Distributed storage method and device for multi-modal data, equipment and medium

Abstract

The embodiment of the invention discloses a distributed storage method and device for multi-modal data, equipment and a medium, and the method comprises the steps: carrying out the analysis of the current multi-modal data based on a preset relevance weight model, and dynamically updating the metadata corresponding to the current multi-modal data based on an analysis result; based on a preset space-time partitioning strategy, performing space-time partitioning on the current multi-modal data to obtain a plurality of current multi-modal data sets; and according to the metadata corresponding to each current multi-modal data set, determining associated data of each current multi-modal data, so as to distribute and store each current multi-modal data according to the storage information corresponding to the associated data.

Description

Technical Field

[0001] This specification relates to the field of distributed data storage technology, and in particular to a distributed storage method, apparatus, device, and medium for multimodal data. Background Technology

[0002] In today's global digital wave, data has become a core production factor, and the explosive growth of multimodal data is profoundly changing the form of information interaction and industrial applications. From the immersive scene construction of the metaverse to the multi-sensor collaborative perception of autonomous driving, multimodal data has become a key carrier supporting the implementation of cutting-edge technologies and industrial upgrading. However, multimodal data complements and verifies each other. Only by performing joint queries, correlation analysis, and fusion processing on them can complex cognitive and decision-making tasks be completed. However, traditional storage architectures do not consider this strongly correlated, cross-modal access mode, so the current demand for distributed storage of multimodal data is quite urgent.

[0003] Currently, multimodal data is typically stored independently, which necessitates multiple cross-node reads during cross-modal joint queries, significantly increasing latency. Secondly, traditional hash-based or range-based block partitioning rules lack semantic awareness and cannot identify the inherent relationships between modalities, resulting in scattered distribution of related data and violating data locality. Furthermore, existing caching strategies such as LRU caching rely solely on access time or frequency for decisions, failing to capture the associated access characteristics between multimodal data, leading to low cache hit rates. Summary of the Invention

[0004] To address the aforementioned technical problems, this specification provides one or more embodiments of a distributed storage method, apparatus, device, and medium for multimodal data.

[0005] One or more embodiments of this specification employ the following technical solutions: This specification provides one or more embodiments of a distributed storage method for multimodal data, the method comprising: The current multimodal data is analyzed based on a pre-set correlation weight model, and the metadata corresponding to the current multimodal data is dynamically updated based on the analysis results. Based on a preset spatiotemporal partitioning strategy, the current multimodal data is spatiotemporally partitioned to obtain multiple current multimodal datasets; Based on the metadata corresponding to each current multimodal dataset, the associated data of each current multimodal dataset is determined, and the storage of each current multimodal dataset is allocated according to the storage information corresponding to the associated data.

[0006] Optionally, in one or more embodiments of this specification, the current multimodal data is analyzed based on a preset correlation weight model to dynamically update the metadata corresponding to the current multimodal data based on the analysis results, specifically including: Collect historical multimodal data monitored in the current scenario to obtain a sample set for the initial correlation weight model; The initial correlation weight model is trained based on the sample set to obtain a preset correlation weight model; The current multimodal data is input into the preset correlation weight model to obtain the correlation weight result output by the preset correlation weight model; Based on the association weight results, the metadata corresponding to the current multimodal data is dynamically updated; wherein, the metadata includes: the association relationship between data and the access frequency between data.

[0007] Optionally, in one or more embodiments of this specification, the current multimodal data is spatiotemporally partitioned based on a preset spatiotemporal partitioning strategy to obtain multiple current multimodal datasets, specifically including: Obtain the spatiotemporal attributes corresponding to each of the current multimodal data; wherein, the spatiotemporal attributes include: timestamp and geographic location; Based on the preset spatiotemporal partitioning strategy and the spatiotemporal attributes corresponding to each current multimodal data, the current multimodal data is spatiotemporally partitioned to obtain multiple current multimodal datasets.

[0008] Optionally, in one or more embodiments of this specification, associated data for each current multimodal dataset is determined based on the metadata corresponding to each current multimodal dataset, and the current multimodal data is allocated and stored according to the storage information corresponding to the associated data, specifically including: Parse the metadata corresponding to each current multimodal dataset to obtain the data association relationships between each current multimodal dataset; Based on the relationships between the data, the associated data among the current multimodal data are determined; wherein, the associated data includes: strongly associated data and weakly associated data; The granularity of the current multimodal data segmentation is determined based on the correlation strength of the associated data. The preset index directory is queried according to the identifier of the associated data to determine the storage information corresponding to each of the associated data. The current multimodal data is segmented according to the segmentation granularity, and then the current multimodal data is allocated to the storage nodes corresponding to the associated data for storage based on the storage information.

[0009] Optionally, in one or more embodiments of this specification, after determining the associated data of each current multimodal data based on the metadata corresponding to each current multimodal dataset, and allocating and storing each current multimodal data according to the storage information corresponding to the associated data, the method further includes: The current multimodal data is updated and verified to determine whether there has been an update. If so, the incremental index corresponding to the current modal data is obtained, and the incremental index is updated based on the updated data corresponding to the current multimodal data; wherein, the incremental index is used to record the latest position and association information of the current modal data.

[0010] Optionally, in one or more embodiments of this specification, after allocating and storing each of the current multimodal data according to the storage information corresponding to the associated data, the method further includes: Based on the association weight results of the current multimodal data, construct the association graph of the current multimodal data; Based on the correlation graph, predict the cross-modal data of the current scene, and load the cross-modal data into the cache space based on a preset caching mechanism.

[0011] Optionally, in one or more embodiments of this specification, after allocating and storing each of the current multimodal data according to the storage information corresponding to the associated data, the method further includes: Receive a data query instruction to extract the spatiotemporal constraints of the query instruction; The spatiotemporal constraints are matched with the spatiotemporal partitions corresponding to each current multimodal dataset to determine the specified current multimodal dataset. Based on the association graph corresponding to the current multimodal dataset, determine the query path and data access order corresponding to the query command; Based on the query path and data access order, the incremental index and storage information corresponding to the current multimodal dataset are searched.

[0012] This specification provides one or more embodiments of a distributed storage device for multimodal data, the device comprising: The analysis unit is used to analyze the current multimodal data based on a preset correlation weight model, so as to dynamically update the metadata corresponding to the current multimodal data based on the analysis results; A partitioning unit is used to perform spatiotemporal partitioning on the current multimodal data based on a preset spatiotemporal partitioning strategy to obtain multiple current multimodal datasets. The storage unit is used to determine the associated data of each current multimodal data based on the metadata corresponding to each current multimodal dataset, and to allocate and store each current multimodal data according to the storage information corresponding to the associated data.

[0013] This specification provides one or more embodiments of a distributed storage device for multimodal data, the device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described above.

[0014] This specification provides one or more embodiments of a non-volatile computer storage medium storing computer-executable instructions, the computer-executable instructions being configured to execute any of the methods described above.

[0015] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: By quantifying the strength of associations between multimodal data using a pre-defined association weight model, metadata is obtained. This facilitates the localization of associated data storage during subsequent storage, avoiding cross-node call delays caused by associated data being scattered across different nodes. Dynamically updating metadata reflects the latest association status in real time, ensuring the accuracy of subsequent storage and retrieval, and avoiding incorrect allocation based on outdated metadata. Partitioning multimodal data based on spatiotemporal characteristics helps ensure that multimodal data from the same spatial scenario is aggregated within the same partition, reducing cross-partition calls and narrowing the query scope for subsequent data retrieval. By using the association information in the metadata and the storage location of associated data, the storage node for the current data is ultimately determined, enabling the allocation of strongly associated different modalities to the same storage node, thus reducing the number of cross-node reads. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating a distributed storage method for multimodal data provided in an embodiment of this specification; Figure 2This is a schematic diagram of the structure of a distributed storage device for multimodal data provided in an embodiment of this specification; Figure 3 This is a schematic diagram of the structure of a distributed storage device for multimodal data provided in an embodiment of this specification; Figure 4 This is a schematic diagram of the structure of a non-volatile storage medium provided in the embodiments of this specification. Detailed Implementation

[0017] This specification provides a method, apparatus, device, and medium for distributed storage of multimodal data.

[0018] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0019] like Figure 1 As shown, this specification provides a flowchart illustrating a distributed storage method for multimodal data. Figure 1 As can be seen, in one or more embodiments of this specification, a distributed storage method for multimodal data specifically includes: S101: Analyze the current multimodal data based on a pre-set correlation weight model, and dynamically update the metadata corresponding to the current multimodal data based on the analysis results.

[0020] To address the issues of data storage isolation, static sharding strategies, and inefficient caching in existing technologies, this specification's embodiments analyze the current multimodal data based on a pre-defined correlation weight model, and then dynamically update the metadata corresponding to the current multimodal data based on the analysis results.

[0021] Specifically, in one or more embodiments of this specification, the current multimodal data is analyzed based on a preset correlation weight model, and the metadata corresponding to the current multimodal data is dynamically updated based on the analysis results. This process includes the following steps: First, historical multimodal data monitored in the current scenario is collected to obtain a sample set for the initial correlation weight model. Then, the initial correlation weight model is trained based on the sample set to obtain a preset correlation weight model. The current multimodal data is input into the preset correlation weight model to obtain the correlation weight results output by the preset correlation weight model. Based on the correlation weight results, the metadata corresponding to the current multimodal data is dynamically updated; the metadata includes: data relationships and data access frequencies. This correlation weight model is based on a machine learning algorithm, taking access logs and data attributes of the multimodal data as input, and outputting the correlation weights between data. Based on its analysis results, the metadata of the multimodal data, including data relationships and access frequencies, can be dynamically updated, providing a foundation for subsequent storage strategies.

[0022] S102: Based on a preset spatiotemporal partitioning strategy, the current multimodal data is spatiotemporally partitioned to obtain multiple current multimodal datasets.

[0023] After determining the analysis results based on the above steps, in order to store the associated data nearby according to the correlation strength shown in the analysis results, thereby reducing the number of cross-node reads, the embodiments of this specification will perform spatiotemporal partitioning on the current multimodal data according to a preset spatiotemporal partitioning strategy to obtain multiple current multimodal datasets. Specifically, in one or more embodiments of this specification, the process of performing spatiotemporal partitioning on the current multimodal data according to a preset spatiotemporal partitioning strategy to obtain multiple current multimodal datasets includes the following steps: First, the spatiotemporal attributes corresponding to each current multimodal data point are obtained; these attributes include timestamps and geographic locations. Then, based on a pre-defined spatiotemporal partitioning strategy and the corresponding spatiotemporal attributes, the current multimodal data is spatiotemporally partitioned, resulting in multiple current multimodal datasets. In other words, in a specific application scenario, based on a pre-defined spatiotemporal partitioning strategy and attributes such as timestamps and geographic locations, multimodal data is divided into different spatiotemporal partitions. Within each spatiotemporal partition, based on the analysis results of the aforementioned model, related data is stored locally; that is, strongly correlated different modalities are allocated to the same storage node to reduce the number of cross-node reads.

[0024] S103: Based on the metadata corresponding to each current multimodal dataset, determine the associated data of each current multimodal dataset, and allocate and store each current multimodal dataset according to the storage information corresponding to the associated data.

[0025] After obtaining the metadata corresponding to each current multimodal dataset based on the above steps, in order to store associated data nearby (i.e., strongly associated different modal data are allocated to the same storage node to reduce the number of cross-node reads), the embodiments of this specification determine the associated data of each current multimodal data based on the metadata corresponding to each current multimodal dataset, and then allocate and store each current multimodal data according to the storage information corresponding to the associated data. Specifically, in one or more embodiments of this specification, determining the associated data of each current multimodal data based on the metadata corresponding to each current multimodal dataset, and allocating and storing each current multimodal data according to the storage information corresponding to the associated data, specifically includes: First, the metadata corresponding to each current multimodal dataset is parsed to obtain the data relationships between them. Then, based on these relationships, the associated data among the current multimodal datasets is determined. It should be noted that associated data includes both strongly and weakly associated data. Next, the granularity of the current multimodal data segmentation is determined based on the strength of the associations. A pre-defined index directory is queried based on the identifiers of the associated data to determine the storage information corresponding to each associated data point. The current multimodal data is then segmented according to the granularity, and subsequently allocated to the storage nodes corresponding to the associated data points based on the storage information.

[0026] Furthermore, to improve data access efficiency, when data is updated or moved, only the relevant information in the incremental index is updated, avoiding the need to rebuild the entire index and improving index update efficiency. In one or more embodiments of this specification, after determining the associated data of each current multimodal dataset based on the metadata corresponding to each current multimodal dataset, and allocating storage for each current multimodal dataset according to the storage information corresponding to the associated data, the method further includes the following process: First, the current multimodal data is updated to determine if any updates have occurred. If so, the incremental index corresponding to the current modality data is obtained, and this incremental index is updated based on the updated data corresponding to the current multimodal data. The incremental index records the latest position and associated information of the current modality data.

[0027] Furthermore, in one or more embodiments of this specification, after allocating and storing each current multimodal data according to the storage information corresponding to the associated data, the method further includes the following process: First, a correlation graph of the current multimodal data is constructed based on the correlation weights of the current multimodal data. Then, cross-modal data for the current scenario is predicted based on the correlation graph, and this cross-modal data is loaded into the cache space using a pre-built caching mechanism. In other words, the correlation graph constructed based on the correlation weights of the previous multimodal data predicts and preloads potentially accessed cross-modal data. By using a cross-modal prefetch caching mechanism, data is loaded into the cache in advance, thereby improving cache hit rate and data access speed.

[0028] Furthermore, in one or more embodiments of this specification, after allocating and storing each current multimodal data according to the storage information corresponding to the associated data, the method further includes: First, a data query instruction is received to extract its spatiotemporal constraints. Then, these constraints are matched against the spatiotemporal partitions corresponding to each current multimodal dataset to determine the specified current multimodal dataset. Based on the association graph corresponding to the current multimodal dataset, the query path and data access order corresponding to the query instruction are determined. Based on the query path and data access order, the incremental index and storage information corresponding to the current multimodal dataset are searched. In other words, this embodiment combines the association and spatiotemporal partitioning characteristics of multimodal data to optimize the query path and data access order, utilizing incremental indexes and prefetch caching mechanisms to accelerate the query process and reduce query latency. This significantly improves the storage efficiency and access performance of multimodal data in the distributed storage system.

[0029] like Figure 2 As shown in the diagram, this specification provides a schematic diagram of the structure of a distributed storage device for multimodal data. Figure 2 As can be seen from one or more embodiments of this specification, a distributed storage device for multimodal data includes: Analysis unit 201 is used to analyze the current multimodal data based on a preset correlation weight model, so as to dynamically update the metadata corresponding to the current multimodal data based on the analysis results; Partitioning unit 202 is used to perform spatiotemporal partitioning on the current multimodal data based on a preset spatiotemporal partitioning strategy to obtain multiple current multimodal datasets; Storage unit 203 is used to determine the associated data of each current multimodal data according to the metadata corresponding to each current multimodal dataset, so as to allocate and store each current multimodal data according to the storage information corresponding to the associated data.

[0030] like Figure 3 As shown in the diagram, this specification provides a schematic diagram of the structure of a distributed storage device for multimodal data. Figure 3As can be seen, in one or more embodiments of this specification, a distributed storage device for multimodal data includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described above.

[0031] like Figure 4 As shown in the diagram, this specification provides a schematic diagram of the structure of a non-volatile storage medium. Figure 4 As can be seen, in one or more embodiments of this specification, a non-volatile storage medium stores computer-executable instructions, which are capable of executing any of the methods described above.

[0032] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0033] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0034] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A distributed storage method for multimodal data, characterized in that, The method includes: The current multimodal data is analyzed based on a pre-set correlation weight model, and the metadata corresponding to the current multimodal data is dynamically updated based on the analysis results. Based on a preset spatiotemporal partitioning strategy, the current multimodal data is spatiotemporally partitioned to obtain multiple current multimodal datasets; Based on the metadata corresponding to each current multimodal dataset, the associated data of each current multimodal dataset is determined, and the storage of each current multimodal dataset is allocated according to the storage information corresponding to the associated data.

2. The distributed storage method for multimodal data according to claim 1, characterized in that, The current multimodal data is analyzed based on a pre-defined correlation weight model, and the metadata corresponding to the current multimodal data is dynamically updated based on the analysis results, specifically including: Collect historical multimodal data monitored in the current scenario to obtain a sample set for the initial correlation weight model; The initial correlation weight model is trained based on the sample set to obtain a preset correlation weight model; The current multimodal data is input into the preset correlation weight model to obtain the correlation weight result output by the preset correlation weight model; Based on the association weight results, the metadata corresponding to the current multimodal data is dynamically updated; wherein, the metadata includes: the association relationship between data and the access frequency between data.

3. The distributed storage method for multimodal data according to claim 1, characterized in that, Based on a pre-defined spatiotemporal partitioning strategy, the current multimodal data is spatiotemporally partitioned to obtain multiple current multimodal datasets, specifically including: Obtain the spatiotemporal attributes corresponding to each of the current multimodal data; wherein, the spatiotemporal attributes include: timestamp and geographic location; Based on the preset spatiotemporal partitioning strategy and the spatiotemporal attributes corresponding to each current multimodal data, the current multimodal data is spatiotemporally partitioned to obtain multiple current multimodal datasets.

4. The distributed storage method for multimodal data according to claim 1, characterized in that, Based on the metadata corresponding to each current multimodal dataset, the associated data of each current multimodal dataset is determined, and the storage of each current multimodal dataset is allocated according to the storage information corresponding to the associated data, specifically including: Parse the metadata corresponding to each current multimodal dataset to obtain the data association relationships between each current multimodal dataset; Based on the relationships between the data, the associated data among the current multimodal data are determined; wherein, the associated data includes: strongly associated data and weakly associated data; The granularity of the current multimodal data segmentation is determined based on the correlation strength of the associated data. The preset index directory is queried according to the identifier of the associated data to determine the storage information corresponding to each of the associated data. The current multimodal data is segmented according to the segmentation granularity, and then the current multimodal data is allocated to the storage nodes corresponding to the associated data for storage based on the storage information.

5. The distributed storage method for multimodal data according to claim 1, characterized in that, Based on the metadata corresponding to each current multimodal dataset, the method determines the associated data for each current multimodal dataset. After allocating and storing each current multimodal dataset according to the storage information corresponding to the associated data, the method further includes: The current multimodal data is updated and verified to determine whether there has been an update. If so, the incremental index corresponding to the current modal data is obtained, and the incremental index is updated based on the updated data corresponding to the current multimodal data; wherein, the incremental index is used to record the latest position and association information of the current modal data.

6. The distributed storage method for multimodal data according to claim 2, characterized in that, After allocating and storing each of the current multimodal data according to the storage information corresponding to the associated data, the method further includes: Based on the association weight results of the current multimodal data, construct the association graph of the current multimodal data; Based on the correlation graph, predict the cross-modal data of the current scene, and load the cross-modal data into the cache space based on a preset caching mechanism.

7. The distributed storage method for multimodal data according to claim 1, characterized in that, After allocating and storing each of the current multimodal data according to the storage information corresponding to the associated data, the method further includes: Receive a data query instruction to extract the spatiotemporal constraints of the query instruction; The spatiotemporal constraints are matched with the spatiotemporal partitions corresponding to each current multimodal dataset to determine the specified current multimodal dataset. Based on the association graph corresponding to the current multimodal dataset, determine the query path and data access order corresponding to the query command; Based on the query path and data access order, the incremental index and storage information corresponding to the current multimodal dataset are searched.

8. A distributed storage device for multimodal data, characterized in that, The device includes: The analysis unit is used to analyze the current multimodal data based on a preset correlation weight model, so as to dynamically update the metadata corresponding to the current multimodal data based on the analysis results; A partitioning unit is used to perform spatiotemporal partitioning on the current multimodal data based on a preset spatiotemporal partitioning strategy to obtain multiple current multimodal datasets. The storage unit is used to determine the associated data of each current multimodal data based on the metadata corresponding to each current multimodal dataset, and to allocate and store each current multimodal data according to the storage information corresponding to the associated data.

9. A distributed storage device for multimodal data, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in any one of claims 1-7.

10. A non-volatile storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of performing the method described in any one of claims 1-7.

Images