Low earth orbit satellite communication terminal power consumption optimization method

Abstract

The invention discloses a low-orbit satellite communication terminal power consumption optimization method, and relates to the technical field of satellite mobile communication and wireless communication terminal power consumption management. A candidate set is generated based on satellite ephemeris and shielding information, so that a terminal only scans a limited frequency band, and energy waste caused by full-frequency-band blind scanning is fundamentally avoided; in a dense urban scene, signal interruption is caused by high-rise buildings, and a traditional terminal needs to repeatedly scan to maintain connection, however, a predictive scanning mechanism of the method can activate scanning in advance before signal attenuation and reduce response delay, so that a power consumption curve is smoothed, and a power consumption peak value caused by sudden switching is avoided; secondly, the energy consumption perceptual sorting mechanism dynamically evaluates the power consumption performance of the candidate satellites through historical data and context features, and preferentially selects low-energy-consumption options, so that the scanning process is more targeted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of satellite mobile communication and wireless communication terminal power consumption management, and in particular to a low-orbit satellite communication terminal power consumption optimization method. BACKGROUND

[0002] With the deployment and business application expansion of low-orbit satellite networks, the use of terminals in dense urban environments is increasing. Affected by high-rise building shielding and limited line-of-sight, the terminal's satellite visible window is short, and the link quality fluctuates greatly, resulting in more frequent synchronization maintenance and switching / reselection processes. In such scenarios, the terminal needs to perform satellite / beam search and measurement to maintain service continuity; when there is a lack of effective assistance information or out-of-sync occurs, the search range and duration increase, which is prone to produce significant power consumption peaks.

[0003] The existing solutions in the prior art mainly develop resource and beam management from the satellite or network side, such as dynamic beam and beam hopping scheduling, load and traffic awareness, and general measurement and power saving mechanisms, such as measurement throttling, conditional switching, etc., but for the specific scenario of high-frequency out-of-sync induced by urban shielding and movement, how to adaptively adjust the scanning timing, range and candidate set on the terminal side to reduce the energy consumption of recapturing still has room for improvement.

[0004] Therefore, a terminal power consumption optimization method for dense urban areas is proposed, which combines environment / motion information and ephemeris / beam plan to reduce unnecessary scanning and suppress power consumption peaks, which has practical significance for improving terminal endurance and user experience. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] The present application provides a low-orbit satellite communication terminal power consumption optimization method to solve the problem that the low-orbit satellite terminal triggers full-band scanning due to frequent switching in dense urban areas, resulting in a sharp increase in power consumption, and the existing solutions focus on satellite-side optimization, lacking adaptive management on the terminal side.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] The embodiment of the present application provides a low-orbit satellite communication terminal power consumption optimization method, which comprises:

[0009] Step S1, based on terminal location information, satellite ephemeris data and shielding information, a candidate set with a visible time interval is generated, and the candidate set contains candidate satellites and their available frequency point / beam identifiers;

[0010] Step S2, limit the frequency point / beam range and timing of measurement and scanning according to the candidate set, set the scanning residence time and measurement period, and perform predictive scanning when the trigger condition is met.

[0011] Step S3, selecting target satellite / beams in the candidate set for access or handover according to the energy-aware scan ranking and multi-target connection scoring;

[0012] Step S4, based on the connection result feedback, using online learning to adaptively update the preset parameters for adjusting the scan ranking and connection scoring;

[0013] Step S5, when the candidate set is empty or the continuous access / maintenance failure reaches a threshold, performing a limited fallback scan to recover the connection, and returning to step S1.

[0014] As a preferred scheme of the low-orbit satellite communication terminal power consumption optimization method, in step S1, the following steps are included:

[0015] Based on the three-dimensional building model or the shielding database and the terminal attitude, the visible elevation angle and the shielding probability of each candidate satellite in the time interval are calculated, and the candidate with the shielding probability exceeding a threshold is removed or its priority is reduced.

[0016] As a preferred scheme of the low-orbit satellite communication terminal power consumption optimization method, in step S1, the following steps are included:

[0017] The terminal motion trajectory prediction is formed by using GNSS and IMU data, and the visible time interval of the candidate satellite is time-aligned and corrected.

[0018] As a preferred scheme of the low-orbit satellite communication terminal power consumption optimization method, in step S2, the predictive scan is performed when at least one of the following conditions is met:

[0019] The channel quality index decreases at a rate exceeding a threshold, the continuous observation time reaches a lower limit, and the hysteresis threshold is broken through;

[0020] After triggering, the scan is performed in advance in the next visible window of the candidate set.

[0021] As a preferred scheme of the low-orbit satellite communication terminal power consumption optimization method, before step S3, the candidate set is energy-aware ranked, and the ranking is based on the estimated value of the unit successful access expected energy consumption of each candidate frequency point / beams, and the scan order is determined from low to high according to the estimated value; the estimated value is calculated using context features, including position, speed, time period and shielding probability.

[0022] The estimated value of the unit successful access expected energy consumption is specifically:

[0023] Construct candidate frequency points / beam entries and historical sample sets. The sample records contextual features, success status, and energy consumption decomposition items. They are managed by time decay and spatial grid to form a nearest neighbor subset of the time to be estimated and the context.

[0024] When at time And the context is At that time, for candidates The estimated energy consumption for a unit to successfully connect is:

[0025] ,

[0026] in, Indicates candidate In context The unit below successfully connected to the expected energy consumption estimate. This represents the estimated expected energy consumption for a single access attempt. This represents an estimate of the probability of a single successful connection. Indicates the context-weighted factor. Indicates the candidate frequency / beam index. Indicates time Context feature vectors, Indicates the current estimated time;

[0027] Energy consumption per attempt based on measurement and model decomposition:

[0028] ,

[0029] in, Indicating in the candidate Receiver equivalent power overhead, Indicating in the candidate The duration of stay on the platform is set. Indicates the energy consumption of the synchronization process. This represents the transmit and listen energy consumption during the random access procedure. This indicates the power consumption of baseband processing and protocol overhead.

[0030] significance level is Confidence interval:

[0031] ,

[0032] in, express The endpoints of the confidence interval at the level are expressed. Represents the standard normal quantile. This represents the variance estimate of energy consumption per transaction. This represents the sample size used for energy consumption estimation. This indicates the number of attempts used for estimating the probability of success.

[0033] Computational pipeline from historical samples to estimators:

[0034] a) Nearest neighbor selection and time weighting;

[0035] b) Perform a weighted regression on the energy consumption decomposition term to obtain the mean and variance;

[0036] c) Context-conditional estimation of the success probability and smoothing with hierarchical priors;

[0037] d) Map business priorities to occlusion risks ;

[0038] e) by The calculation formula is used to synthesize estimates and the scan order is given in ascending order;

[0039] f) Incremental updates and uncertainty gating are performed when new samples arrive.

[0040] As a preferred embodiment of the low-orbit satellite communication terminal power consumption optimization method described in this invention, the connection scoring in step S3 simultaneously considers: the estimated power consumption required for access and maintenance, link stability indicators, expected dwell time, and service priority, and selects the target satellite / beam from the candidate set accordingly.

[0041] The multi-objective scores for the estimated power consumption, link stability metrics, expected dwell time, and service priority are as follows:

[0042] At any moment With context Next, for the candidates Data collection or estimation: Expected energy consumption upon successful unit connection Link stability indicators Expected length of stay Business Priority ; with monotonic mapping Unify all items ;

[0043] Entries within the candidate set Calculate the overall score:

[0044] ,

[0045] in, Indicates candidate Overall score Indicates time The weights of energy consumption, stability, residency, and priority. This represents the normalized mapping of the corresponding indicator;

[0046] Convex loss is constructed using online feedback and mirror descent is performed on the probabilistic simplex:

[0047] ,

[0048] ,

[0049] in, Indicates time No. Item weight, A set of indexes representing energy consumption, stability, residency, and priority. Indicates the learning rate. Represents the loss function. Indicates the selected entry The first observed Normalized returns Represents the weight regularization coefficient. Represents the L2 norm, Represents the prior weight vector. Indicates time Candidate indexes for the final join, Represents a set Component index within;

[0050] Weights satisfy and Set the lower bound Upper Realm , Class mappings maintain monotonicity and boundaries, and for ties, expected residency is prioritized, followed by stability as the tie-breaking criterion.

[0051] Monitor the stability of weight changes and business returns within a sliding window:

[0052] ,

[0053] ,

[0054] in, Indicates the number of evaluations within the window. Indicates the first The time index of the next evaluation. Describes the norm, Indicates the threshold for weight changes. Indicates normalized business returns. Indicates the hysteresis step size of the control window. Indicates the threshold for change in return; freezes when the criterion is met. Or switch to small-step fine-tuning.

[0055] As a preferred embodiment of the low-orbit satellite communication terminal power consumption optimization method described in this invention, step S4 involves online learning to adaptively update the following parameters:

[0056] (i) Parameters for energy consumption sensing scanning sorting, including the unit expected energy consumption estimate for successful access and its update coefficient for each candidate frequency point / beam;

[0057] (ii) Weighting parameters for multi-target connectivity scoring, including weights used to characterize estimated power consumption, link stability, expected dwell time, and service priority;

[0058] (iii) Predictive scan triggering parameters, including channel quality degradation rate threshold, continuous observation duration and hysteresis threshold;

[0059] (iv) Scan throttling parameters, including the maximum number of frequency points, minimum dwell time, and minimum measurement cycle in a single scan;

[0060] The online learning process uses incremental updates based on contextual features and connection result feedback, and sets exploration ratios and fallback conditions to ensure connection stability.

[0061] As a preferred embodiment of the low-orbit satellite communication terminal power consumption optimization method of the present invention, step S2 implements throttling control for scanning, including:

[0062] Limit the maximum number of frequency points in a single scan, set the minimum dwell time and minimum measurement cycle, and extend the scan interval when the number of consecutive failures reaches a threshold.

[0063] As a preferred embodiment of the low-orbit satellite communication terminal power consumption optimization method described in this invention, the ephemeris data and frequency / beam mapping table are periodically updated via offline packets or network assistance.

[0064] When updates are unavailable, the method runs independently based on local cache and online learning statistics.

[0065] As a preferred embodiment of the low-orbit satellite communication terminal power consumption optimization method described in this invention, step S5, the restricted backtracking scan, includes a two-level process:

[0066] First, perform a narrowband back-back scan in the neighborhood of the most recently successfully connected frequency point / beam. If it still fails, perform a wideband back-back scan under the conditions of limited maximum duration and maximum number of frequency points.

[0067] The beneficial effects of this invention are:

[0068] This invention generates a candidate set based on satellite ephemeris and obstruction information, enabling the terminal to scan only a limited frequency band, fundamentally avoiding the energy waste of blind scanning across the entire frequency band. In dense urban scenarios, tall buildings cause signal intermittent issues, requiring traditional terminals to repeatedly scan to maintain connectivity. The predictive scanning mechanism of this method activates scanning before signal attenuation, reducing response latency and smoothing the power consumption curve, thus avoiding power peaks caused by sudden handovers. Secondly, the energy consumption-aware ranking mechanism dynamically evaluates the power consumption performance of candidate satellites using historical data and contextual features, prioritizing low-energy options and making the scanning process more targeted. Combined with a multi-objective scoring system, the terminal balances power consumption, stability, dwell time, and service priority when making connection decisions, avoiding simply pursuing signal strength while ignoring energy efficiency. This is particularly important in mobile scenarios, effectively extending battery life. The online learning module further enhances adaptability, adjusting parameters such as scan trigger thresholds and scoring weights in real time based on connection feedback, enabling the terminal to autonomously adapt to non-stationary changes in the urban environment, reducing reliance on satellite commands and improving robustness. In addition, throttling control and limited back-off scan mechanisms ensure connection restoration in extreme cases while preventing energy dissipation caused by overscanning.

[0069] This invention shifts the optimization focus from the satellite side to the terminal side, reduces system interaction overhead through localized processing, enables the terminal to maintain low power consumption in frequently switching scenarios, and maintains link quality, providing practical support for IoT devices and mobile terminals. Attached Figure Description

[0070] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0071] Figure 1 This is a flowchart of a low-orbit satellite communication terminal power consumption optimization method in one of the embodiments. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0073] All terms used in this application (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0074] For example, the terms “first” and “second” used in this application are only used to distinguish and describe similar objects, to differentiate the first object from another object, and are not used to describe a specific order or sequence, nor should they be interpreted as indicating or implying relative importance.

[0075] This application proposes a power consumption optimization method for low-Earth orbit satellite communication terminals, combining... Figure 1 As shown, the method includes:

[0076] Step S1: Based on terminal location information, satellite ephemeris data and occlusion information, generate a candidate set with visible time intervals. The candidate set includes candidate satellites and their available frequency points / beam identifiers.

[0077] In this embodiment, the location information is the geographic coordinates and attitude of the GNSS fused IMU, which is periodically output by the terminal; the ephemeris data is the offline or network-synchronized ephemeris of the target constellation; and the occlusion information is the local 3D building model or occlusion probability database. The default visible time interval is set to a rolling window of 30-120 seconds, with an update cycle of 1-5 seconds, determined based on ephemeris drift, terminal speed, and computational resource experience. The frequency / beam identifier is provided by the network-side mapping table or local cache. If ephemeris loss or positioning degradation occurs, the upper limit of the window can be optionally converged to 30 seconds and the update cycle maintained at 1 second to reduce timing errors. When the candidate set is empty, a window period is retained as the baseline using the most recent valid candidate set to avoid jitter.

[0078] Step S2: Based on the candidate set, limit the frequency / beam range and timing of measurement and scanning, set the scanning dwell time and measurement cycle, and perform predictive scanning when the triggering conditions are met;

[0079] Similarly, the default scan dwell time is 20-80 ms, adjustable from 10-120 ms; the default measurement period is 200-500 ms, adjustable from 100-1000 ms, adjusted according to service priority and processor load. Channel quality metrics are derived from a single bounded metric measured at the terminal's physical layer, selected by the implementation as SNR or received power, etc., with the sampling period consistent with the measurement period. When the candidate size N is large, measurements are performed on the top K candidates, with K defaulting to min(5,N) to control the duty cycle. In case of measurement failure, the dwell time is amplified by 1.2 times once, not exceeding two amplifications.

[0080] Step S3: Based on the energy consumption-aware scan ranking and multi-target connection score, select a target satellite / beam from the candidate set for access or handover. Further, historical samples are obtained from access attempts and session maintenance records locally on the terminal, including timestamps, contextual features, and statistics on energy consumption decomposition items. These are saved for the past 24-72 hours by default and decay with time weighting. The ranking and scoring calculation cycle is aligned with the measurement cycle, defaulting to 200-500 ms. When the difference in candidate scores is less than a set minor difference threshold, a parallel decision is made prioritizing expected dwell time, and the last successful connection is reviewed to ensure session continuity. When resources are scarce, the number of score updates within a single cycle is limited to no more than two.

[0081] Step S4: Based on the connection result feedback, online learning is used to adaptively update the preset parameters to adjust the scan sorting and connection score;

[0082] Step S5: When the candidate set is empty or the continuous access / hold failure reaches the threshold, perform a restricted backoff scan to restore the connection and return to step S1;

[0083] In one embodiment, step S1 includes:

[0084] Based on 3D building models or occlusion databases and terminal attitude, calculate the visible elevation angle and occlusion probability of each candidate satellite in the time interval, and remove candidates whose occlusion probability exceeds the threshold or reduce their priority.

[0085] Specifically, the occlusion probability is obtained through empirical mapping based on the intersection ratio of satellite line of sight and building voxels or the contour occlusion angle, while the terminal attitude is derived from the IMU pitch and azimuth angles. The default occlusion probability threshold is 0.6, with an adjustment range of 0.4 to 0.8, set according to city density and false detection rate statistics; the lower limit of the elevation angle is 10° by default, with a range of 5° to 20°, to reduce false triggers caused by ground object occlusion. Optionally, a degradation configuration with a threshold increase of 0.05 is used for devices with high pose noise to improve robustness; when the occlusion database is missing data, candidates are retained but their priority is lowered to avoid excessive rejection.

[0086] In one embodiment, step S1 further includes:

[0087] Terminal motion trajectory prediction is generated using GNSS and IMU data. The visible time intervals of candidate satellites are time-aligned and corrected to adapt to mobile scenarios. For example, trajectory prediction uses a constant velocity or constant acceleration motion model to extrapolate the position 1-5 seconds into the future, with an extrapolation period of 200 ms to 1 s, set according to the IMU sampling rate and GNSS availability. Time alignment recalculates the visible elevation angle based on the extrapolated position at the visible window boundary and fine-tunes the start and end times of the window, with each correction not exceeding ±2 s to avoid excessive oscillation. In the event of GNSS intermittent periods or IMU saturation, the previous extrapolation state is maintained for no more than 3 seconds, and the correction amplitude is reduced to ±1 s.

[0088] In one embodiment, the predictive scan of step S2 is performed when at least one of the following conditions is met:

[0089] The rate of decline of channel quality indicators exceeds the threshold, the duration of continuous observation reaches the lower limit, and the hysteresis threshold is breached.

[0090] After triggering, a scan is performed in advance in the next visible window of the candidate set to reduce the occurrence of blind scans across the entire frequency band;

[0091] Optionally, the descent rate threshold is set to an equivalent value of 3 dB / s by default, ranging from 1 to 5 dB / s; the lower limit of continuous observation duration is set to 500 ms by default, ranging from 200 to 1000 ms; and the hysteresis threshold is set to 2 dB by default, ranging from 1 to 4 dB. The relative lead of the advance scan is set to 200 to 600 ms before the start of the next visible window to ensure synchronization and avoid conflict with regular measurements. If the scan is triggered more than 3 times in a short period of time, a 1 to 2 s cooling-off period is entered to suppress excessive triggering.

[0092] In one embodiment, before step S3, the candidate set is sorted by energy consumption perception. The sorting is based on the estimated value of the unit successful access expected energy consumption for each candidate frequency point / beam, and the scanning order is determined from low to high according to the estimated value. The calculation of the estimated value uses contextual features, including location, velocity, time period and occlusion probability.

[0093] The specific method for estimating the expected energy consumption upon successful grid connection is as follows:

[0094] Construct candidate frequency points / beam entries and historical sample sets. The sample records contextual features, success status, and energy consumption decomposition items. They are managed by time attenuation and spatial grid to form a nearest neighbor subset of the time and context to be estimated, which is used for subsequent estimation and calibration.

[0095] When at time And the context is At that time, for candidates The estimated energy consumption for a unit to successfully connect is:

[0096] ,

[0097] in, Indicates candidate In context The unit below successfully connected to the expected energy consumption estimate. This represents the estimated expected energy consumption for a single access attempt. This represents an estimate of the probability of a single successful connection. This represents the context-weighted factor; a positive value indicates that 1 represents no weighting. Indicates the candidate frequency / beam index. Indicates time Context feature vectors, Indicates the current estimated time;

[0098] In this embodiment, the success probability estimate is given by contextual condition statistics, typically output by the nearest neighbor subset frequency or equivalent discriminant model; the decomposition term of the energy consumption of a single attempt is obtained by measuring the receiver operating current and process duration; the context weighting factor takes a smooth value around 1 to reflect service priority or occlusion risk; the estimated value is enhanced in sparse data regions through time decay and spatial nearest neighbor sharing to improve stability.

[0099] Energy consumption per attempt based on measurement and model decomposition:

[0100] ,

[0101] in, Indicating in the candidate Receiver equivalent power overhead, Indicating in the candidate The duration of stay on the platform is set. This indicates the energy consumption of the synchronization process, capture, and initial tracking. This represents the transmit and listen energy consumption during the random access procedure. This indicates the power consumption of baseband processing and protocol overhead.

[0102] Specifically, the receiver's equivalent power overhead is derived from the average current conversion of the front-end and baseband measurements, and the dwell time is the dwell configuration set in step S2; the synchronization process energy consumption corresponds to the total power consumption during the acquisition and initial tracking phases; the random access process energy consumption includes the total energy of the transmit burst and the listening cycle; and the baseband processing and protocol overhead covers channel estimation, demodulation, control signaling parsing, and other stages. The above components are estimated using a sliding window within the sample pipeline to measure the mean and variance, thus supporting the uncertainty measurement.

[0103] significance level is Confidence interval:

[0104] ,

[0105] in, express The endpoints of the confidence interval at the level are expressed. Represents the standard normal quantile. This represents the variance estimate of energy consumption per transaction. This represents the sample size used for energy consumption estimation. This indicates the number of attempts used for estimating the probability of success.

[0106] Computational pipeline from historical samples to estimators:

[0107] a) Nearest neighbor selection and time weighting;

[0108] b) Perform a weighted regression on the energy consumption decomposition term to obtain the mean and variance;

[0109] c) Context-conditional estimation of the success probability and smoothing with hierarchical priors;

[0110] d) Map business priorities to occlusion risks ;

[0111] e) by The calculation formula is used to synthesize estimates and the scan order is given in ascending order;

[0112] f) Incremental updates and uncertainty gating are performed when new samples arrive;

[0113] Specifically, by using a joint modeling approach combining data and context, an actionable estimate of the energy consumption required for a unit to successfully access the network is provided, supplementing the description of uncertainties at the interval level. The core idea of ​​the measurement is to measure the cost by the ratio of energy consumption per attempt to the probability of success, combined with strategic weighting, so that the ranking not only conforms to physical resource consumption but also aligns with business priorities. Energy consumption is decomposed into the portion of receiver overhead that accumulates linearly with dwell time, as well as the discrete process overhead of synchronization, random access, and baseband processing, facilitating the synchronous collection of samples from the metering and protocol sides. The success probability is given by contextual condition statistics, relying on a discriminative model in areas with sufficient data, and outputting a stable output through hierarchical priors in sparse areas. Interval estimation provides a basis for setting throttling, backoff, and exploration ratios, which can be used to adjust the scanning interval and dwell time to reduce energy consumption fluctuations under abnormal conditions. It can continuously output stable ranking signals in scenarios with moving occlusion and dynamic changes in ephemeris.

[0114] In one embodiment, the connectivity score in step S3 simultaneously considers: the estimated power consumption required for access and retention, link stability metrics, expected dwell time, and service priority, and selects the target satellite / beam from the candidate set accordingly.

[0115] The multi-objective scores for estimated power consumption, link stability metrics, expected dwell time, and service priority are as follows:

[0116] At any moment With context Next, for the candidates Data collection or estimation: Expected energy consumption upon successful unit connection Link stability indicators Expected length of stay Business Priority ; with monotonic mapping Unify all items Lower energy consumption is preferable, while higher stability, dwell time, and priority are preferable.

[0117] For example, the normalization mapping uses a monotonically compressed function to limit the impact of extreme values; the weights satisfy the constraints of non-negativity and summation to one and are in a state of equilibrium. Inside, Used to prevent weight collapse To prevent single-objective dominance; the learning rate is adaptively decreased based on online feedback, with an initial small value to ensure convergence stability; when the score update fails or observations are missing, the weights from the previous time step are retained and the learning rate is reduced in the next round.

[0118] Entries within the candidate set Calculate the overall score:

[0119] ,

[0120] in, Indicates candidate Overall score Indicates time The weights of energy consumption, stability, residency, and priority. This represents the normalized mapping of the corresponding indicator;

[0121] Convex loss is constructed using online feedback and mirror descent is performed on the probabilistic simplex:

[0122] ,

[0123] ,

[0124] in, Indicates time No. Item weight, A set of indexes representing energy consumption, stability, residency, and priority. Indicates the learning rate. Represents the loss function. Indicates the selected entry The first observed Normalized returns, such as The same applies to the rest. Represents the weight regularization coefficient. Represents the L2 norm, Represents the prior weight vector. Indicates time Candidate indexes for the final join, Represents a set The component index is used within the range; similarly, the normalized return is taken from the observed return of the currently selected candidate, including the same-domain values ​​of components such as energy consumption, stability, and residence; the prior weight vector is used as a bias term to improve the convergence speed when the business model changes; when the weight change and return change within the window simultaneously satisfy the stability criterion, the learning rate is frozen or a small step size is switched to fine-tuning to reduce the disturbance of short-term fluctuations on the score.

[0125] Weights satisfy and Set the lower bound Upper Realm To prevent collapse, Class mappings maintain monotonicity and boundaries, and for ties, expected residency is prioritized, followed by stability as the tie-breaking criterion.

[0126] Monitor the stability of weight changes and business returns within a sliding window:

[0127] ,

[0128] ,

[0129] in, Indicates the number of evaluations within the window. Indicates the first The time index of the next evaluation. Describes the norm, Indicates the threshold for weight changes. This represents a normalized business return (e.g., a bounded mapping of throughput / energy consumption). Indicates the hysteresis step size of the control window. Indicates the threshold for change in return; freezes when the criterion is met. Or switch to small-step fine-tuning;

[0130] Specifically, the multi-objective problem is transformed into a linearly separable model that can be optimized online. An interpretable score is obtained by summing the energy consumption, stability, residence, and priority after normalization within the same domain. The weight evolution adopts an exponentially weighted mirror descent form, which not only meets the constraints of non-negativity and summation, but also maintains responsiveness when feedback fluctuates. Prior terms are added to the loss to facilitate maintaining reasonable preferences under changes in business models or sparse data regions. Constraint and decision rules ensure that the score is consistent with the system strategy and avoid jitter caused by extreme weights. The convergence and stability criteria are determined by two indicators: weight change and reward stability. Once these are met, a small step size stage is entered, which facilitates joint operation with the online learning module.

[0131] The overall coupling is geared towards real-time scenarios: the score is used for immediate selection, the feedback enters the weight update closed loop, and the updated weight directly affects the next round of ranking, so that the method can maintain a balance between low energy consumption and stable connection under the conditions of dynamic occlusion and rapid changes in track visibility.

[0132] In one embodiment, the online learning in step S4 adaptively updates the following parameters:

[0133] (i) Parameters for energy consumption sensing scanning sorting, including the unit expected energy consumption estimate for successful access and its update coefficient for each candidate frequency point / beam;

[0134] (ii) Weighting parameters for multi-target connectivity scoring, including weights used to characterize estimated power consumption, link stability, expected dwell time, and service priority;

[0135] (iii) Predictive scan triggering parameters, including channel quality degradation rate threshold, continuous observation duration and hysteresis threshold;

[0136] (iv) Scan throttling parameters, including the maximum number of frequency points, minimum dwell time, and minimum measurement cycle in a single scan;

[0137] Online learning uses incremental updates based on contextual features and connection result feedback, and sets exploration ratios and rollback conditions to ensure connection stability. Optionally, the parameter update cycle is aligned with the scoring calculation cycle. The default exploration ratio is 0.05~0.15 to balance stability and convergence; when the consecutive failure threshold is exceeded, the exploration ratio is automatically reduced by one level. When the rollback condition is triggered, the parameters are rolled back to the most recent stable snapshot, and further rollback is prohibited for three cycles. In case of missing tests or feedback delays, only observable parameters are updated, and the remaining parameters are frozen until they are filled.

[0138] In one embodiment, step S2 implements throttling control on the scan, including:

[0139] Limit the maximum number of frequency points in a single scan, set the minimum dwell time and minimum measurement cycle, and extend the scan interval when the number of consecutive failures reaches a threshold;

[0140] In one embodiment, the ephemeris data and frequency / beam mapping table are periodically updated via offline packets or network assistance;

[0141] When updates are unavailable, the method runs independently based on local cache and online learning statistics;

[0142] In this embodiment, the default update cycle for the ephemeris and mapping table is 30~120 minutes. When the network is unavailable, the most recent valid version is used and marked as read-only. If the cached version exceeds 24 hours and still cannot be updated, the upper limit of the prediction time window is reduced to 30 seconds and the backslide scan frequency is appropriately increased to maintain a basic balance between availability and energy consumption.

[0143] In one embodiment, the restricted rollback scan in step S5 includes a two-level process:

[0144] First, a narrowband back-off scan is performed in the neighborhood of the most recently successfully connected frequency point / beam. If it still fails, a wideband back-off scan is performed under the conditions of limited maximum duration and maximum number of frequency points. Further, the neighborhood range is defined according to spectral adjacency and beam adjacency, and the nearest 2 to 3 entries are selected by default for each. The single duration of narrowband back-off is not more than 300 ms by default, and the total number of attempts is not more than 2. The maximum duration of wideband back-off is 1 to 2 s by default, and the maximum number of frequency points is not more than 50% of the total number of candidates. If there is no connection after the back-off, a cooling period of 500 to 1000 ms is entered before returning to step S1 to avoid excessive duty cycle.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0146] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of this application and form different embodiments. For example, all the embodiments above can be used in any combination. The information disclosed in this background section is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for optimizing power consumption of a low earth orbit satellite communication terminal, applied to a terminal comprising a processor, a memory, a radio frequency front end and an antenna, characterized in that, Comprise: Step S1, based on terminal location information, satellite ephemeris data and shielding information, generate a candidate set with visible time interval, the candidate set contains candidate satellites and its available frequency point / beam identifier; Step S2, according to the candidate set, limit the frequency point / beam range and timing of measurement and scanning, set the scanning residence time and measurement period, and perform predictive scanning when the trigger condition is met; Step S3, according to the scanning order of energy consumption perception and multi-objective connection score, select the target satellite / beam in the candidate set for access or handover; Step S4, based on the connection result feedback, use online learning to adaptively update the preset parameters for adjusting the scanning order and connection score; Step S5, when the candidate set is empty or the continuous access / failure reaches the threshold, perform limited fallback scanning to restore the connection, and return to step S1.

2. The method of claim 1, wherein the method further comprises: Step S1 includes: Based on three-dimensional building model or shielding database and terminal attitude, the visible elevation angle and shielding probability of each candidate satellite in the time interval are calculated, and the candidates with shielding probability exceeding the threshold are removed or their priority is reduced.

3. The method of claim 2, wherein the method further comprises: Step S1 also includes: Using GNSS and IMU data to form terminal motion trajectory prediction, time alignment and correction are performed on the visible time interval of the candidate satellite.

4. The method of claim 1, wherein, The predictive scanning of step S2 is performed when at least one of the following trigger conditions is met: The decline rate of channel quality indicator exceeds the threshold, the continuous observation time reaches the lower limit, and the hysteresis threshold is broken; Scan in advance in the next visible window of the candidate set after triggering.

5. The method of claim 1, wherein, Before step S3, the candidate set is energy consumption perception sorted, the sorting is based on the estimated value of the unit successful access expected energy consumption of each candidate frequency point / beam, and the scanning order is determined from low to high according to the estimated value; the calculation of the estimated value uses context features, including location, speed, time period and shielding probability; The estimated method of the unit successful access expected energy consumption is as follows: Construct candidate frequency point / beam entry and historical sample set, sample record context features, success or not and energy consumption decomposition item, time decay and space grid management, form the neighbor subset of the estimated time and context; When at time and the context is the candidate unit successful access is expected to consume energy is estimated as: ， wherein, represents a candidate In the context of a unit successful access expected energy consumption estimator, represents a single access attempt expected energy consumption estimate, represents a single access success probability estimate, represents a context weighting factor, represents a candidate frequency point / beam index, represents a context feature vector at time point in time, represents a current estimation point in time; Based on measurement and model decomposition single attempt energy consumption: ， wherein, denotes the receiver equivalent power overhead at candidate denotes the camping duration setting at candidate denotes the synchronization procedure energy consumption, denotes the transmission and listening energy consumption of the random access procedure, denotes the baseband processing and protocol overhead energy consumption;​​ The level of significance is the confidence interval: ， wherein, denotes horizontal confidence interval endpoints, denotes standard normal quantile, denotes variance estimate of single energy consumption, denotes sample size for energy consumption estimation, denotes number of attempts for success probability estimation; From the calculation pipeline of historical samples to the estimated value: a) neighbor sample selection and time weight; b) weighted regression of energy consumption decomposition item to get mean and variance; c) context condition estimation of success probability and hierarchical prior smoothing; d) mapping the service priority to a risk of occlusion ; e) by The calculation formula synthesizes the estimates and gives the scan order in ascending order. f) incremental update and uncertainty gating when new samples arrive.

6. The method of claim 1, wherein, The connection score of step S3 considers the estimated energy consumption required for access and maintenance, link stability index, expected residence time and business priority, and selects the target satellite / beam from the candidate set accordingly; The multi-objective score of the estimated energy consumption, link stability index, expected residence time and business priority is as follows: At time With context Next, candidate Collected or estimated: unit success access expected energy consumption , link stability index , expected residence time , service priority ; In a monotone mapping Each item is normalized to ; Within the candidate set Compute an overall score: ， wherein, represents a comprehensive score of the candidate , represents the weight of energy consumption, stability, residence and priority at the time , represents the normalized mapping of the corresponding indicators; Construct convex loss with online feedback and do mirror descent on probability simplex: ， ， wherein, denotes a time instant a first item weight, denotes a set of indices of energy consumption, stability, dwell and priority, denotes a learning rate, denotes a loss function, denotes a first item normalized reward observed on a selected item denotes a second weight regularization coefficient, denotes a two-norm, denotes a prior weight vector, denotes a time instant a final connected candidate index, denotes a component index within a set ​ The weights satisfy and , set lower bounds upper bounds , Class mappings preserve monotonicity and bounds, and tie-breaking among equally scored entries uses expected residency preference, followed by stability. Monitor the stability of weight change and business return in the sliding window: ， ， wherein, represents the number of evaluations within the window, represents the index of the time instance of the evaluation, represents a norm, represents a weight change threshold, represents a normalized business return, represents a lag step size against the window, represents a return change threshold; criterion fulfilled to freeze or to enter small step fine tuning.

7. The method of claim 1, wherein, The online learning of step S4 adaptively updates the following parameters: (i) parameters for energy-aware scan ordering, including the estimated energy consumption per successful access and its update coefficient for each candidate frequency point / beams; (ii) weight parameters for multi-objective connection scoring, including the weights for estimated energy consumption, link stability, expected residence time and traffic priority; (iii) predictive scan triggering parameters, including the channel quality degradation rate threshold, continuous observation duration and hysteresis threshold; (iv) scan throttling parameters, including the maximum number of frequency points per scan, minimum residence time and minimum measurement period; The online learning is based on the incremental update of context features and connection result feedback, and sets the exploration ratio and back-off condition to ensure the stability of the connection.

8. The method of claim 1, wherein, Step S2 performs throttling control on the scan, including: limiting the maximum number of frequency points per scan, setting the minimum residence time and minimum measurement period, and lengthening the scan interval when the number of consecutive failures reaches the threshold.

9. The method of claim 1, wherein, Ephemeris data and frequency point / beams mapping table are periodically updated through offline package or network assistance; When the update is unavailable, the method is independently run based on the local cache and the statistical quantity of online learning.

10. The method of claim 1, wherein, The limited back-off scan of step S5 includes two levels of processes: First, narrowband back-off scan is performed in the neighborhood of the frequency point / beams of the latest successful connection, and if it still fails, wideband back-off scan is performed under the condition of limiting the maximum duration and the maximum number of frequency points.

Images