION
[0011] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.
[0012] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.
[0013] Combine the following Figure 1 to Figure 2 The multi-scale set adjustment Kalman filtering method suitable for small samples in an embodiment of the present invention is specifically described.
[0014] like Figure 1 , Figure 2 As shown in FIG. 1 , the present invention proposes a multi-scale ensemble adjustment Kalman filter method suitable for small samples, based on a numerical model, requiring the simulation results of the model to be representative of the process of interest in the actual problem. At the same time, observation data are collected to meet the requirements of correcting the model state using assimilation algorithms and observation data. At the same time, using the model, generate During the operation of the model, an interface is designed to ensure that the state variables of the model can be modified dynamically using the observation information, which includes the following steps: Step S1: Use the interface to read the mode state variables that need to be modified into the assimilation module; in the assimilation module, for each dynamic ensemble member (Dynamical Ensemble, DE), a queue structure is used to record the integral value of each dynamic ensemble member mode in the recent period of time to construct Static Ensemble (SE) of time scales, each of which has The static set is the Members are the distance from the current integration step to the previous The average of the step integration results, where s is the time scale of the τth static set; for each static set, the time scale is different, but the update method is similar: when the pattern is integrated forward, each step will be recorded first, and when the set is full of s steps, the mean of the s time scale is calculated, and then the static set member arrays from 2 to N are added. e members replace numbers 1 to N in turn e -1 data, and then replace the Nth data with the value just calculated e The whole process complies with the principle of "first in, first out" and completes an online update of a time scale static set. Time scale parameters It should be determined based on specific practical problems. Generally speaking, it should be able to reflect the multi-time scale characteristics of real problems.
[0015] Step S2: Calculate the observation increment at each scale using an efficient multi-scale assimilation algorithm: in, It is The mode value of the dynamic set members at the observation point (obtained by interpolation), represents the observed value at the observation point, It is The first member of the dynamic set The standard deviation calculated from the static set of groups, is the standard deviation of the observation, which represents the uncertainty of the observation. In the calculation process of the observation increment of the i-th dynamic set member, the static set is only used to calculate the uncertainty and error covariance, and the model background field uses the model value of the single dynamic set member. After the observation increment is calculated, it is projected into the state space of the pattern to obtain the adjusted value of the pattern state variable of each dynamic set: It is the weight coefficient corresponding to the observation increment of each scale, which is determined according to the actual problem; Is to use the The dynamic ensemble members construct the static ensemble members of the τ time scale, and the calculated mode grid points ( ) and the observation grid ( ) The error covariance between Step S3: Use each updated dynamic set member to perform set Kalman filtering: Step S3-1, calculate the observation increment based on the multi-scale algorithm: in, is the standard deviation of the dynamic set after being updated by formula (2), is the mean of the dynamic set; Step S3-2, project the observation increment into the pattern space: Select the numerical model that needs to be assimilated and generate a certain number of dynamic ensemble members, and prepare the corresponding observation data. Then build the corresponding assimilation module, which can collect the state variables output by the model to construct the static ensemble members. Finally, in the assimilation module, the model state is estimated according to formulas (1), (2), (3), and (4), and the updated model state is returned.
[0016] Example 1: Estimation of global sea surface pressure Figure 1 The results of the estimation of global sea surface pressure after using this assimilation method in the twin experiment framework are shown. The experiment is conducted in the coupled model framework and is divided into atmospheric data assimilation (ADA), ocean data assimilation (ODA) and coupled data assimilation (CDA). Since the sea surface atmospheric pressure reflects the weight of the positive layer air column, in the assimilation process, not only the sea surface pressure is modified by the assimilation method, but also the adjusted results are projected onto other isobaric surfaces through the geostrophic relationship to achieve three-dimensional adjustment.
[0017] The experimental results show that coupled data assimilation and atmospheric data assimilation can provide more accurate state estimates than ocean assimilation. Due to the uncertainty of the land model, the improved results are mostly located in the ocean area. Such improvements are global. After adding this method, the global sea surface pressure is closer to the true value. Compared with the effect of multi-scale assimilation, it is comparable.
[0018] Example 2: Estimation of global ocean temperature and salinity Figure 2 The results show the effect of ocean temperature and salinity estimation after coupled data assimilation using this assimilation method in the twin experiment framework. The final results show that this assimilation method has not only improved significantly compared to the control experiment (Figures a, b, c, d), but also improved by more than 10% on average over 2000 meters in the ocean compared to the original multi-scale assimilation algorithm, and can reach 50% to 70% in some areas. For the ocean below 2000 meters, due to the scarcity of observational data, the results presented by the two assimilation methods are not much different.
[0019] The data results obtained by the present invention can be used for the initialization of forecasts and predictions, and can also provide data support for tasks including earth system forecasts and predictions, seamless forecasts, etc. A coordinated and balanced initial field is crucial for forecasts, especially for coupled system forecasts, and can provide technical support for business units and be applied to atmospheric and marine disaster prevention and mitigation.
[0020] In the description of the present invention, the term "plurality" refers to two or more than two. Unless otherwise clearly defined, the orientation or positional relationship indicated by the terms "upper" and "lower" is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention; the terms "connection", "installation", "fixation", etc. should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0021] In the description of this specification, the description of the terms "one embodiment", "some embodiments", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0022] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. An efficient multi-scale ensemble adjustment Kalman filter algorithm suitable for small samples, characterized by ,Based on a numerical model, during the model operation, an interface is designed, which includes the following steps: Step S1: Use the interface to read the mode state variables that need to be modified into the assimilation module; for each dynamic set member, a queue structure is used to record the integral value of each dynamic set member mode in the recent period of time to construct static collections, each with The static set is the Members are the distance from the current integration step to the previous The average of the step integration results, where s is the time scale of the τth static set; Step S2: Calculate the observation increment at each scale using an efficient multi-scale assimilation algorithm: in, It is The mode value of the dynamic set members at the observation point (obtained by interpolation), represents the observed value at the observation point, It is The first member of the dynamic set The standard deviation calculated from the static set of groups, is the standard deviation of the observation, which represents the uncertainty of the observation. In the calculation process of the observation increment of the i-th dynamic set member, the static set is only used to calculate the uncertainty and error covariance, and the model background field uses the model value of the single dynamic set member. After the observation increment is calculated, it is projected into the state space of the pattern to obtain the adjusted value of the pattern state variable of each dynamic set: It is the weight coefficient corresponding to the observation increment of each scale, which is determined according to the actual problem; Is to use the The dynamic ensemble members construct the static ensemble members of the τ time scale, and the calculated mode grid points ( ) and the observation grid ( ) The error covariance between Step S3: Use each updated dynamic set member to perform set Kalman filtering: Step S3-1, calculate the observation increment based on the multi-scale algorithm: in, is the standard deviation of the dynamic set after being updated by formula (2), is the mean of the dynamic set; Step S3-2, project the observation increment into the pattern space: 。 2. According to claim 1, an efficient multi-scale ensemble adjustment Kalman filter algorithm suitable for small samples is characterized in that ,The update method for each static set in step S1 is: when the pattern is integrated forward, each step will be recorded first, and when the set is full of s steps, the mean of the s time scale is calculated, and then the static set member array from 2 to N is added. e members replace numbers 1 to N in turn e-1 data, and then replace the Nth data with the value just calculated e The whole process complies with the "first in, first out" principle and completes an online update of a static collection of a time scale.
