Method and device for acquiring regional weighted average temperature information based on LSTM model

Abstract

The present invention discloses a method for obtaining regional weighted average temperature information based on an LSTM model, comprising: collecting ERA5 reanalysis data of a region; constructing an atmospheric weighted average temperature model related to an annual cycle, a semi-annual cycle, and a daily cycle, and calculating the amplitude of each coefficient of the atmospheric weighted average temperature model based on the least squares method; obtaining a calculated value of the atmospheric weighted average temperature model and a residual value of the atmospheric weighted average temperature model; constructing an LSTM neural network error compensation model to obtain a residual forecast value of the atmospheric weighted average temperature model; and obtaining the atmospheric weighted average temperature based on the atmospheric weighted average temperature model and the residual forecast value of the atmospheric weighted average temperature model. The technology of the present invention effectively compensates for the difficulties of high computational difficulty and low precision in obtaining atmospheric weighted average temperature values ​​in the prior art by constructing an error compensation model, and can be widely applied with significant accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of global navigation satellite systems and meteorology, and in particular to a method and device for obtaining regional weighted average temperature information based on an LSTM model. Background Art

[0002] Water vapor is an important component of the atmosphere and the most active meteorological parameter in the atmosphere. Changes in water vapor directly affect the amount of precipitation and have a very important impact on climate change. The water vapor content in the atmosphere can be obtained by inverting the precipitable water (PWV) through ground-based GNSS technology. PWV is calculated from the zenith tropospheric wet delay (ZWD) and the water vapor conversion coefficient (Π) from ZWD to PWV, and the relationship is PWV = Π·ZWD. The accuracy of PWV mainly includes the accuracy of ZWD calculated from GNSS observation data and the accuracy of Π. The atmospheric weighted mean temperature Tm is a key parameter in the calculation of Π, so the accuracy of Π mainly depends on the accuracy of the atmospheric weighted mean temperature Tm.

[0003] The method for calculating the weighted average temperature Tm generally uses meteorological data obtained from sounding observation stations to integrate and solve the atmospheric weighted average temperature. However, this method has low accuracy and poor applicability in areas where sounding data is difficult to obtain. Among the methods for calculating the weighted average temperature Tm, the most commonly used is the linear regression model established by Bevis, which relates the atmospheric weighted average temperature Tm to the surface temperature Ts. This method substitutes the ground temperature Ts into the formula Tm = 0.72·Ts + 70.2 to calculate the atmospheric weighted average temperature Tm. Because the Bevis formula was established using sounding data from North America, its accuracy is low when used in other regions. Therefore, a method for obtaining regional weighted average temperature information with strong applicability and high accuracy is urgently needed to address this problem. Summary of the Invention

[0004] In order to solve the above-mentioned deficiencies in the prior art, the purpose of the present invention is to overcome the existing deficiencies and provide a method for obtaining regional weighted average temperature information based on an LSTM model, comprising:

[0005] S01, collecting ERA5 reanalysis data of the region, wherein the data types include gravity potential, temperature, and specific humidity at multiple pressures at each grid point in the region; and performing data preprocessing on the reanalysis data;

[0006] S02, construct the atmospheric weighted mean temperature T related to the annual cycle, semi-annual cycle and daily cycle m1 The model calculates the atmospheric weighted average temperature T based on the least squares method. m1 The amplitude of each coefficient of the model;

[0007] S03, obtain the atmospheric weighted average temperature T m1The calculated values ​​of the model and the weighted average temperature of the atmosphere T m1 Model residual value ΔT m1 ;

[0008] S04, build an LSTM neural network error compensation model to obtain the atmospheric weighted average temperature T m1 Model residual forecast value ΔT m2 ;

[0009] S05, based on the atmospheric weighted average temperature T m1 Model and atmospheric weighted mean temperature T m1 Model residual prediction value ΔT m2 , get the atmospheric weighted average temperature T m2 .

[0010] As a further optimization of the above solution, the data preprocessing operation of the reanalysis data includes:

[0011] Convert the gravity potential into the height of the pressure layer, and convert the specific humidity into atmospheric water vapor pressure;

[0012] Calculation of weighted average temperature T based on discrete integration method ERA5 :

[0013]

[0014] In formula (1), They are respectively expressed as the average value of the sum of the atmospheric water vapor pressure and the absolute atmospheric temperature between two adjacent pressure layers, and e is the atmospheric water vapor pressure, that is, Δh i is the height difference between two adjacent pressure layers, i.e. Δh i =h j+1 -h j ; n is the number of pressure layers.

[0015] As a further optimization of the above solution, the weighted average temperature T ERA5 The atmospheric water vapor pressure e in the calculation formula is calculated based on the conversion formula of specific humidity-air pressure-water vapor pressure:

[0016]

[0017] In formula (2), q and p represent the specific humidity and atmospheric pressure of the same pressure layer, respectively;

[0018] The weighted average temperature T ERA5 The pressure layer height h in the calculation formula is calculated as follows:

[0019] h=GP / g (3)

[0020] In formula (3), h represents the potential height; GP represents the gravitational potential; and g represents the gravitational acceleration.

[0021] As a further optimization of the above scheme, the atmospheric weighted average temperature T m1 The model construction method is as follows:

[0022]

[0023] In formula (4), A0 represents the weighted average temperature of the atmosphere T m1 The annual average value; (A1, A2), (A3, A4) and (A5, A6) represent the weighted average temperature of the atmosphere T m1 The coefficients of the annual, semi-annual, and daily cycles; doy represents the cumulative day of the year; hod represents the UTC hour.

[0024] As a further optimization of the above scheme, the atmospheric weighted average temperature T m1 Model residual value ΔT m1 is the weighted average temperature T ERA5 Calculated value and atmospheric weighted average temperature T m1 Calculate the difference of the values, that is:

[0025] ΔT m1 =|T ERA5 -T m1 | (5)

[0026] As a further optimization of the above solution, the LSTM neural network error compensation model is a four-layer network structure of 5×P×Q×4:

[0027] The input layer parameters of the compensation model include ΔT m1 , the sin function related to the week sin(doy×2π / 7), the cos function related to the week cos(doy×2π / 7), the sin function related to the hour sin(hod×2π / 24), and the cos function related to the hour cos(hod×2π / 24);

[0028] The output layer parameters of the compensation model include the atmospheric weighted average temperature T for the next 6 hours, 12 hours, 18 hours and 24 hours. m1 Model residual prediction value ΔT m2 result.

[0029] As a further optimization of the above scheme, the atmospheric weighted average temperature T m1 Model calculated value and atmospheric weighted average temperature T m1 Model residual prediction value ΔT m2 Add up to get the final result: atmospheric weighted average temperature T m2 :

[0030] T m2 =T m1 +ΔT m2 (6).

[0031] As a further optimization of the above scheme, the ERA5 reanalysis data in the region were collected from ERA5 reanalysis data with a horizontal resolution of 0.25°×0.25° and a temporal resolution of 6 hours.

[0032] As a further optimization of the above solution, an electronic device includes:

[0033] processor;

[0034] a memory for storing processor-executable instructions;

[0035] The processor implements the method for acquiring regional weighted average temperature information based on the LSTM model by running the executable instructions.

[0036] As a further optimization of the above solution, a computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the steps of the method for obtaining regional weighted average temperature information based on the LSTM model.

[0037] The present invention adopts the above-mentioned technical solution. Compared with the existing technology, the technology of the present invention performs preprocessing such as data summarization and analysis based on the existing data system to obtain the model parameters to be used in the present technical solution, and combines mathematical calculation methods such as the least squares method to construct an atmospheric weighted average temperature model; and further processes the constructed atmospheric weighted average temperature model to obtain the atmospheric weighted average temperature model residual value, and then constructs an LSTM neural network error compensation model based on the calculation result, so as to obtain the atmospheric weighted average temperature after processing; the technology of the present invention effectively compensates for the high difficulty and low precision in obtaining the atmospheric weighted average temperature value in the existing technology by constructing the error compensation model, can be widely used, and has obvious accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0039] Figure 1 This is a flow chart of a method for obtaining regional weighted average temperature information based on an LSTM model of the present invention;

[0040] Figure 2 This is another flow chart of a method for obtaining regional weighted average temperature information based on an LSTM model according to the present invention;

[0041] Figure 3 This is a line graph of the RMS distribution probability of two models of a method for obtaining regional weighted average temperature information based on an LSTM model of the present invention. DETAILED DESCRIPTION

[0042] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0043] like Figure 1-3 As shown, the embodiment of the present invention discloses a method for obtaining regional weighted average temperature information based on an LSTM model. The specific process of the method is as follows:

[0044] S01, collecting ERA5 reanalysis data of the region, wherein the data types include gravity potential, temperature, and specific humidity at multiple pressures at each grid point in the region; and performing data preprocessing on the reanalysis data;

[0045] S02, construct the atmospheric weighted mean temperature T related to the annual cycle, semi-annual cycle and daily cycle m1 The model calculates the atmospheric weighted average temperature T based on the least squares method. m1 The amplitude of each coefficient of the model;

[0046] S03, obtain the atmospheric weighted average temperature T m1 The calculated values ​​of the model and the weighted average temperature of the atmosphere T m1 Model residual value ΔT m1 ;

[0047] S04, build an LSTM neural network error compensation model to obtain the atmospheric weighted average temperature T m1 Model residual prediction value ΔT m2 ;

[0048] S05, based on the atmospheric weighted average temperature T m1 Model and atmospheric weighted mean temperature T m1 Model residual prediction value ΔT m2 , get the atmospheric weighted average temperature T m2 .

[0049] The technology of the present invention performs preprocessing such as summarization and analysis of data based on an existing data system to obtain model parameters to be used in the present technical solution, and combines mathematical calculation methods such as the least squares method to construct an atmospheric weighted average temperature model; and further processes the constructed atmospheric weighted average temperature model to obtain the atmospheric weighted average temperature model residual value, and then constructs an LSTM neural network error compensation model based on the calculation result, so as to obtain the atmospheric weighted average temperature after processing; the technology of the present invention effectively compensates for the difficulties of high computational difficulty and low precision in obtaining the atmospheric weighted average temperature value in the existing technology by constructing the error compensation model, and can be widely used with obvious accuracy.

[0050] It should be noted here that the regional ERA5 reanalysis data are data collected from the Chinese region from 2014 to 2021, that is, ERA5 reanalysis data with a horizontal resolution of 0.25°×0.25° and a temporal resolution of 6 hours. The technical solution of the present invention adopts ERA5 reanalysis data with a horizontal resolution of 0.25°×0.25° and a temporal resolution of 6 hours, and uses the data from 2014 to 2019 as training data, and the data from 2020 to 2021 as validation data. In addition, part of the data of the region selected by this technical solution is for scheme research. The technical achievements of the present invention include but are not limited to part of the data of the illustrated area. That is, based on the technology disclosed in this technical solution, those skilled in the art should easily know that when applying this solution, by taking other longitude and latitude regional data included in the ERA5 database, such as the gravity potential, temperature, and specific humidity of the region's multiple pressure parameters, and combining them with the patented technical solution, a weighted average temperature calculation method based on the LSTM model for the corresponding region can be obtained, which effectively improves the accuracy of the calculation results. Therefore, this technical solution has good adaptability and can be applied to multiple regions.

[0051] More specifically, the data preprocessing operations for reanalysis data include:

[0052] Convert the gravity potential to the height of the pressure layer, and convert the specific humidity to the atmospheric water vapor pressure. It should be noted that the data types of this technical solution include the gravity potential, temperature, and specific humidity of multiple pressures at each grid point from 1000hPa to 100hPa with an interval of 25hPa;

[0053] Calculation of weighted average temperature T based on discrete integration method ERA5 :

[0054]

[0055] In formula (1), They are respectively expressed as the average value of the sum of the atmospheric water vapor pressure and the absolute atmospheric temperature between two adjacent pressure layers, and e is the atmospheric water vapor pressure, that is, Δhi is the height difference between two adjacent pressure layers, i.e. Δh i =h j+1 -h j ; n is the number of pressure layers.

[0056] Weighted average temperature T ERA5 The atmospheric water vapor pressure e in the calculation formula is calculated based on the conversion formula of specific humidity-air pressure-water vapor pressure:

[0057]

[0058] In formula (2), q and p represent the specific humidity and atmospheric pressure of the same pressure layer, respectively;

[0059] Weighted average temperature T ERA5 The pressure layer height h in the calculation formula is calculated as follows:

[0060] h=GP / g (3)

[0061] In formula (3), h represents the potential height; GP represents the gravitational potential; and g represents the gravitational acceleration.

[0062] More specifically, the weighted mean temperature of the atmosphere T m1 The model construction method is as follows:

[0063]

[0064] In formula (4), A0 represents the weighted average temperature of the atmosphere T m1 The annual average value; (A1, A2), (A3, A4) and (A5, A6) represent the weighted average temperature of the atmosphere T m1 The coefficients of the annual, semi-annual, and daily cycles; doy represents the cumulative day of the year; hod represents the UTC hour.

[0065] An embodiment of the present application further provides an electronic device, comprising:

[0066] processor;

[0067] a memory for storing processor-executable instructions;

[0068] The processor executes the executable instructions to implement the method for obtaining regional weighted average temperature information based on the LSTM model provided in the above embodiment.

[0069] It is understood that the memory of the electronic device includes a non-volatile storage medium and 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 computer program in the non-volatile storage medium. When the computer program is executed by the processor, it implements the above-mentioned method for obtaining regional weighted average temperature information based on the LSTM model. Furthermore, the electronic device also has a network interface for communicating with an external terminal via a network connection.

[0070] The embodiment of the present application also provides a computer-readable storage medium having computer instructions stored thereon, characterized in that when the instructions are executed by the processor, the steps of the method for obtaining regional weighted average temperature information based on the LSTM model as provided in the above embodiment are implemented. The computer-readable storage medium includes: permanent and non-permanent, removable and non-removable media, which are tangible devices that can retain and store instructions for use by instruction execution devices. Computer-readable storage media include: electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, and any suitable combination of the above. Specifically, computer-readable storage media include, but are not limited to, read-only memory (ROM), random access memory (RAM), read-only compact disc (CD-ROM), phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), etc.

[0071] The present invention provides a specific embodiment, which is to use the weighted average temperature T m1 The performance of the model is compared with the existing Bevis model. The analysis uses the 2020-2021 ERA5 reanalysis data of the China map area as the test sample and uses the root mean square (RMS) as the evaluation indicator. The accuracy comparison of the two models is shown in Table 1:

[0072]

[0073] Table 1: Bevis model and atmospheric weighted mean temperature T m1 Model Accuracy Comparison Table

[0074] More specifically, the embodiment of the present invention further compares the atmospheric weighted average temperature T m1 The performance of the model and the Bevis model are analyzed by statistically analyzing the RMS of the grid points. The values ​​are divided into different ranges and the probability line graphs of the RMS distribution of the two models are constructed, as shown in Figure 2. Figure 3 As shown ( Figure 3 The dotted line is the Bevis model).

[0075] More specifically, the weighted mean temperature of the atmosphere T m1 Model residual value ΔTm1 is the weighted average temperature T ERA5 Calculated value and atmospheric weighted average temperature T m1 Calculate the difference of the values, that is:

[0076] ΔT m1 =|T ERA5 -T m1 | (5)

[0077] More specifically, the LSTM neural network error compensation model is a four-layer network structure of 5×P×Q×4:

[0078] The input layer parameters of the compensation model include ΔT m1 , the sin function related to the week sin(doy×2π / 7), the cos function related to the week cos(doy×2π / 7), the sin function related to the hour sin(hod×2π / 24), and the cos function related to the hour cos(hod×2π / 24);

[0079] The output layer parameters of the compensation model include the atmospheric weighted average temperature T for the next 6 hours, 12 hours, 18 hours and 24 hours. m1 Model residual prediction value ΔT m2 result.

[0080] The weighted mean temperature of the atmosphere T m1 Model calculated value and atmospheric weighted average temperature T m1 Model residual prediction value ΔT m2 Add up to get the final result: atmospheric weighted average temperature T m2 :

[0081] T m2 =T m1 +ΔT m2 (6).

[0082] As another embodiment of the present invention, in order to compare the performance of the atmospheric weighted average temperature model and the atmospheric weighted average temperature model after LSTM neural network residual compensation, the 2020-2021 ERA5 reanalysis data in the Chinese region are used as test samples, and the root mean square (RMS) is used as the evaluation indicator. The accuracy comparison of the two models is shown in Table 2:

[0083] Table 2 shows the LSTM neural network error compensation model for the atmospheric weighted average temperature T m1 Comparison of model accuracy before and after residual compensation, T m1 1 is before compensation, and 6 hours, 12 hours, 18 hours and 24 hours are after compensation.

[0084]

[0085] Table 2: Weighted mean atmospheric temperature T m1 Accuracy comparison table before and after model residual compensation

[0086] By comparing Table 1, Table 2 and Figure 3 The comprehensive analysis can be obtained as follows:

[0087] (1) Weighted mean atmospheric temperature T m1 The mean RMS of the model is 3.12K, the mean RMS of the Bevis model is 3.60K, and the atmospheric weighted mean temperature T m1 The mean RMS of the model is less than the mean RMS of the Bevis model, and the atmospheric weighted mean temperature T m1 The model is improved by 13.3% compared with the Bevis model, and the atmospheric weighted mean temperature T m1 The model is better than the Bevis model;

[0088] (2) Weighted average atmospheric temperature T after LSTM neural network residual compensation m1 The RMS mean of the model was reduced from the original 3.12K to 1.26K, 1.79K, 2.12K and 2.35K, respectively, with improvements of 59.6%, 42.6%, 32.1% and 24.7%. The weighted mean temperature of the atmosphere after residual compensation is T m1 The model is better than the atmospheric weighted mean temperature T without residual compensation. m1 Model;

[0089] (3) Weighted average atmospheric temperature T after LSTM neural network residual compensation m1 Compared with the Bevis model, this model is more suitable for calculating the weighted average temperature in China.

[0090] Based on the above technical solutions and embodiments, the present invention provides a method for obtaining regional weighted average temperature information based on an LSTM model. By constructing an LSTM neural network error compensation model, it effectively compensates for the difficulties in the prior art in obtaining the numerical value of the atmospheric weighted average temperature, such as high computational difficulty and low precision. The method can be widely used and has obvious accuracy.

[0091] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the inventive concept. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for obtaining regional weighted average temperature information based on an LSTM model, characterized in that: include: S01, collect ERA5 reanalysis data of the region, the data type includes gravity potential, temperature and specific humidity at multiple pressures of each grid point in the region; Perform data preprocessing on reanalysis data; S02, construct the atmospheric weighted mean temperature T related to the annual cycle, semi-annual cycle and daily cycle m1 The model calculates the atmospheric weighted average temperature T based on the least squares method. m1 The amplitude of each coefficient of the model; S03, obtain the atmospheric weighted average temperature T m1 The calculated values ​​of the model and the weighted average temperature of the atmosphere T m1 Model residual value ΔT m1 ; The atmospheric weighted average temperature T m1 Model residual value ΔT m1 is the weighted average temperature T ERA5 Calculated value and atmospheric weighted average temperature T m1 Calculate the difference of the values, that is: ΔT m1 =|T ERA5 -T m1 | (5) S04, build an LSTM neural network error compensation model to obtain the atmospheric weighted average temperature T m1 Model residual forecast value ΔT m2 ; The LSTM neural network error compensation model is a four-layer network structure of 5×P×Q×4: The input layer parameters of the compensation model include ΔT m1 , the sin function related to the week sin(doy×2π / 7), the cos function related to the week cos(doy×2π / 7), the sin function related to the hour sin(hod×2π / 24), and the cos function related to the hour cos(hod×2π / 24); The output layer parameters of the compensation model include the atmospheric weighted average temperature T for the next 6 hours, 12 hours, 18 hours and 24 hours. m1 Model residual forecast value ΔT m2 result; S05, based on the atmospheric weighted average temperature T m1 Model and atmospheric weighted mean temperature T m1 Model residual forecast value ΔT m2 , get the atmospheric weighted average temperature T m2 .

2. The method for obtaining regional weighted average temperature information based on the LSTM model according to claim 1, characterized in that: The data preprocessing operation of the reanalysis data includes: Convert the gravity potential into the height of the pressure layer, and convert the specific humidity into atmospheric water vapor pressure; Calculation of weighted average temperature T based on discrete integration method ERA5 : In formula (1), They are respectively expressed as the average value of the sum of the atmospheric water vapor pressure and the absolute atmospheric temperature between two adjacent pressure layers, and e is the atmospheric water vapor pressure, that is, Δh i is the height difference between two adjacent pressure layers, i.e. Δh i =h j+1 -h j ; n is the number of pressure layers.

3. The method for obtaining regional weighted average temperature information based on the LSTM model according to claim 2, characterized in that: The weighted average temperature T ERA5 The atmospheric water vapor pressure e in the calculation formula is calculated based on the conversion formula of specific humidity-air pressure-water vapor pressure: In formula (2), q and p represent the specific humidity and atmospheric pressure of the same pressure layer, respectively; The weighted average temperature T ERA5 The pressure layer height h in the calculation formula is calculated as follows: h=GP / g(3) In formula (3), h represents the potential height; GP represents the gravitational potential; and g represents the gravitational acceleration.

4. The method for obtaining regional weighted average temperature information based on the LSTM model according to claim 1, characterized in that: The atmospheric weighted average temperature T m1 The model construction method is as follows: In formula (4), A0 represents the weighted average temperature of the atmosphere T m1 The annual average value of (A1, A2), (A3, A4) and (A5, A6) represent the weighted average temperature of the atmosphere T m1 The coefficients of the annual, semi-annual, and daily cycles; doy represents the cumulative day of the year; hod represents the UTC hour.

5. The method for obtaining regional weighted average temperature information based on the LSTM model according to claim 1, wherein: The atmospheric weighted average temperature T m1 Model calculated value and atmospheric weighted average temperature T m1 Model residual forecast value ΔT m2 Add up to get the final result, the weighted average atmospheric temperature T m2 : T m2 =T m1 +ΔT m2 (6)。 6. The method for obtaining regional weighted average temperature information based on the LSTM model according to claim 1, characterized in that: The ERA5 reanalysis data for the region were collected from ERA5 reanalysis data with a horizontal resolution of 0.25°×0.25° and a temporal resolution of 6 hours.

7. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing processor-executable instructions; The processor executes the executable instructions to implement the method for obtaining regional weighted average temperature information based on the LSTM model as described in any one of claims 1 to 6.

8. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by the processor, the steps of the method for obtaining regional weighted average temperature information based on the LSTM model as described in any one of claims 1 to 6 are implemented.

Images