A multi-link transmission and synchronization method for data of an overwater photovoltaic inspection system

Abstract

This invention discloses a multi-link transmission and synchronization method for data from a floating photovoltaic inspection system. It involves uniformly time-calibrating and encoding image, video, equipment operation, and environmental monitoring data generated during the inspection process to form structured data units. Simultaneously, it acquires status information such as bandwidth, latency, packet loss, and stability of various communication links. A transmission strategy space is constructed based on data characteristics and link status, and a preferred strategy is selected through hierarchical optimization. An iterative optimization mechanism is used to dynamically adjust the strategy, generating a stable transmission scheme. Based on this scheme, data is fragmented, distributed, and transmitted in parallel across multiple links, with link switching and retransmission control during transmission. Finally, data synchronization and transmission result output are achieved through reception confirmation and statistical analysis. This method enables collaborative scheduling of multi-source data and multi-link resources, improving the stability and real-time performance of data transmission.

Description

Technical Field

[0001] This invention relates to the field of data communication and intelligent inspection technology, and in particular to a multi-link transmission and synchronization method for data of a waterborne photovoltaic inspection system. Background Technology

[0002] With the continuous expansion of the scale of floating photovoltaic power stations and the increasing demand for intelligent operation and maintenance, inspection methods based on drones and various types of sensing devices are gradually becoming mainstream. The multi-source data generated during inspections, including images, videos, equipment operation data, and environmental monitoring data, needs to be transmitted back to the control center in real time via wireless communication links for centralized analysis and operation and maintenance decision-making. Currently, multi-link communication technology has been introduced into floating photovoltaic inspection scenarios, using various links such as cellular networks, satellite communication, and private networks for data transmission to improve coverage and transmission reliability.

[0003] Existing data transmission technologies still have significant shortcomings in complex aquatic environments. On the one hand, multi-source inspection data lacks a unified time alignment and structured processing mechanism during generation, resulting in time deviations and format differences between different data, making it difficult to efficiently integrate and synchronize the data, thus affecting the accuracy of subsequent analysis. On the other hand, existing multi-link transmission methods mostly rely on static or empirical rules for link selection and switching, lacking comprehensive modeling and dynamic optimization capabilities for link bandwidth, latency, packet loss, and stable states. This easily leads to problems such as increased transmission delay, higher packet loss rates, or frequent link switching in environments with fluctuating links. Furthermore, traditional transmission strategies lack systematic optimization and iterative adjustment mechanisms, making it difficult to obtain stable and effective transmission solutions under complex constraints. Moreover, the lack of fine-grained scheduling control during data fragmentation, diversion, and retransmission results in insufficient data transmission efficiency and synchronization performance, making it difficult to meet the high reliability and real-time requirements of aquatic photovoltaic inspection scenarios.

[0004] Therefore, how to provide a multi-link transmission and synchronization method for data from a floating photovoltaic inspection system is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention proposes a multi-link transmission and synchronization method for data from a floating photovoltaic inspection system. By constructing a unified encoding and time alignment mechanism for multi-source inspection data, combined with multi-link state perception and modeling, a transmission strategy space is established, and a stable transmission strategy is generated using hierarchical optimization and iterative search methods.

[0006] A method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system according to an embodiment of the present invention includes the following steps: S1. Collect image data, video data, equipment operation data and environmental monitoring data generated by the inspection of the water photovoltaic system, and perform timestamp calibration and unified encoding and encapsulation on the multi-source data to generate structured inspection data units; S2. Obtain the real-time status parameters of cellular communication links, satellite communication links, wireless private network links, and self-organizing network relay links to form a link status vector; S3. Receive structured inspection data units and link status vectors, establish a multi-link transmission optimization model, map link selection parameters, link switching timing parameters and transmission redundancy parameters into unified strategy variables, and generate a candidate transmission strategy space. S4. Perform a nested partitioning optimization algorithm on the candidate transmission strategy space, and divide and shrink the candidate transmission strategy space hierarchically according to the comprehensive evaluation index, and output the optimal strategy subspace. S5. Execute the delayed acceptance hill climbing algorithm in the preferred strategy subspace, make acceptance decisions based on historical window evaluation results and iteratively update strategy variables to generate a stable transmission strategy. S6. Based on the stable transmission strategy, perform fragmentation, diversion scheduling, link switching and parallel transmission on the structured inspection data unit, and output the transmission results.

[0007] Optionally, S2 includes: S21. Generate link identifiers for cellular communication links, satellite communication links, wireless private network links, and self-organizing network relay links respectively, and establish a status acquisition context for each link. The status acquisition context includes the acquisition period, statistical window length, and probe message format. S22. In each acquisition cycle, send a bandwidth probe message to each link, record the sending time and receiving confirmation time of the bandwidth probe message, calculate the bandwidth value according to the number of effective payload bytes that have been confirmed within the statistical window, and bind the bandwidth value to the link identifier. S23. In each acquisition cycle, send a delay probe message to each link, record the sending time and receiving confirmation time of the delay probe message, calculate the round-trip delay and obtain the one-way transmission delay value according to the preset mapping rule, and bind the transmission delay value with the link identifier. S24. In each acquisition cycle, accumulate the probe message transmission count and reception acknowledgment count for each link, generate a packet loss rate value based on the unacknowledged message count within the statistical window, and bind the packet loss rate value to the link identifier. S25. In N consecutive collection cycles, the bandwidth, transmission delay and packet loss rate of each link are time-series merged, the volatility of each index is calculated and synthesized into a link stability value according to a preset weight. The link stability value is obtained by reverse mapping of the normalized volatility. S26. For each link, encapsulate the bandwidth value, transmission delay value, packet loss rate value, and link stability value in a fixed field order to form a link state vector, and write the collection period timestamp into the link state vector. The link state vector is represented as follows: ,in For link identification, This is the bandwidth value. This is the transmission delay value. This is the packet loss rate value. This represents the link stability value. S27. Align the link state vectors corresponding to each link according to the timestamp of the acquisition period, write them into the link state buffer, and output the link state vector set.

[0008] Optionally, S3 includes: S31. In each decision cycle, retrieve the structured inspection data unit from the pending transmission buffer queue, read the data type identifier, timestamp, data length and priority identifier, establish a data sub-queue according to the data type identifier, and write the deadline transmission time and target arrival rate to each data sub-queue. S32. Within the same decision-making cycle, retrieve the link state vectors corresponding to each link from the link state buffer, select the latest set of link state vectors based on the collection cycle timestamp, and remove link state vectors with bandwidth values ​​of zero or missing reception confirmations to obtain the set of available links. S33. For each link in the set of available links, calculate the upper limit of bytes that can be sent within the decision period based on the bandwidth value, calculate the number of available transmission time slots based on the transmission delay value, and write the upper limit of bytes that can be sent and the number of available transmission time slots into the link parameter table. S34. For each data sub-queue, select the data unit at the head of the queue in sequence and calculate the latest transmission start point. Match the latest transmission start point with the transmission delay value in the link parameter table item by item. Delete the link that causes the data unit with the latest transmission start point to be earlier than the current time to be unable to meet the requirement, and obtain the list of optional links corresponding to the data sub-queue. S35. For each data subqueue, generate a set of link selection parameter values ​​in the optional link list. The set of link selection parameter values ​​consists of single link selection items and parallel link combination items. Write the number of parallel links and the splitting ratio for each combination item. S36. Generate a set of link switching timing parameter values ​​for each data subqueue. The set of link switching timing parameter values ​​includes three types of values: trigger threshold, hold duration, and cooldown duration. The trigger threshold is composed of a threshold pair of link stability value and packet loss rate value. The hold duration and cooldown duration are represented by decision cycle count. S37. Generate a set of transmission redundancy parameter values ​​for each data sub-queue. The set of transmission redundancy parameter values ​​includes the redundancy fragmentation ratio value and the retransmission count limit value. The redundancy fragmentation ratio value corresponds to the splitting ratio of the parallel link combination item and meets the preset limit. S38. Enumerate and generate policy configuration records by combining "data sub-queue - link selection parameter value - link switching timing parameter value - transmission redundancy parameter value", assign policy identifiers, and write all policy configuration records into the candidate policy library to form a candidate transmission policy space.

[0009] Optionally, S4 includes: S41. Read all policy configuration records from the candidate transmission policy space from the candidate policy library, extract the data sub-queue identifier, target link or parallel link combination, switching trigger threshold, hold duration, cooldown duration, redundancy fragmentation ratio and retransmission limit for each policy configuration record, and establish a policy record table. S42. Construct a constraint signature for each policy configuration record in the policy record table. The constraint signature is obtained by concatenating the deadline transmission time satisfaction flag, link capacity satisfaction flag, packet loss constraint satisfaction flag, and handover cooling satisfaction flag of each data sub-queue in a fixed order, and write the constraint signature into the corresponding policy configuration record. S43. Perform first-level nested partitioning on the policy record table according to constraint signatures to generate N signature partitions. Each signature partition contains only policy configuration records with completely identical constraint signatures. Delete any policy configuration record in each signature partition whose constraint satisfaction flag is not found. This yields a set of first-level feasible partitions. S44. For each partition in the first-level feasible partition set, generate a partition skeleton description. The partition skeleton description includes: the set of data sub-queues involved in the partition, the set of allowed link combinations in the partition, the range of allowed redundancy shard ratios in the partition, and the range of allowed switching thresholds in the partition. Write the partition skeleton description into the partition index table. S45. For each partition in the partition index table, construct a set of partition representative points according to the partition skeleton description; S46. Calculate the comprehensive evaluation index value for each representative point set of each partition. The comprehensive evaluation index value is obtained by summarizing the estimated average transmission delay, estimated packet loss rate, link bandwidth occupancy ratio and synchronization deviation in a fixed field order, and write the comprehensive evaluation index value of each representative point into the partition index table. S47. Generate a partition evaluation value for each partition. The partition evaluation value is obtained by sorting the comprehensive evaluation index values ​​of the partition representative point set and taking the median value as the main value. The "count of any constraint violation within the representative point" is added to form a penalty term. The partition evaluation value is then written into the partition index table. S48. Select the partition with the best partition evaluation value from the partition index table as the current active partition. Decompose the policy variable domain in the current active partition into the structure domain and the parameter domain. The structure domain includes the link combination and the number of parallel links, and the parameter domain includes the handover trigger threshold, hold duration, cooldown duration, redundant fragmentation ratio and retransmission limit. S49. Perform a second-level nested partitioning on the currently active partition. The construction process of the second-level nested partitioning includes: Within the structural domain, policy configuration records are divided into structural sub-partitions based on the "number of parallel links"; Within each structural sub-partition, the policy configuration records are divided into parameter sub-partitions by interval partitioning according to the "redundancy sharding ratio"; Within each parameter sub-partition, the policy configuration records are divided into intervals according to the "switch trigger threshold pair" to obtain the threshold sub-partition; Within each threshold sub-partition, leaf partitions are obtained by meshing according to the ordered pairs of "hold duration and cooling duration"; S410. Repeat the representative point construction and partition evaluation value generation steps for each leaf partition obtained from the second-level nested partition, and set the leaf partition with the best partition evaluation value as the next round of active partition. S411. During each round of active partitioning updates, perform boundary locking on the partition skeleton description of the previous round of active partitioning. Boundary locking includes: preventing the values ​​of the structure fields that have been deleted in the previous round of active partitioning from reappearing in the next round of active partitioning; marking the parameter intervals that have been determined to be infeasible in the previous round of active partitioning as disabled intervals and skipping them in the next round of partitioning. S412. When the number of policy configuration records in the active partition is lower than the preset threshold or the partition level reaches the preset upper limit, stop partitioning, output all policy configuration records in the active partition as the preferred policy subspace, and write the preferred policy subspace into the preferred policy library.

[0010] Optionally, the construction process of the partition representative point set includes: generating a structure key using "data sub-queue set - link combination set" as the key; performing deterministic mapping on the structure key to obtain a fixed sequence number set; configuring records as representative points according to the strategy of locating the corresponding records in the partition according to the fixed sequence number set; when the number of locatable records in the partition is insufficient, supplementing the representative points in the order of "link combination contains fewer parallel links - smaller redundant sharding ratio - shorter handover retention time".

[0011] Optionally, S5 includes: S51. Read the set of policy configuration records in the preferred policy subspace from the preferred policy library, generate a structure signature and a parameter signature for each policy configuration record. The structure signature is obtained by concatenating "data subqueue identifier - target link or parallel link combination - number of parallel links" in a fixed order. The parameter signature is obtained by concatenating "switching trigger threshold - hold duration - cooldown duration - redundancy fragmentation ratio - maximum number of retransmissions" in a fixed order. Write the structure signature and parameter signature into the corresponding policy configuration record. S52. Read the capacity limit, latency limit and packet loss limit of this decision cycle from the link parameter table and data sub-queue constraints. Perform an executability review on each policy configuration record. The policy configuration records that pass the review constitute an iterable set. Select the policy configuration record with the smallest comprehensive evaluation index value from the iterable set as the initial current solution. S53. Establish a history window table. The history window table contains sequential slots of a preset integer L and a delayed reference table indexed by the structure signature. The sequential slots are used to write the comprehensive evaluation index value and structure signature of the current solution according to the iteration number. S54. Calculate the comprehensive evaluation index value for the initial current solution and write it into the first sequential slot of the history window table. At the same time, register the delayed reference evaluation value and delayed reference iteration number in the delayed reference table with the structural signature of the initial current solution. S55. Construct a conflict location table. The process of constructing the conflict location table includes: calculating the capacity margin and latency margin of each link in the link combination corresponding to the initial current solution, counting the unmet count and trigger switching count of each data sub-queue, and writing "the link identifier with the smallest capacity margin, the link identifier with the smallest latency margin, the sub-queue identifier with the largest trigger switching count, and the link identifier with the largest packet loss rate" into the conflict location table. S56. At the beginning of each iteration, a neighborhood generation list is generated based on the conflict location table. The neighborhood generation list lists the policy variables to be disturbed in a fixed priority order. The fixed priority order includes: the diversion ratio associated with the link with the smallest capacity margin, the target link selection associated with the link with the smallest latency margin, the handover trigger threshold pair associated with the sub-queue with the largest handover count, and the redundant fragmentation ratio and retransmission limit associated with the link with the largest packet loss rate. S57. Generate a set of candidate solutions item by item based on the neighborhood generation list; S58. Calculate the comprehensive evaluation index value for each candidate solution set in the order of generation, extract the structural signature for each candidate solution, read the delayed reference evaluation value from the delayed reference table according to the structural signature, and read the sequential slot evaluation value before the Lth one in the history window table as the delayed reference evaluation value when the structural signature is not hit. S59. Perform a delayed acceptance determination, which satisfies the following steps: First determination: If the comprehensive evaluation index value of the candidate solution is not greater than the delay reference evaluation value of the corresponding structural signature, the candidate solution is set as a solution to be accepted; otherwise, the candidate solution is set as a solution to be rejected. Second determination: When the structural signature of the solution to be accepted is different from the structural signature of the current solution, calculate the structural change mark of the solution to be accepted relative to the current solution, and read the structural change mark count of the most recent L sequential slots in the history window table. When the structural change mark count is greater than the preset upper limit K, set the solution to be accepted as a rejected solution; when the structural change mark count is not greater than the preset upper limit K, retain the solution to be accepted. The delay reference evaluation value is generated using the delay reference table of the structure signature index and updated according to the following formula: ; in, For structural signature, The iteration number, The interval from the current iteration number in the history window table is... The sequential slot evaluation value, For structural signature In iteration number The corresponding delay reference evaluation value, In recent Structural signature within each sequential slot The count of structural changes that occurred. This is the penalty coefficient; S510. When there is a solution to be accepted, replace the current solution with the solution to be accepted, and write the comprehensive evaluation index value and structural signature of the solution to be accepted into the next sequential slot of the history window table. At the same time, update the delayed reference evaluation value and delayed reference iteration number in the delayed reference table with the structural signature of the solution to be accepted. When there is no solution to be accepted, write the comprehensive evaluation index value and structural signature of the current solution into the next sequential slot of the history window table. S511. Update the conflict location table after each write to the history window table. The update process includes: recalculating the capacity margin and latency margin with the link combination corresponding to the current solution, re-counting the unmet count and the trigger switch count, and rewriting the rules to overwrite the conflict location table. S512. When the number of iterations reaches the preset upper limit or the current solution is not updated for a preset number of consecutive iterations in the history window table, the iteration is terminated. The current solution at the time of termination is output as a stable transmission strategy and written into the stable strategy library.

[0012] Optionally, a set of candidate solutions is generated item by item based on the neighborhood generation list. The generation rules include: The link selection parameters are modified only once, either by replacing the target link with the next candidate link in the list of available links, or by replacing the parallel link combination with the next candidate combination with the same number of parallel links. The splitting ratio is adjusted in steps only if the parallel link combination remains unchanged. The ratio adjustment uses a preset step table and ensures that the ratio sums are 1. For the switching trigger threshold pair, only one-sided step adjustment of the threshold pair is performed, and the hold duration and cooldown duration are only adjusted once in an ordered pair step. The redundant fragmentation ratio and the upper limit of retransmission times are only adjusted once. The adjustment meets the preset linkage table that "the upper limit of retransmission times is lowered when the redundant fragmentation ratio is increased, and the upper limit of retransmission times is increased when the redundant fragmentation ratio is lowered". Boundary constraint verification is performed immediately after each candidate solution is generated. Candidate solutions that fail the boundary constraint verification are not included in the candidate solution set.

[0013] Optionally, S6 includes: S61. Read the structured inspection data units from the queue to be transmitted in the order of timestamps, generate the transmission batch number, determine the number of fragments allocated to each link according to the target link or parallel link combination and diversion ratio in the stable transmission strategy, calculate the total number of fragments, and write the transmission batch number and the total number of fragments into the transmission record table. S62. Perform sequential segmentation of the structured inspection data unit according to the preset maximum load length of the segment to generate segmented data units. Write the segment sequence number, total number of segments, sending batch number and target link identifier to each segmented data unit, and register the segment enqueue timestamp. S63. Write the fragmented data unit into the corresponding link transmission buffer. Record the transmission timestamp and start the acknowledgment timer when transmitting. Fragments whose acknowledgment timer expires are marked as fragments to be retransmitted. Fragments whose retransmission count reaches the upper limit of the retransmission count are written into the loss record table. S64. When the switching trigger threshold is met and the cooldown time count is zero, the target link identifier for the unsent fragments is recalculated and migrated to the new link sending buffer. At the same time, the link switching timestamp and the migrated fragment sequence number are recorded. S65. The receiving end establishes a reassembly record table according to the sending batch number, writes it into the reassembly buffer according to the fragment sequence number and updates the arrival bitmap. When the arrival bitmap is complete, it generates a reception acknowledgment message and returns it to the sending end. S66. After receiving the acknowledgment message, the sending end writes an acknowledgment timestamp for the corresponding fragment. S67. At the end of the preset statistical period, count the total number of fragments sent. Confirm the number of fragments Number of lost fragments Generate packet loss rate information The average transmission delay is calculated by summarizing the transmission delay samples of each fragment, and the transmission result record is formed and output together with the sending batch number and the number of confirmed fragments.

[0014] The beneficial effects of this invention are: This invention introduces a nested partitioning optimization algorithm to hierarchically divide and gradually shrink the candidate transmission strategy space. Under the premise of satisfying multiple constraints, it achieves fast screening and effective compression of the strategy space, avoids the problem of excessive computational complexity caused by global traversal, and improves strategy search efficiency and global optimization capability.

[0015] This invention employs a delayed acceptance hill-climbing algorithm, which combines historical window evaluation results with the strategy iteration process to make acceptance decisions. This effectively avoids getting stuck in local optima and improves optimization accuracy while ensuring search stability. As a result, the generated transmission strategy has higher stability and adaptability in complex dynamic environments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a multi-link transmission and synchronization method for data in a waterborne photovoltaic inspection system proposed in this invention; Figure 2 This is a schematic diagram illustrating the iterative optimization process of the delayed acceptance hill-climbing algorithm proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figure 1 - Figure 2 A method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system, comprising the following steps: S1. Collect image data, video data, equipment operation data and environmental monitoring data generated by the inspection of the water photovoltaic system, and perform timestamp calibration and unified encoding and encapsulation on the multi-source data to generate structured inspection data units; S2. Obtain real-time status parameters of cellular communication links, satellite communication links, wireless private network links and self-organizing network relay links, and form a link status vector including bandwidth, transmission delay, packet loss rate and link stability. S3. Receive structured inspection data units and link status vectors, establish a multi-link transmission optimization model, map link selection parameters, link switching timing parameters and transmission redundancy parameters into unified strategy variables, and generate a candidate transmission strategy space. S4. Perform a nested partitioning optimization algorithm on the candidate transmission strategy space, and divide and shrink the candidate transmission strategy space hierarchically according to the comprehensive evaluation index, and output the optimal strategy subspace. S5. Execute the delayed acceptance hill climbing algorithm in the preferred strategy subspace, make acceptance decisions based on historical window evaluation results and iteratively update strategy variables to generate a stable transmission strategy. S6. Based on the stable transmission strategy, perform fragmentation, diversion scheduling, link switching and parallel transmission on the structured inspection data unit, and output the transmission results.

[0019] In this embodiment, S1 includes: S11. When the inspection task is started, an inspection task identifier is generated, and device identifiers are assigned to the UAV payload terminal, floating body monitoring terminal and edge acquisition node, and a mapping relationship between the inspection task identifier and the device identifier is established. S12. Collect image data, video data, equipment operation data and environmental monitoring data generated during the inspection of the waterborne photovoltaic system, and write the corresponding data type identifier according to the data source to form the original inspection data; S13. Based on the BeiDou time signal or network time protocol, the data acquisition terminal is uniformly synchronized to generate a standard time reference, and a millisecond-level timestamp is written for each piece of raw inspection data to achieve time alignment of multi-source data. S14. Perform standardization processing on the time-aligned raw inspection data, where image data is converted into image frames of uniform resolution, video data is divided into continuous data segments according to key frames, and equipment operation data and environmental monitoring data are converted into field-aligned structured records. S15. Construct a data encapsulation header according to the preset encapsulation rules, and write the inspection task identifier, equipment identifier, data type identifier, timestamp, data sequence number and data length in sequence, and combine it with the standardized data payload to form an initial data unit. S16. Generate integrity verification information for the initial data unit and append it to the end of the data. Sort the data according to the timestamp and data sequence number, write the sorted data into the buffer queue to be transmitted, and generate structured inspection data units.

[0020] In this embodiment, S2 includes: S21. Generate link identifiers for cellular communication links, satellite communication links, wireless private network links and self-organizing network relay links respectively, and establish a status acquisition context for each link. The status acquisition context shall at least include the acquisition period, statistical window length and probe message format. S22. In each acquisition cycle, send a bandwidth probe message to each link, record the sending time and receiving confirmation time of the bandwidth probe message, calculate the bandwidth value according to the number of effective payload bytes that have been confirmed within the statistical window, and bind the bandwidth value to the link identifier. S23. In each acquisition cycle, send a delay probe message to each link, record the sending time and receiving confirmation time of the delay probe message, calculate the round-trip delay and obtain the one-way transmission delay value according to the preset mapping rule, and bind the transmission delay value with the link identifier. S24. In each acquisition cycle, accumulate the probe message transmission count and reception acknowledgment count for each link, generate a packet loss rate value based on the unacknowledged message count within the statistical window, and bind the packet loss rate value to the link identifier. S25. In multiple consecutive collection cycles, the bandwidth, transmission delay and packet loss rate of each link are time-series merged, the volatility of each indicator is calculated and synthesized into a link stability value according to a preset weight. The link stability value is obtained by reverse mapping of the normalized volatility. S26. For each link, encapsulate the bandwidth value, transmission delay value, packet loss rate value, and link stability value in a fixed field order to form a link state vector, and write the collection period timestamp into the link state vector. The link state vector is represented as follows: ,in For link identification, This is the bandwidth value. This is the transmission delay value. This is the packet loss rate value. This represents the link stability value. S27. Align the link state vectors corresponding to each link according to the timestamp of the acquisition period, write them into the link state buffer, and output the link state vector set.

[0021] In this embodiment, S3 includes: S31. In each decision cycle, retrieve the structured inspection data unit from the pending transmission buffer queue, read the data type identifier, timestamp, data length and priority identifier, establish a data sub-queue according to the data type identifier, and write the deadline transmission time and target arrival rate to each data sub-queue. S32. Within the same decision-making cycle, retrieve the link state vectors corresponding to each link from the link state buffer, select the latest set of link state vectors based on the collection cycle timestamp, and remove link state vectors with bandwidth values ​​of zero or missing reception confirmations to obtain the set of available links. S33. For each link in the set of available links, calculate the upper limit of bytes that can be sent within the decision period based on the bandwidth value, calculate the number of available transmission time slots based on the transmission delay value, and write the upper limit of bytes that can be sent and the number of available transmission time slots into the link parameter table. S34. For each data sub-queue, select the data unit at the head of the queue in sequence and calculate the latest transmission start point. Match the latest transmission start point with the transmission delay value in the link parameter table item by item. Delete the link that causes the data unit with the latest transmission start point to be earlier than the current time to be unable to meet the requirement, and obtain the list of optional links corresponding to the data sub-queue. S35. For each data subqueue, generate a set of link selection parameter values ​​in the optional link list. The set of link selection parameter values ​​consists of single link selection items and parallel link combination items. Write the number of parallel links and the splitting ratio for each combination item. S36. Generate a set of link switching timing parameter values ​​for each data subqueue. The set of link switching timing parameter values ​​shall include at least three types of values: trigger threshold, hold duration and cooldown duration. The trigger threshold is composed of a threshold pair of link stability value and packet loss rate value. The hold duration and cooldown duration are represented by decision cycle count. S37. Generate a set of transmission redundancy parameter values ​​for each data subqueue. The set of transmission redundancy parameter values ​​shall include at least the redundancy fragmentation ratio value and the retransmission count limit value. The redundancy fragmentation ratio value corresponds to the splitting ratio of the parallel link combination item and meets the preset limit. S38. Enumerate and generate policy configuration records by combining "data sub-queue - link selection parameter value - link switching timing parameter value - transmission redundancy parameter value", and assign policy identifiers. Write all policy configuration records into the candidate policy library to form a candidate transmission policy space. Each policy configuration record in the candidate transmission policy space shall at least include a policy identifier, a data sub-queue identifier, a target link or parallel link combination, a switching trigger threshold, a hold duration, a cooldown duration, a redundancy fragmentation ratio, and a maximum number of retransmissions.

[0022] In this embodiment, S4 includes: S41. Read all policy configuration records from the candidate transmission policy space from the candidate policy library, extract the data sub-queue identifier, target link or parallel link combination, switching trigger threshold, hold duration, cooldown duration, redundancy fragmentation ratio and retransmission limit for each policy configuration record, and establish a policy record table. S42. Construct a constraint signature for each policy configuration record in the policy record table. The constraint signature is obtained by concatenating the deadline transmission time satisfaction flag, link capacity satisfaction flag, packet loss constraint satisfaction flag, and handover cooling satisfaction flag of each data sub-queue in a fixed order, and write the constraint signature into the corresponding policy configuration record. S43. Perform first-level nested partitioning on the policy record table according to constraint signatures to generate multiple signature partitions. Each signature partition contains only policy configuration records with completely identical constraint signatures. Delete any policy configuration record in each signature partition whose constraint satisfaction flag is not set to "no". This yields a set of first-level feasible partitions. S44. For each partition in the first-level feasible partition set, generate a partition skeleton description. The partition skeleton description shall include at least: the set of data sub-queues involved in the partition, the set of allowed link combinations in the partition, the range of allowed redundancy shard ratios in the partition, and the range of allowed switching thresholds in the partition. Write the partition skeleton description into the partition index table. S45. For each partition in the partition index table, construct a set of partition representative points according to the partition skeleton description. The process of constructing the set of partition representative points includes: Generate a structure key using "data subqueue set - link combination set" as the key; A fixed set of sequence numbers is obtained by deterministically mapping the structural keys; The corresponding strategy configuration record within the partition is located using a fixed sequence number set as a representative point. When the number of locatable records in a partition is insufficient, representative points are supplemented in the following order: "link combination contains fewer parallel links - smaller proportion of redundant fragments - shorter handover hold time". S46. Calculate the comprehensive evaluation index value for each representative point set of each partition. The comprehensive evaluation index value is obtained by summarizing the estimated average transmission delay, estimated packet loss rate, link bandwidth occupancy ratio and synchronization deviation in a fixed field order. Write the comprehensive evaluation index value of each representative point into the partition index table. S47. Generate a partition evaluation value for each partition. The partition evaluation value is obtained by sorting the comprehensive evaluation index values ​​of the partition representative point set and taking the median value as the main value. The "count of any constraint violation within the representative point" is added to form a penalty term. The partition evaluation value is then written into the partition index table. S48. Select the partition with the best partition evaluation value from the partition index table as the current active partition. Decompose the policy variable domain in the current active partition into the structure domain and the parameter domain. The structure domain includes the link combination and the number of parallel links, and the parameter domain includes the handover trigger threshold, hold duration, cooldown duration, redundant fragmentation ratio and retransmission limit. S49. Perform a second-level nested partitioning on the currently active partition. The construction process of the second-level nested partitioning includes: Within the structural domain, policy configuration records are divided into structural sub-partitions based on the "number of parallel links"; Within each structural sub-partition, the policy configuration records are divided into parameter sub-partitions by interval partitioning according to the "redundancy sharding ratio"; Within each parameter sub-partition, the policy configuration records are divided into intervals according to the "switch trigger threshold pair" to obtain the threshold sub-partition; Within each threshold sub-partition, leaf partitions are obtained by meshing according to the ordered pairs of "hold duration and cooling duration"; S410. Repeat the representative point construction and partition evaluation value generation steps for each leaf partition obtained from the second-level nested partition, and set the leaf partition with the best partition evaluation value as the next round of active partition. S411. During each round of active partitioning updates, perform boundary locking on the partition skeleton description of the previous round of active partitioning. Boundary locking includes: preventing the values ​​of the structure fields that have been deleted in the previous round of active partitioning from reappearing in the next round of active partitioning; marking the parameter intervals that have been determined to be infeasible in the previous round of active partitioning as disabled intervals and skipping them in the next round of partitioning. S412. When the number of policy configuration records in the active partition is lower than the preset threshold or the partition level reaches the preset upper limit, stop partitioning, output all policy configuration records in the active partition as the preferred policy subspace, and write the preferred policy subspace into the preferred policy library.

[0023] In this embodiment, S5 includes: S51. Read the set of policy configuration records in the preferred policy subspace from the preferred policy library, generate a structure signature and a parameter signature for each policy configuration record. The structure signature is obtained by concatenating "data subqueue identifier - target link or parallel link combination - number of parallel links" in a fixed order. The parameter signature is obtained by concatenating "switching trigger threshold - hold duration - cooldown duration - redundancy fragmentation ratio - maximum number of retransmissions" in a fixed order. Write the structure signature and parameter signature into the corresponding policy configuration record. S52. Read the capacity limit, latency limit and packet loss limit of this decision cycle from the link parameter table and data sub-queue constraints. Perform an executability review on each policy configuration record. The policy configuration records that pass the review constitute an iterable set. Select the policy configuration record with the smallest comprehensive evaluation index value from the iterable set as the initial current solution. S53. Establish a history window table. The history window table contains sequential slots of a preset integer L and a delayed reference table indexed by the structure signature. The sequential slots are used to write the comprehensive evaluation index value and structure signature of the current solution according to the iteration number. The delayed reference table is used to record the triplet of "structure signature - delayed reference evaluation value - delayed reference iteration number". S54. Calculate the comprehensive evaluation index value for the initial current solution and write it into the first sequential slot of the history window table. At the same time, register the delayed reference evaluation value and delayed reference iteration number in the delayed reference table with the structural signature of the initial current solution. S55. Construct a conflict location table. The process of constructing the conflict location table includes: calculating the capacity margin and latency margin of each link in the link combination corresponding to the initial current solution, counting the unmet count and trigger switching count of each data sub-queue, and writing "the link identifier with the smallest capacity margin, the link identifier with the smallest latency margin, the sub-queue identifier with the largest trigger switching count, and the link identifier with the largest packet loss rate" into the conflict location table. S56. At the beginning of each iteration, a neighborhood generation list is generated based on the conflict location table. The neighborhood generation list lists the policy variables to be disturbed in a fixed priority order. The fixed priority order includes at least: the diversion ratio associated with the link with the smallest capacity margin, the target link selection associated with the link with the smallest latency margin, the handover trigger threshold pair associated with the sub-queue with the largest handover count, and the redundant fragmentation ratio and retransmission limit associated with the link with the largest packet loss rate. S57. Generate a set of candidate solutions item by item based on the neighborhood generation list. The generation rules include: The link selection parameters are modified only once, either by replacing the target link with the next candidate link in the list of available links, or by replacing the parallel link combination with the next candidate combination with the same number of parallel links. The splitting ratio is adjusted in steps only if the parallel link combination remains unchanged. The ratio adjustment uses a preset step table and ensures that the ratio sums are 1. For the switching trigger threshold pair, only one-sided step adjustment of the threshold pair is performed, and the hold duration and cooldown duration are only adjusted once in an ordered pair step. The redundant fragmentation ratio and the upper limit of retransmission times are only adjusted once. The adjustment meets the preset linkage table that "the upper limit of retransmission times is lowered when the redundant fragmentation ratio is increased, and the upper limit of retransmission times is increased when the redundant fragmentation ratio is lowered". Boundary constraint verification is performed immediately after each candidate solution is generated. Candidate solutions that fail the boundary constraint verification are not included in the candidate solution set. S58. Calculate the comprehensive evaluation index value for each candidate solution set in the order of generation, extract the structural signature for each candidate solution, read the delayed reference evaluation value from the delayed reference table according to the structural signature, and read the sequential slot evaluation value before the Lth one in the history window table as the delayed reference evaluation value when the structural signature is not hit. S59. Perform a delayed acceptance determination, which satisfies the following steps: First determination: If the comprehensive evaluation index value of the candidate solution is not greater than the delay reference evaluation value of the corresponding structural signature, the candidate solution is set as a solution to be accepted; otherwise, the candidate solution is set as a solution to be rejected. Second determination: When the structural signature of the solution to be accepted is different from the structural signature of the current solution, calculate the structural change mark of the solution to be accepted relative to the current solution, and read the structural change mark count of the most recent L sequential slots in the history window table. When the structural change mark count is greater than the preset upper limit K, set the solution to be accepted as a rejected solution; when the structural change mark count is not greater than the preset upper limit K, retain the solution to be accepted. The delay reference evaluation value is generated using the delay reference table of the structure signature index and updated according to the following formula: ; in, For structural signature, The iteration number, The interval from the current iteration number in the history window table is... The sequential slot evaluation value, For structural signature In iteration number The corresponding delay reference evaluation value, In recent Structural signature within each sequential slot The count of structural changes that occurred. This is the penalty coefficient; S510. When there is a solution to be accepted, replace the current solution with the solution to be accepted, and write the comprehensive evaluation index value and structural signature of the solution to be accepted into the next sequential slot of the history window table. At the same time, update the delayed reference evaluation value and delayed reference iteration number in the delayed reference table with the structural signature of the solution to be accepted. When there is no solution to be accepted, write the comprehensive evaluation index value and structural signature of the current solution into the next sequential slot of the history window table. S511. Update the conflict location table after each write to the history window table. The update process includes: recalculating the capacity margin and latency margin with the link combination corresponding to the current solution, re-counting the unmet count and the trigger switch count, and rewriting the rules to overwrite the conflict location table. S512. When the number of iterations reaches the preset upper limit or the current solution is not updated for a preset number of consecutive iterations in the history window table, the iteration is terminated. The current solution at the time of termination is output as a stable transmission strategy and written into the stable strategy library.

[0024] In this embodiment, S6 includes: S61. Read the structured inspection data units from the queue to be transmitted in the order of timestamps, generate the transmission batch number, determine the number of fragments allocated to each link according to the target link or parallel link combination and diversion ratio in the stable transmission strategy, calculate the total number of fragments, and write the transmission batch number and the total number of fragments into the transmission record table. S62. Perform sequential segmentation of the structured inspection data unit according to the preset maximum load length of the segment to generate segmented data units. Write the segment sequence number, total number of segments, sending batch number and target link identifier to each segmented data unit, and register the segment enqueue timestamp. S63. Write the fragmented data unit into the corresponding link transmission buffer. Record the transmission timestamp and start the acknowledgment timer when transmitting. Fragments whose acknowledgment timer expires are marked as fragments to be retransmitted. Fragments whose retransmission count reaches the upper limit of the retransmission count are written into the loss record table. S64. When the switching trigger threshold is met and the cooldown time count is zero, the target link identifier for the unsent fragments is recalculated and migrated to the new link sending buffer. At the same time, the link switching timestamp and the migrated fragment sequence number are recorded. S65. The receiving end establishes a reassembly record table according to the sending batch number, writes it into the reassembly buffer according to the fragment sequence number and updates the arrival bitmap. When the arrival bitmap is complete, it generates a reception acknowledgment message and returns it to the sending end. S66. After receiving the acknowledgment message, the sending end writes an acknowledgment timestamp for the corresponding fragment. The transmission delay of a single fragment is calculated using the following formula. ,in For the first The transmission delay of each segment For sending timestamps, To confirm the timestamp; S67. At the end of the preset statistical period, count the total number of fragments sent. Confirm the number of fragments Number of lost fragments Generate packet loss rate information The average transmission delay is calculated by summarizing the transmission delay samples of each fragment, and the transmission result record is formed and output together with the sending batch number and the number of confirmed fragments.

[0025] Example: To verify the feasibility of this invention, it was applied to a data transmission system for inspection of a typical floating photovoltaic power station. This type of power station consists of numerous photovoltaic module arrays distributed in open water environments. Inspection operations primarily rely on the collaborative efforts of UAV payload terminals, floating monitoring terminals, and distributed edge acquisition nodes. During the inspection process, image data, video data, equipment operation data, and environmental monitoring data are continuously generated. These data are characterized by large scale, dispersed sources, and high real-time requirements, necessitating transmission to a backend system for centralized processing via multiple wireless communication links.

[0026] In practical operation, it has been found that existing technologies generally suffer from inconsistent timing and data structures among multiple data sources. Data collected by different devices is difficult to align on the same timeline, leading to biased analysis results. Furthermore, during multi-link communication, link bandwidth and latency fluctuate significantly. Traditional methods, employing only fixed links or simple priority switching strategies, cannot dynamically adjust based on link status, easily resulting in link congestion, frequent switching, and data loss. In addition, in complex environments, traditional strategies lack effective optimization mechanisms, making it difficult to obtain stable transmission schemes under multiple constraints, resulting in low overall transmission efficiency.

[0027] In this embodiment, a unified data processing mechanism is built within each inspection terminal. This mechanism performs unified time calibration on the collected image data, video data, equipment operation data, and environmental monitoring data. A unified time base is used to write millisecond-level timestamps to each data entry. Different data types are standardized: image data is converted into image frames of uniform resolution, video data is divided into continuous data segments, and equipment operation data and environmental monitoring data are converted into structured records. According to preset encapsulation rules, information such as inspection task identifier, equipment identifier, data type identifier, timestamp, and data sequence number are written into the data encapsulation header. This is combined with the data payload to generate structured inspection data units. Simultaneously, integrity verification information is added to the data units, and they are written to a cache queue in chronological order.

[0028] During the link status awareness process, a status acquisition mechanism is established for cellular communication links, satellite communication links, wireless private network links, and self-organizing network relay links. By periodically sending probe messages, link bandwidth, transmission latency, and packet loss rate are obtained. Data from multiple acquisition periods are time-series merged to calculate the fluctuation of link status and generate link stability values. Bandwidth, latency, packet loss rate, and stability are encapsulated in a fixed format to form a link status vector, which is then stored in the link status buffer after being aligned by time.

[0029] During the data transmission strategy construction process, structured inspection data units are read from the cache queue in each decision cycle. Data sub-queues are divided according to data type and priority, and the set of available links is filtered in conjunction with the link state vector. For each link, the maximum number of bytes that can be sent and the available transmission time slots are calculated to form a link parameter table. Based on the data sub-queues and the link parameter table, a set of values ​​for link selection parameters, link switching timing parameters, and transmission redundancy parameters is generated, and a candidate transmission strategy space is constructed through a combination enumeration method.

[0030] In the strategy optimization phase, a nested partitioning optimization algorithm is used to process the candidate transmission strategy space. The strategies are divided into multiple partitions according to constraints, and a representative point set is constructed within each partition. Partitions are then screened and narrowed using comprehensive evaluation metrics, gradually reducing the strategy space to obtain the optimal strategy subspace. Based on this, a delayed acceptance hill-climbing algorithm is further employed for strategy iteration. By establishing a historical window table and a delay reference table, candidate strategies are evaluated and accepted. This dynamic optimization of the strategy, while ensuring search stability, ultimately generates a stable transmission strategy.

[0031] During data transmission, structured inspection data units are fragmented according to a stable transmission strategy. Fragmented data is then distributed to different links for parallel transmission according to a distribution ratio. Transmission time is recorded and an acknowledgment timing mechanism is set. When the link status meets the switching conditions, untransmitted data is reallocated to a new link's transmission buffer. Simultaneously, retransmission control is implemented for unacknowledged fragments, and the number of retransmissions is limited to avoid network congestion. The receiving end reassembles the data according to the fragment sequence number and returns an acknowledgment message after complete reception. The sending end calculates the transmission delay and analyzes packet loss based on the acknowledgment message, forming a transmission result record.

[0032] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system, characterized in that, include: S1. Collect image data, video data, equipment operation data and environmental monitoring data generated by the inspection of the water photovoltaic system, and perform timestamp calibration and unified encoding and encapsulation on the multi-source data to generate structured inspection data units; S2. Obtain the real-time status parameters of cellular communication links, satellite communication links, wireless private network links, and self-organizing network relay links to form a link status vector; S3. Receive structured inspection data units and link status vectors, establish a multi-link transmission optimization model, map link selection parameters, link switching timing parameters and transmission redundancy parameters into unified strategy variables, and generate a candidate transmission strategy space. S4. Perform a nested partitioning optimization algorithm on the candidate transmission strategy space, and divide and shrink the candidate transmission strategy space hierarchically according to the comprehensive evaluation index, and output the optimal strategy subspace. S5. Execute the delayed acceptance hill climbing algorithm in the preferred strategy subspace, make acceptance decisions based on historical window evaluation results and iteratively update strategy variables to generate a stable transmission strategy. S6. Based on the stable transmission strategy, perform fragmentation, diversion scheduling, link switching and parallel transmission on the structured inspection data unit, and output the transmission results.

2. The method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system according to claim 1, characterized in that, S2 includes: S21. Generate link identifiers for cellular communication links, satellite communication links, wireless private network links, and self-organizing network relay links respectively, and establish a status acquisition context for each link. The status acquisition context includes the acquisition period, statistical window length, and probe message format. S22. In each acquisition cycle, send a bandwidth probe message to each link, record the sending time and receiving confirmation time of the bandwidth probe message, calculate the bandwidth value according to the number of effective payload bytes that have been confirmed within the statistical window, and bind the bandwidth value to the link identifier. S23. In each acquisition cycle, send a delay probe message to each link, record the sending time and receiving confirmation time of the delay probe message, calculate the round-trip delay and obtain the one-way transmission delay value according to the preset mapping rule, and bind the transmission delay value with the link identifier. S24. In each acquisition cycle, accumulate the probe message transmission count and reception acknowledgment count for each link, generate a packet loss rate value based on the unacknowledged message count within the statistical window, and bind the packet loss rate value to the link identifier. S25. In N consecutive collection cycles, the bandwidth, transmission delay and packet loss rate of each link are time-series merged, the volatility of each index is calculated and synthesized into a link stability value according to a preset weight. The link stability value is obtained by reverse mapping of the normalized volatility. S26. For each link, encapsulate the bandwidth value, transmission delay value, packet loss rate value, and link stability value in a fixed field order to form a link state vector, and write the collection period timestamp into the link state vector. The link state vector is represented as follows: ,in For link identification, This is the bandwidth value. This is the transmission delay value. This is the packet loss rate value. This represents the link stability value. S27. Align the link state vectors corresponding to each link according to the timestamp of the acquisition period, write them into the link state buffer, and output the link state vector set.

3. The method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system according to claim 1, characterized in that, S3 includes: S31. In each decision cycle, retrieve the structured inspection data unit from the pending transmission buffer queue, read the data type identifier, timestamp, data length and priority identifier, establish a data sub-queue according to the data type identifier, and write the deadline transmission time and target arrival rate to each data sub-queue. S32. Within the same decision-making cycle, retrieve the link state vectors corresponding to each link from the link state buffer, select the latest set of link state vectors based on the collection cycle timestamp, and remove link state vectors with bandwidth values ​​of zero or missing reception confirmations to obtain the set of available links. S33. For each link in the set of available links, calculate the upper limit of bytes that can be sent within the decision period based on the bandwidth value, calculate the number of available transmission time slots based on the transmission delay value, and write the upper limit of bytes that can be sent and the number of available transmission time slots into the link parameter table. S34. For each data sub-queue, select the data unit at the head of the queue in sequence and calculate the latest transmission start point. Match the latest transmission start point with the transmission delay value in the link parameter table item by item. Delete the link that causes the data unit with the latest transmission start point to be earlier than the current time to be unable to meet the requirement, and obtain the list of optional links corresponding to the data sub-queue. S35. For each data subqueue, generate a set of link selection parameter values ​​in the optional link list. The set of link selection parameter values ​​consists of single link selection items and parallel link combination items. Write the number of parallel links and the splitting ratio for each combination item. S36. Generate a set of link switching timing parameter values ​​for each data subqueue. The set of link switching timing parameter values ​​includes three types of values: trigger threshold, hold duration, and cooldown duration. The trigger threshold is composed of a threshold pair of link stability value and packet loss rate value. The hold duration and cooldown duration are represented by decision cycle count. S37. Generate a set of transmission redundancy parameter values ​​for each data sub-queue. The set of transmission redundancy parameter values ​​includes the redundancy fragmentation ratio value and the retransmission count limit value. The redundancy fragmentation ratio value corresponds to the splitting ratio of the parallel link combination item and meets the preset limit. S38. Enumerate and generate policy configuration records by combining "data sub-queue - link selection parameter value - link switching timing parameter value - transmission redundancy parameter value", and assign policy identifiers. Write all policy configuration records into the candidate policy library to form a candidate transmission policy space.

4. The method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system according to claim 1, characterized in that, S4 includes: S41. Read all policy configuration records from the candidate transmission policy space from the candidate policy library, extract the data sub-queue identifier, target link or parallel link combination, switching trigger threshold, hold duration, cooldown duration, redundancy fragmentation ratio and retransmission limit for each policy configuration record, and establish a policy record table. S42. Construct a constraint signature for each policy configuration record in the policy record table. The constraint signature is obtained by concatenating the deadline transmission time satisfaction flag, link capacity satisfaction flag, packet loss constraint satisfaction flag, and handover cooling satisfaction flag of each data sub-queue in a fixed order, and write the constraint signature into the corresponding policy configuration record. S43. Perform first-level nested partitioning on the policy record table according to constraint signatures to generate N signature partitions. Each signature partition contains only policy configuration records with completely identical constraint signatures. Delete any policy configuration record in each signature partition whose constraint satisfaction flag is not found. This yields a set of first-level feasible partitions. S44. For each partition in the first-level feasible partition set, generate a partition skeleton description. The partition skeleton description includes: the set of data sub-queues involved in the partition, the set of allowed link combinations in the partition, the range of allowed redundancy shard ratios in the partition, and the range of allowed switching thresholds in the partition. Write the partition skeleton description into the partition index table. S45. For each partition in the partition index table, construct a set of partition representative points according to the partition skeleton description; S46. Calculate the comprehensive evaluation index value for each representative point set of each partition. The comprehensive evaluation index value is obtained by summarizing the estimated average transmission delay, estimated packet loss rate, link bandwidth occupancy ratio and synchronization deviation in a fixed field order, and write the comprehensive evaluation index value of each representative point into the partition index table. S47. Generate a partition evaluation value for each partition. The partition evaluation value is obtained by sorting the comprehensive evaluation index values ​​of the partition representative point set and taking the median value as the main value. The "count of any constraint violation within the representative point" is added to form a penalty term. The partition evaluation value is then written into the partition index table. S48. Select the partition with the best partition evaluation value from the partition index table as the current active partition. Decompose the policy variable domain in the current active partition into the structure domain and the parameter domain. The structure domain includes the link combination and the number of parallel links, and the parameter domain includes the handover trigger threshold, hold duration, cooldown duration, redundant fragmentation ratio and retransmission limit. S49. Perform a second-level nested partitioning on the currently active partition. The construction process of the second-level nested partitioning includes: Within the structural domain, policy configuration records are divided into structural sub-partitions based on the "number of parallel links"; Within each structural sub-partition, the policy configuration records are divided into parameter sub-partitions by interval partitioning according to the "redundancy sharding ratio"; Within each parameter sub-partition, the policy configuration records are divided into intervals according to the "switch trigger threshold pair" to obtain the threshold sub-partition; Within each threshold sub-partition, leaf partitions are obtained by meshing according to the ordered pairs of "hold duration and cooling duration". S410. Repeat the representative point construction and partition evaluation value generation steps for each leaf partition obtained from the second-level nested partition, and set the leaf partition with the best partition evaluation value as the next round of active partition. S411. During each round of active partitioning updates, perform boundary locking on the partition skeleton description of the previous round of active partitioning. Boundary locking includes: preventing the values ​​of the structure fields that have been deleted in the previous round of active partitioning from reappearing in the next round of active partitioning; marking the parameter intervals that have been determined to be infeasible in the previous round of active partitioning as disabled intervals and skipping them in the next round of partitioning. S412. When the number of policy configuration records in the active partition is lower than the preset threshold or the partition level reaches the preset upper limit, stop partitioning, output all policy configuration records in the active partition as the preferred policy subspace, and write the preferred policy subspace into the preferred policy library.

5. The method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system according to claim 4, characterized in that, The process of constructing the partition representative point set includes: generating a structure key using "data sub-queue set - link combination set" as the key; performing deterministic mapping on the structure key to obtain a fixed sequence number set; configuring records as representative points according to the strategy of locating the corresponding records in the partition based on the fixed sequence number set; when the number of locatable records in the partition is insufficient, supplementing the representative points in the order of "link combination contains fewer parallel links - smaller redundancy shard ratio - shorter handover retention time".

6. The method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system according to claim 1, characterized in that, S5 includes: S51. Read the set of policy configuration records in the preferred policy subspace from the preferred policy library, generate a structure signature and a parameter signature for each policy configuration record. The structure signature is obtained by concatenating "data subqueue identifier - target link or parallel link combination - number of parallel links" in a fixed order. The parameter signature is obtained by concatenating "switching trigger threshold - hold duration - cooldown duration - redundancy fragmentation ratio - maximum number of retransmissions" in a fixed order. Write the structure signature and parameter signature into the corresponding policy configuration record. S52. Read the capacity limit, latency limit and packet loss limit of this decision cycle from the link parameter table and data sub-queue constraints. Perform an executability review on each policy configuration record. The policy configuration records that pass the review constitute an iterable set. Select the policy configuration record with the smallest comprehensive evaluation index value from the iterable set as the initial current solution. S53. Establish a history window table. The history window table contains sequential slots of a preset integer L and a delayed reference table indexed by the structure signature. The sequential slots are used to write the comprehensive evaluation index value and structure signature of the current solution according to the iteration number. S54. Calculate the comprehensive evaluation index value for the initial current solution and write it into the first sequential slot of the history window table. At the same time, register the delayed reference evaluation value and delayed reference iteration number in the delayed reference table with the structural signature of the initial current solution. S55. Construct a conflict location table. The process of constructing the conflict location table includes: calculating the capacity margin and latency margin of each link in the link combination corresponding to the initial current solution, counting the unmet count and trigger switching count of each data sub-queue, and writing "the link identifier with the smallest capacity margin, the link identifier with the smallest latency margin, the sub-queue identifier with the largest trigger switching count, and the link identifier with the largest packet loss rate" into the conflict location table. S56. At the beginning of each iteration, a neighborhood generation list is generated based on the conflict location table. The neighborhood generation list lists the policy variables to be disturbed in a fixed priority order. The fixed priority order includes: the diversion ratio associated with the link with the smallest capacity margin, the target link selection associated with the link with the smallest latency margin, the handover trigger threshold pair associated with the sub-queue with the largest handover count, and the redundant fragmentation ratio and retransmission limit associated with the link with the largest packet loss rate. S57. Generate a set of candidate solutions item by item based on the neighborhood generation list; S58. Calculate the comprehensive evaluation index value for each candidate solution set in the order of generation, extract the structural signature for each candidate solution, read the delayed reference evaluation value from the delayed reference table according to the structural signature, and read the sequential slot evaluation value before the Lth one in the history window table as the delayed reference evaluation value when the structural signature is not hit. S59. Perform a delayed acceptance determination, which satisfies the following steps: First determination: If the comprehensive evaluation index value of the candidate solution is not greater than the delay reference evaluation value of the corresponding structural signature, the candidate solution is set as a solution to be accepted; otherwise, the candidate solution is set as a solution to be rejected. Second determination: When the structural signature of the solution to be accepted is different from the structural signature of the current solution, calculate the structural change mark of the solution to be accepted relative to the current solution, and read the structural change mark count of the most recent L sequential slots in the history window table. When the structural change mark count is greater than the preset upper limit K, set the solution to be accepted as a rejected solution; when the structural change mark count is not greater than the preset upper limit K, retain the solution to be accepted. The delay reference evaluation value is generated using the delay reference table of the structure signature index and updated according to the following formula: ； in, For structural signature, The iteration number, The interval from the current iteration number in the history window table is... The sequential slot evaluation value, For structural signature In iteration number The corresponding delay reference evaluation value, In recent Structural signature within each sequential slot The count of structural changes that occurred. This is the penalty coefficient; S510. When there is a solution to be accepted, replace the current solution with the solution to be accepted, and write the comprehensive evaluation index value and structural signature of the solution to be accepted into the next sequential slot of the history window table. At the same time, update the delayed reference evaluation value and delayed reference iteration number in the delayed reference table with the structural signature of the solution to be accepted. When there is no solution to be accepted, write the comprehensive evaluation index value and structural signature of the current solution into the next sequential slot of the history window table. S511. Update the conflict location table after each write to the history window table. The update process includes: recalculating the capacity margin and latency margin with the link combination corresponding to the current solution, re-counting the unmet count and the trigger switch count, and rewriting the rules to overwrite the conflict location table. S512. When the number of iterations reaches the preset upper limit or the current solution is not updated for a preset number of consecutive iterations in the history window table, the iteration is terminated. The current solution at the time of termination is output as a stable transmission strategy and written into the stable strategy library.

7. The method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system according to claim 6, characterized in that, Candidate solution sets are generated item by item based on the neighborhood generation list. The generation rules include: The link selection parameters are modified only once, either by replacing the target link with the next candidate link in the list of available links, or by replacing the parallel link combination with the next candidate combination with the same number of parallel links. The splitting ratio is adjusted in steps only if the parallel link combination remains unchanged. The ratio adjustment uses a preset step table and ensures that the ratio sums are 1. For the switching trigger threshold pair, only one-sided step adjustment of the threshold pair is performed, and the hold duration and cooldown duration are only adjusted once in an ordered pair step. The redundant fragmentation ratio and the upper limit of retransmission times are adjusted only once. The adjustment meets the preset linkage table that "the upper limit of retransmission times is lowered when the redundant fragmentation ratio is increased, and the upper limit of retransmission times is increased when the redundant fragmentation ratio is lowered". Boundary constraint verification is performed immediately after each candidate solution is generated. Candidate solutions that fail the boundary constraint verification are not included in the candidate solution set.

8. The method for multi-link transmission and synchronization of data in a floating photovoltaic inspection system according to claim 1, characterized in that, S6 includes: S61. Read the structured inspection data units from the queue to be transmitted in the order of timestamps, generate the transmission batch number, determine the number of fragments allocated to each link according to the target link or parallel link combination and diversion ratio in the stable transmission strategy, calculate the total number of fragments, and write the transmission batch number and the total number of fragments into the transmission record table. S62. Perform sequential segmentation of the structured inspection data unit according to the preset maximum load length of the segment to generate segmented data units. Write the segment sequence number, total number of segments, sending batch number and target link identifier to each segmented data unit, and register the segment enqueue timestamp. S63. Write the fragmented data unit into the corresponding link transmission buffer. Record the transmission timestamp and start the acknowledgment timer when transmitting. Fragments whose acknowledgment timer expires are marked as fragments to be retransmitted. Fragments whose retransmission count reaches the upper limit of the retransmission count are written into the loss record table. S64. When the switching trigger threshold is met and the cooldown time count is zero, the target link identifier for the unsent fragments is recalculated and migrated to the new link sending buffer. At the same time, the link switching timestamp and the migrated fragment sequence number are recorded. S65. The receiving end establishes a reassembly record table according to the sending batch number, writes it into the reassembly buffer according to the fragment sequence number and updates the arrival bitmap. When the arrival bitmap is complete, it generates a reception acknowledgment message and returns it to the sending end. S66. After receiving the acknowledgment message, the sending end writes an acknowledgment timestamp for the corresponding fragment. S67. At the end of the preset statistical period, count the total number of fragments sent. Confirm the number of fragments Number of lost fragments Generate packet loss rate information The average transmission delay is calculated by summarizing the transmission delay samples of each fragment, and the transmission result record is formed and output together with the sending batch number and the number of confirmed fragments.

Images