A satellite network cache migration method based on regional feature prediction

Abstract

The application relates to the technical field of satellite communication network cache management, in particular to a satellite network cache migration method based on regional characteristic prediction. The method divides a satellite service range into multiple regions, counts user content requests in a completed period of a target region, constructs a regional characteristic vector and fuses historical period characteristics to obtain a regional preference characteristic vector of a next service period; the importance of source satellite cache content is calculated according to the regional preference characteristic vector, and a cache migration candidate set is screened; a missing content set is generated through candidate content announcement and hash fingerprint comparison, and an intersatellite migration transmission channel is established when a remaining residence time meets a pre-trigger condition, and a successor satellite preferentially pulls and updates the cache according to the importance. The application is suitable for cache migration and content distribution in a low-orbit satellite network.

Description

Technical Field

[0001] This invention relates to the field of satellite communication network cache management technology, specifically a satellite network cache migration method based on regional feature prediction. Background Technology

[0002] Low Earth Orbit (LEO) satellite networks are commonly used for remote area communication, mobile access, emergency communication, and large-capacity content distribution. To reduce latency and link load caused by the backhaul of duplicate content, existing satellite networks typically incorporate a data center network architecture, setting up caches in satellite nodes or satellite-to-ground links so that user requests can be retrieved from nearby nodes. Existing cache migration methods generally transfer some cached content from satellites about to leave their service area to satellites that will subsequently take over the service, based on content popularity, cache status, satellite orbital position, or node switching relationships.

[0003] However, the high-speed movement of low-Earth orbit satellites causes continuous changes in service areas and user groups. Simply migrating based on global popularity or historical request counts fails to reflect the true content preferences of the next service area and easily leads to the migration of cached content with low value to subsequent areas. Furthermore, if the cached content of subsequent satellites and migration window constraints are not fully identified, problems such as duplicate migrations, invalid use of inter-satellite link bandwidth, and unreasonable cache replacement may occur. Therefore, a satellite network cache migration method based on regional feature prediction is needed to solve these problems. Summary of the Invention

[0004] To address the above problems, this invention provides the following technical solution: a satellite network cache migration method based on regional feature prediction, comprising: S1. Divide the satellite service area into multiple regions, and statistically analyze user content requests in the target region within the current completed statistical period according to a fixed length cycle. This yields the set of popular content corresponding to the current completed statistical period, the number of requests for each popular content, and the [data / information / resources] for each popular content. 3D content feature vector; S2. Based on the number of requests for each popular content and the... The region feature vector for the current completed statistical period is constructed from the dimensional content feature vector, and the region feature vector for the current completed statistical period is fused with the region feature vector for the historical period through time smoothing to obtain the region preference feature vector for the next service period. S3. When the source satellite is expected to leave the target area or enter the migration window, the region preference feature vector of the next service cycle is used to determine the content objects in the cached content set of the source satellite. The content feature vector is used to calculate the content importance of each content object, and a cache migration candidate set is obtained by filtering based on the content importance. S4. The source satellite generates a candidate content announcement and sends the candidate content announcement to the successor satellite via inter-satellite control signaling. The candidate content announcement includes the content name, content hash fingerprint, and content importance of each candidate content in the cache migration candidate set. The successor satellite compares the content hash fingerprint with the local cache hash set and removes the candidate content already cached by the successor satellite to obtain the missing content set. S5. Calculate the remaining dwell time based on the current position of the source satellite, the boundary position of the target area, and the operating speed. When the remaining dwell time meets the pre-triggering condition, establish a migration transmission channel between the source satellite and the successor satellite through an inter-satellite link. S6. The successor satellite generates a priority retrieval queue according to the content importance of each content object in the missing content set, sends an Interest request to the source satellite based on the priority retrieval queue, the source satellite returns a corresponding Data message, and the successor satellite writes the content objects in the Data message into its local cache and completes the cache update.

[0005] Furthermore, the step of dividing the satellite service area into multiple regions and statistically analyzing user content requests in the target region within the currently completed statistical period according to a fixed length cycle includes: The geographical area covered by the satellite is discretized into multiple grid regions with regional identifiers, and the target area is determined as at least one of the multiple grid regions; Record user request logs in the target area within the currently completed statistical period. The user request logs include the area identifier, content name, request time, and content feature description. The user request logs are aggregated and counted according to content name, and the top requests are selected from highest to lowest number. The content is used as the set of popular content; The aggregate count of each popular content in the set of popular content is taken as the number of requests for that popular content.

[0006] Furthermore, the method based on the number of requests for each popular content and the... The dimensional content feature vector constructs the regional feature vector for the currently completed statistical period, including: The number of requests for each popular content is normalized to obtain the content weight corresponding to each popular content. The content weight corresponding to each popular content and its corresponding The feature vectors of the content in dimension are weighted and summed to obtain the feature vector of the region whose statistical period has been completed; where, the first... The content weight of popular content satisfies: ,in, For the first The number of requests for popular content. The number of items in the popular content collection; The th region feature vector that has completed the current statistical period The preference strengths in each dimension satisfy: ,in, For the first Popular content The content feature vector in the dimensional dimension is at the 1st dimension. Feature values ​​in each dimension.

[0007] Furthermore, the process of performing time-smooth fusion of the regional feature vector of the currently completed statistical period with the regional feature vector of the historical period to obtain the regional preference feature vector for the next service period satisfies: ,in, This will be the regional preference feature vector for the next service cycle. This represents the feature vector of the region that has completed the current statistical period. This represents the regional feature vectors of the historical cycle. This is a smoothing factor.

[0008] Furthermore, the step of basing the regional preference feature vector of the next service cycle on the content objects in the cached content set of the source satellite... A dimensional content feature vector is used to calculate the content importance of each content object, including: Retrieve each content object from the cached content set of the source satellite. 3D content feature vector; Calculate the regional preference feature vector for the next service cycle and the relationship between each content object. The inner product of the feature vectors of the content dimension; The vector inner product is used as the content importance of the corresponding content object relative to the target region in the next service cycle.

[0009] Furthermore, the step of filtering the cache migration candidate set based on the content importance includes: The importance threshold is determined based on migration resource constraints, which include at least one of the following: available migration bandwidth for inter-satellite links, estimated remaining dwell time, available cache capacity of successor satellites, and content object size. Content objects whose content importance is greater than or equal to the importance threshold are added to the cache migration candidate set; When the amount of data corresponding to the cache migration candidate set is greater than the available cache capacity of the subsequent satellite, the content objects in the cache migration candidate set are truncated according to their content importance from high to low.

[0010] Furthermore, the successor satellite compares the content hash fingerprint with the local cache hash set, removes candidate content already cached by the successor satellite, and obtains a set of missing content, including: The subsequent satellite generates the local cache hash set based on the content hash fingerprint of the content already stored in the local cache. The subsequent satellite parses the candidate content announcement to obtain the content name, content hash fingerprint, and content importance of each candidate content in the cache migration candidate set; When the content hash fingerprint of a candidate content exists in the local cache hash set, the candidate content is marked as cached content; When the content hash fingerprint of a candidate content does not exist in the local cache hash set, the candidate content is added to the missing content set.

[0011] Furthermore, the pre-triggering condition is that the remaining dwell time is less than or equal to the pre-triggering time threshold. The pre-triggering time threshold is determined based on the total amount of data in the missing content set, the allocated bandwidth of the migration transmission channel, and the migration protection time. The remaining dwell time is determined based on the ratio of the distance from the current position of the source satellite along the running direction to the boundary position of the target area to the running speed of the source satellite.

[0012] Furthermore, sending an Interest request to the source satellite based on the priority retrieval queue includes: The subsequent satellite sorts the content objects in the missing content set in descending order according to the content importance of each content object in the missing content set, and generates the priority retrieval queue. The subsequent satellite, according to the order of the content objects in the priority retrieval queue, sequentially constructs an Interest request carrying the content name and region identifier and sends it to the source satellite; The source satellite searches for the corresponding content object in its local cache based on the content name in the Interest request, and constructs a data structure carrying the content name, content hash fingerprint, and other information. Data messages containing dimensional content feature vectors and content payloads.

[0013] Furthermore, the subsequent satellite writes the content object from the Data message into its local cache and completes the cache update, including: After receiving the Data message, the subsequent satellite confirms that the received content matches the content object in the missing content set based on the content hash fingerprint in the Data message. Once consistency is confirmed and writable space exists in the local cache, the corresponding content object is written to the local cache; If consistency is confirmed and there is no writable space in the local cache, cache replacement is performed based on the content importance of the local cache content and the content importance of the object to be written. The migration transmission channel is closed when all content objects in the missing content set have been migrated, or when the preset migration time limit has been reached.

[0014] Compared with the prior art, the present invention has the following beneficial effects: A satellite network cache migration method based on regional feature prediction is proposed. This method statistically analyzes historical content requests in the target area into a regional feature vector and then smoothly integrates it with historical periodic features to form a regional preference feature vector for the next service cycle. This allows cache migration to no longer rely solely on global popularity or the number of single historical requests, but rather to select content to be migrated based on changes in content demand in subsequent service areas. By matching the regional preference feature vector with the content feature vector of the source satellite's cached content, a cache migration candidate set that better meets the needs of subsequent areas is selected, reducing the invalid transfer of low-value content during satellite handover. Candidate content announcements, content hash fingerprint comparisons, and the generation of missing content sets prevent the repeated migration of cached content on subsequent satellites, reducing bandwidth consumption caused by duplicate data packet transmissions in inter-satellite links. The migration transmission channel is triggered by the remaining dwell time, and subsequent satellites generate priority retrieval queues based on content importance, ensuring that high-value cached content updates are completed first within a limited migration window. This improves the matching degree between cached content and regional request preferences during satellite coverage area handover, reducing the impact of invalid cache migration and unreasonable cache replacement on the continuity of content distribution. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall process of the method provided by the present invention; Figure 2 This is a schematic diagram of the ICN-LEO satellite network cache migration scenario provided by the present invention; Figure 3 This is a schematic diagram of the extended structure of the Interest request message provided by the present invention; Figure 4 This is a schematic diagram of the extended structure of the Data message provided by the present invention; Figure 5 The timing diagram of the collaborative migration between the source satellite and the successor satellite provided by this invention; Figure 6 This is a diagram illustrating the impact of cache migration on average cache hit rate provided by the present invention. Figure 7 The diagram showing the impact of cache migration on the average number of hops provided by this invention; Figure 8 This invention provides a diagram illustrating the impact of cache migration on average request latency. Figure 9 The diagram showing the impact of content quantity on average cache hit rate provided by this invention; Figure 10 The diagram showing the impact of the amount of content on average request latency provided by this invention; Figure 11 The diagram showing the impact of the amount of content provided by this invention on the average number of hops; Figure 12 A graph illustrating the impact of node cache capacity on average cache hit rate provided by this invention; Figure 13 A diagram illustrating the impact of node cache capacity on average request latency provided by this invention; Figure 14 The diagram illustrates the impact of node cache capacity on average hop count, as provided by this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] Please see Figure 1 This invention provides a satellite network cache migration method based on regional feature prediction, comprising: S1. Divide the satellite service area into multiple regions, and statistically analyze user content requests in the target region within the current completed statistical period according to a fixed length cycle. This yields the set of popular content corresponding to the current completed statistical period, the number of requests for each popular content, and the [data / information / resources] for each popular content. 3D content feature vector; Specifically, in combination Figure 1 and Figure 2 As shown, low-Earth orbit satellites move continuously along their orbits and periodically cover different geographical areas. The satellite service area can be divided into multiple regions according to latitude and longitude, ground grid numbering, or service coverage units. The target area is the area currently being served by the source satellite or about to be handed over to a successor satellite for continued service. Within a fixed-length statistical period, the source satellite receives content requests from ground users, ground stations, or edge access nodes and records the content name, request time, area identifier, and content feature description in the request. The content feature description can consist of content category, service type, semantic tags, file type, or service attributes, and is encoded as follows: A multidimensional content feature vector enables comparisons of different contents within the same feature space.

[0019] Furthermore, in one embodiment provided in this application, the step of dividing the satellite service area into multiple regions and statistically analyzing user content requests in the target region within the currently completed statistical period according to a fixed length cycle includes: The geographical area covered by the satellite is discretized into multiple grid regions with regional identifiers, and the target area is determined as at least one of the multiple grid regions; Record user request logs in the target area within the currently completed statistical period. The user request logs include the area identifier, content name, request time, and content feature description. The user request logs are aggregated and counted according to content name, and the top requests are selected from highest to lowest number. The content is used as the set of popular content; The aggregate count of each popular content in the set of popular content is taken as the number of requests for that popular content.

[0020] Specifically, each grid area has a unique area identifier, which can be written into an extended field of the interest request, such as... Figure 3 As shown. In addition to the content name, random number, and lifecycle fields, interest requests can also carry region identifiers, preference information, and feature information in extended fields. The region identifier indicates the target region from which the request originated; the preference information carries user-side or region-side preferences; and the feature information carries a description of the content's characteristics. After receiving an interest request, the satellite node assigns the request to the corresponding grid region based on the region identifier and performs an aggregation count based on the content name field. At the end of the statistical period, the satellite node selects the top-performing content based on the aggregation count. The set of popular content is used as the current set of popular content for the completed statistical period. For content with the same number of requests, it can be sorted according to the most recent request time, content object size, or content feature completeness to ensure that the set of popular content has a definite generation result.

[0021] S2. Based on the number of requests for each popular content and the... The region feature vector for the current completed statistical period is constructed from the dimensional content feature vector, and the region feature vector for the current completed statistical period is fused with the region feature vector for the historical period through time smoothing to obtain the region preference feature vector for the next service period. Specifically, the regional feature vector is used to express the strength of preference for different content feature dimensions in the target region within the current completed statistical period. Since user requests within the same region can change over time, directly using the statistical results of a single period is easily affected by sudden surges in requests. Therefore, the regional feature vector of the current completed statistical period is fused with the regional feature vectors of historical periods through a time-smoothing process, ensuring that the regional preference feature vector for the next service period simultaneously reflects recent request changes and historical preference continuations. This regional preference feature vector is then used to evaluate the migration value of source satellite cached content to the target region in the next service period.

[0022] Furthermore, in one embodiment provided in this application, the step of basing the request count of each popular content on the... The dimensional content feature vector constructs the regional feature vector for the currently completed statistical period, including: The number of requests for each popular content is normalized to obtain the content weight corresponding to each popular content. The content weight corresponding to each popular content and its corresponding The feature vectors of the content in dimension are weighted and summed to obtain the feature vector of the region whose statistical period has been completed; where, the first... The content weight of popular content satisfies: ,in, For the first The number of requests for popular content. The number of items in the popular content collection; The th region feature vector that has completed the current statistical period The preference strengths in each dimension satisfy: ,in, For the first Popular content The content feature vector in the dimensional dimension is at the 1st dimension. Feature values ​​in each dimension.

[0023] Specifically, content weight Indicates the first The relative proportion of popular content in the request distribution of the target region. By multiplying this weight by the content feature vector dimension by dimension and summing the results, the region feature vector can be obtained. When a certain feature dimension has a high value in multiple popular content items, and these content items are requested frequently, the corresponding value of that dimension... The increase indicates that the target region has a stronger preference for this type of content feature during the statistical period. If the total number of requests during the statistical period is zero, the regional feature vector from the previous period can be used, or the regional feature vector can be set as the default equilibrium vector to avoid subsequent migration decisions being unable to be executed due to a short period of no requests.

[0024] Furthermore, in one embodiment provided in this application, the process of performing time-smooth fusion of the regional feature vector of the currently completed statistical period and the regional feature vector of the historical period to obtain the regional preference feature vector for the next service period satisfies: ,in, This will be the regional preference feature vector for the next service cycle. This represents the feature vector of the region that has completed the current statistical period. This represents the regional feature vectors of the historical cycle. This is a smoothing factor.

[0025] Specifically, smoothing factor This is used to adjust the degree to which historical preferences influence the prediction results. When user requests in the target area change rapidly, a smaller adjustment can be made. This allows the prediction results to better reflect the regional characteristics of the currently completed statistical period; when user requests in the target region are periodic or stable, a larger [percentage] can be used. This process allows the prediction results to retain more of the preference characteristics from historical periods. Through the above fusion processing, the regional preference feature vector for the next service period will not completely depend on the peak value of a single request, thereby reducing the misleading effect of sudden content requests on the selection of cache migration candidates.

[0026] S3. When the source satellite is expected to leave the target area or enter the migration window, the region preference feature vector of the next service cycle is used to determine the content objects in the cached content set of the source satellite. The content feature vector is used to calculate the content importance of each content object, and a cache migration candidate set is obtained by filtering based on the content importance. Specifically, in combination Figure 2 As shown, the source satellite gradually leaves the current target area during its orbital motion, and subsequent satellites will take over the service in that area. The source satellite can determine whether it is about to leave the target area or enter the migration window based on its orbital ephemeris, current position, target area boundary, and preset migration window. After entering the migration window, the source satellite evaluates the content objects in its locally cached content set and assigns each content object to a specific location. The content feature vector is matched with the regional preference feature vector for the next service period to obtain the content importance. The higher the content importance, the closer the content object is to the predicted request preference of the target region in the next service period, and the more suitable it is to be migrated to the subsequent satellite.

[0027] Furthermore, in one embodiment provided in this application, the step of basing the regional preference feature vector of the next service period on each content object in the cached content set of the source satellite... A dimensional content feature vector is used to calculate the content importance of each content object, including: Retrieve each content object from the cached content set of the source satellite. 3D content feature vector; Calculate the regional preference feature vector for the next service cycle and the relationship between each content object. The inner product of the feature vectors of the content dimension; The vector inner product is used as the content importance of the corresponding content object relative to the target region in the next service cycle.

[0028] Specifically, for any content object in the source satellite cache content set Its content feature vector is The regional preference feature vector for the next service cycle is The importance of content can be expressed as:

[0029] in, For content objects Relative to the importance of the content in the next service cycle for the target region, The regional preference feature vector for the next service cycle is in the 1st... Preference intensity across dimensions For content objects In the Feature values ​​in each dimension. Calculated. Used to filter and sort cached content, so that the migration process prioritizes content objects that match the predicted preferences of the target region.

[0030] Furthermore, in one embodiment provided in this application, the step of filtering the cache migration candidate set based on the content importance includes: The importance threshold is determined based on migration resource constraints, which include at least one of the following: available migration bandwidth for inter-satellite links, estimated remaining dwell time, available cache capacity of successor satellites, and content object size. Content objects whose content importance is greater than or equal to the importance threshold are added to the cache migration candidate set; When the amount of data corresponding to the cache migration candidate set is greater than the available cache capacity of the subsequent satellite, the content objects in the cache migration candidate set are truncated according to their content importance from high to low.

[0031] Specifically, the importance threshold can be set based on the amount of data that can be transmitted within the migration window and the available buffer space of the successor satellite. When the available migration bandwidth of the inter-satellite link is low, the remaining dwell time is short, or the available buffer capacity of the successor satellite is small, the importance threshold is increased so that only higher-value content enters the buffer migration candidate set. When migration resources are sufficient, the importance threshold is decreased so that more content related to regional preferences enters the buffer migration candidate set. If the total amount of candidate content data exceeds the available buffer capacity of the successor satellite, it is truncated according to the content importance from high to low, so that the content objects retained in the candidate set have higher regional adaptation value under limited buffer capacity.

[0032] S4. The source satellite generates a candidate content announcement and sends the candidate content announcement to the successor satellite via inter-satellite control signaling. The candidate content announcement includes the content name, content hash fingerprint, and content importance of each candidate content in the cache migration candidate set. The successor satellite compares the content hash fingerprint with the local cache hash set and removes the candidate content already cached by the successor satellite to obtain the missing content set. Specifically, in combination Figure 4 and Figure 5 As shown, after the source satellite forms a cache migration candidate set, it does not immediately send all content payloads, but first sends candidate content announcements. These announcements inform subsequent satellites which content has migration value, as well as the content name, hash fingerprint, and importance of each candidate. Upon receiving the candidate content announcements, the subsequent satellite first performs deduplication based on its local cache hash set to prevent content already in its local cache from being transmitted again via inter-satellite links. Through this processing order of announcement, comparison, and retrieval, the migration process only targets content objects that are actually missing by the subsequent satellite and have regional value.

[0033] Furthermore, in one embodiment provided in this application, the successor satellite compares the content hash fingerprint with the local cache hash set, removes candidate content already cached by the successor satellite, and obtains a missing content set, including: The subsequent satellite generates the local cache hash set based on the content hash fingerprint of the content already stored in the local cache. The subsequent satellite parses the candidate content announcement to obtain the content name, content hash fingerprint, and content importance of each candidate content in the cache migration candidate set; When the content hash fingerprint of a candidate content exists in the local cache hash set, the candidate content is marked as cached content; When the content hash fingerprint of a candidate content does not exist in the local cache hash set, the candidate content is added to the missing content set.

[0034] Specifically, the content hash fingerprint can be generated based on the content payload, content name, or the unique encoding of the content object. Subsequent satellites maintain a set of hashes for stored content in their local cache. Upon arrival of candidate content notifications, they read the content hash fingerprints of each candidate content and match them against the local cache hash set. If a match is successful, it means the subsequent satellite already possesses the content object, and no further pull request is generated. If a match fails, it means the content object is missing for the subsequent satellite, and its content name, content hash fingerprint, and content importance are written into the missing content set. Figure 4 The extended fields in the data message shown can carry content features, hash identifiers, preference information, and popularity information. The hash identifier is used for content consistency identification, the content features are used to carry content feature vectors, and the preference information and popularity information are used to help represent the importance or popularity of the content.

[0035] S5. Calculate the remaining dwell time based on the current position of the source satellite, the boundary position of the target area, and the operating speed. When the remaining dwell time meets the pre-triggering condition, establish a migration transmission channel between the source satellite and the successor satellite through an inter-satellite link. Specifically, the source satellite periodically acquires its current position and calculates the remaining distance required to reach the target area boundary along its operating direction, combining this with the target area boundary position. This remaining dwell time is then used to determine whether immediate migration transmission needs to be initiated. When the remaining dwell time is relatively long, the source satellite can continue to perform area request statistics and update candidate content. When the remaining dwell time is less than or equal to the pre-trigger time threshold, the source satellite and the successor satellite establish a migration transmission channel via an inter-satellite link. This migration transmission channel can be logically managed differently from ordinary service forwarding links to ensure the migration of high-value content is completed within a limited handover time.

[0036] Furthermore, in one embodiment provided in this application, the pre-triggering condition is that the remaining dwell time is less than or equal to a pre-triggering time threshold. The pre-triggering time threshold is determined based on the total amount of data in the missing content set, the allocated bandwidth of the migration transmission channel, and the migration protection time. The remaining dwell time is determined based on the ratio of the distance from the current position of the source satellite along the running direction to the boundary position of the target area to the running speed of the source satellite.

[0037] Specifically, let the distance from the current position of the source satellite along its direction of travel to the boundary of the target area be... The source satellite's orbital speed is The remaining stay time can be expressed as:

[0038] Let the total amount of data in the missing content set be... The allocated bandwidth of the migration transmission channel is The migration protection period is The pre-trigger time threshold can be expressed as:

[0039] When the following conditions are met: At that time, the migration transmission channel establishment process is initiated. The migration protection time is used to absorb the overhead of inter-satellite link establishment, control signaling interaction, and transmission retries, ensuring that the migration trigger does not occur too close to the moment the source satellite leaves the target area. If If changes occur, the pre-trigger time threshold can be updated based on the real-time allocable bandwidth to ensure that the triggering judgment is consistent with the current link conditions.

[0040] S6. The successor satellite generates a priority retrieval queue according to the content importance of each content object in the missing content set, sends an Interest request to the source satellite based on the priority retrieval queue, the source satellite returns a corresponding Data message, and the successor satellite writes the content objects in the Data message into its local cache and completes the cache update.

[0041] Specifically, in combination Figure 3 , Figure 4 and Figure 5 As shown, after the follow-up satellite generates the set of missing content, it sorts the missing content according to its importance and sends interest requests sequentially. The content name field in the interest request indicates the name of the content to be retrieved, the region identifier indicates the target region corresponding to the request, and the preference information and feature information carry region preference information or content feature information. Upon receiving the interest request, the source satellite searches for the corresponding content object in its local cache and constructs a data packet to return to the follow-up satellite. The content name field in the data packet corresponds to the content name, the content payload field carries the content payload, the signature field is used for content integrity verification, and the extended fields can carry content features, hash identifiers, preference information, and popularity information, so that the follow-up satellite can confirm the content and update its cache after receiving the data.

[0042] Furthermore, in one embodiment provided in this application, sending an Interest request to the source satellite based on the priority retrieval queue includes: The subsequent satellite sorts the content objects in the missing content set in descending order according to the content importance of each content object in the missing content set, and generates the priority retrieval queue. The subsequent satellite, according to the order of the content objects in the priority retrieval queue, sequentially constructs an Interest request carrying the content name and region identifier and sends it to the source satellite; The source satellite searches for the corresponding content object in its local cache based on the content name in the Interest request, and constructs a data structure carrying the content name, content hash fingerprint, and other information. Data messages containing dimensional content feature vectors and content payloads.

[0043] Specifically, subsequent satellites sort the content objects in the missing content set in descending order of content importance. Content objects with the same content importance can be prioritized based on their size, the most recent request time, or their order in the candidate content announcement. The content at the head of the priority retrieval queue is prioritized to generate interest requests, ensuring that even if the migration window is insufficient to transfer all missing content, high-value content can still be migrated first. Figure 5 In the migration sequence shown, the successor satellite enters the migration mode after the triggering conditions are met. It first receives the candidate content name and hash digest, then obtains the missing content set through local matching, then requests the first missing content and receives the corresponding Data message, writes it into the cache, and continues to request the next missing content until the missing content set is processed or the migration window ends.

[0044] Furthermore, in one embodiment provided in this application, the subsequent satellite writes the content object in the Data message into its local cache and completes the cache update, including: After receiving the Data message, the subsequent satellite confirms that the received content matches the content object in the missing content set based on the content hash fingerprint in the Data message. Once consistency is confirmed and writable space exists in the local cache, the corresponding content object is written to the local cache; If consistency is confirmed and there is no writable space in the local cache, cache replacement is performed based on the content importance of the local cache content and the content importance of the object to be written. The migration transmission channel is closed when all content objects in the missing content set have been migrated, or when the preset migration time limit has been reached.

[0045] Specifically, after receiving a Data packet, the follower satellite first reads the content hash fingerprint in the Data packet and checks its consistency with the content hash fingerprint of the corresponding content object in the missing content set. If they do not match, the Data packet is discarded or the corresponding Interest request is resent; if they match, it checks whether there is writable space in the local cache. If writable space exists, the content object is directly written to the local cache; if no writable space exists, the importance of the content object to be written is compared with the importance of replaceable content in the local cache, and the content object with higher importance is retained first. When the entire missing content set has been migrated, the follower satellite closes the migration transmission channel; if the preset migration time limit is reached but there is still unmigrated content, the follower satellite retains the content objects that have been written and ends the current migration process to avoid the migration transmission crowding out normal service forwarding resources in subsequent service cycles.

[0046] To further verify the technical effectiveness of this method, combined with Figures 6 to 14 This describes the experimental results of this embodiment. Figure 6 The impact of cache migration on average cache hit rate is shown. Under the same cache capacity, the average cache hit rate using this method is higher than that without it, and the hit rate increases overall with increasing cache capacity. This result indicates that simply increasing cache capacity cannot solve the problem of mismatch between cached content and user preferences in the new area after satellite coverage area switching; this method, through regional feature prediction and candidate content migration, ensures that subsequent satellites have cached content more in line with the preferences of that area before taking over the target area, thus making it easier for user requests to hit on the serving satellite or neighboring satellites.

[0047] Figure 7 The impact of cache migration on average hop count is shown. Under different cache capacities, the average hop count using this method is lower than that without it. This result indicates that without region-preference-based cache migration, user requests are more likely to continue being forwarded to remote satellites or ground servers after satellite handover; this method migrates high-value content to subsequent satellites in advance, enabling user requests to receive responses within a shorter path, thereby reducing inter-satellite multi-hop forwarding.

[0048] Figure 8 The impact of cache migration on average request latency is shown. As cache capacity increases, average request latency decreases in both scenarios, but the average request latency using this method remains consistently low. This result is consistent with... Figure 7 The change in the average number of hops corresponds to the fact that after the request path is shortened, the number of inter-satellite links traversed by interest requests and data messages is reduced, and the queuing, propagation and forwarding processing time is reduced accordingly.

[0049] Figure 9The impact of content quantity on average cache hit rate is shown. As the total content volume increases, the average cache hit rate of all caching strategies decreases because the cache space needs to cover more content objects, reducing the probability of a single piece of content being cached. Compared to strategies such as LCE, LCD, Prob 0.3, Prob 0.5, Prob 0.7, and Betw, this method maintains a high hit rate across different content volumes. This result indicates that this method does not simply place content based on path-passing nodes or topological centrality, but rather selects migration objects based on the regional preference feature vector of the target region for the next service cycle. Therefore, even with increased content diversity, it can still prioritize retaining content more relevant to regional needs.

[0050] Figure 10 The impact of content quantity on average request latency is shown. As the total content volume increases, the average request latency for all strategies generally rises, but the latency curve corresponding to this method remains at a lower level. This result indicates that when the content volume is large and cache contention is more pronounced, traditional path caching or probabilistic caching strategies are prone to issues such as duplicate caching or low-value content occupying cache space, requiring requests to return to the origin server or perform multi-hop lookups. This method, through content importance filtering and hash deduplication, allocates limited migration bandwidth and cache space to content objects with higher regional adaptability, thus suppressing the latency increase caused by the increase in content volume.

[0051] Figure 11 The impact of content quantity on average hop count is shown. As the total content volume increases, the average hop count increases for all strategies, but the average hop count of this method remains low. This result indicates that with a larger content set, user requests are less likely to be found using the nearest caching strategy. This method, through candidate content announcements, missing content sets, and priority retrieval queues, prioritizes subsequent satellites for content objects more likely to be requested in the target area, thereby reducing the need for requests to search for content across multiple satellite nodes.

[0052] Figure 12 The impact of node cache capacity on average cache hit rate is shown. As cache capacity increases, the average cache hit rate of each strategy improves, and our proposed method exhibits a high hit rate under all capacity conditions. This result demonstrates that, under the same cache resource conditions, our method can improve the matching degree between cached content and user requests through region preference prediction, ensuring that newly added cache capacity is used more to store high-value content in subsequent regions, rather than being occupied by path redundancy content or expired region preference content.

[0053] Figure 13The impact of node cache capacity on average request latency is shown. As node cache capacity increases, the average request latency of each strategy decreases overall, with the proposed method exhibiting the lowest average request latency. This result indicates that increasing cache capacity improves the probability of responding from the nearest node, while the proposed method further enhances this by using pre-migration filtering and priority retrieval during migration to ensure that high-value content is written to the local cache of subsequent satellites earlier, reducing the likelihood of users waiting for responses from remote nodes.

[0054] Figure 14 The impact of node cache capacity on average hop count is shown. As node cache capacity increases, the average hop count for each strategy decreases overall, with the average hop count corresponding to this method being at a low level. This result indicates that this method can translate cache capacity gains into path compression effects, enabling more requests to complete hits within the serving satellite or neighboring satellites. Compared to caching methods based solely on content path, probability, or topological centrality, this method combines regional feature prediction, hash deduplication, and priority retrieval within the migration window, ensuring that cached content maintains spatiotemporal matching with regional request preferences even when satellite movement causes changes in the service area.

[0055] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0056] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A satellite network cache migration method based on regional feature prediction, characterized in that, include: S1. Divide the satellite service area into multiple regions, and statistically analyze user content requests in the target region within the current completed statistical period according to a fixed length cycle. This yields the set of popular content corresponding to the current completed statistical period, the number of requests for each popular content, and the [data / information / resources] for each popular content. 3D content feature vector; S2, based on the number of requests for each of the popular content items and the... The region feature vector for the current completed statistical period is constructed from the dimensional content feature vector, and the region feature vector for the current completed statistical period is fused with the region feature vector for the historical period through time smoothing to obtain the region preference feature vector for the next service period. S3. When the source satellite is expected to leave the target area or enter the migration window, the region preference feature vector of the next service cycle is used to determine the content objects in the cached content set of the source satellite. The content feature vector is used to calculate the content importance of each content object, and a cache migration candidate set is obtained by filtering based on the content importance. S4. The source satellite generates a candidate content announcement and sends the candidate content announcement to the successor satellite via inter-satellite control signaling. The candidate content announcement includes the content name, content hash fingerprint, and content importance of each candidate content in the cache migration candidate set. The successor satellite compares the content hash fingerprint with the local cache hash set and removes the candidate content already cached by the successor satellite to obtain the missing content set. S5. Calculate the remaining dwell time based on the current position of the source satellite, the boundary position of the target area, and the operating speed. When the remaining dwell time meets the pre-triggering condition, establish a migration transmission channel between the source satellite and the successor satellite through an inter-satellite link. S6. The successor satellite generates a priority retrieval queue according to the content importance of each content object in the missing content set, sends an Interest request to the source satellite based on the priority retrieval queue, the source satellite returns a corresponding Data message, and the successor satellite writes the content objects in the Data message into the local cache and completes the cache update. The subsequent satellite compares the content hash fingerprint with the local cache hash set, removes candidate content already cached by the subsequent satellite, and obtains a set of missing content, including: The subsequent satellite generates the local cache hash set based on the content hash fingerprint of the content already stored in the local cache. The subsequent satellite parses the candidate content announcement to obtain the content name, content hash fingerprint, and content importance of each candidate content in the cache migration candidate set; When the content hash fingerprint of a candidate content exists in the local cache hash set, the candidate content is marked as cached content; When the content hash fingerprint of a candidate content does not exist in the local cache hash set, the candidate content is added to the missing content set. The pre-triggering condition is that the remaining dwell time is less than or equal to the pre-triggering time threshold. The pre-triggering time threshold is determined based on the total amount of data in the missing content set, the allocated bandwidth of the migration transmission channel, and the migration protection time. The remaining dwell time is determined based on the ratio of the distance from the current position of the source satellite along the running direction to the boundary position of the target area to the running speed of the source satellite. The pre-trigger time threshold satisfies ,in, The total amount of data in the set of missing content. Allocate bandwidth for the migration transmission channel. The migration protection time is defined as follows: when the allocated bandwidth of the migration transmission channel changes, the pre-trigger time threshold is updated based on the real-time allocable bandwidth.

2. The satellite network cache migration method based on regional feature prediction according to claim 1, characterized in that, The process of dividing the satellite service area into multiple regions and statistically analyzing user content requests in the target region within the current statistical period according to a fixed length cycle includes: The geographical area covered by the satellite is discretized into multiple grid regions with regional identifiers, and the target area is determined as at least one of the multiple grid regions; Record user request logs in the target area within the currently completed statistical period. The user request logs include the area identifier, content name, request time, and content feature description. The user request logs are aggregated and counted according to content name, and the top requests are selected from highest to lowest number. The content is used as the set of popular content; The aggregate count of each popular content in the set of popular content is taken as the number of requests for that popular content.

3. The satellite network cache migration method based on regional feature prediction according to claim 1, characterized in that, The number of requests for each popular content and the The dimensional content feature vector constructs the regional feature vector for the currently completed statistical period, including: The number of requests for each popular content is normalized to obtain the content weight corresponding to each popular content. The content weight corresponding to each popular content and its corresponding The weighted summation of the feature vectors of the content in dimension 1 yields the feature vector of the region whose statistical period has been completed; where, the th... The content weight of popular content satisfies: ,in, For the first The number of requests for popular content. The number of items in the popular content collection; The th region feature vector that has completed the current statistical period The preference strengths in each dimension satisfy: ,in For the first Popular content The content feature vector in the dimensional dimension is at the 1st dimension. Feature values ​​in each dimension.

4. The satellite network cache migration method based on regional feature prediction according to claim 3, characterized in that, The process of time-smoothingly fusing the regional feature vector of the currently completed statistical period with the regional feature vector of the historical period to obtain the regional preference feature vector for the next service period satisfies: ,in, This will be the regional preference feature vector for the next service cycle. This represents the feature vector of the region that has completed the current statistical period. This represents the regional feature vectors of the historical cycle. This is a smoothing factor.

5. The satellite network cache migration method based on regional feature prediction according to claim 1, characterized in that, The method is based on the regional preference feature vector of the next service cycle and each content object in the cached content set of the source satellite. A dimensional content feature vector is used to calculate the content importance of each content object, including: Retrieve each content object from the cached content set of the source satellite. 3D content feature vector; Calculate the regional preference feature vector for the next service cycle and the relationship between each content object. The inner product of the feature vectors of the content dimension; The vector inner product is used as the content importance of the corresponding content object relative to the target region in the next service cycle.

6. The satellite network cache migration method based on regional feature prediction according to claim 1, characterized in that, The step of filtering the cache migration candidate set based on the content importance includes: The importance threshold is determined based on migration resource constraints, which include at least one of the following: available migration bandwidth for inter-satellite links, estimated remaining dwell time, available cache capacity of successor satellites, and content object size. Content objects whose content importance is greater than or equal to the importance threshold are added to the cache migration candidate set; When the amount of data corresponding to the cache migration candidate set is greater than the available cache capacity of the subsequent satellite, the content objects in the cache migration candidate set are truncated according to their content importance from high to low.

7. The satellite network cache migration method based on regional feature prediction according to claim 1, characterized in that, Sending an Interest request to the source satellite based on the priority retrieval queue includes: The subsequent satellite sorts the content objects in the missing content set in descending order according to the content importance of each content object in the missing content set, and generates the priority retrieval queue. The subsequent satellite, according to the order of the content objects in the priority retrieval queue, sequentially constructs an Interest request carrying the content name and region identifier and sends it to the source satellite; The source satellite searches for the corresponding content object in its local cache based on the content name in the Interest request, and constructs a data structure carrying the content name, content hash fingerprint, and other information. Data messages containing dimensional content feature vectors and content payloads.

8. The satellite network cache migration method based on regional feature prediction according to claim 1, characterized in that, The subsequent satellite writes the content object from the Data message into its local cache and completes the cache update, including: After receiving the Data message, the subsequent satellite confirms that the received content matches the content object in the missing content set based on the content hash fingerprint in the Data message. Once consistency is confirmed and writable space exists in the local cache, the corresponding content object is written to the local cache; If consistency is confirmed and there is no writable space in the local cache, cache replacement is performed based on the content importance of the local cache content and the content importance of the object to be written. The migration transmission channel is closed when all content objects in the missing content set have been migrated, or when the preset migration time limit has been reached.

Images