Method and apparatus of forecasting the atmospheric radio refractivity index based on lstm

Abstract

The present disclosure relates to a method and apparatus for predicting time-series atmospheric refractive index based on an LSTM. According to an embodiment of the present disclosure, the LSTM-based time-series atmospheric refractive index prediction method is a Long Short-Term Memory Network (LSTM)-based time-series atmospheric refractive index prediction method performed by a computing device including at least one processor, and may include the steps of: collecting weather environment data from at least one weather station; connecting the weather environment data in chronological order; processing the collected weather environment data into weather environment data of a specific altitude interval; calculating the atmospheric refractive index of a specific altitude interval using the processed weather environment data of the specific altitude interval; and inputting the calculated atmospheric refractive index into an LSTM model to generate a predicted atmospheric refractive index at a specific time point.

Description

Technology Field

[0001] The present disclosure relates to a method and apparatus for predicting atmospheric refractive index based on a recurrent neural network, and more specifically, to a method and apparatus for predicting atmospheric refractive index based on an LSTM, which processes meteorological observation data collected during a specific past period into the form of training data for a deep learning model and trains an LSTM neural network to predict atmospheric refractive index at a future point in time. Background Technology

[0002] Electromagnetic waves propagating through the atmosphere exhibit characteristics such as refraction and reflection at the boundary between two media with different refractive indices. The Earth's atmosphere, through which waves travel, consists of multiple layers with different refractive indices. The propagation characteristics of electromagnetic waves reaching the ground from a satellite or from a satellite through the atmosphere via satellite uplink and downlink communication are determined by the atmospheric refractive index of the Earth's atmosphere. While the precise value of the atmospheric refractive index can be determined by measuring temperature, pressure, and relative humidity using a rawinsonde, obtaining real-time measurements from every point presents difficulties due to physical and temporal limitations.

[0003] Therefore, a method is used to predict atmospheric refractive index at locations without observations by utilizing measurements from select points and spatial interpolation (Inverse Distance Weighting, IDW). However, using spatial interpolation results in lower accuracy compared to the method of measuring precise values ​​using radiosondes. Since the accuracy of predictions for unobserved points using spatial interpolation is proportional to the number of measurement data points used, values ​​measured at a wider variety of locations are required to improve the accuracy of spatial interpolation. The number of weather stations providing the meteorological observations necessary for atmospheric refractive index prediction is very limited.

[0004] In other words, in order to improve the accuracy of atmospheric refractive index predictions at conventional unobserved points, it is required to secure an LSTM model that increases the number of data used in spatial interpolation. The problem to be solved

[0005] To solve the conventional problem of predicting atmospheric refractive index, the present disclosure can solve this problem by utilizing past measurement data from weather stations that are no longer in operation. That is, by predicting future measurement values ​​using a deep learning model trained on the time-series patterns of measurements from weather stations that were in operation in the past or time-series data collected over a specific period in some regions, the number of measurement values ​​available for spatial interpolation can be increased. In order to obtain an accurate distribution of atmospheric refractive index, it is necessary to secure the number of measurement values ​​using deep learning, and to achieve this purpose, the present disclosure aims to provide a method and apparatus for predicting time-series atmospheric refractive index based on Long Short-Term Memory (LSTM).

[0006] The technical problems to be solved by the present disclosure are not limited to those mentioned above. Other objects and advantages of the present disclosure can be understood from the following description and will become more clearly apparent from the embodiments of the present disclosure. Furthermore, it will be readily apparent that the objects and advantages of the present disclosure can be realized by the means and combinations thereof set forth in the claims. means of solving the problem

[0007] A time-series atmospheric refractive index prediction method based on an LSTM according to one aspect of the present disclosure is a time-series atmospheric refractive index prediction method based on a Long Short-Term Memory Network (LSTM) performed by a computing device including at least one processor, and may include the steps of: collecting weather environment data from at least one weather station; connecting the weather environment data in chronological order; processing the collected weather environment data into weather environment data at specific altitude intervals; calculating the atmospheric refractive index at specific altitude intervals using the processed weather environment data at specific altitude intervals; and inputting the calculated atmospheric refractive index into an LSTM model to generate a predicted atmospheric refractive index at a specific time point.

[0008] In the LSTM-based time-series atmospheric refractive index prediction method according to the present disclosure, the meteorological environment data may include at least one of atmospheric pressure, temperature, relative humidity, wind speed, wind direction, snow, rain, clouds, and aerosols.

[0009] In the LSTM-based time-series atmospheric refractive index prediction method according to the present disclosure, the step of processing the weather environment data into weather environment data with specific altitude intervals may be to process the weather environment data having specific altitude intervals using a piecewise cubic Hermite interpolation polynomial.

[0010] In the LSTM-based time-series atmospheric refractive index prediction method according to the present disclosure, the atmospheric refractive index of a specific altitude interval can be calculated using MPM93 (Millimeter-wave Propagation Model) on meteorological environment data of a specific altitude interval.

[0011] In the LSTM-based time-series atmospheric refractive index prediction method according to the present disclosure, the atmospheric refractive index at a specific altitude interval can be calculated using the ITU-R model (International Telecommunication Union - Radiocommunication Model) on meteorological environment data at a specific altitude.

[0012] In the LSTM-based time-series atmospheric refractive index prediction method according to the present disclosure, prior to the step of generating the predicted atmospheric refractive index at the specific time point, the method may further include the step of generating training data from the calculated atmospheric refractive index and training an LSTM model that generates the predicted atmospheric refractive index at the specific time point.

[0013] An LSTM-based time-series atmospheric refractive index prediction device according to another aspect of the present disclosure may include a processor comprising a weather data receiving unit that receives weather environment data from at least one weather station, a computing device comprising at least one computer process including a step of connecting the weather environment data in chronological order or a step of processing the collected weather environment data into weather environment data at specific altitude intervals, and a time-series atmospheric refractive index generating unit that calculates the atmospheric refractive index at specific altitude intervals using the processed weather environment data at specific altitude intervals and inputs the calculated atmospheric refractive index into an LSTM model to generate the predicted atmospheric refractive index at a specific time point.

[0014] In the LSTM-based time-series atmospheric refractive index prediction device according to the present disclosure, the weather environment data may include at least one of atmospheric pressure, temperature, relative humidity, wind speed, wind direction, snow, rain, clouds, and aerosols.

[0015] In the LSTM-based time-series atmospheric refractive index prediction device according to the present disclosure, the meteorological environment data may include at least one of the meteorological environment data provided by the meteorological information provision portal of the Korea Meteorological Administration or the database of the Department of Meteorological Sciences webpage of the University of Wyoming.

[0016] In the LSTM-based time-series atmospheric refractive index prediction device according to the present disclosure, the step of processing the weather environment data into weather environment data with specific altitude intervals may be to process the weather environment data having specific altitude intervals using a piecewise cubic Hermite interpolation polynomial.

[0017] In the LSTM-based time-series atmospheric refractive index prediction device according to the present disclosure, the atmospheric refractive index of the specific altitude interval can be calculated using MPM93 (Millimeter-wave Propagation Model) on meteorological environment data of the specific altitude interval.

[0018] In the LSTM-based time-series atmospheric refractive index prediction device according to the present disclosure, the atmospheric refractive index of the specific altitude interval can be calculated using the ITU-R model (International Telecommunication Union - Radiocommunication Model) on meteorological environment data of the specific altitude. Effects of the invention

[0019] According to the LSTM-based time-series atmospheric refractive index prediction method and apparatus according to the embodiments of the present disclosure, the embodiments of the present disclosure can provide an atmospheric refractive index prediction method and apparatus capable of predicting atmospheric refractive index values ​​at unobserved points at future points in time based on LSTM using meteorological environment data measured by each of the meteorological stations.

[0020] In addition, according to the present disclosure, data applicable to spatial interpolation can be obtained to increase the accuracy of the prediction of atmospheric refractive index at unobserved points.

[0021] The various and beneficial advantages and effects of the present disclosure are not limited to those described above and may be more easily understood in the process of describing specific embodiments of the present disclosure. Brief explanation of the drawing

[0022] FIG. 1 is a simplified diagram illustrating an atmospheric refractive index prediction device according to the present disclosure communicating with each virtual observatory through a network. FIG. 2 is a simplified block diagram illustrating an LSTM-based time-series atmospheric refractive index generation device according to one embodiment of the present disclosure. FIG. 3 is a flowchart briefly illustrating the execution process of an LSTM-based time-series atmospheric refractive index prediction method according to the present disclosure. FIG. 4 is a graph showing the distribution of temperature, humidity, and relative humidity over time of meteorological environment data collected according to the present disclosure. FIG. 5 is a flowchart that simply illustrates the process of calculating the atmospheric refractive index at specific altitude intervals according to the present disclosure. FIG. 6 is a diagram illustrating the process of generating training data for an LSTM network using a sliding window technique according to the present disclosure. FIG. 7 is a diagram showing the structure of an LSTM network memory cell according to the present disclosure. FIG. 8 is a diagram showing the loss function of a training set and a validation set according to the number of training iterations according to an embodiment of the present disclosure. Specific details for implementing the invention

[0023] Hereinafter, embodiments of the present disclosure are described in detail with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present disclosure may be embodied in various different forms and is not limited to the embodiments described herein.

[0024] In describing the embodiments of the present disclosure, if it is determined that a detailed description of known configurations or functions could obscure the essence of the present disclosure, such detailed description is omitted. Additionally, parts of the drawings unrelated to the description of the present disclosure have been omitted, and similar parts are denoted by similar reference numerals.

[0025] In the present disclosure, when a component is described as being "connected," "combined," or "joined" with another component, this may include not only a direct connection but also an indirect connection in which another component exists in between. Furthermore, when a component is described as "comprising" or "having" another component, this means that, unless specifically stated otherwise, it does not exclude the other component but may include additional components.

[0026] In the present disclosure, terms such as first, second, etc. are used solely for the purpose of distinguishing one component from another component and do not limit the order or importance of the components unless specifically stated otherwise. Accordingly, within the scope of the present disclosure, a first component in one embodiment may be referred to as a second component in another embodiment, and likewise, a second component in one embodiment may be referred to as a first component in another embodiment.

[0027] In this disclosure, distinct components are intended to clearly describe their respective features and do not imply that the components are separate. That is, multiple components may be integrated to form a single hardware or software unit, or a single component may be distributed to form multiple hardware or software units. Accordingly, such integrated or distributed embodiments are included within the scope of this disclosure, even if not otherwise mentioned.

[0028] In the present disclosure, each of the phrases such as “A or B”, “at least one of A and B”, “at least one of A or B”, “A, B or C”, “at least one of A, B and C”, and “at least one of A, B, C or combination thereof” may include any one of the items listed together in the corresponding phrase, or all possible combinations thereof.

[0029] In the present disclosure, the components described in various embodiments do not necessarily mean essential components, and some may be optional components. Accordingly, embodiments consisting of a subset of the components described in one embodiment are also included within the scope of the present disclosure. Furthermore, embodiments including other components in addition to the components described in various embodiments are also included within the scope of the present disclosure.

[0030] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments presented below but can be implemented in various different forms, and these embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention.

[0031] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached pages. In describing the embodiments of this specification, if it is determined that a detailed description of related known configurations or functions may obscure the gist of this specification, such detailed description will be omitted, and the same reference numerals will be used for identical components in the drawings, and redundant descriptions of identical components will be omitted.

[0032] FIG. 1 is a simplified diagram illustrating an atmospheric refractive index prediction device according to the present disclosure communicating with each virtual observatory through a network.

[0033] As illustrated in FIG. 1, the atmospheric refractive index prediction device (100) can collect weather data from at least one weather station (110, 120, 130, 140, 150) through a network.

[0034] The communication method between the above atmospheric refractive index prediction device (100) and the weather station (110, 120, 130, 140, 150) is not limited and, for example, a mobile communication network, wired internet, wireless internet, broadcasting network, satellite network, etc. may be utilized. In addition, it is not limited thereto and may also include short-range wireless communication between devices. As an example, the network may include any one or more network topologies including a bus network, a star network, a ring network, a mesh network, a star-bus network, a tree or hierarchical network, but is not limited thereto. In addition, the network may include a PAN (Personal Area Network), LAN (Local Area Network), CAN (Campus Area Network), MAN (Metropolitan Area Network), WAN (Wide Area Network), BBN (Broadband Network), etc.

[0035] Weather stations (110, 120, 130, 140, 150) can measure weather data and store the measured weather data along with the date, time, location, etc. In addition, weather stations (110, 120, 130, 140, 150) can process or transform the information described above into other forms. Weather stations (110, 120, 130, 140, 150) can collect weather data at various altitudes, such as the troposphere and stratosphere, by launching weather observation equipment, such as a radiosonde, from the surface of the earth twice a day. Weather stations (110, 120, 130, 140, 150) can collect weather data at predetermined time intervals using radiosondes, and the weather data may include at least one of temperature, atmospheric pressure, partial pressure of water vapor, dew point, relative humidity, wind speed, wind direction, snow, rain, clouds, and aerosols according to altitude.

[0036] The above meteorological data can be utilized as LSTM-based training data for predicting atmospheric refractive index according to the present disclosure. That is, the atmospheric refractive index prediction device can predict the atmospheric refractive index at a future point in time through the trained LSTM, and can increase the accuracy of atmospheric refractive index prediction for points where observed meteorological data does not exist by using data on the augmented atmospheric refractive index. This will be explained in detail through FIG. 3.

[0037] An atmospheric refractive index prediction device (100) can predict the atmospheric refractive index at a specific point in time by processing weather data measured over a certain period by at least one weather station (110, 120, 130, 140, 150) into weather data at specific altitude intervals in a time series order, calculating the atmospheric refractive index at specific altitude intervals using the processed weather data, and inputting the atmospheric refractive index into an LSTM neural network to generate the predicted atmospheric refractive index at a specific point in time.

[0038] FIG. 2 is a simplified block diagram illustrating an LSTM-based time-series atmospheric refractive index generation device according to one embodiment of the present disclosure.

[0039] Referring to FIG. 2, an LSTM-based time-series atmospheric refractive index generating device (200) may include a weather data receiving unit (210), an AI server (220) composed of a computing device including at least one computer process, a time-series atmospheric refractive index generating unit (230), and a memory (240). Each of the above components is not an essential component and may be omitted by including one or more components. Additionally, it may be configured to include other components.

[0040] The weather data receiving unit (210) can receive weather data from the atmospheric refractive index prediction device (100) and weather stations (110, 120, 130, 140, 150) through a network as seen in FIG. 1.

[0041] The AI ​​server (220) can perform overall control of the atmospheric refractive index generating device (200). The AI ​​server (220) can be configured to execute instructions and applications stored in memory (240). In connection with the present disclosure, the AI ​​server (220) can train an LSTM neural network to generate a predicted atmospheric refractive index using instructions, applications, and data stored in memory (240), and can generate an atmospheric refractive index at a specific point in time in the time-series atmospheric refractive index generating unit (230) through the trained LSTM neural network.

[0042] The memory (240) stores applications and various data for controlling the LSTM-based time-series atmospheric refractive index generator (200), and can load applications or read and write data upon the request of the AI ​​server (220). The memory (240) may be included as a component of the atmospheric refractive index generator (200) or may be located separately outside the atmospheric refractive index generator (200).

[0043] FIG. 3 is a flowchart briefly illustrating the execution process of an LSTM-based time-series atmospheric refractive index prediction method according to the present disclosure.

[0044] Referring to Figure 3, the LSTM-based time-series atmospheric refractive index prediction method is a method that processes meteorological observation data collected during a specific past period into the form of training data for a deep learning model and trains an LSTM neural network to predict the atmospheric refractive index at a future point in time.

[0045] First, meteorological environment data measured from specific weather stations from a distant past point in time up to the day to which the value is to be predicted is collected (S310). Specific weather stations provide meteorological environment data by measuring atmospheric conditions at specific time intervals up to an altitude of 30 km by launching radiosondes from the surface twice a day. As an example according to the present disclosure, the number of meteorological data available from the weather observation start date (March 31, 1957) to the collection end date (December 2, 2023) at High-Air Weather Station No. 47122 located in Osan may be 52,832. Subsequently, the collected meteorological environment data is connected in chronological order (S320).

[0046] The same specific formula is applied to the measured weather environment data for each date to process the data so that it has the same altitude interval, thereby ensuring that the weather environment data collected for each date has values ​​at the same altitude interval (S330). Once the processing into weather environment data with specific altitude intervals is completed by the above method, the atmospheric refractive index at a specific altitude is calculated from the processed weather environment data with specific altitude intervals (S340).

[0047] Observed meteorological data possesses the characteristics of time-series data, where each data point is recorded sequentially at regular time intervals. A suitable deep learning model for learning the patterns of such time-series data is a Recurrent Neural Network (RNN), and an RNN is fundamentally a Multi-Layered Perceptron (MLP) structure in which the previous time (t n-1 The hidden layer output of the ) step is the next time (t n It has the function of remembering past information in a form in which a path reused as an input to the hidden layer of ) is added. Long Short Term Memory (LSTM) is a neural network in which a gate mechanism is added to an RNN, and since long-term memory ability is enhanced and the vanishing gradient problem of RNNs can be mitigated, it may be more suitable for learning long-term patterns of time-series data. Because the characteristics of such LSTM neural networks are suitable for learning long-term patterns of atmospheric refractive indices accumulated over decades, the present disclosure may use a neural network with an LSTM structure for time-series prediction of atmospheric refractive indices.

[0048] Training data for an LSTM neural network can be obtained by processing time-series data such that the correct answer is labeled according to the input and the input variables and targets are configured. An LSTM neural network can be trained using the above training data (S350). Finally, the atmospheric refractive index value at an unobserved point can be predicted more accurately through the trained LSTM neural network (S360).

[0049] FIG. 4 is a graph showing the distribution of temperature, humidity, and relative humidity over time of meteorological environment data collected according to the present disclosure.

[0050] Referring to Fig. 4, meteorological environment data is a visualization of all data collected from a specific observation station from the start of observation to the end of collection over time. The meteorological environment data may include temperature, atmospheric pressure, humidity, dew point, relative humidity, etc., according to altitude, and among them, graphs of temperature (410), humidity (Atmospheric Pressure, 420), and relative humidity (Relative Humidity, 430) according to a continuous time sequence (Time Sequence, 440) are shown. This can be used as data to be processed into meteorological environment data at specific altitude intervals, as in Fig. 3.

[0051] FIG. 5 is a flowchart that simply illustrates the process of calculating the atmospheric refractive index at specific altitude intervals according to the present disclosure.

[0052] Referring to FIG. 5, the same specific formula is applied to the measured weather environment data for each date to process the data so that it has the same altitude interval, thereby ensuring that the weather environment data collected for each date has values ​​at the same altitude interval (S510). The specific formula may be a Piecewise Cubic Hermite Interpolation Polynomial (PCHIP). PCHIP is one of the mathematical techniques for performing interpolation between data, and it can divide the intervals and use a third-order Hermite polynomial in each interval to perform interpolation. The coefficients of the polynomial in each interval may be determined by the function value and derivative value at the endpoint of the corresponding interval.

[0053] When the processing of meteorological environment data with specific altitude intervals is completed using the above method, the atmospheric refractive index at a specific altitude is calculated from the processed meteorological environment data with specific altitude intervals (S520). Here, the atmospheric refractive index can be calculated, for example, as shown in Equation 1 below, by applying the meteorological environment data at a specific altitude to the International Telecommunication Union Radiocommunication Sector (ITU-R) model.

[0054] [Mathematical Formula 1]

[0055] N=77.6 +3.73

[0056] In the above mathematical formula 1, N represents the atmospheric refractive index [N-unit], P represents the atmospheric pressure [mbar], T represents the temperature [K], and e represents the partial pressure of water vapor [mbar].

[0057] FIG. 6 is a diagram illustrating the process of generating training data for an LSTM network using a sliding window technique according to the present disclosure.

[0058] Referring to Figure 6, the training data for an LSTM network can be processed using the Sliding Window Technique. The Sliding Window Technique is one of the techniques used in data processing, which involves applying a fixed-size window to a data stream to process data in sequence over a certain interval. Types of sliding windows may include fixed-size sliding windows, variable-size sliding windows, and overlap sliding windows.

[0059] The time series raw data (600) represents a graph of temperature over a continuous period, which is one of the meteorological environment data. At this time, the size of the window can change dynamically (610). Using the above data, the entire dataset for machine learning can be selected (620). More specifically, the time series data collected from the observation start date to the collection end date at a specific observation station can be generated into multiple data with a specific number of input windows (630) and output windows (640) using a sliding window method. In a supervised learning problem, the input window (630) can be the input data (feature) used when training the model, and the output window (640) can be the value (target) that the model intends to predict or classify.

[0060] FIG. 7 is a diagram illustrating the structure of an LSTM network memory cell according to the present disclosure. The gate of the first stage of the LSTM is called the forget gate layer, and in this layer, what information to discard from the cell state can be determined by the sigmoid layer, which is the forget gate layer. Input value X at time step (t). t (702) and the hidden state (h) from the previous step (t-1) t-1 Calculating the input state with , 703) as input gives the output(f t , 701) can be calculated, for example, as in the following mathematical formula 2.

[0061] [Mathematical Formula 2]

[0062] f t =σ(W f [h t-1 , x t ]+b f )

[0063] In the second step, it is determined which of the incoming information will be stored in the cell state. At this point, the sigmoid layer, which acts as the input gate layer, determines which value to update; calculating this yields the output (i t , 705) can be calculated, for example, as in the following mathematical formula 3.

[0064] [Mathematical Formula 3]

[0065] i t =σ(W i [h t-1 , x t ]+b i )

[0066] In the tanh layer, go through tanh(706) t A vector of (707) is created, and the state is updated by combining the information from the second step. Output t (707) can be calculated, for example, as in the following mathematical formula 4.

[0067] [Mathematical Formula 4]

[0068] t =tanh(W c [h t-1 , x t ]+b c )

[0069] C, the past state t-1 Update (704) to the new cell state C t The relationship of (709) can be calculated, for example, as shown in the following mathematical formula 5.

[0070] [Mathematical Formula 5]

[0071] C t =f t C t-1 +i t t

[0072] Finally, the output gate layer, which is responsible for the step of deciding what to output, feeds the input into the sigmoid layer to determine which part of the cell state to output, and the output (h t , 708, 710) can be calculated, for example, as in the following mathematical formula 6.

[0073] [Mathematical Formula 6]

[0074] o t =σ(W o [h t-1 , x t ]+b o )

[0075] h t =o t tanh(C t )

[0076] As such, LSTM neural networks enable long-term prediction by considering short-term information as well as old past information.

[0077] FIG. 8 is a diagram showing the loss function of a training set and a validation set according to the number of training iterations according to an embodiment of the present disclosure.

[0078] The direction of learning for the connection weights of each node undergoing an update can be determined by the learning rate. The neural network's computation on the training data and the backpropagation of errors can constitute a single training cycle (epoch). An epoch in a machine learning model refers to the period during which the entire training dataset is trained once. The learning rate is applied differently depending on the number of iterations in the neural network's training cycle.

[0079] In the training of LSTM neural networks, the training data is generally a subset of the actual data (i.e., the data intended to be processed using the trained network). Consequently, a training cycle may exist where errors decrease on the training data but increase on the actual data. Overfitting is a phenomenon where the model learns excessively on the training data, leading to increased errors on the actual data; it can act as a cause for increased algorithmic errors. To prevent such overfitting, methods such as regularization, dropout (which intentionally disables some nodes in the network during training), or the use of batch normalization layers can be applied. To avoid underfitting or overfitting models during the evaluation of the trained model, the final model can be selected through cross-validation.

[0080] For cross-validation, the training set can be divided into 90% of the total training data, a validation set, and a test set. To prevent overfitting or underfitting, the number of training iterations can be determined by the intersection point (809) of the loss (801) graphs of the training set (807) and the validation set (805) according to the change in Epoch (803). To prevent overfitting or underfitting, the number of training iterations can be determined as 141, which is the intersection point (809) of the two graphs. The prediction performance showed the best performance when the input window size (N) was 20 and the output window size (T) was 4. Accordingly, the optimized LSTM neural network can have a structure with 20 nodes in the input layer, 4 nodes in the output layer, and 30 nodes in the hidden layer. In other words, the neural network operates by using 20 days' worth of refractive indices as input values ​​in the input layer, and outputting 4 days' worth of refractive indices calculated through 30 hidden layers (drop out=0.2) in the output layer. The optimizer used for training is RMSprop, the loss function is MSE (Mean Squared Error), batch size=32, and learning rate=0.001.

[0081] As described above, the time-series atmospheric refractive index prediction method based on an LSTM neural network according to the present disclosure has the advantage of predicting the atmospheric refractive index at a future point in time by processing weather environment data collected during a specific past period into the form of training data for a deep learning model and training an LSTM neural network with it.

[0082] The embodiments described above include examples of various forms. While it is not possible to describe all possible combinations for representing various forms, those skilled in the art will recognize that other combinations are possible. Accordingly, the present invention shall be deemed to include all other substitutions, modifications, and changes falling within the scope of the following claims.

[0083] Although specific details such as specific components of the present disclosure and limited embodiments and drawings have been described above, this is provided only to aid in a more comprehensive understanding of the present disclosure, and the present disclosure is not limited to the above embodiments, and a person skilled in the art to which the present invention pertains can make various modifications and variations from this description.

[0084] Accordingly, the scope of the present disclosure is not limited to the embodiments described above, and the claims set forth below and all equivalent or equivalent modifications thereof shall be considered to fall within the scope of the concept of the present invention.

Claims

Claim 1 A Long Short-Term Memory Network (LSTM)-based time-series atmospheric refractive index prediction method performed by a computing device comprising at least one processor, comprising: collecting weather environment data from at least one weather station; connecting the weather environment data in chronological order; processing the collected weather environment data into weather environment data of a specific altitude interval; calculating the atmospheric refractive index of a specific altitude interval using the processed weather environment data of the specific altitude interval; and inputting the calculated atmospheric refractive index into an LSTM model to generate a predicted atmospheric refractive index at a specific time point. Claim 2 A time-series atmospheric refractive index prediction method based on LSTM according to claim 1, wherein the meteorological environment data includes at least one of atmospheric pressure, temperature, relative humidity, wind speed, wind direction, snow, rain, clouds, and aerosols. Claim 3 In claim 1, the step of processing the weather environment data into weather environment data with specific altitude intervals is to process the weather environment data having specific altitude intervals using a piecewise cubic Hermite interpolation polynomial, an LSTM-based time-series atmospheric refractive index prediction method. Claim 4 In claim 1, an LSTM-based time-series atmospheric refractive index prediction method that calculates the atmospheric refractive index of a specific altitude interval using MPM93 (Millimeter-wave Propagation Model) on meteorological environment data of a specific altitude interval. Claim 5 In claim 1, an LSTM-based time-series atmospheric refractive index prediction method that calculates the atmospheric refractive index of a specific altitude interval using the ITU-R model (International Telecommunication Union - Radiocommunication Model) on meteorological environment data of a specific altitude. Claim 6 An LSTM-based time-series atmospheric refractive index prediction method according to claim 1, further comprising, prior to the step of generating the predicted atmospheric refractive index at the specific point in time, the step of generating training data from the calculated atmospheric refractive index to train an LSTM model that generates the predicted atmospheric refractive index at the specific point in time. Claim 7 An LSTM-based time-series atmospheric refractive index prediction device comprising: a weather data receiving unit that receives weather environment data from at least one weather station; a processor comprising a computing device including at least one computer process among the steps of connecting the weather environment data in chronological order or processing the collected weather environment data into weather environment data of a specific altitude interval; and a time-series atmospheric refractive index generating unit that calculates the atmospheric refractive index of a specific altitude interval using the processed weather environment data of a specific altitude interval and inputs the calculated atmospheric refractive index into an LSTM model to generate a predicted atmospheric refractive index at a specific time point. Claim 8 In claim 7, the above meteorological environment data includes at least one of atmospheric pressure, temperature, relative humidity, wind speed, wind direction, snow, rain, clouds, and aerosols, an LSTM-based time-series atmospheric refractive index prediction device. Claim 9 In claim 8, the above-mentioned meteorological environment data includes meteorological environment data provided by the Korea Meteorological Administration’s weather information provision portal, an LSTM-based time-series atmospheric refractive index prediction device. Claim 10 In claim 7, the step of processing the weather environment data into weather environment data with specific altitude intervals is to process the weather environment data having specific altitude intervals using a piecewise cubic Hermite interpolation polynomial, an LSTM-based time-series atmospheric refractive index prediction device. Claim 11 In claim 7, an LSTM-based time-series atmospheric refractive index prediction device that calculates the atmospheric refractive index of the specific altitude interval using MPM93 (Millimeter-wave Propagation Model) on meteorological environment data of the specific altitude interval. Claim 12 In claim 7, an LSTM-based time-series atmospheric refractive index prediction device that calculates the atmospheric refractive index of the specific altitude interval using the ITU-R model (International Telecommunication Union - Radiocommunication Model) on meteorological environment data of the specific altitude.

Images