Suspended waveguide intensity short-term forecasting method and device based on CEEMDAN decomposition

Abstract

The invention discloses a suspended waveguide intensity short-term forecasting method based on CEEMDAN decomposition, and the method comprises the following steps: constructing a suspended waveguide intensity prediction model, collecting suspended waveguide correlation parameter data, and carrying out the preprocessing of the collected data, and obtaining an original time series data set for training the prediction model; each data in the data set is decomposed into the sum of n IMF components and one residual signal by adopting a CEEMDAN algorithm, the IMF components are obtained through decomposition, and n IMF component time sequence data sets are obtained; respectively extracting one piece of IMF component time sequence data from the n pieces of IMF component time sequence data; sequentially inputting the taken-out time sequence data into a prediction model, or inputting the taken-out time sequence data into n prediction models with the same structure in a one-to-one correspondence manner; and carrying out superposition synthesis on prediction results output by the prediction model to obtain a comprehensive prediction result of the suspended waveguide intensity. The recognition probability of the suspended waveguide is improved.

Description

Technical Field

[0001] The invention relates to the field of marine radio meteorological characteristic analysis, and in particular to a method and device for short-term forecasting of suspended waveguide strength based on CEEMDAN decomposition. Background Art

[0002] Atmospheric waveguides can cause radio signals to undergo super-refraction and propagate to a farther range along the curved surface of the earth. In particular, suspended waveguides have an important impact on the 6G ground-sea-air-space integrated communication network system due to their wide occurrence height and area. At present, atmospheric waveguides are an abnormal atmospheric refraction phenomenon that occurs in the troposphere. Due to the particularity of the marine atmospheric environment, the probability of occurrence of marine atmospheric waveguides is high, the duration is long, and they are relatively stable. In tropical and subtropical waters, under the influence of subtropical high pressure, there is a large-scale sinking movement in the upper atmosphere. The sinking and warming cause dry and hot air to cover the lower atmosphere of the cold and humid marine boundary layer, forming a suspended inversion layer, which is conducive to the formation of suspended waveguides.

[0003] Atmospheric waveguides at sea can significantly change the propagation path and intensity of electromagnetic waves, reduce propagation losses, increase propagation distance, and achieve over-the-horizon detection. However, atmospheric waveguides can also lead to detection blind spots and increase errors in distance, angle, and speed measurements.

[0004] At present, the research on suspended waveguides mainly focuses on the statistics of spatiotemporal distribution, the hydrological and meteorological conditions and physical models generated by atmospheric waveguides, and the measurement of atmospheric waveguide propagation characteristics. However, it is difficult to accurately predict the characteristics of suspended waveguides due to the wide distribution of suspended waveguides and the large range of parameter variations. Summary of the invention

[0005] The present invention provides a method and device for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition in order to solve the technical problems existing in the known technology.

[0006] The technical solution adopted by the present invention to solve the technical problems existing in the known technology is:

[0007] A short-term prediction method for suspended waveguide strength based on CEEMDAN decomposition, the method comprising the following steps:

[0008] Step 1, constructing a prediction model for the strength of a suspended waveguide based on time series data, collecting the associated parameter data of the suspended waveguide and preprocessing the collected data to obtain an original time series data set for training the prediction model;

[0009] Step 2: Use the CEEMDAN algorithm to decompose each data in the original time series data set into the sum of n IMF components and a residual signal. Combine the IMF components obtained by decomposition according to the time order of the data before decomposition, and obtain n IMF component time series data sets from the original time series data set.

[0010] Step 3, sorting by the same time, extracting one IMF component time series data from each of the n IMF component time series data sets; inputting the extracted n IMF component time series data into a prediction model in sequence, or inputting them one by one into n prediction models with the same structure; superimposing and synthesizing the prediction results output by the prediction model to obtain a comprehensive prediction result of the suspended waveguide strength.

[0011] Furthermore, the CEEMDAN algorithm is used to decompose the real-time collected parameter data of each suspended waveguide into the sum of n IMF components and a residual signal. The time series data of the n IMF components obtained by decomposition are input into the trained prediction model, and the prediction results output by the prediction model are superimposed and synthesized to obtain the comprehensive prediction result of the suspended waveguide strength.

[0012] Furthermore, a prediction model for the strength of the suspended waveguide is constructed based on the LSTM model.

[0013] Furthermore, the method for collecting the suspended waveguide associated parameter data in step 1 includes the following method steps:

[0014] The suspended waveguide associated parameters include the altitude, temperature, pressure, and relative humidity of the target area. The suspended waveguide associated parameter data are measured twice a day, each measurement lasts 1 to 2 hours, and the data sampling interval is 1 second.

[0015] Furthermore, the method for preprocessing the collected suspended waveguide associated parameter data in step 1 includes the following method steps:

[0016] The atmospheric corrected refractive index values ​​at different altitudes are calculated from the collected data; if the corrected refractive index value at a certain moment differs by 100 M-unit from that at the previous moment, or if the altitude detected decreases with time due to measurement problems with the detection instrument, the data at that moment is considered invalid and is filled with the data at the same moment of the previous day.

[0017] Furthermore, the difference between the atmospheric corrected refractive index values ​​corresponding to the waveguide layer top height and the waveguide bottom height is defined as the waveguide intensity. Based on the characteristic that the corrected refractive index decreases with increasing altitude under the condition of the existence of a suspended waveguide, it is judged whether there is a suspended waveguide on a certain day. If there is a suspended waveguide on that day, the waveguide intensity at the corresponding measurement time is calculated and recorded. If no suspended waveguide appears on that day, the data at the same time of the previous day is used to fill in the gap to ensure the continuity of the time series. At the same time, a timetable of the waveguide intensity is established to analyze the accuracy of the suspended waveguide associated parameter data at each time.

[0018] Furthermore, in step 1, the original time series dataset is divided into a training set and a validation set.

[0019] Furthermore, in step 2, when the CEEMDAN algorithm is used to decompose the data in the original time series data set, a limited number of adaptive white noises are added at each stage to suppress modal confusion and avoid distortion of the original signal, thereby achieving a smaller reconstruction error. The function expression of the CEEMDAN algorithm is as follows:

[0020] IMF i k =F C (Strength i ),k=1,…,n;

[0021] Where:

[0022] F C () represents the CEEMDAN algorithm function;

[0023] i is the serial number of the original time series data;

[0024] Strength i is the waveguide intensity of the i-th original time series data;

[0025] k is the serial number of the IMF component of the waveguide intensity of the i-th original time series data;

[0026] n is the number of IMF components decomposed from the original time series data;

[0027] IMF i k is the kth IMF component of the waveguide intensity of the i-th original time series data.

[0028] Furthermore, step 3 includes the following sub-steps:

[0029] Step A1, inputting multiple IMF component time series data obtained by decomposing the i-th original time series data into the prediction model;

[0030] Step A2, adding up the multiple prediction result components output by the prediction model to obtain the waveguide intensity prediction value M of the i-th original time series data i p , M i p The calculation formula is as follows:

[0031]

[0032] Where:

[0033] F L () represents the prediction function of the prediction model;

[0034] F a () means adding multiple prediction result components;

[0035] F C () represents the CEEMDAN algorithm function;

[0036] m is the prediction component number output by the prediction model;

[0037] Step A3: After the model training is completed, the performance of the prediction model is evaluated using the validation set.

[0038] The present invention also provides a device for a method for short-term prediction of the strength of a suspended waveguide based on CEEMDAN decomposition, comprising a memory and a processor, wherein the memory is used to store a computer program; the processor is used to execute the computer program and implement the steps of the method for short-term prediction of the strength of a suspended waveguide based on CEEMDAN decomposition as described above when executing the computer program.

[0039] The advantages and positive effects of the present invention are:

[0040] The present invention provides a short-term prediction method for the strength of suspended waveguides based on CEEMDAN decomposition. Aiming at the nonlinear and non-stationary characteristics of time series, a hybrid deep learning model combined with a decomposition algorithm is proposed, which improves the prediction accuracy of waveguide parameters and effectively improves the performance of the model.

[0041] The present invention constructs a prediction model for the strength of a suspended waveguide based on the LSTM model, and combines the LSTM model with the prediction of the strength of the suspended waveguide. The suspended waveguide strength data can capture more details after being decomposed by the CEEMDAN algorithm, and the LSTM model is suitable for time series. The combination of the two improves the accuracy of model prediction.

[0042] The present invention utilizes the high vertical resolution radiosonde data of the high-altitude station to perform statistical analysis, and the use of the data of the high-precision radiosonde instrument can significantly improve the recognition probability of the suspended waveguide.

[0043] The present invention further proposes a complete processing flow and method for the original waveguide data: judging whether the data is valid, solving the atmospheric correction refractive index, judging whether a suspended waveguide appears in the data, how to solve the waveguide intensity after the waveguide appears, how to identify the waveguide characteristic parameters corresponding to the maximum intensity when a composite waveguide appears, and establishing a timetable to store data to complete the waveguide parameter calibration.

[0044] The present invention can provide data and technical support for marine radio meteorological analysis, ocean wireless communication support, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of a method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition of the present invention.

[0046] Figure 2 It is a flowchart for preprocessing the collected data to obtain the original time series data set.

[0047] Figure 3 It is a flowchart of processing the original time series data set using the CEEMDAN algorithm.

[0048] Figure 4 It is the training flow chart of the prediction model of suspended waveguide strength. DETAILED DESCRIPTION

[0049] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0050] The Chinese meanings of the following English words, phrases and abbreviations are as follows:

[0051] CEEMDAN: It is an improved algorithm based on the empirical mode decomposition algorithm. The empirical mode decomposition algorithm can decompose complex signals into several intrinsic mode functions, thereby realizing multi-scale analysis of the signal. However, the empirical mode decomposition algorithm is prone to modal aliasing when processing noisy signals.

[0052] IMF: IMF is the abbreviation of Intrinsic Mode Functions, which means intrinsic mode function. In empirical mode decomposition, a complex time series can be decomposed into multiple IMF components. Each IMF component represents a different frequency component or mode in the original data. These components are localized and physically meaningful patterns that can reveal the inherent dynamic characteristics and changing laws of the data. In the field of economics, IMF component analysis helps to better understand the volatility characteristics of economic data, thereby making more accurate predictions and decisions.

[0053] LSTM model: LSTM (Long Short-Term Memory) is a special RNN (Recurrent Neural Network) model that aims to solve the problem of gradient dispersion in RNN models. LSTM enhances the long-term memory capacity of RNN by introducing storage units to store memory.

[0054] See also Figures 1 to 4 A short-term prediction method for suspended waveguide strength based on CEEMDAN decomposition is provided, the method comprising the following steps:

[0055] Step 1, construct a prediction model for the strength of a suspended waveguide based on time series data, collect the associated parameter data of the suspended waveguide and preprocess the collected data to obtain an original time series data set for training the prediction model; the initial time series data set can be divided into a training set, a validation set and a test set.

[0056] Step 2: Use the CEEMDAN algorithm to decompose each data in the original time series data set into the sum of n IMF components and a residual signal. The IMF components obtained by decomposition are combined according to the time order of the data before decomposition, and n IMF component time series data sets are obtained from the original time series data set.

[0057] Step 3, sorting by the same time, extracting one IMF component time series data from each of the n IMF component time series data sets; inputting the extracted n IMF component time series data into a prediction model in sequence, or inputting them one by one into n prediction models with the same structure; superimposing and synthesizing the prediction results output by the prediction model to obtain a comprehensive prediction result of the suspended waveguide strength.

[0058] According to the same time sorting, one IMF component time series data is taken out from each of the n IMF component time series data sets; that is, according to the same start and end time sequence, one IMF component time series data is taken out from each of the n IMF component time series data sets synchronously, and the start and end time sequences of the taken out n IMF component time series data are the same.

[0059] The suspended waveguide associated parameter data collected by each sampling node can be taken as a vector data; a number of vector data are arranged in time sequence to compile an original time series data set.

[0060] Based on empirical mode decomposition, the CEEMDAN algorithm effectively suppresses the influence of noise on the decomposition results by introducing the ideas of adaptive noise and integrated averaging, thereby improving the accuracy and stability of signal decomposition.

[0061] Adaptability: CEEMDAN can adaptively adjust the intensity and distribution of noise according to the characteristics of the signal itself, thereby ensuring the accuracy of the decomposition results.

[0062] Completeness: Compared with the traditional empirical mode decomposition algorithm, CEEMDAN can more completely retain the effective information in the signal and avoid information loss.

[0063] Noise resistance: By introducing adaptive noise and integrated averaging methods, CEEMDAN is more robust when processing noisy signals.

[0064] Preferably, the CEEMDAN algorithm can be used to decompose the real-time collected parameter data of each suspended waveguide into the sum of n IMF components and a residual signal, and the time series data of the n IMF components obtained by decomposition are input into the trained prediction model, and the prediction results output by the prediction model are superimposed and synthesized to obtain a comprehensive prediction result of the suspended waveguide strength.

[0065] Preferably, a prediction model for the strength of the suspended waveguide can be constructed based on the LSTM model.

[0066] Preferably, the method for collecting the suspended waveguide associated parameter data in step 1 may include the following method steps:

[0067] The suspended waveguide associated parameters include the altitude, temperature, pressure, and relative humidity of the target area. The suspended waveguide associated parameter data are measured twice a day, each measurement lasts 1 to 2 hours, and the data sampling interval is 1 second.

[0068] Preferably, the method for preprocessing the collected suspended waveguide associated parameter data in step 1 may include the following method steps:

[0069] The atmospheric corrected refractive index values ​​at different altitudes are calculated from the collected data; if the atmospheric corrected refractive index value at a certain moment differs by 100 M-unit from that at the previous moment, or if the altitude detected decreases with time due to measurement problems with the detection instrument, the data at that moment is considered invalid and is filled with the data at the same moment of the previous day.

[0070] Preferably, the difference between the atmospheric corrected refractive index values ​​corresponding to the waveguide layer top height and the waveguide bottom height can be defined as the waveguide intensity. According to the characteristic that the atmospheric corrected refractive index decreases with increasing altitude under the condition that a suspended waveguide exists, it is judged whether there is a suspended waveguide on a certain day. If there is a suspended waveguide on that day, the waveguide intensity at the corresponding measurement time is calculated and recorded. If no suspended waveguide appears on that day, the data at the same time of the previous day is used to fill in the gap to ensure the continuity of the time series. At the same time, a timetable of the waveguide intensity is established to analyze the accuracy of the suspended waveguide associated parameter data at each time.

[0071] Assuming that the difference in the corrected refractive index corresponding to the top height of the waveguide layer and the bottom height of the waveguide layer is the waveguide strength, the calculation formula of the waveguide strength Strength is as follows:

[0072] Strength i =f(press,T,Rh,h) i

[0073] Where:

[0074] Strength i is the waveguide intensity of the i-th original time series data, in M-unit;

[0075] f() represents the suspended waveguide strength function;

[0076] T is temperature, in K;

[0077] press is the air pressure, in hpa;

[0078] h is the altitude above sea level, in meters;

[0079] Rh is the relative humidity of the atmosphere, in %.

[0080] The above calculation formula can adopt the calculation formula in the prior art, or use Matlab to implement arbitrary order nonlinear fitting based on historical data to obtain the suspended waveguide strength function.

[0081] Preferably, in step 1, the original time series data set can be divided into a training set and a validation set.

[0082] Preferably, in step 2, when the CEEMDAN algorithm is used to decompose the data in the original time series data set, a limited number of adaptive white noises can be added at each stage to suppress modal confusion and avoid distortion of the original signal, thereby achieving a smaller reconstruction error. The function expression of the CEEMDAN algorithm is as follows:

[0083] IMF i k =F C (Strength i ),k=1,…,n;

[0084] Where:

[0085] F C () represents the CEEMDAN algorithm function;

[0086] i is the serial number of the original time series data;

[0087] Strength iis the waveguide intensity of the i-th original time series data;

[0088] k is the serial number of the IMF component of the waveguide intensity of the i-th original time series data;

[0089] n is the number of IMF components decomposed from the original time series data;

[0090] IMF i k is the kth IMF component of the waveguide intensity of the i-th original time series data.

[0091] Preferably, step 3 may include the following sub-steps:

[0092] Step A1, inputting multiple IMF component time series data obtained by decomposing the i-th original time series data into the prediction model;

[0093] Step A2, adding up the multiple prediction result components output by the prediction model to obtain the waveguide intensity prediction value M of the i-th original time series data i p , M i p The calculation formula is as follows:

[0094]

[0095] Where:

[0096] F L () represents the prediction function of the prediction model;

[0097] F a () means adding multiple prediction result components;

[0098] F C () represents the CEEMDAN algorithm function;

[0099] m is the prediction component number output by the prediction model;

[0100] Step A3: After the model training is completed, the performance of the prediction model is evaluated using the validation set.

[0101] The present invention also provides a device for a method for short-term prediction of the strength of a suspended waveguide based on CEEMDAN decomposition, comprising a memory and a processor, wherein the memory is used to store a computer program; the processor is used to execute the computer program and implement the steps of the method for short-term prediction of the strength of a suspended waveguide based on CEEMDAN decomposition as described above when executing the computer program.

[0102] The following is a preferred embodiment of the present invention to further illustrate the working process and working principle of the present invention:

[0103] The present invention aims to predict the strength of the regional tropospheric suspended waveguide, and it is an indispensable process to establish a regional tropospheric suspended waveguide strength prediction model. The present invention uses the statistical learning method to determine the three elements of model, algorithm and strategy to realize the construction and training of the prediction model:

[0104] (1) Determine the data needed for predictive model training and verification. The premise of statistical learning is that the data used to train and verify the model have the same, certain statistical characteristics. The modeling process is to find a suitable model to describe this statistical characteristic.

[0105] (2) Determine the appropriate prediction model structure. This requires that before modeling, we should first have an in-depth understanding of the purpose of modeling, the principles and scope of application of different models, their advantages and disadvantages, and the data characteristics of the training data set.

[0106] (3) Determine the parameters of the prediction model. The present invention selects the traversal method to analyze and determine the relevant parameters and unknown coefficients of the regional tropospheric suspended waveguide prediction model.

[0107] (4) Evaluate the accuracy of the model. In order to determine the target model and its related parameters and unknown coefficients from the hypothesis space containing many models, it is necessary to formulate a unified standard to evaluate the model. The present invention selects the root mean square error and the relative root mean square error as the strategy evaluation model.

[0108] In summary, the present invention proposes a modeling idea for the tropospheric suspended waveguide prediction model through correlation analysis of the suspended waveguide strength prediction model, analyzes the suspended waveguide characteristics and establishes its corresponding time and space model.

[0109] See also Figure 1 A short-term prediction method for suspended waveguide strength based on CEEMDAN decomposition includes the following steps:

[0110] S1. Obtain the initial suspended waveguide data and preprocess it. During the processing, determine the content and storage format of the data used in the modeling process, ensure the continuity of the time series, and obtain standardized modeling data.

[0111] See also Figure 2 , S1 includes the following specific steps:

[0112] S101. Obtain the suspended waveguide data of the target area. The data is measured twice a day, each measurement lasts about 1.5 hours, the data sampling interval is 1s, and the measurement parameters include altitude h, temperature T, pressure press, atmospheric relative humidity Rh, etc. After obtaining the raw data, calculate the atmospheric corrected refractive index value M at different altitudes in each data according to these parameters.

[0113] The atmospheric refractive index is a parameter that corrects the atmospheric refractive index for the curvature of the earth. It is used to evaluate the atmospheric refractive index gradient and its impact on radio wave propagation. The specific value is determined by the atmospheric refractive index N, the altitude h, and the average earth radius r. e Sure.

[0114] S102. Perform data preprocessing. There may be some abnormal values ​​in the original measurement data. If the adjacent corrected refractive index values ​​M in a data item differ by 100 M-unit, or if the detection data height decreases with time due to measurement problems of the detection instrument, it is determined to be invalid data and filled with the data at the same time of the previous day.

[0115] S103. Waveguide layer top height h t and the waveguide bottom height h b The difference between the corresponding modified refractive index M is the waveguide strength Strength. According to the characteristic that the modified refractive index M decreases with the increase of height h when a suspended waveguide appears, it is judged whether there is a suspended waveguide on that day and the waveguide strength Strength is calculated. The specific expression of this process is:

[0116] Strength i =f(press,T,Rh,h) i ;

[0117] Where:

[0118] Strength i is the waveguide intensity of the i-th original time series data, in M-unit;

[0119] f() represents the suspended waveguide strength function;

[0120] T is temperature, in K;

[0121] press is the air pressure, in hpa;

[0122] h is the altitude above sea level, in meters;

[0123] Rh is the relative humidity of the atmosphere, in %.

[0124] If a waveguide exists, relevant parameters such as waveguide strength are recorded. If no waveguide appears on that date, the data at the same time of the previous day are used to fill in the gaps to ensure the continuity of the time series. At the same time, a waveguide timetable is established to analyze the accuracy of the suspended waveguide parameters at each moment.

[0125] S2. The modeling data obtained in step S1 is decomposed using the CEEMDAN algorithm to decompose the signal into the sum of several IMF components and a residual signal, and features are extracted from the decomposed IMF components. These features will be used for subsequent analysis and modeling.

[0126] See also Figure 3 , S2 includes the following specific steps:

[0127] S201. Determine a training data set and a validation data set, and select an appropriate ratio to divide the training set and the validation set so as to perform subsequent model training on the training set and performance evaluation on the validation set.

[0128] S202. The time series data is decomposed using the CEEMDAN algorithm. By adding a limited number of adaptive white noises at each stage, modal confusion is suppressed and distortion of the original signal is avoided, thereby achieving a smaller reconstruction error. The specific formula is expressed as follows:

[0129]

[0130] Where:

[0131] F C () represents the CEEMDAN algorithm function;

[0132] i is the serial number of the original time series data;

[0133] Strength i is the waveguide intensity of the i-th original time series data;

[0134] k is the serial number of the IMF component of the waveguide intensity of the i-th original time series data;

[0135] n is the number of IMF components decomposed from the original time series data;

[0136] IMF i k is the kth IMF component of the waveguide intensity of the i-th original time series data.

[0137] S3. Use the IMF time series obtained in step S2 to input them into the LSTM model for training, and superimpose several obtained prediction sequences to obtain the final prediction result.

[0138] See also Figure 4 , S3 includes the following specific steps:

[0139] S301. Bring the n IMF time series of waveguide intensity in the training set into the LSTM model for training.

[0140] S302. Superimpose several predicted components of each data to obtain the waveguide strength prediction value M of the i-th data i p , the specific expression of this process is:

[0141]

[0142] in,

[0143] F L () represents the prediction function of the prediction model;

[0144] F a () means adding multiple prediction result components;

[0145] F C () represents the CEEMDAN algorithm function.

[0146] S303. After the model is trained, the model performance index is calculated based on the measured values ​​and predicted values ​​of the validation set to evaluate the model performance.

[0147] The above-mentioned CEEMDAN algorithm, LSTM model and other functional modules, models and algorithms can all adopt applicable functional modules, models and algorithms of the existing technology, or adopt functional modules, models and algorithms of the existing technology and construct them by conventional technical means.

[0148] The embodiments described above are only used to illustrate the technical ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. The patent scope of the present invention cannot be limited only by this embodiment, that is, any equivalent changes or modifications made to the spirit disclosed by the present invention still fall within the patent scope of the present invention.

Claims

1. A short-term prediction method for suspended waveguide strength based on CEEMDAN decomposition, characterized in that: The method comprises the following steps: Step 1, constructing a prediction model for the strength of a suspended waveguide based on time series data, collecting the associated parameter data of the suspended waveguide and preprocessing the collected data to obtain an original time series data set for training the prediction model; Step 2: Use the CEEMDAN algorithm to decompose each data in the original time series data set into the sum of n IMF components and a residual signal. Combine the IMF components obtained by decomposition according to the time order of the data before decomposition, and obtain n IMF component time series data sets from the original time series data set. Step 3, sorting by the same time, extracting one IMF component time series data from each of the n IMF component time series data sets; inputting the extracted n IMF component time series data into a prediction model in sequence, or inputting them one by one into n prediction models with the same structure; superimposing and synthesizing the prediction results output by the prediction model to obtain a comprehensive prediction result of the suspended waveguide strength.

2. The method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition according to claim 1 is characterized in that: The CEEMDAN algorithm is used to decompose the real-time collected parameter data of each suspended waveguide into the sum of n IMF components and a residual signal. The time series data of the n IMF components obtained by decomposition are input into the trained prediction model. The prediction results output by the prediction model are superimposed and synthesized to obtain the comprehensive prediction result of the suspended waveguide strength.

3. The method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition according to claim 1 is characterized in that: A prediction model for the strength of suspended waveguide is constructed based on the LSTM model.

4. The method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition according to claim 1 is characterized in that: The method for collecting the suspended waveguide associated parameter data in step 1 includes the following method steps: The suspended waveguide associated parameters include the altitude, temperature, pressure, and relative humidity of the target area. The suspended waveguide associated parameter data are measured twice a day, each measurement lasts 1 to 2 hours, and the data sampling interval is 1 second.

5. The method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition according to claim 1 is characterized in that: The method for preprocessing the collected suspended waveguide associated parameter data in step 1 includes the following method steps: The atmospheric corrected refractive index values ​​at different altitudes are calculated from the collected data; if the corrected refractive index value at a certain moment differs by 100 M-unit from that at the previous moment, or if the altitude detected decreases with time due to measurement problems with the detection instrument, the data at that moment is considered invalid and is filled with the data at the same moment of the previous day.

6. The method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition according to claim 5 is characterized in that: The difference between the atmospheric corrected refractive index values ​​corresponding to the waveguide layer top height and the waveguide bottom height is defined as the waveguide intensity. Based on the characteristic that the corrected refractive index decreases with increasing altitude under the condition of the existence of a suspended waveguide, it is judged whether there is a suspended waveguide on a certain day; if there is a suspended waveguide on that day, the waveguide intensity at the corresponding measurement time is calculated and recorded; if no suspended waveguide appears on that day, the data at the same time of the previous day is used to fill in the gap to ensure the continuity of the time series; at the same time, a timetable of waveguide intensity is established to analyze the accuracy of the suspended waveguide associated parameter data at each time.

7. The method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition according to claim 1 is characterized in that: In step 1, the original time series dataset is divided into a training set and a validation set.

8. The method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition according to claim 7 is characterized in that: In step 2, when the CEEMDAN algorithm is used to decompose the data in the original time series data set, a limited number of adaptive white noises are added at each stage to suppress modal confusion and avoid distortion of the original signal, thereby achieving a smaller reconstruction error. The function expression of the CEEMDAN algorithm is as follows: IMF i k =F C (Strength i ),k=1,…,n; Where: F C () represents the CEEMDAN algorithm function; i is the serial number of the original time series data; Strength i is the waveguide intensity of the i-th original time series data; k is the serial number of the IMF component of the waveguide intensity of the i-th original time series data; n is the number of IMF components decomposed from the original time series data; IMF i k is the kth IMF component of the waveguide intensity of the i-th original time series data.

9. The method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition according to claim 8, characterized in that: Step 3 includes the following sub-steps: Step A1, inputting multiple IMF component time series data obtained by decomposing the i-th original time series data into the prediction model; Step A2, adding up the multiple prediction result components output by the prediction model to obtain the waveguide intensity prediction value M of the i-th original time series data i p , M i p The calculation formula is as follows: Where: F L () represents the prediction function of the prediction model; F a () means adding multiple prediction result components; F C () represents the CEEMDAN algorithm function; m is the prediction component number output by the prediction model; Step A3: After the model training is completed, the performance of the prediction model is evaluated using the validation set.

10. A device for a short-term prediction method of suspended waveguide strength based on CEEMDAN decomposition, comprising a memory and a processor, characterized in that: The memory is used to store a computer program; the processor is used to execute the computer program and implement the steps of the method for short-term prediction of suspended waveguide strength based on CEEMDAN decomposition as described in any one of claims 1 to 9 when executing the computer program.

Images