Distributed storage-based digital management method for catering business data

Abstract

This invention discloses a digital management method for restaurant merchant business data based on distributed storage. This invention relates to the fields of cloud computing and distributed data storage management technology. In an edge-cloud collaborative distributed storage system, this method integrates restaurant business tags and access frequency time decay scores to calculate a comprehensive temperature score for data blocks. Data with business tag scores exceeding a circuit breaker threshold is forcibly classified as hot data, achieving business semantic overwriting of the underlying hot / cold layering logic. A cross-modal hybrid index tree is constructed using order transaction identifiers, physical space coordinates, and timestamps as anchor points, enabling bidirectional penetration retrieval of structured order data and unstructured audio / video data via single hashing. Historical tidal maps and real-time access heat prediction are combined to pre-distribute access hotspots, and high-concurrency writes are handled using a lock-free method of sequential appending logs and asynchronous merging. This invention improves the accuracy of restaurant merchant data layering and the efficiency of cross-modal retrieval.

Description

Technical Field

[0001] This invention relates to the field of cloud computing and distributed data storage management technology, specifically a method for digital management of business data of catering merchants based on distributed storage. Background Technology

[0002] Distributed storage systems typically consist of clusters of multiple storage nodes. They achieve high availability storage through data sharding and multi-replica replication. Distributed databases are used to carry structured data, and log structure merging trees enable sequential writing and background merging and compression of data. Distributed object storage is used to carry unstructured audio and video data. Common hot and cold data tiering techniques are generally based on least recently used or least frequently used algorithms. These algorithms count the number of times and access time of data blocks within a time window, migrating low-frequency data blocks to low-cost storage media and retaining high-frequency data blocks in high-speed storage media.

[0003] The closest existing technical solution to this invention is a restaurant data management system built on a general-purpose distributed storage system. In this solution, the edge devices of the store collect structured order data and unstructured audio and video data respectively and write them into different storage channels. The structured data is uploaded to a cloud-based distributed database cluster, and the unstructured data is uploaded to a cloud-based distributed object storage cluster. The cloud storage management module periodically scans the historical access records of each data segment based on access frequency statistics to determine the data's hotness level. When it is necessary to retrieve the monitoring video associated with a certain order, manual retrieval and comparison are required in the object storage system based on the order time and store number.

[0004] In the existing solutions mentioned above, the hot and cold data stratification relies solely on general access frequency statistics and does not take into account the business semantics specific to the catering industry. This results in low-frequency but high-business-value data being incorrectly archived. Structured and unstructured data are stored separately and lack underlying correlation indexes. Cross-modal retrieval relies on manual comparison, which is inefficient. In addition, extreme concurrency during the lunch and evening peak periods can easily lead to hot spot tilting and node jitter in the storage cluster, affecting the real-time performance of front-end services. Summary of the Invention

[0005] The technical problem this invention aims to solve is that when existing distributed storage technology is applied to the digital management of multimodal business data of chain catering merchants, the stratification of hot and cold data relies only on general statistical indicators such as access frequency, which cannot perceive the business semantics unique to the catering industry. This results in low-frequency but high-business-value data being incorrectly archived and retrieved slowly. The extreme spatiotemporal concurrency during the lunch and dinner peak hours in the catering industry causes hot data skew and node jitter in the distributed cluster, affecting the real-time performance of the store's front-end business. The separate storage of structured transaction data and unstructured media data, without a unified underlying association index, leads to low efficiency in cross-modal data retrieval.

[0006] Existing distributed storage general cold and hot tiering technologies (such as LRU and LFU algorithms) generally use access frequency as the sole or primary criterion for determining the value of data storage. This is a long-standing technical bias in the field. According to this approach, data with high business value but low access frequency will inevitably be archived to low-speed storage media. This application breaks through this technical bias by introducing a business tag circuit breaker threshold forced overwrite mechanism. This ensures that data with high business value tags such as food safety risks, refund complaints, and supply chain anomalies that exceed the threshold will not be incorrectly archived as cold data due to low access frequency. This achieves the data tiering accuracy improvement effect that general solutions in this field have failed to achieve.

[0007] To address the aforementioned technical problems, this invention provides a method for digital management of catering merchant business data based on distributed storage. This method operates on an edge-cloud collaborative distributed storage system, comprising a store edge acquisition layer and cloud-based distributed structured and unstructured storage clusters. The method includes the following:

[0008] Firstly, the store edge acquisition layer collects structured order transaction data and unstructured video data, and generates a transaction identifier, physical space coordinates and timestamp for each order transaction. At the same time, it extracts the business tags associated with the order transaction, including food safety risk tags, refund complaint tags and supply chain anomaly tags.

[0009] Secondly, the transaction identifier, physical space coordinates, and timestamp are concatenated in a preset field order and then hashed to generate an anchor index key. The anchor index key is then used as the leaf node to construct a cross-modal hybrid index tree. Each leaf node simultaneously carries a storage address pointer to the corresponding order record in the structured storage cluster and an offset pointer to the corresponding video frame slice in the unstructured storage cluster. This enables bidirectional penetration and location of associated structured and unstructured data with a single hash lookup.

[0010] Third, the access frequency time decay score is calculated based on the historical access count of the data block and the timestamp of each access. The dynamic business weight factor is calculated based on the business tag. The two are weighted and summed according to the preset fusion coefficient to obtain the comprehensive temperature score. When any score in the business tag exceeds the preset circuit breaker threshold, the corresponding data block is forcibly determined as hot data. This forced determination takes precedence over the hot / cold level determination based on the comprehensive temperature score. Otherwise, the hot / cold level of the data block is determined based on the comparison result of the comprehensive temperature score with the preset high temperature threshold and low temperature threshold.

[0011] Fourth, a historical tidal map is constructed based on historical access logs. The data access popularity at future times is predicted by combining the real-time access popularity at the current moment. When the predicted access popularity exceeds the preset sharding threshold, a pre-sharding operation is performed on the corresponding data shards before the predicted peak period arrives. Data blocks are migrated in layers according to the aforementioned hot and cold levels. The migration scheduling avoids the peak period identified by the tidal prediction. When the write concurrency of the storage node is detected to exceed the preset concurrency threshold during the actual peak period, the write request is first written to the memory buffer and recorded in a sequential append log manner. After the buffer reaches the preset merge trigger condition, it is asynchronously merged and written to the main storage structure. This merging process does not lock the storage node or trigger physical redistribution.

[0012] Fifth, when an abnormal backtracking request for a specified transaction identifier is received, the corresponding leaf node is located according to the cross-modal hybrid index tree, and the corresponding structured order record and unstructured video frame slice are read simultaneously according to the two pointers in the leaf node to complete cross-modal consistency traceability.

[0013] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0014] First, data temperature is calculated by fusing business tags with access frequency and time decay scores, and high business value data that exceeds the circuit breaker threshold is forcibly marked as hot data, so that low-frequency but high business value data can be accurately identified and retained in the high-speed storage layer, avoiding critical data from being incorrectly archived.

[0015] Secondly, by constructing a cross-modal hybrid index tree using transaction identifiers, physical space coordinates, and timestamps, a unified association index is established at the underlying level for structured transaction data and unstructured media data, enabling bidirectional penetration and positioning with a single hash lookup, thereby improving the retrieval efficiency of cross-modal data.

[0016] Third, by integrating historical tide maps and real-time access popularity for pre-predictive sharding, and combining lock-free merged write mechanism to handle high-concurrency write requests, the probability of front-end device response lag caused by hotspot tilt and node redistribution during peak store business periods is reduced.

[0017] Fourth, it enables the scheduling of hierarchical migration actions to proactively avoid peak periods of tidal prediction and identification, and combines hierarchical judgment results with cross-modal index trees to serve anomaly backtracking scenarios, achieving collaborative and efficient management of structured and unstructured data. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the system architecture of the digital management method for catering merchant business data based on distributed storage as described in this invention;

[0019] Figure 2This is a flowchart illustrating the data temperature dynamic evaluation method for business value semantic perception as described in this invention.

[0020] Figure 3 This is a schematic diagram of the cross-modal spatiotemporal anchor hybrid index tree described in this invention;

[0021] Figure 4 This is a flowchart illustrating the pre-predictive elastic slicing and lock-free peak-shaving method based on catering tidal characteristics described in this invention. Detailed Implementation

[0022] The following is in conjunction with the appendix Figure 1-4 The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0023] To make the technical solution and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments. The method of the present invention runs on a distributed storage system with edge-cloud collaboration. The overall architecture of the system, from bottom to top, includes a store edge acquisition layer, an edge metadata tagging and preprocessing layer, an edge-cloud transmission layer, a cloud distributed storage layer, a cross-modal hybrid index layer, a business value semantic evaluation and dynamic hierarchical management layer, and a tidal prediction elastic scheduling layer. The store edge acquisition layer includes POS cash register terminals and kitchen open-kitchen monitoring cameras, which are used to generate structured order data and unstructured video data, respectively. The edge preprocessing layer includes a transaction identifier generation module, a spatial coordinate calibration module, a timestamp tagging module, and a business identifier generation module. The tag extraction module includes a business tag extraction module for extracting food safety risk tags, refund complaint tags, and supply chain anomaly tags. This layer outputs data packets with unified metadata. The edge-cloud transmission layer uploads the preprocessed data packets to the cloud via the network. The cloud distributed storage layer includes a distributed database cluster carrying structured data and a distributed object storage cluster carrying unstructured audio and video data. The two are linked through a cross-modal hybrid index layer. The business value semantic assessment and dynamic hierarchical management layer reads metadata tags from the two types of storage clusters, calculates data temperature, and outputs hierarchical migration instructions. The tidal prediction elastic scheduling layer receives historical access logs and real-time access logs, and outputs pre-sharding instructions and lock-free merge write instructions.

[0024] I. A Dynamic Evaluation Method for Data Temperature Based on Semantic Awareness of Business Value

[0025] The core objective of this section is to calculate a dynamic data temperature value that better reflects the true business value by integrating restaurant-specific business semantic tags into traditional access frequency statistics. Based on this, the original hot and cold tiering logic of the underlying distributed storage is overwritten. The comprehensive data temperature evaluation formula is as follows:

[0026] ;

[0027] in, Represents data block At any moment The overall temperature score, Represents data block Time decay score based on historical access frequency Represents data block Dynamic business weighting factors, and These are the fusion coefficients of the access frequency score and the business weight factor, respectively, and satisfy the following conditions: Both are pre-configured by the storage management module based on historical tuning experience.

[0028] The formula for calculating the time decay of access frequency is as follows:

[0029] ;

[0030] in, Represents data block The total number of times the data was accessed within the statistics window. Indicates the first The timestamp of the next access. This represents the time decay rate coefficient; the larger the value, the faster the contribution of historical visits to the current rating decays.

[0031] The formula for calculating the dynamic business weight factor is as follows:

[0032] ;

[0033] in, Represents data block The associated order is rated on whether it contains labels indicating high-risk allergens or food safety risks, with a value range of [0,1]. Represents data block The rating for refunds or customer complaints associated with a particular order ranges from [0,1]. Represents data block The associated supply chain batch anomaly label score has a value range of [0,1]. , , These are the weight coefficients for the three types of labels mentioned above, and they satisfy... The rules are pre-configured by the business rules engine according to industry regulatory requirements and support dynamic adjustment.

[0034] The data stratification determination rules are as follows:

[0035] Set high temperature threshold With low temperature threshold ,when At that time, data block Data classified as hot data is retained on high-speed storage media;

[0036] when At that time, the data is identified as warm data and stored in medium-speed storage media;

[0037] when When the data is deemed cold, it is allowed to be migrated to a low-cost storage medium.

[0038] In particular, as long as , , If any one of the three tags scores exceeds the preset business circuit breaker threshold, the business circuit breaker will be triggered. If the data block is deemed to be hot data, then this determination takes precedence over the criteria used in the data processing. The routine hot / cold determination is performed to override the business semantics of the original hot / cold layering logic at the underlying level.

[0039] The specific implementation of the algorithm is as follows: When the edge preprocessing layer generates data, it adds the aforementioned business tag metadata to each order transaction and its associated audio and video data blocks. The cloud business value semantic evaluation module periodically reads the historical access logs of each data block in the distributed database cluster and the distributed object storage cluster, counts the number of accesses of each data block within the statistical window and the timestamp of each access, and calculates the access frequency time decay score according to the aforementioned formula. The cloud business value semantic evaluation module then reads the business tag metadata and calculates the dynamic business weight factor according to the aforementioned formula. The access frequency time decay score and the dynamic business weight factor are weighted and summed according to the fusion coefficient to obtain the comprehensive temperature score. It is determined whether the business tag score exceeds the circuit breaker threshold. If it does, the data block is directly marked as hot data. If it does not, the comprehensive temperature score is compared with the preset threshold to determine the hot / cold level of the data block. The storage tier execution unit issues a data migration instruction to the merge compression process of the log structure merge tree storage engine of the distributed database or the lifecycle management strategy of the distributed object storage according to the obtained hot / cold level. The timing of this migration action is arranged to be executed during the period when the store business access volume is lower than the preset low threshold, thus completing the migration of the data block between different storage media tiers.

[0040] II. Construction and Retrieval Methods of Cross-Modal Spatiotemporal Anchor Hybrid Index Tree

[0041] This section addresses the problem of the disconnect between structured transaction data and unstructured media data, hindering rapid correlation and retrieval. The anchor point generation formula for the cross-modal spatiotemporal anchor hybrid index tree is as follows:

[0042] ;

[0043] in, Represents data block anchor index key, Represents data block The associated order transaction identifier, This indicates the physical space coordinates corresponding to when the data was generated, such as the specific workbench number or table number. The timestamp representing the data generation, symbol This indicates a field concatenation operation. This indicates the preset hash function.

[0044] The cross-modal spatiotemporal anchor hybrid index tree adopts an inverted tree structure: the root node of the tree is the root hash node, whose value is the aggregate digest of the hash values ​​of all its child nodes. The root node splits downwards into several intermediate layer nodes. Each intermediate layer node is divided according to the store number and date partition. The node stores the hash aggregate value of all transactions under that partition. The intermediate layer nodes continue to split downwards to the leaf node level. Each leaf node corresponds to a specific anchor index key. The leaf node lists three fields simultaneously: transaction identifier, physical space coordinates, and timestamp, along with two pointers. One pointer points to the physical address of the structured order record corresponding to the transaction in the distributed database cluster, and the other pointer points to the specific frame slice offset of the corresponding video file in the distributed object storage cluster. This enables a retrieval logic of hash lookup and bidirectional penetration positioning.

[0045] The specific implementation method for constructing and retrieving this index tree is as follows:

[0046] When generating an order transaction, the store edge terminal simultaneously obtains the physical space coordinate information corresponding to the transaction and the timestamp of the transaction. This coordinate information is collected from the fixed identifier of the operating table or table.

[0047] At the edge, the transaction identifier, physical space coordinates, and timestamp are concatenated in a preset field order, and a hash function is called to calculate and generate the anchor index key;

[0048] At the edge, the anchor index key, the storage address pointer of the structured order record, and the frame slice offset pointer of the corresponding time period video file are encapsulated together as leaf node data and uploaded to the cloud along with the data packet;

[0049] After receiving the leaf node data, the cloud-based cross-modal hybrid index layer attaches the leaf nodes to the corresponding intermediate layer nodes according to the store number and date partitioning rules, and recalculates the hash aggregation value of the intermediate layer node.

[0050] The cloud-based cross-modal hybrid index layer recalculates and updates the aggregated digest value of the root hash node layer by layer from bottom to top, completing the dynamic update of the index tree;

[0051] When the business system triggers an exception backtracking request, the transaction identifier to be queried is entered. The cross-modal hybrid index layer locates the corresponding intermediate layer node based on the store number and date to which the transaction identifier belongs. Then, under the intermediate layer node, the transaction identifier field of the leaf node is traversed and compared to lock the target anchor index key.

[0052] Based on the two pointers within the locked leaf node, the structured order records in the distributed database cluster and the corresponding video frame slices in the distributed object storage cluster are read respectively, and both types of data are returned to the business system simultaneously to complete cross-modal consistency traceability.

[0053] III. Predictive Elastic Segmentation and Lock-Free Peak Shaving Method Based on Catering Tidal Characteristics

[0054] This section addresses the storage cluster hotspot skew and jitter issues caused by extreme concurrency during the lunch and dinner peak hours in the catering industry. The formula for predicting the heat of data sharding is as follows:

[0055] ;

[0056] in, Indicates fragmentation In the future Predicted visit popularity This indicates the segmentation obtained based on historical tidal maps. Historical average visit popularity during the same period Indicates fragmentation At the present moment Nearby real-time access popularity This is a smoothing coefficient between historical patterns and real-time popularity, with a value range of [0,1]. This represents the lead time for prediction, which is the time window reserved for the system to perform pre-segmentation operations.

[0057] The specific implementation of this method is as follows: The tidal prediction elastic scheduling layer continuously collects historical access logs from each store, and constructs a historical tidal map according to the data sharding dimension and time sharding dimension to obtain the historical average access popularity. At the same time, the tidal prediction elastic scheduling layer collects the access popularity of each shard at the current moment in real time, calculates the predicted popularity of each shard at future moments according to the aforementioned prediction popularity formula and combined with the smoothing coefficient, and compares the predicted popularity with the preset sharding threshold. If the predicted popularity is not lower than the threshold, the shard is determined to be a hot shard in the future. For the determined hot shard, the shard management unit is triggered to perform a pre-splitting operation on the shard within the time window corresponding to the lead time before the predicted peak arrives, breaking it down into multiple sub-shards and distributing them to different storage nodes to complete the pre-elastic sharding.

[0058] During peak periods, the storage write monitoring unit monitors the instantaneous write concurrency of each storage node in real time. When it detects that the write concurrency of a certain node exceeds the preset concurrency threshold, the unit will take action. When the peak period ends, the lock-free peak shaping mode is triggered. In this mode, write requests are no longer directly written to the main storage structure of the distributed database. Instead, they are first written to a micro-batch buffer in memory, and the written content is recorded in a sequential append log manner, thereby avoiding locking operations on the storage nodes. When the micro-batch buffer reaches the preset merge trigger condition, the scheduled merge thread asynchronously merges the sequential log in the buffer into the main storage structure. During the merge process, the physical redistribution of storage nodes and master node migration operations are not triggered. After the peak period ends, the system continuously monitors the load of each storage node. After confirming that the load has recovered to below the stable threshold, the lock-free peak shaping mode ends and the normal write process resumes.

[0059] IV. The Synergistic Relationship Among the Three Mechanisms

[0060] The aforementioned dynamic data temperature assessment mechanism, cross-modal hybrid index tree mechanism, and tidal prediction elastic sharding and lock-free peak shaving mechanism operate in the same cloud-coordinated distributed storage system. The three have the following synergistic relationship: the scheduling timing of the hierarchical migration action actively avoids the peak periods identified by the tidal prediction elastic scheduling layer, so that the data hierarchical migration will not compete with the high-concurrency writes during peak business periods for system resources; the hierarchical judgment results output by the business value semantic assessment and dynamic hierarchical management layer, together with the index tree built by the cross-modal hybrid index layer, serve the anomaly backtracking scenario; high business value data that is forcibly marked as hot data is always retained in the high-speed storage layer. With the bidirectional penetration and positioning capability of the cross-modal index tree, cross-modal evidence retrieval in scenarios such as food safety traceability and regulatory spot checks can be completed in milliseconds.

[0061] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for digital management of catering merchant business data based on distributed storage, characterized in that: The method for operating an edge-cloud collaborative distributed storage system comprising a store edge acquisition layer, a cloud-based distributed structured storage cluster, and a cloud-based unstructured storage cluster includes: Data collection steps: The store edge collection layer collects structured order transaction data and unstructured video data, and generates a transaction identifier, physical space coordinates and timestamp for each order transaction. At the same time, it extracts the business tags associated with the order transaction, including food safety risk tags, refund complaint tags and supply chain abnormality tags. Index construction steps: The transaction identifier, physical space coordinates and timestamp are concatenated in the order of preset fields and then hashed to generate anchor index keys. A cross-modal hybrid index tree is constructed using the anchor index keys as leaf nodes. Each leaf node is simultaneously attached with a storage address pointer pointing to the corresponding order record in the structured storage cluster and an offset pointer pointing to the corresponding video frame slice in the unstructured storage cluster. Temperature assessment steps: Calculate the access frequency time decay score based on the historical access count of the data block and the timestamp of each access; calculate the dynamic business weight factor based on the business tag; and weight and sum the access frequency time decay score and the dynamic business weight factor according to a preset fusion coefficient to obtain the comprehensive temperature score of the data block. When the score of any item in the business tag exceeds the preset circuit breaker threshold, the corresponding data block is forcibly identified as hot data. This forced identification takes precedence over the hot / cold level identification based on the comprehensive temperature score. Otherwise, the hot / cold level of the data block is determined based on the comparison results of the comprehensive temperature score with the preset high temperature threshold and the preset low temperature threshold. The hot / cold level includes hot data, warm data, and cold data. Tidal prediction steps: Construct a historical tidal map based on historical access logs, and predict the data access popularity in the future by combining the real-time access popularity at the current moment. When the predicted access popularity exceeds the preset sharding threshold, perform a pre-sharding operation on the corresponding data shards before the predicted peak period arrives. Layered migration step: Based on the hot and cold levels obtained from the temperature assessment step, the data blocks are migrated between different storage media layers corresponding to the structured storage cluster and the unstructured storage cluster, and the scheduling of the migration avoids the peak periods identified by the tidal prediction step. Retrieval steps: When an abnormal backtracking request for a specified transaction identifier is received, the corresponding leaf node is located according to the cross-modal hybrid index tree, and the corresponding structured order record and unstructured video frame slice are read simultaneously according to the two pointers in the leaf node.

2. The method according to claim 1, characterized in that, The access frequency time decay score is obtained by summing the time decay values ​​corresponding to each access of the data block within the statistical window, wherein the time decay value corresponding to each access decreases as the time interval between the time of the access and the current time increases.

3. The method according to claim 1, characterized in that, The dynamic business weighting factor is obtained by weighting and summing the scores corresponding to the food safety risk label, the refund complaint label, and the supply chain anomaly label according to preset weighting coefficients, and the sum of the preset weighting coefficients is 1.

4. The method according to claim 1, characterized in that, The cross-modal hybrid index tree includes a root hash node, intermediate layer nodes, and leaf node layers. The intermediate layer nodes are divided according to store number and date partition and store the hash aggregation value of all transactions under that partition. The root hash node stores the aggregated summary of the hash values ​​of all intermediate layer nodes. When a new leaf node is added, the leaf node is attached to the corresponding intermediate layer node according to its store number and date partition, and the aggregated value of the corresponding intermediate layer node and the root hash node is recalculated and updated from bottom to top.

5. The method according to claim 1, characterized in that, The pre-splitting operation includes breaking down the data fragments predicted to be hotspots into multiple sub-fragments and distributing the multiple sub-fragments to different storage nodes.

6. The method according to claim 1, characterized in that, Also includes: During peak periods, the write concurrency of each storage node is monitored in real time. When the write concurrency exceeds the preset concurrency threshold, the write request is first written to the buffer in memory and recorded in the order of appending to the log. After the buffer reaches the preset merge trigger condition, the write content recorded in the buffer is asynchronously merged and written to the main storage structure of the corresponding storage cluster. The asynchronous merge process does not perform locking operations on the storage nodes, nor does it trigger the physical redistribution of the storage nodes.

7. The method according to claim 1, characterized in that, In the hierarchical migration step, the data block is migrated from its current storage media level to the storage media level corresponding to its hot / cold status, based on the data block's hot / cold status. The migration is scheduled when the store's business access volume is lower than a preset low threshold.

8. A digital management system for restaurant merchant business data based on distributed storage, characterized in that: include: The store edge acquisition layer is used to collect structured order transaction data and unstructured video data; An edge preprocessing layer, connected to the store edge acquisition layer, is used to generate a transaction identifier, physical space coordinates, and timestamp for each order transaction, and to extract the business tags associated with the order transaction. The business tags include food safety risk tags, refund complaint tags, and supply chain anomaly tags. The cloud-based distributed storage layer includes a distributed database cluster for storing structured order transaction data and a distributed object storage cluster for storing unstructured audio and video data. A cross-modal hybrid index layer, connected to the edge preprocessing layer and the cloud distributed storage layer, is used to concatenate the transaction identifier, the physical space coordinates and the timestamp and perform a hash operation to generate an anchor index key. The anchor index key is used as the leaf node to construct a cross-modal hybrid index tree. Each leaf node is simultaneously attached with a storage address pointer to the corresponding order record in the distributed database cluster and an offset pointer to the corresponding video frame slice in the distributed object storage cluster. The business value semantic assessment and dynamic hierarchical management module is connected to the cloud distributed storage layer. It is used to calculate the access frequency time decay score based on the historical access count of the data block and the timestamp of each access, calculate the dynamic business weight factor based on the business tag, obtain the comprehensive temperature score by weighted summation of the access frequency time decay score and the dynamic business weight factor, determine the hot and cold level of the data block based on the comprehensive temperature score and the business tag, and issue a hierarchical migration instruction to the cloud distributed storage layer. The tidal prediction elastic scheduling layer, connected to the cloud distributed storage layer, is used to construct a historical tidal map based on historical access logs and combine it with real-time access popularity to predict the data access popularity at future moments. Before the predicted peak period arrives, it triggers the pre-splitting operation of the corresponding data shards and ensures that the execution of the hierarchical migration instruction avoids the predicted peak period.

9. The system according to claim 8, characterized in that, The tidal prediction elastic scheduling layer is also used to trigger a lock-free peak-shaving mode when the write concurrency of the storage node exceeds the preset concurrency threshold. The write request is converted into sequential append logs written to the memory buffer, and asynchronously merged and written to the cloud distributed storage layer after the buffer reaches the preset merge trigger condition. The asynchronous merge process does not trigger the physical redistribution of the storage node.

Images