A satellite access traffic prediction method based on LSTM and access strategy

Abstract

The application discloses a satellite access traffic prediction method based on LSTM and access strategy and belongs to the technical field of communication.The application regards the satellite access traffic of a certain ground coverage area as time series data, and utilizes a long short-term memory network model to predict the traffic at the next moment.In addition, the application analyzes the satellite access strategy of the ground terminal, that is, analyzes which satellite the ground traffic accesses, and distributes the ground traffic to the corresponding satellite, so that the satellite traffic prediction problem is effectively solved.The application is aimed at the problems of low prediction accuracy or high prediction complexity caused by the fact that the periodicity of the satellite-to-ground movement is not obvious or too long, effectively predicts the ground fixed area first, and then utilizes the satellite movement and satellite traffic access characteristics to distribute the ground traffic to the satellite nodes, so that simple, accurate and efficient prediction is realized.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and specifically refers to a satellite access traffic prediction method based on LSTM and access strategy. Background Technology

[0002] As the ITU and 3GPP gradually advance the development of non-terrestrial network technologies in areas such as frequency allocation, standards development, and technological research, satellite networks, represented by low-Earth orbit satellite constellations, will break through the limitations of terrestrial networks as supplementary technologies and become an important component of future networks, playing a significant role in remote areas, uninhabited regions, and aviation and maritime sectors. However, on-board resources such as frequency, power, and computing power are subject to significant limitations and cannot be used as freely as terrestrial networks; therefore, the rational and optimized use of on-board resources is essential.

[0003] Meanwhile, with the development of intelligent technologies, deep learning and other methods can effectively handle time series forecasting with high accuracy. An efficient and accurate forecasting mechanism may help satellite networks allocate and optimize resources. However, low-Earth orbit satellites typically orbit at altitudes of 200-2000 km, with an orbital period of 90 to 120 minutes. Since their nadir does not return to its starting point, the satellite's motion around the Earth can be considered a non-periodic or extremely long-period event. Therefore, using time series data to predict satellite access traffic may lead to two problems: low accuracy or exceptionally high computational complexity. Currently, a more reasonable and accurate forecasting mechanism has not yet been developed. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a satellite access traffic prediction method based on LSTM and access strategy. This method achieves access traffic allocation by reasonably and accurately predicting satellite access traffic in the ground area and combining it with the characteristics of satellite access in the area, thus realizing an access traffic prediction mechanism for low-Earth orbit satellites.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A satellite access traffic prediction method based on LSTM and access policies includes the following steps:

[0007] Step 1: Divide all satellite coverage areas into N sub-regions, and establish a set of sub-regions based on the N sub-regions. Where R N This represents the Nth block region;

[0008] Step 2: Traverse the set of partitioned regions When the segmented region R is selected i At that time, for the segmented region R iThe amount of uploaded satellite data within the specified time is sampled and recorded M times at equal time intervals. A set of uploaded satellite data is then established based on the recorded M data volumes. in This indicates that at the Mth sampling, the block region R i The amount of satellite data uploaded within the specified range, r, is 1 ≤ i ≤ N;

[0009] Step 3: Construct a long short-term memory network model, dividing the block region R... i The set of uploaded satellite data is input into the Long Short-Term Memory network model to obtain the block region R for the next time step. i The predicted amount of satellite data to be uploaded within the scope

[0010] Step 4: Based on the satellite data for the segmented region R i Pitch angle, communication signal strength, and communication connection time are used to filter the block region R. i Access satellites within the country;

[0011] Step 5: According to the satellite access strategy, divide the segmented area R i The predicted amount of satellite data to be uploaded within the scope Assigned to block region R i All satellites connected within the region;

[0012] Step 6: Calculate the distribution of predicted satellite data volume in all segmented regions, and sum the predicted satellite data volume uploaded to the same satellite to finally complete the access traffic prediction for different satellites.

[0013] Furthermore, the specific method for constructing the Long Short-Term Memory network model in step 3 is as follows:

[0014] (301) Collect public datasets based on MAWI workgroup traffic archives to obtain multiple sets of network traffic data as training datasets;

[0015] (302) Construct a long short-term memory network model, which includes a forgetting gate, an input gate, and an output gate;

[0016] The formula for calculating the forgetting gate is as follows:

[0017] f t =σ(W f [h t-1 ,x t ]+b f )

[0018] Where σ is the Sigmoid activation function, W f Let h be the weight matrix of the forget gate. t-1 x represents the neuron state of the forget gate at the previous moment.t b is the input value at the current moment. f Forgetting bias term; f t f represents the degree to which information is retained in network neurons. t ∈[0,1];

[0019] The formula for calculating the input gate is:

[0020] i t =σ(W i [h t-1 ,x t ]+b i )

[0021]

[0022]

[0023] Among them, i t b is an intermediate variable. i and b c For the input bias term, W i W is the first input weight matrix. c This is the second input weight matrix. c represents the candidate values ​​selected by the tanh activation function. t-1 This represents the neuron state of the input gate at the previous time step;

[0024] The formula for calculating the output gate is as follows:

[0025] o t =σ(W[h t-1 ,x t ]+b o )

[0026] h t =o t tanh(c t )

[0027] Among them, o t b represents the output information filtered through the Sigmoid activation function. o The output bias term is given by W, and the output weight matrix is ​​given by h. t This is the final output of the Long Short-Term Memory network model;

[0028] (303) Based on the output h of the Long Short-Term Memory Network Model t and the next time step data y in the training dataset t Calculate the mean absolute error loss L pre :

[0029]

[0030] Where K is the number of input data in one training process, K = M;

[0031] Based on the average absolute error loss L pre The parameters of the Long Short-Term Memory network model are updated using backpropagation until the mean absolute error loss L is reached. pre Upon convergence, training ends and the parameters of the Long Short-Term Memory network model are saved, completing the deep learning training.

[0032] Furthermore, step 4 is specifically implemented as follows:

[0033] (401) Based on the block region R i Ground information and satellite ephemeris information are used to obtain the R block region. i Information on Y satellites within the visible range is denoted as

[0034] (402) in Of the Y satellites, select the region R. i During communication, satellites with a communication elevation angle greater than the lowest elevation angle are added to the satellite ensemble. middle, and

[0035] (403) in e Y′ Among the satellites, select the block region R i During communication, satellites with communication signal power greater than the minimum communication power are added to the satellite ensemble. middle, and

[0036] (404) in p Y” Among the satellites, select the block region R i During communication, satellites whose communication connection time is longer than the time required for a single satellite data upload are added to the satellite set. middle, and

[0037] Furthermore, step 5 is specifically implemented as follows:

[0038] (501)Judgment Is it equal to 1? If so, then... All uploaded to satellite collection Satellites in the middle, otherwise the region R will be divided. i It is divided into J multi-satellite overlay coverage sub-regions and G single-satellite independent coverage sub-regions;

[0039] Traverse G single-satellite independent coverage sub-regions, when a single-satellite independent coverage sub-region is selected. At that time, a single satellite will independently cover a sub-region. The area of ​​the block region R i area ratio multiplied by Obtain independent coverage sub-regions for a single satellite The predicted amount of satellite data to be uploaded to the single access satellite. in Represents the block region R i The g-th independent coverage sub-region of a single satellite within the region, 1≤g≤G;

[0040] Traverse J multi-satellite overlay coverage sub-regions, when a multi-satellite overlay coverage sub-region is selected. At that time, multiple satellites will be superimposed to cover the sub-region. The area of ​​the block region R i area ratio multiplied by Obtain multi-satellite overlay coverage sub-region Internal upload to multiple access satellites Predicted amount of satellite data to be uploaded Represents the block region R i The j-th multi-satellite overlay coverage sub-region within the region, 1≤j≤J;

[0041] For multi-satellite overlay coverage sub-regions If the satellite adopts a uniformly distributed random access strategy, then proceed to step (502); if the satellite adopts an elevation angle optimal access strategy, then proceed to step (503); if the satellite adopts a communication power optimal access strategy, then proceed to step (504).

[0042] (502) Overlay multiple satellites to cover the sub-region The predicted amount of satellite data to be uploaded within the scope Equally distributed among multiple access satellites

[0043] (503) Overlay multiple satellites to cover the sub-region The system is divided into multiple elevation angle strategy regions, each of which is connected to a satellite with the optimal elevation angle. The area of ​​each elevation angle strategy region is then superimposed with the coverage area of ​​multiple satellites to form a sub-region. The proportion of the area is multiplied by respectively Obtain the strategy area for each elevation angle in the docking satellite in the multi-satellite overlay coverage sub-region. The predicted amount of satellite data to be accessed within the region;

[0044] (504) Overlay multiple satellites to cover the sub-region The system is divided into multiple communication power strategy regions, each of which is connected to a satellite with the highest communication power. The area of ​​each communication power strategy region is then superimposed with the coverage area of ​​multiple satellites. The proportion of the area is multiplied by respectively Obtain the communication power strategy area docking satellite in the multi-satellite superimposed coverage sub-region. The predicted amount of satellite data to be accessed within the region.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. To address the problem of low prediction accuracy or excessive prediction complexity caused by the high dynamism of low-orbit satellites, this invention adopts a method of dividing the region. Based on the characteristic that traffic prediction within a fixed region is basically a time series, and combined with satellite access characteristics, traffic in different regions is allocated to different satellites, thereby solving the above problems.

[0047] 2. This invention proposes different traffic allocation models for different access strategies, such as random access, optimal elevation angle, and optimal signal power. By analyzing different access strategies, the accuracy and applicability of this invention are improved. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of low-orbit satellite traffic access in an embodiment of the present invention. Detailed Implementation

[0049] The present invention will now be described in further detail with reference to the accompanying drawings.

[0050] A satellite access traffic prediction method based on LSTM and access strategy, such as Figure 1 The image shows a low-Earth orbit (LEO) satellite access scenario. Typically, LEO satellites are located at an altitude of 200-2000 kilometers and orbit the Earth once every 90 to 120 minutes. From the perspective of the nadir point, the satellite does not return to its starting point (the nadir point's position on the Earth's surface) after orbiting the Earth once. Moreover, it usually does not have a stable periodicity, or the periodicity is long, resulting in low accuracy or a large computational load.

[0051] Typically, satellite network traffic within a fixed area is a periodic time series, usually with a period of one day (24 hours). In this case, it is usually possible to predict the network traffic in that area quite well. At the same time, by querying ephemeris data to obtain satellite position information, the set of satellites covering the area can be obtained. Combined with satellite access strategies, the access traffic generated in this area can be accurately distributed to different satellites, thereby achieving accurate prediction of satellite traffic.

[0052] Specifically, this embodiment includes the following steps:

[0053] Step 1: Divide all satellite coverage areas into N sub-regions, and establish a set of sub-regions based on the N sub-regions. Where R N This represents the Nth block region;

[0054] Step 2: Traverse the set of partitioned regions When the segmented region R is selected i At that time, for the block region R in the most recent period i The amount of uploaded satellite data within the specified time is sampled and recorded M times at equal time intervals. A set of uploaded satellite data is then established based on the recorded M data volumes. in This indicates that at the Mth sampling, the block region R i The amount of satellite data uploaded within the specified range, r, is 1 ≤ i ≤ N;

[0055] Step 3: Construct a long short-term memory network model, dividing the block region R... i The set of uploaded satellite data is input into the Long Short-Term Memory network model to obtain the block region R for the next time step. i The predicted amount of satellite data to be uploaded within the scope

[0056] Step 4: Based on the satellite data for the segmented region R i Pitch angle, communication signal strength, and communication connection time are used to filter the block region R. i Access satellites within the country;

[0057] Step 5: According to the satellite access strategy, divide the segmented area R i The predicted amount of satellite data to be uploaded within the scope Assigned to block region R i All satellites connected within the region;

[0058] Step 6: Calculate the distribution of predicted satellite data volume in all segmented regions, and sum the predicted satellite data volume uploaded to the same satellite to finally complete the access traffic prediction for different satellites.

[0059] Furthermore, the specific method for constructing the Long Short-Term Memory network model in step 3 is as follows:

[0060] (301) Collect public datasets based on MAWI workgroup traffic archives to obtain multiple sets of network traffic data as training datasets;

[0061] (302) Construct a long short-term memory network model, which includes a forgetting gate, an input gate, and an output gate;

[0062] The formula for calculating the forgetting gate is as follows:

[0063] f t =σ(W f [h t-1 ,x t ]+b f )

[0064] Where σ is the Sigmoid activation function, W f Let h be the weight matrix of the forget gate. t-1 x represents the neuron state of the forget gate at the previous moment. t b is the input value at the current moment. f Forgetting bias term; f t f represents the degree to which information is retained in network neurons. t ∈[0,1];

[0065] The formula for calculating the input gate is:

[0066] i t =σ(W i [h t-1 ,x t ]+b i )

[0067]

[0068]

[0069] Among them, i t b is an intermediate variable. i and b c For the input bias term, W i W is the first input weight matrix. c This is the second input weight matrix. c represents the candidate values ​​selected by the tanh activation function. t-1 This represents the neuron state of the input gate at the previous time step;

[0070] The formula for calculating the output gate is as follows:

[0071] o t =σ(W[h t-1 ,x t ]+b o )

[0072] h t =o t tanh(c t )

[0073] Among them, o t b represents the output information filtered through the Sigmoid activation function. oThe output bias term is given by W, and the output weight matrix is ​​given by h. t This is the final output of the Long Short-Term Memory network model;

[0074] (303) Based on the output h of the Long Short-Term Memory Network Model t and the next time step data y in the training dataset t Calculate the mean absolute error loss L pre :

[0075]

[0076] Where K is the number of input data in one training process, K = M;

[0077] Based on the average absolute error loss L pre The parameters of the Long Short-Term Memory network model are updated using backpropagation until the mean absolute error loss L is reached. pre Upon convergence, training ends and the parameters of the Long Short-Term Memory network model are saved, completing the deep learning training.

[0078] Furthermore, step 4 is specifically implemented as follows:

[0079] (401) Based on the block region R i Ground information and satellite ephemeris information are used to obtain the R block region. i Information on Y satellites within the visible range is denoted as

[0080] (402) in Of the Y satellites, select the region R. i During communication, satellites with a communication elevation angle greater than the lowest elevation angle are added to the satellite ensemble. middle, and

[0081] (403) in e Y′ Among the satellites, select the block region R i During communication, satellites with communication signal power greater than the minimum communication power are added to the satellite ensemble. middle, and

[0082] (404) in p Y” Among the satellites, select the block region R i During communication, satellites whose communication connection time is longer than the time required for a single satellite data upload are added to the satellite set. middle, and

[0083] Furthermore, step 5 is specifically implemented as follows:

[0084] (501)Judgment Is it equal to 1? If so, then... All uploaded to satellite collection Satellites in the middle, otherwise the region R will be divided. i It is divided into J multi-satellite overlay coverage sub-regions and G single-satellite independent coverage sub-regions;

[0085] Traverse G single-satellite independent coverage sub-regions, when a single-satellite independent coverage sub-region is selected. At that time, a single satellite will independently cover a sub-region. The area of ​​the block region R i area ratio multiplied by Obtain independent coverage sub-regions for a single satellite The predicted amount of satellite data to be uploaded to the single access satellite. in Represents the block region R i The g-th independent coverage sub-region of a single satellite within the region, 1≤g≤G;

[0086] Traverse J multi-satellite overlay coverage sub-regions, when a multi-satellite overlay coverage sub-region is selected. At that time, multiple satellites will be superimposed to cover the sub-region. The area of ​​the block region R i area ratio multiplied by Obtain multi-satellite overlay coverage sub-region Internal upload to multiple access satellites Predicted amount of satellite data to be uploaded Represents the block region R i The j-th multi-satellite overlay coverage sub-region within the region, 1≤j≤J;

[0087] For multi-satellite overlay coverage sub-regions If the satellite adopts a uniformly distributed random access strategy, then proceed to step (502); if the satellite adopts an elevation angle optimal access strategy, then proceed to step (503); if the satellite adopts a communication power optimal access strategy, then proceed to step (504).

[0088] (502) Overlay multiple satellites to cover the sub-region The predicted amount of satellite data to be uploaded within the scope Equally distributed among multiple access satellites

[0089] (503) Overlay multiple satellites to cover the sub-region The system is divided into multiple elevation angle strategy regions, each of which is connected to a satellite with the optimal elevation angle. The area of ​​each elevation angle strategy region is then superimposed with the coverage area of ​​multiple satellites to form a sub-region. The proportion of the area is multiplied by respectively Obtain the strategy area for each elevation angle in the docking satellite in the multi-satellite overlay coverage sub-region. The predicted amount of satellite data to be accessed within the region;

[0090] (504) Overlay multiple satellites to cover the sub-region The system is divided into multiple communication power strategy regions, each of which is connected to a satellite with the highest communication power. The area of ​​each communication power strategy region is then superimposed with the coverage area of ​​multiple satellites. The proportion of the area is multiplied by respectively Obtain the communication power strategy area docking satellite in the multi-satellite superimposed coverage sub-region. The predicted amount of satellite data to be accessed within the region.

[0091] In summary, this invention addresses the problems of low prediction accuracy or high prediction complexity caused by the unclear or excessively long periodicity of satellite motion around the Earth. By first effectively predicting a fixed area on the ground, and then utilizing the characteristics of satellite motion and satellite traffic access, ground traffic is distributed to satellite nodes, thereby achieving simple, accurate, and efficient prediction.

[0092] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and does not limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the present invention as defined in the appended claims should be included within the protection scope of the present invention.

Claims

1. A satellite access traffic prediction method based on LSTM and access strategy, characterized in that, Includes the following steps: Step 1: Divide all satellite coverage areas into N sub-regions, and establish a set of sub-regions based on the N sub-regions. Where R N This represents the Nth block region; Step 2: Traverse the set of partitioned regions When the segmented region R is selected i At that time, for the segmented region R i The amount of uploaded satellite data within the specified time is sampled and recorded M times at equal time intervals. A set of uploaded satellite data is then established based on the recorded M data volumes. in This indicates that at the Mth sampling, the block region R i The amount of satellite data uploaded within the specified range, r, is 1 ≤ i ≤ N; Step 3: Construct a long short-term memory network model, dividing the block region R... i The set of uploaded satellite data is input into the Long Short-Term Memory network model to obtain the block region R for the next time step. i The predicted amount of satellite data to be uploaded within the scope Step 4: Based on the satellite data for the segmented region R i Pitch angle, communication signal strength, and communication connection time are used to filter the block region R. i Access satellites within the country; Step 5: According to the satellite access strategy, divide the segmented area R i The predicted amount of satellite data to be uploaded within the scope Assigned to block region R i All satellites connected within the region; Step 6: Calculate the distribution of predicted satellite data volume in all segmented regions, and sum the predicted satellite data volume uploaded to the same satellite to finally complete the access traffic prediction for different satellites.

2. The satellite access traffic prediction method based on LSTM and access strategy according to claim 1, characterized in that, The specific method for constructing the Long Short-Term Memory network model in step 3 is as follows: (301) Collect public datasets based on MAWI workgroup traffic archives to obtain multiple sets of network traffic data as training datasets; (302) Construct a long short-term memory network model, which includes a forgetting gate, an input gate, and an output gate; The formula for calculating the forgetting gate is as follows: f t =σ(W f [h t-1 ,x t ]+b f ) Where σ is the Sigmoid activation function, W f Let h be the weight matrix of the forget gate. t-1 x represents the neuron state of the forget gate at the previous moment. t b is the input value at the current moment. f Forgetting bias term; f t f represents the degree to which information is retained in network neurons. t ∈[0,1]; The formula for calculating the input gate is: i t =σ(W i [h t-1 ,x t ]+b i ) Among them, i t b is an intermediate variable. i and b c For the input bias term, W i W is the first input weight matrix. c This is the second input weight matrix. c represents the candidate values ​​selected by the tanh activation function. t-1 This represents the neuron state of the input gate at the previous time step; The formula for calculating the output gate is as follows: the t =σ(W[h t-1 ,x t ]+b o ) h t = no t fishy(c) t ) Among them, o t b represents the output information filtered through the Sigmoid activation function. o The output bias term is given by W, and the output weight matrix is ​​given by h. t This is the final output of the Long Short-Term Memory network model; (303) Based on the output h of the Long Short-Term Memory Network Model t and the next time step data y in the training dataset t Calculate the mean absolute error loss L pre : Where K is the number of input data in one training process, K = M; Based on the average absolute error loss L pre The parameters of the Long Short-Term Memory network model are updated using backpropagation until the mean absolute error loss L is reached. pre Upon convergence, training ends and the parameters of the Long Short-Term Memory network model are saved, completing the deep learning training.

3. The satellite access traffic prediction method based on LSTM and access strategy according to claim 2, characterized in that, The specific method for step 4 is as follows: (401) Based on the block region R i Ground information and satellite ephemeris information are used to obtain the R block region. i Information on Y satellites within the visible range is denoted as (402) in Of the Y satellites, select the region R. i During communication, satellites with a communication elevation angle greater than the lowest elevation angle are added to the satellite ensemble. middle, and (403) in e Y′ Among the satellites, select the block region R i During communication, satellites with communication signal power greater than the minimum communication power are added to the satellite ensemble. middle, and (404) in p Y” Among the satellites, select the block region R i During communication, satellites whose communication connection time is longer than the time required for a single satellite data upload are added to the satellite set. middle, and 4. The satellite access traffic prediction method based on LSTM and access strategy according to claim 3, characterized in that, The specific method for step 5 is as follows: (501)Judgment Is it equal to 1? If so, then... All uploaded to satellite collection Satellites in the middle, otherwise the region R will be divided. i It is divided into J multi-satellite overlay coverage sub-regions and G single-satellite independent coverage sub-regions; Traverse G single-satellite independent coverage sub-regions, when a single-satellite independent coverage sub-region is selected. At that time, a single satellite will independently cover a sub-region. The area of ​​the block region R i area ratio multiplied by Obtain independent coverage sub-regions for a single satellite The predicted amount of satellite data to be uploaded to this single satellite in Represents the block region R i The g-th independent coverage sub-region of a single satellite within the region, 1≤g≤G; Traverse J multi-satellite overlay coverage sub-regions, when a multi-satellite overlay coverage sub-region is selected. At that time, multiple satellites will be superimposed to cover the sub-region. The area of ​​the block region R i area ratio multiplied by Obtain multi-satellite overlay coverage sub-region Internal upload to multiple access satellites Predicted amount of satellite data to be uploaded Represents the block region R i The j-th multi-satellite overlay coverage sub-region within the region, 1≤j≤J; For multi-satellite overlay coverage sub-regions If the satellite adopts a uniformly distributed random access strategy, then proceed to step (502); if the satellite adopts an elevation angle optimal access strategy, then proceed to step (503); if the satellite adopts a communication power optimal access strategy, then proceed to step (504). (502) Overlay multiple satellites to cover the sub-region The predicted amount of satellite data to be uploaded within the scope Equally distributed among multiple access satellites (503) Overlay multiple satellites to cover the sub-region The system is divided into multiple elevation angle strategy regions, each of which is connected to a satellite with the optimal elevation angle. The area of ​​each elevation angle strategy region is then superimposed with the coverage area of ​​multiple satellites to form a sub-region. The proportion of the area is multiplied by respectively Obtain the strategy area for each elevation angle in the docking satellite in the multi-satellite overlay coverage sub-region. The predicted amount of satellite data to be accessed within the region; (504) Overlay multiple satellites to cover the sub-region The system is divided into multiple communication power strategy regions, each of which is connected to a satellite with the highest communication power. The area of ​​each communication power strategy region is then superimposed with the coverage area of ​​multiple satellites. The proportion of the area is multiplied by respectively Obtain the communication power strategy area docking satellite in the multi-satellite superimposed coverage sub-region. The predicted amount of satellite data to be accessed within the region.

Images