Satellite-ground cooperative resource scheduling method

Abstract

The application relates to a kind of star-ground cooperative resource scheduling methods, comprising: satellite side receives terminal uplink data packet and analysis, adopts dynamic sliding window to extract physical layer feature and service layer key field, runs quantized compression isolated forest model, generates exception event label and is associated with original data packet after, as observation data is downloaded to ground side;Ground side receives observation data, combines historical data to form multi-star joint observation data, constructs space-time cubic grid, based on space-time cubic grid, carries out space-time correlation analysis to multi-star joint observation data, calculates the space-time correlation coefficient of abnormal event time sequence and spatial aggregation degree sequence, when space-time correlation coefficient matches preset threshold value, identify event type and generate structured alarm;According to event type, generate adaptive scheduling strategy, upload scheduling strategy instruction to satellite through ground side and satellite's instruction upload link, and resource allocation is executed according to scheduling strategy by satellite when overtop.

Description

Technical Field

[0001] This invention relates to the field of satellite communication technology, and in particular to a resource scheduling method for satellite-ground coordination. Background Technology

[0002] Low-Earth orbit (LEO) narrowband IoT constellations are widely used in asset tracking, environmental monitoring, and emergency response due to their advantages of global coverage, low power consumption, and low cost. A typical system employs a store-and-forward architecture: a large number of terminals communicate with overhead satellites via low-power wide-area network technologies such as CSS (linear frequency modulation spread spectrum). The satellites receive the data, store it temporarily, and then download it when the satellites approach the gateway station. Existing systems have the following technical shortcomings in anomaly detection and resource scheduling:

[0003] First, existing technologies struggle to distinguish between single-point terminal failures and regional events (such as interference or severe weather), leading to excessive alarms or inaccurate response strategies. Second, satellite monitoring resources (beam duration, scheduling priority) are statically configured and cannot be dynamically adjusted based on abnormal terminal states or regional events, resulting in low resource utilization and weak anomaly response capabilities.

[0004] Therefore, how to improve a space-ground collaborative resource scheduling scheme to identify different types of events and perform adaptive dynamic resource scheduling accordingly is a technical problem that needs to be solved. Summary of the Invention

[0005] In view of this, the present invention provides a satellite-ground collaborative resource scheduling method that can identify different types of event conditions and perform adaptive dynamic resource scheduling accordingly.

[0006] To achieve the above objectives, the first aspect of this application provides a resource scheduling method for satellite-ground coordination, comprising the following steps:

[0007] S1: The satellite receives and parses the uplink data packets from the terminal, uses a dynamic sliding window to extract physical layer features and key fields of the service layer, runs a quantized and compressed isolated forest model, generates abnormal event labels, associates them with the original data packets, and then transmits them to the ground side as observation data.

[0008] S2: The ground side receives the observation data, combines it with historical data to form multi-satellite joint observation data, constructs a spatiotemporal cube grid, performs spatiotemporal correlation analysis on the multi-satellite joint observation data based on the spatiotemporal cube grid, calculates the spatiotemporal correlation coefficient between the time series of abnormal events and the spatial clustering sequence, and identifies the event type and generates a structured alarm when the spatiotemporal correlation coefficient matches a preset threshold.

[0009] S3: Generate an adaptive scheduling strategy based on the event type, and upload the scheduling strategy instruction to the satellite through the instruction upload link between the ground side and the satellite. The satellite will then perform resource allocation according to the scheduling strategy when it passes overhead.

[0010] As one possible implementation of the first aspect, in step S1, running the quantized compressed isolated forest model includes: extracting terminal identifier, arrival time, Doppler frequency shift residual value, signal-to-noise ratio, and beam identifier in real time based on the original data packet; calculating the current window size N=min(K,M) based on the number of historical valid reports M of the terminal in the current overpass period, where K is a preset upper limit of the window, and then calculating the mean μ, standard deviation σ, and linear regression slope β of the physical layer features within the window; mapping the floating-point feature vector composed of μ, σ, and β to an integer feature vector, inputting it into the pre-trained isolated forest model for integer operations, outputting anomaly scores, and then dequantizing and mapping it back to the floating-point domain to generate anomaly event labels.

[0011] As one possible implementation of the first aspect, in step S2, the construction of the spatiotemporal cube grid includes: determining the spatial coordinates (Lon, Lat) using beam identifiers, determining the time coordinates T using arrival time t, and filling the spatiotemporal cube grid Cube(T, Lon, Lat) with the anomaly score s from the historical observation data of multi-satellite joint data as feature values; for spatiotemporal grids without satellite observation data, an interpolation algorithm is used to fill the grid based on neighboring data.

[0012] As one possible implementation of the first aspect, it also includes: the update frequency of the spatiotemporal cube grid is synchronized with the satellite overpass event.

[0013] As one possible implementation of the first aspect, in step S2, the calculation of the spatiotemporal correlation coefficient between the time series of the abnormal event and the spatial clustering sequence includes: for the spatiotemporal cube grid, extracting the time series X of the abnormal event and the spatial clustering sequence Y within the same geographic grid Cube(Lon,Lat), and calculating the Pearson correlation coefficient ρ between the two; if ρ>ρth1, it is identified as a regional clustering event; if ρ≤ρth2, it is identified as a single-point terminal anomaly; wherein, the threshold ρth2<threshold ρth1.

[0014] As one possible implementation of the first aspect, the method for obtaining the time series X of the abnormal event includes: for each geographic grid of the spatiotemporal cube, extracting all feature values ​​at each moment, taking the average or maximum value of all feature values, and forming the time series X of the abnormal event over time.

[0015] As one possible implementation of the first aspect, the method for obtaining the spatial clustering sequence Y includes: for each time slice of the spatiotemporal cube, extracting the feature values ​​in all grids at each time moment, calculating the similarity index of the feature values ​​between adjacent grids using a global spatial autocorrelation algorithm, and forming a spatial clustering sequence Y that changes over time.

[0016] As one possible implementation of the first aspect, in step S3, generating an adaptive scheduling strategy based on the event type includes: when a regional clustering event is identified, generating an energy-saving scheduling strategy, including broadcasting instructions to reduce the reporting frequency of terminals in non-critical areas and instructing satellites to adjust beam resources to cover key areas; when a single-point terminal is identified as abnormal, generating an enhanced listening strategy, including instructing satellites to increase the receiving gain in a specified beam and allocate additional time slot resources, or triggering data retransmission of a specified terminal.

[0017] As one possible implementation of the first aspect, in step S3, when the satellite passes overhead, it performs resource allocation according to the scheduling strategy, including at least one of the following: the satellite adjusts the listening duration, receiving gain, or scheduling priority of the specified beam according to the instruction content; the satellite forwards data retransmission instructions or instructions to adjust the reporting cycle to the terminal.

[0018] As one possible implementation of the first aspect, in step S1, the frequency of setting the dynamic sliding window is synchronized with the satellite overpass time window, adapting to the duration window of a single satellite overpass.

[0019] In resource-constrained low-Earth orbit narrowband IoT, this application's solution enables satellite-to-ground closed-loop anomaly detection and resource scheduling based on an end-edge-cloud collaborative architecture, significantly improving the system's intelligence level. Specifically, this application has the following technical effects:

[0020] On the one hand, by coordinating satellite and ground, inference can be performed on the satellite, avoiding the minute-level delay of transmitting raw data to the ground for processing (satellite-to-ground transmission + ground queuing processing), shortening the delay time between terminal reporting and response, and better adapting to the needs of real-time monitoring services.

[0021] On the other hand, for the problem that relying solely on on-board models lacks continuous observation data over a long period of time for the same region, and is limited by computing power to store massive amounts of historical global data for comparison, this application addresses the issue by combining satellite and ground systems, introducing a spatiotemporal cube model on the ground, and using multi-satellite joint observation data to make up for the spatiotemporal limitations of single-satellite observations, thereby significantly improving detection accuracy and significantly reducing the false alarm rate.

[0022] On the other hand, in view of the short satellite overpass time (usually only 5-10 minutes), a dynamic sliding window feature extraction mechanism is designed. The window size is adaptively adjusted according to the terminal reporting cycle, which solves the problem that the traditional fixed window cannot adapt to the changing business cycle.

[0023] On the other hand, the proposed solution is adaptable to situations where the processor computing power and memory are limited for low-Earth orbit satellite payloads. Specifically, for environments with limited onboard computing power (<100MB memory), the isolated forest model is compressed using 8-bit quantization, and an integer inference pipeline including dequantization is designed, reducing the model's memory usage from 40MB to below 10MB, adapting to onboard processors with a maximum memory of 64MB. This reduces computing power requirements, shortens the inference time for the onboard processor, and accommodates a single satellite overpass lasting 5-10 minutes. Attached Figure Description

[0024] Figure 1 A flowchart of the satellite-ground collaborative resource scheduling method provided in the first embodiment of this application;

[0025] Figure 2 A schematic diagram of the resource scheduling method for satellite-ground coordination provided in the second embodiment of this application;

[0026] Figure 3 This is a schematic diagram of the spatiotemporal correlation analysis of multi-satellite observations in this application.

[0027] It should be understood that the dimensions and shapes of the blocks in the above structural diagrams are for reference only and should not constitute an exclusive interpretation of the embodiments of the present invention. The relative positions and inclusion relationships between the blocks presented in the structural diagrams are only schematic representations of the structural relationships between the blocks, and are not intended to limit the physical connection methods of the embodiments of the present invention. Detailed Implementation

[0028] To make this application easier to understand, specific embodiments are described below to further illustrate this application. The technical solutions provided by this application are further explained below with reference to the accompanying drawings and examples. It should be understood that the system architecture and business scenarios provided in the embodiments of this application are mainly for illustrating possible implementations of the technical solutions of this application and should not be construed as the sole limitation on the technical solutions of this application. Those skilled in the art will recognize that, with the evolution of system architecture and the emergence of new business scenarios, the technical solutions provided by this application are equally applicable to similar technical problems.

[0029] It should be understood that the satellite-ground collaborative resource scheduling scheme provided in the embodiments of this application includes a satellite-ground collaborative resource scheduling method, scheduling device, scheduling system, and its application. Since these technical solutions solve problems based on the same or similar principles, some repetitive details may not be repeated in the following descriptions of specific embodiments. However, these specific embodiments should be considered as mutually referencing each other and can be combined with each other.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. In case of any inconsistency, the meaning set forth in this specification or derived from the content described herein shall prevail. Furthermore, the terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application. To accurately describe the technical content of this application and to accurately understand the invention, the following explanations or definitions of the terms used in this specification are provided before describing specific embodiments:

[0031] 1) 8-bit asymmetric quantization compression of the isolated forest model: 8-bit asymmetric quantization compression is a model compression technique that compresses the original floating-point isolated forest model using 8-bit asymmetric quantization compression technology. It maps the threshold of floating-point decision tree nodes to integer fields, which can directly reduce the size of the model to 1 / 4 of the original size and improve the model's computing speed.

[0032] 2) Spatiotemporal Cube Grid: Used for spatiotemporal correlation analysis, the spatiotemporal cube grid is a three-dimensional cube constructed with a time axis T, a spatial axis S, and a feature axis F. In this application, based on the slice granularity ΔT of the time axis T, the timestamps of the on-board transmitted anomalous event tags are aligned with the original data packets. The beam identifiers in the tags are extracted and mapped to spatial axis coordinates Si, the arrival times are extracted and mapped to time axis coordinates Tj, and the anomaly scores are extracted as feature values ​​F to fill the cube grid (Tj, Si). For spatiotemporal grids without satellite observation data, interpolation algorithms can be used to fill the grid based on data from nearby satellite overpass times, or the grid can be marked as null values.

[0033] 3) Spatial Clustering Sequence: This is a quantitative indicator used to measure the degree of clustering (or dispersion) of anomalous events in space within each time slice. For each time slice of the spatiotemporal cube, feature values ​​from all grids at each time point are extracted, and then calculated using spatial statistical algorithms. Common calculation methods include global spatial autocorrelation (such as Moran's I index) algorithms, which derive an index value by calculating the similarity of feature values ​​between adjacent grids.

[0034] 4) Pearson correlation coefficient ρ(X,Y): A statistic used to measure the degree and direction of linear correlation between two continuous variables X and Y. In this application, it can be understood as a measure of the synchronicity of X and Y.

[0035] This application discloses a satellite-ground collaborative resource scheduling scheme, specifically a lightweight anomaly detection and satellite-ground closed-loop scheduling system and method for resource-constrained low-Earth orbit satellites. This application constructs a satellite-ground closed-loop anomaly detection and resource scheduling system based on an edge-cloud collaborative architecture, employing a spatiotemporal cube model for multi-satellite data fusion. A quantized compressed isolated forest model runs on the satellite, and a dynamic sliding window is used to extract physical and operational layer features, taking advantage of the short satellite overpass time. Suspected anomalies are pre-screened and transmitted with spatiotemporal tags. Ground gateway stations calculate the spatiotemporal correlation coefficients of multi-satellite joint observation data based on the spatiotemporal cube model, identify regional clustered events, and generate structured alarms. An adaptive scheduling decision module adjusts terminal reporting strategies, satellite monitoring resource allocation, or triggers data retransmission based on alarm type via satellite command links. This application, through collaboration between on-board edge nodes and the ground-cloud, evolves anomaly response from post-ground analysis to pre-processing via satellite-ground collaboration, significantly improving response real-time performance, resource utilization efficiency, and system intelligence.

[0036] The present application will now be described in detail with reference to the accompanying drawings.

[0037] The first embodiment of this application provides a resource scheduling method for satellite-ground cooperation, such as... Figure 1 The flowchart shown includes the following steps S110-S130:

[0038] S110: The satellite receives and parses the uplink data packets from the terminal, uses a dynamic sliding window to extract physical layer features and key fields of the service layer, runs a quantized and compressed isolated forest model, generates abnormal event labels, associates them with the original data packets, and then transmits them to the ground side as observation data.

[0039] In some embodiments, the frequency of setting the dynamic sliding window is synchronized with the satellite overpass time window, adapting to the duration window of a single satellite overpass.

[0040] In some embodiments, running the quantized compressed isolated forest model includes the following steps S111-S113:

[0041] S111: Based on the raw data packet, extract the terminal identifier, arrival time, Doppler frequency shift residual value, signal-to-noise ratio and beam identifier in real time;

[0042] S112 calculates the current window size N=min(K,M) based on the number of valid reports M made by the terminal in the current overpass cycle, where K is the preset upper limit of the window, and then calculates the mean μ, standard deviation σ and linear regression slope β of the physical layer features within the window;

[0043] S113: Map the floating-point feature vector composed of μ, σ and β to an integer feature vector, input it into the pre-trained isolated forest model for integer operations, output anomaly scores, dequantize and map it back to the floating-point domain, and generate anomaly event labels.

[0044] S120: The ground side receives the observation data, combines it with historical data to form multi-satellite joint observation data, constructs a spatiotemporal cube grid, performs spatiotemporal correlation analysis on the multi-satellite joint observation data based on the spatiotemporal cube grid, calculates the spatiotemporal correlation coefficient between the time series of abnormal events and the spatial clustering sequence, and identifies the event type and generates a structured alarm when the spatiotemporal correlation coefficient matches a preset threshold.

[0045] In some embodiments, constructing a spatiotemporal cube grid includes: determining spatial coordinates (Lon, Lat) using beam identifiers, determining temporal coordinates T using arrival time t, and filling the spatiotemporal cube grid Cube(T, Lon, Lat) with anomaly scores s from multi-satellite joint historical observation data as feature values; for spatiotemporal grids without satellite observation data, an interpolation algorithm is used to fill the grid based on neighboring data.

[0046] The update frequency of the spatiotemporal cube grid can be synchronized with the satellite overpass event.

[0047] In some embodiments, calculating the spatiotemporal correlation coefficient between the time series of anomaly events and the spatial clustering sequence includes: for the spatiotemporal cube grid, extracting the time series X of anomaly events and the spatial clustering sequence Y within the same geographic grid Cube(Lon,Lat), and calculating the Pearson correlation coefficient ρ between the two; if ρ>ρth1, it is identified as a regional clustering event; if ρ≤ρth2, it is identified as a single-point terminal anomaly; wherein, the threshold ρth2<th threshold ρth1.

[0048] In some embodiments, the method for obtaining the time series X of the anomalous event includes: for each geographic grid of the spatiotemporal cube, extracting all feature values ​​at each moment, taking the average or maximum value of all feature values, and forming the time series X of the anomalous event over time.

[0049] In some embodiments, the method for obtaining the spatial clustering sequence Y includes: for each time slice of the spatiotemporal cube, extracting feature values ​​from all grids at each time point, calculating the similarity index of feature values ​​between adjacent grids using a global spatial autocorrelation algorithm, and forming a spatial clustering sequence Y that changes over time.

[0050] S130: Generate an adaptive scheduling strategy based on the event type, and upload the scheduling strategy instruction to the satellite through the instruction upload link between the ground side and the satellite. The satellite will then perform resource allocation according to the scheduling strategy when passing overhead.

[0051] In some embodiments, generating an adaptive scheduling strategy based on the event type includes:

[0052] When a regional clustering event is identified, an energy-saving scheduling strategy is generated, including broadcasting instructions to reduce the reporting frequency of terminals in non-critical areas and instructing satellites to adjust beam resources to cover key areas;

[0053] When a single-point terminal is identified as abnormal, an enhanced listening strategy is generated, including instructing the satellite to increase the receiving gain in a specified beam and allocate additional time slot resources, or triggering data retransmission of a specified terminal.

[0054] In some embodiments, when the satellite passes overhead, it performs resource allocation according to the scheduling strategy, including at least one of the following: the satellite adjusts the listening duration, receiving gain, or scheduling priority of a specified beam according to the instruction content; the satellite forwards a data retransmission instruction or an instruction to adjust the reporting cycle to the terminal.

[0055] To better understand the solution of this application, a resource scheduling method for satellite-ground cooperation provided in the second embodiment of this application is further described below, which can also be found in [reference 1]. Figure 2 The schematic diagram shown illustrates that the method includes the following steps S10-S40:

[0056] S10: The satellite receives and parses the uplink data packets from the terminal, extracts physical layer features and key fields of the service layer using a dynamic sliding window, runs a quantized and compressed isolated forest model, generates abnormal event labels, associates them with the original data packets, and then transmits them as observation data to the ground side.

[0057] This step can be performed by the satellite-side lightweight detection unit, specifically including the following steps S11-S15:

[0058] S11: The satellite first receives the raw data packets (containing service data) from the terminal and retrieves the preset window upper limit K (usually 5~15) and the quantized compressed isolated forest model.

[0059] S12: Based on the original data packet, the onboard lightweight detection unit parses the data packet in real time and extracts physical layer features and key service layer fields such as terminal identifier, arrival time (t), Doppler frequency shift residual value (Δf), signal-to-noise ratio (SNR), and beam identifier (indicating which beam of the satellite the terminal is currently covered by).

[0060] Among them, the aforementioned physical layer features can be extracted based on the physical layer (PHY) and link layer (MAC) signals at the bottom layer of the satellite receiver.

[0061] S13: Based on the number of valid reports M made by the terminal in the current overpass period, calculate the current window size N = min(K,M), where K is the preset upper limit of the window length. Also calculate the mean (μ), standard deviation (σ), and linear regression slope (β) of the physical layer features within the statistical window.

[0062] Here, instead of using a fixed-length window, the system adaptively adjusts the time based on the number of valid reports made by the terminal within the current overpass period, which is suitable for the short satellite overpass time.

[0063] Among them, the three statistics μ, σ, and β represent the baseline level, fluctuation stability, and trend of the signal, respectively, which can reflect the overall status and trend of the communication link, as detailed below:

[0064] The mean (μ) represents the average state of a physical layer feature within a statistical window (e.g., the data from the most recent 5 to 15 reports), reflecting the basic quality of the communication link.

[0065] Standard deviation (σ) measures the degree to which data within a window deviates from the mean, that is, the dispersion or jitter of the data, reflecting the stability and anti-interference ability of the communication link.

[0066] The linear regression slope (β) is calculated by linearly fitting the time series data within the window (the horizontal axis represents time or reporting order, and the vertical axis represents the feature value). It represents the rate at which the feature changes over time and reflects the direction of the evolution of the communication link quality (whether it is deteriorating or recovering).

[0067] S14: Map the floating-point feature vector composed of μ, σ, and β to an 8-bit integer feature vector, input it into the pre-trained isolated forest model for integer operations, and output an anomaly score (s). Then, dequantize the anomaly score and map it back to the floating-point domain.

[0068] The mapping to 8-bit integers is to adapt the model to 8-bit integers. This model is small in size to fit the limited memory capacity of the satellite, and saves computing power to fit the inference speed of the onboard computer, so as to fit the short window of the satellite passing overhead.

[0069] The anomaly score(s) represents the degree of anomaly of the current data point (i.e., the feature vector composed of μ, σ, and β). The higher the value, the greater the likelihood that the data point is an anomaly.

[0070] When dequantizing back to the floating-point domain, the dequantization formula scale×(q-zero_point) can be used to map back to the floating-point domain, outputting the anomaly score s in the floating-point domain. Here, scale is 0.01, zero_point is 128, and q represents the anomaly score in the integer domain output by the isolated forest model. The scale and zero_point parameters are calculated offline using an asymmetric quantization formula based on the statistical distribution range of the terminal uplink data packet characteristics (time of arrival, Doppler shift, signal-to-noise ratio) over a month of onboard history, and are then embedded in the onboard model.

[0071] S15: Construct an anomaly event label based on the floating-point anomaly score, terminal identifier, and beam identifier, and associate the label with the original data packet to transmit it as observation data to the ground gateway station via the satellite-to-ground link.

[0072] S20: The ground side receives abnormal event tags and raw data packets, that is, it receives observation data, performs spatiotemporal correlation analysis on multi-satellite joint observation data based on the spatiotemporal cube model, calculates spatiotemporal correlation coefficients, and when the correlation coefficient matches a preset threshold, it identifies the corresponding event and generates a structured alarm.

[0073] This step can be performed by the ground side (such as a ground gateway server with a deep analysis engine), and specifically includes the following steps S21-S24:

[0074] S21: The ground-side gateway station receives anomalous event tags and raw data packets transmitted from the satellite, which are used as observation data for that satellite. For example, Figure 3 The diagram illustrates that each time a satellite passes over a ground gateway station, it receives observation data including anomaly tags and raw data packets sent by each satellite. This allows the ground gateway station to acquire continuous observation data records of the same area across different time segments (i.e., aggregated observation data from multiple satellites), thus forming a combined historical observation data set from multiple satellites.

[0075] S22: Construct a spatiotemporal cube grid Cube(T,Lon,Lat), determine the spatial coordinates (Lon,Lat) using the beam identifier, determine the time coordinate T using the arrival time t (which is the t in the original data packet in step S12), and fill the spatiotemporal cube grid Cube(T,Lon,Lat) with the anomaly score (s) in the historical observation data of multi-satellite joint as the feature value.

[0076] When constructing the spatiotemporal cube grid Cube(T,Lon,Lat), its time slice granularity ΔT (can be 1s~30s). For spatiotemporal grids without satellite observation data, linear, polynomial or spline interpolation algorithms can be used to fill the grid based on neighboring data.

[0077] The spatiotemporal cube grid is constructed at a frequency synchronized with satellite overpass events, and the cube grid data is updated once for each satellite overpass event.

[0078] S23: For the spatiotemporal cube grid, extract the time series X and spatial clustering sequence Y of anomalous events within the same geographic grid Cube(Lon,Lat), and calculate the Pearson correlation coefficient ρ(X,Y) between them.

[0079] In some embodiments, sequence X and sequence Y can be calculated based on feature values ​​(anomaly scores) at different times within the same geographic grid Cube(Lon,Lat). Examples are as follows:

[0080] Assuming a time slice granularity ΔT = 10 seconds, within each 10-second time slice, all anomaly scores within the geographic grid are statistically analyzed. These scores can be the average or maximum value of all anomaly scores within that time period. Over time, this forms a broken line X = {x1, x2, ..., xn} reflecting the severity of anomalies in the region, which is the time series of the anomaly event. Similarly, within each 10-second time slice, the distribution of these anomaly scores on the spatiotemporal cube grid is evaluated. A spatial statistical algorithm (such as the Moran index) can be used to calculate a value, forming a sequence Y = {y1, y2, ..., yn} over time, which is the spatial clustering sequence.

[0081] S24: Based on the calculated ρ value, output structured alarm information and send it to the ground side, which may include the following:

[0082] If ρ > ρth1 (the value can be 0.5~0.9, such as 0.7), then determine and output the event type representing "regional clustering event";

[0083] If ρ≤ρth2 (the value range can be 0.2~0.3, such as 0.3), then determine and output the event type representing "single point terminal abnormality".

[0084] Among them, the threshold ρth2 is less than the threshold ρth1, and the output alarm information can include the event type and ρ value.

[0085] It should be noted that in the above empty cube grid Cube(T,Lon,Lat), the feature value is filled with the anomaly score(s) value in the historical observation data of multiple stars. In other embodiments, it may also include signal-to-noise ratio, Doppler shift residual value, etc., such as the feature value is a vector value (i.e. an array) composed of (anomaly score, signal-to-noise ratio, Doppler shift residual value).

[0086] S30: Generate an adaptive scheduling strategy based on the alarm type.

[0087] This step can be executed by the adaptive scheduling decision module on the ground side, specifically including the following steps S31-S33:

[0088] S31: The adaptive scheduling decision module receives structured alarm information (including event type and ρ) from the previous step, as well as the satellite overpass time window table as input.

[0089] S32: Perform strategy matching based on ρ value or event type, including:

[0090] When a regional clustering event is identified (ρ>0.7), an energy-saving scheduling strategy is generated, such as broadcasting instructions to reduce the reporting frequency of terminals in non-critical areas and instructing satellites to adjust beam resources to cover key areas.

[0091] When a single-point terminal is determined to be abnormal (ρ≤0.3), an enhanced listening strategy is generated, such as instructing the satellite to increase the receiving gain (e.g., +3dB) in a specified beam and allocating additional time slot resources.

[0092] S33: After completing the strategy matching, the corresponding strategy is converted into specific satellite instructions (such as data retransmission instructions and parameter adjustment instructions), and finally the encapsulated scheduling strategy instruction package is output and sent to the ground control system.

[0093] The instructions corresponding to different strategies can be pre-stored for instruction conversion in this step.

[0094] S40: The scheduling strategy instructions are uploaded to the satellite via the command upload link between the ground side and the satellite, and the satellite performs resource allocation according to the scheduling strategy when it passes overhead.

[0095] This step is executed by the ground control system on the ground side and the satellite side in cooperation, and specifically includes the following steps S41-S43:

[0096] S41: The ground control system receives the scheduling strategy instruction packet and tracks the satellite in real time. When the satellite enters the line of sight of the gateway station (for example, within 30 seconds after entering the line of sight), it completes the uploading of the scheduling strategy instruction.

[0097] S42: After receiving the instruction, the satellite performs resource allocation according to the instruction content.

[0098] Among them, the execution of resource allocation includes satellite-side resource allocation, such as adjusting the listening duration, receiving gain, or scheduling priority of a specified beam.

[0099] The execution of resource allocation may also include terminal-side resource allocation, such as forwarding instructions to the terminal or broadcasting instructions at a specified time. Instructions may be, for example, data retransmission instructions or instructions to adjust the reporting cycle.

[0100] S43: After receiving the instruction, the terminal performs data retransmission (such as re-reporting unconfirmed data or data for a specified period) or adjusts the reporting cycle to achieve successful retransmission of abnormal terminal data and reduce power consumption of regional terminals.

[0101] Then, a new round of terminal uplink data is generated and fed back to step S10, forming a complete closed loop.

[0102] The following section describes the application of this application's solution to an anomaly detection and adaptive scheduling in a global container tracking scenario, using the third embodiment as an example. Specifically, this embodiment uses a low-Earth orbit narrowband IoT constellation as an example, with a container tracker as the terminal, reporting location, temperature, and vibration data every 4 hours. The onboard processor has 128MB of memory and a computing power of 200MHz.

[0103] The onboard lightweight detection unit runs an isolated forest model compressed with 8-bit quantization, occupying approximately 8MB of memory, meeting the requirement of a model file size of less than 10MB. The inference phase follows a "mapping-operation-dequantization" pipeline: the input feature vector is mapped to an 8-bit integer, and integer operations are performed with the quantized decision tree node thresholds; after the leaf node path length is calculated, the 8-bit integer result q is mapped back to the floating-point domain using the dequantization formula r=scale×(q-zero_point), outputting an anomaly score s. Here, scale is 0.01 and zero_point is 128; the scale and zero_point parameters are calculated offline using an asymmetric quantization formula based on the statistical distribution range of terminal uplink data packet characteristics (time of arrival, Doppler shift, signal-to-noise ratio) over a historical month on the satellite, and are then embedded in the onboard model. The entire inference process has a latency of approximately 5ms, meeting the requirement of a single inference latency of less than 10ms.

[0104] Dynamic sliding window settings: Preset window length upper limit K=10, window size N=min(10, terminal's historical reporting count), extract the mean, standard deviation, and trend of the signal-to-noise ratio (SNR) of the most recent 10 reports. When a terminal (ID: 5678) reports an SNR lower than the baseline (mean decrease >5dB) three consecutive times, and the reporting time deviation exceeds 30 minutes, the model outputs an anomaly score of 0.85 (threshold 0.7), generates a "suspected antenna fault or obstruction" anomaly event label, and downloads it after associating it with the original data packet.

[0105] The ground-based depth analysis engine receives the anomaly event tag and, through the multi-satellite collaborative analysis module, constructs a spatiotemporal cubic grid with a time slice granularity of ΔT=2s. For grids without observation data, a linear interpolation algorithm is used to fill them. The cubic grid update is synchronized with the satellite overpass event. Records of other satellite overpasses (4 in total) in the area where the terminal is located over the past 24 hours are retrieved, and the spatiotemporal correlation coefficient ρ=0.12 is calculated (below the single-point threshold ρth2=0.3), indicating a single-point terminal anomaly. The business logic verification module further checks the data and finds that the temperature value is still within a reasonable range (-20℃~50℃), but the vibration value is abnormally high (peak value reaches 5g, normal <0.5g). A structured alarm is generated: "Terminal 5678 suspected hardware failure, it is recommended to trigger data retransmission and alarm."

[0106] The adaptive scheduling decision module, based on the comparison between ρ=0.12 and the threshold ρth2=0.3, determines that a single-point terminal is abnormal and generates an enhanced monitoring strategy: 1) It sends a data retransmission command to the satellite that will soon cover the area via the ground control system, requiring terminal 5678 to re-report the data that was unsuccessfully confirmed the last two times; 2) It pushes an alarm message to the operation and maintenance platform. The scheduling command is injected within 30 seconds after the satellite enters the gateway station's line of sight. When the satellite passes overhead, a retransmission command is sent, and the terminal immediately retransmits historical data upon receiving it. After the gateway station receives the retransmitted data, the ground depth analysis engine confirms that the data content is consistent with the previous one, but the signal quality is still poor, so it escalates the alarm and recommends on-site inspection. The entire satellite-ground-satellite closed loop takes about 2 minutes from the occurrence of the abnormality to the issuance of the command, which is significantly shorter than the traditional method (several hours).

[0107] The following description, using the fourth embodiment as an example, illustrates the application of this application's solution to regional event identification and adaptive scheduling in a mountainous hydrological monitoring scenario. Specifically, in this embodiment, 200 hydrological sensors are deployed in a certain watershed, reporting water level, rainfall, and battery voltage data every 30 minutes. This area is alternately covered by multiple low-orbit satellites, and the on-board detection model is the same as in Embodiment 1.

[0108] The onboard lightweight detection unit detected that more than 60% of the terminals within the beam coverage area of ​​the basin simultaneously exhibited abnormal event tags of "abnormal battery voltage drop" (voltage drop > 0.5V) and "synchronous reduction in signal-to-noise ratio" (drop > 8dB) within multiple consecutive orbital cycles, generated a large number of abnormal event tags and transmitted them down.

[0109] The multi-satellite collaborative analysis module of the ground-based depth analysis engine fuses anomalous event labels from three different satellites, constructing a spatiotemporal cube model with a time slice granularity of ΔT=5s. The cube grid update is synchronized with satellite overpass events, and linear interpolation is used to fill grids without observation data. The spatiotemporal correlation coefficient is calculated: the correlation coefficient ρ between the anomalous event time series X and the spatial clustering sequence Y is 0.92 (exceeding the regional event threshold ρth1=0.7), thus classifying it as a "regional event". The business logic verification module, combined with external meteorological data (accessed from the local meteorological station), discovers that the region is experiencing continuous heavy rainfall (cumulative rainfall >100mm), resulting in insufficient solar charging and signal attenuation.

[0110] The adaptive scheduling decision module, based on the comparison between ρ=0.92 and the threshold ρth1=0.7, determines it to be a regional clustering event and generates an energy-saving scheduling strategy: 1) Broadcast instructions to all terminals within the basin, temporarily adjusting the reporting cycle from 30 minutes to 90 minutes to reduce power consumption; calculations show this can extend terminal battery life by approximately 35%; 2) Upload instructions to the satellite to improve the beam's listening sensitivity (gain increased by 3dB) and extend the listening time (from 30 seconds to 60 seconds) to cope with signal attenuation; 3) Push "Regional Weather Impact Warning" to the water management platform. The scheduling instructions are injected within 30 seconds of the satellite entering the gateway station's line of sight.

[0111] The dispatch instructions are uploaded to subsequent overhead satellites via the ground control system and executed via satellite broadcast. After 24 hours, the system automatically restores its original configuration. This strategy effectively prevents terminals from losing contact due to power depletion, and the integrity rate of critical hydrological data remains above 95%.

[0112] The terms "first, second, third, etc." or similar terms such as module A, module B, and module C used in the specification and claims are only used to distinguish similar objects and do not represent a specific ordering of objects. It is understood that a specific order or sequence may be interchanged where permitted so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0113] In the above description, the labels of the steps involved, such as S110, S120, etc., do not mean that the steps will necessarily be executed. The order of the steps can be interchanged or executed simultaneously if permitted.

[0114] The term "comprising" as used in the specification and claims should not be construed as limiting itself to what follows; it does not exclude other elements or steps. Therefore, it should be interpreted as specifying the presence of the mentioned feature, integral, step, or component, but does not exclude the presence or addition of one or more other features, integrals, steps, or components, or groups thereof. Thus, the statement "device comprising means A and B" should not be limited to a device consisting solely of components A and B.

[0115] The terms "an embodiment" or "an embodiment" as used in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in at least one embodiment of this application. Therefore, the terms "in one embodiment" or "in an embodiment" appearing throughout this specification do not necessarily refer to the same embodiment, but may refer to the same embodiment. Furthermore, in one or more embodiments, the particular features, structures, or characteristics can be combined in any suitable manner, as will be apparent to those skilled in the art from this disclosure.

[0116] Note that the above are merely preferred embodiments and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of this application, all of which fall within the scope of protection of this application.

Claims

1. A resource scheduling method for satellite-ground coordination, characterized in that, Includes the following steps: S1: The satellite receives and parses the uplink data packets from the terminal, uses a dynamic sliding window to extract physical layer features and key fields of the service layer, runs a quantized and compressed isolated forest model, generates abnormal event labels, associates them with the original data packets, and then transmits them to the ground side as observation data. S2: The ground side receives the observation data, combines it with historical data to form multi-satellite joint observation data, constructs a spatiotemporal cube grid, performs spatiotemporal correlation analysis on the multi-satellite joint observation data based on the spatiotemporal cube grid, calculates the spatiotemporal correlation coefficient between the time series of abnormal events and the spatial clustering sequence, and identifies the event type and generates a structured alarm when the spatiotemporal correlation coefficient matches a preset threshold. S3: Generate an adaptive scheduling strategy based on the event type, and upload the scheduling strategy instruction to the satellite through the instruction upload link between the ground side and the satellite. The satellite will then perform resource allocation according to the scheduling strategy when it passes overhead.

2. The method according to claim 1, characterized in that, In step S1, running the quantized compressed isolated forest model includes: Based on the raw data packets, the terminal identifier, arrival time, Doppler frequency shift residual value, signal-to-noise ratio and beam identifier are extracted in real time. Based on the number of valid reports M made by the terminal in the current overpass cycle, the current window size N = min(K,M) is calculated, where K is the preset upper limit of the window. Then, the mean μ, standard deviation σ, and linear regression slope β of the physical layer features within the window are calculated. The floating-point feature vector composed of μ, σ, and β is mapped to an integer feature vector, input into a pre-trained isolated forest model for integer operations, and the anomaly score is output and dequantized back to the floating-point domain to generate anomaly event labels.

3. The method according to claim 1, characterized in that, In step S2, constructing the spatiotemporal cubic grid includes: Spatial coordinates (Lon,Lat) are determined using beam identifiers, and time coordinates T are determined using arrival time t. Anomaly scores s from historical observation data of multi-satellite joint observations are used as feature values ​​to fill the spatiotemporal cube grid Cube(T,Lon,Lat). For spatiotemporal grids without satellite observation data, an interpolation algorithm is used to fill the grid based on neighboring data.

4. The method according to claim 3, characterized in that, Also includes: The update frequency of the spatiotemporal cube grid is synchronized with the satellite overpass event.

5. The method according to claim 1, characterized in that, In step S2, calculating the spatiotemporal correlation coefficient between the time series of anomalous events and the spatial clustering sequence includes: For the spatiotemporal cube grid, extract the time series X and spatial clustering sequence Y of anomalous events within the same geographic grid Cube(Lon,Lat), and calculate the Pearson correlation coefficient ρ between them; If ρ > ρth1, then it is identified as a regional clustering event; If ρ≤ρth2, it is identified as a single-point terminal anomaly; Where, threshold ρth2 < threshold ρth1.

6. The method according to claim 5, characterized in that, The method for obtaining the time series X of the abnormal event includes: For each geographic grid of the spatiotemporal cube, extract all feature values ​​at each moment, and take the average or maximum value of all feature values ​​to form a time series X of anomalous events over time.

7. The method according to claim 5, characterized in that, The method for obtaining the spatial clustering sequence Y includes: For each time slice of the spatiotemporal cube, feature values ​​from all grids at each time point are extracted. The similarity index of feature values ​​between adjacent grids is calculated using a global spatial autocorrelation algorithm, forming a spatial clustering sequence Y that changes over time.

8. The method according to claim 5, characterized in that, In step S3, generating an adaptive scheduling strategy based on the event type includes: When a regional clustering event is identified, an energy-saving scheduling strategy is generated, including broadcasting instructions to reduce the reporting frequency of terminals in non-critical areas and instructing satellites to adjust beam resources to cover key areas; When a single-point terminal is identified as abnormal, an enhanced listening strategy is generated, including instructing the satellite to increase the receiving gain in a specified beam and allocate additional time slot resources, or triggering data retransmission of a specified terminal.

9. The method according to claim 8, characterized in that, In step S3, when the satellite passes overhead, it performs resource allocation according to the scheduling strategy, including at least one of the following: The satellite adjusts the listening duration, receiving gain, or scheduling priority of the designated beam according to the instructions. The satellite forwards data retransmission instructions or instructions to adjust the reporting cycle to the terminal.

10. The method according to claim 1, characterized in that, In step S1, the frequency of setting the dynamic sliding window is synchronized with the satellite overpass time window, and is adapted to the duration window of a single satellite overpass.

Images