System and method for smoothly updating Internet of Things service

Abstract

The invention discloses a system and a method for smoothly updating an internet of things service, relates to the field of intelligent updating, and particularly relates to the following steps: an updating controller starts an updating process, sends a start message to a cache preheating processor and a connection converter and monitors the progress; and the cache preheating processor performs cache preheating on the new service according to the cache weight table of the old service, so that the data access efficiency is improved. Meanwhile, the connection converter processes the old service equipment connection pool according to a predefined strategy, and health examination and life cycle management are carried out. The normal connection is converted and added into the connection pool of the new service, and the abnormal connection is correspondingly processed. The process ensures effective loading of data and stable transition of connection during service updating, reduces the problems of data access delay and connection interruption caused by service updating, enhances the stability and availability of the system, and realizes efficient and seamless service migration.

Description

Technical Field

[0001] The present application relates to the field of intelligent updating, and more specifically, to a system and method for smoothly updating Internet of Things services. Background Art

[0002] With the rapid development of the Internet of Things (IoT) technology, ensuring efficient updates and upgrades for IoT services has become a key challenge. The rapid increase in the number of IoT devices and the continuous expansion of their application scenarios have placed higher demands on service continuity and stability. Traditional service update methods often carry significant risks, such as potential service interruptions, data loss, and system instability. These issues not only impact the normal operation of IoT devices but can also lead to security risks and unforeseen hazards. Especially in highly stability-reliant fields such as industrial automation, smart homes, and health monitoring, even the slightest failure can have serious consequences.

[0003] Therefore, an optimized solution for smooth update of IoT services is desired, which can realize a more intelligent, agile, secure and reliable IoT service update strategy, thereby ensuring the continuous and stable operation of the IoT system. Summary of the Invention

[0004] In order to solve the above technical problems, this application is proposed.

[0005] According to one aspect of the present application, a method for smoothly updating an Internet of Things service is provided, comprising: The update controller sends a start update message to the cache warmup processor and the connection switch and starts monitoring the update progress; The cache warming processor warms up the cache for the new service based on the cache weight table generated in the old service; The connection converter reads the device connection pool in the old service according to the predefined connection conversion strategy; The connection converter performs health checks and connection lifecycle management on each connection in the device connection pool. If the connection is normal, the corresponding connection is converted into a connection available for the new service and added to the device connection pool of the new service; if the connection is abnormal, the corresponding connection is handled abnormally.

[0006] According to another aspect of the present application, a system for smoothly updating an Internet of Things service is provided, comprising: An update module is used for the update controller to send a start update message to the cache warm-up processor and the connection converter and start monitoring the update progress; The preheating module is used for the cache preheating processor to preheat the cache to the new service according to the cache weight table generated in the old service; A reading module, configured to read a device connection pool in an old service through a connection converter according to a predefined connection conversion strategy; The inspection module is used to perform health checks and connection lifecycle management on each connection in the device connection pool through the connection converter. If the connection is normal, the corresponding connection is converted into a connection available for the new service and added to the device connection pool of the new service; if the connection is abnormal, the corresponding connection is handled abnormally.

[0007] Compared with the prior art, the present application provides a system and method for smooth updates of IoT services. Specifically, the update controller starts the update process, sends a start message to the cache preheating processor and the connection converter, and monitors the progress. The cache preheating processor implements cache preheating for the new service based on the cache weight table of the old service to improve data access efficiency. At the same time, the connection converter processes the old service device connection pool according to the predefined strategy, and performs health checks and lifecycle management. Normal connections are converted and added to the connection pool of the new service, and abnormal connections are processed accordingly. This process ensures the effective loading of data and the smooth transition of connections during service updates, reduces data access delays and connection interruptions caused by service upgrades, enhances the stability and availability of the system, and realizes efficient and seamless service migration. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 The present invention is a flowchart of a method for smoothly updating an Internet of Things service according to an embodiment of the present application.

[0010] Figure 2 Schematic diagram of data flow of a method for smooth update of IoT services according to an embodiment of the present application.

[0011] Figure 3 This is a flowchart of step S120 in the method for smooth updating of IoT services according to an embodiment of the present application.

[0012] Figure 4 This is a flowchart of step S121 in the method for smooth update of IoT services according to an embodiment of the present application.

[0013] Figure 5 The block diagram is a system for smoothly updating IoT services according to an embodiment of the present application. DETAILED DESCRIPTION

[0014] The following describes embodiments of the present disclosure in more detail with reference to the accompanying drawings. While the drawings illustrate certain embodiments of the present disclosure, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0015] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in a different order and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0016] In response to the technical problems in the above background technology, this application proposes a method for smooth updating of Internet of Things services. Figure 1 The present invention is a flowchart of a method for smoothly updating an Internet of Things service according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the method for smooth update of IoT service according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the method for smooth IoT service updates includes: S110, an update controller sends a start-update message to a cache preheating processor and a connection converter and begins monitoring the update progress; S120, the cache preheating processor preheats the cache for the new service based on a cache weight table generated in the old service; S130, the connection converter reads the device connection pool of the old service according to a predefined connection conversion policy; S140, the connection converter performs health checks and connection lifecycle management on each connection in the device connection pool. If a connection is normal, the corresponding connection is converted to a connection available for the new service and added to the device connection pool of the new service; if a connection is abnormal, the corresponding connection is handled. This solution enables smooth IoT service updates, effectively avoiding service interruptions and data loss, and ensuring highly stable system operation. By enhancing the stability and reliability of IoT services, it not only ensures the continued normal operation of devices but also significantly improves application security in key areas such as industrial automation, smart homes, and health monitoring. Furthermore, it reduces manual intervention and system maintenance costs during the update process, significantly improving management efficiency. Taken together, these measures optimize the overall performance of IoT services and build an efficient, secure, and reliable equipment operating environment, which not only reduces operating costs but also speeds up response and improves service quality.

[0017] Specifically, the functions of the above components are: Cache preheating processor: monitors the cache in the old service and performs heat-weighted processing. After receiving the service update instruction, it preheats the new service according to the pre-set preheating strategy. Old service connection pool: maintains the device connection pool of the old version of the IoT service, including functions such as connection creation, management and release. New service connection pool: maintains the database connection pool of the new version of the IoT service, which has the same interface and functions as the old service connection pool. Connection converter: through the connection converter, the connection obtained from the old service connection pool is converted into a connection available for the new service. During the conversion process, the connection is checked for health to detect whether the connection is available. If the connection is found to be invalid, it is removed from the connection pool in time. Update controller: used to control the update process of the IoT service, including starting the update, monitoring the update progress, and completing the update.

[0018] Specifically, in step S110, the update controller sends a start-update message to the cache preheat processor and the connection converter and begins monitoring the update progress. It should be understood that in this solution, each component has different and interrelated tasks. The cache preheat processor is responsible for preheating the cache of the new service based on the weighted cache usage of the old service, while the connection converter is responsible for converting connections from the old service connection pool to the new service connection pool. These components require precise synchronization. As the control core of the system, the update controller sends start-update messages to the cache preheat processor and the connection converter. This is a necessary means to ensure that these components work together, avoid conflicts and disorder, and guarantee the orderly operation of the entire update system. Specifically, cache preheating must be completed before the new service is started to ensure that the new service can quickly load commonly used data and improve response speed. Connection conversion should be executed when the new service is ready to receive connections to achieve a smooth transition of device connections. By precisely controlling the initiation timing of these two steps and adhering to a pre-set logical sequence, the update controller ensures a smooth update process.

[0019] Before initiating an update, the update controller must initialize and test communication connections with various system components to ensure stable and reliable links with the cache warmup processor and connection switch. It also retrieves key information related to the update task from system configuration files or databases, including but not limited to the update version number, detailed information about the target service, and specific strategies for cache warmup and connection switchover. This information provides an accurate basis for subsequent message delivery and progress monitoring.

[0020] After the preparations are completed, the update controller starts sending startup update messages to the cache preheating processor and the connection converter. To ensure that the messages are delivered accurately and promptly, the update controller can use a reliable message queue mechanism, such as Kafka or RabbitMQ. These message queues not only guarantee the reliable delivery of messages, but also alleviate the communication pressure of the system to a certain extent. The update controller sends messages containing update instructions and update-related parameters to the message queue, and the cache preheating processor and the connection converter respectively obtain the messages from the queue and parse them. In the message, the cache preheating processor will be clearly informed of the storage location of the old service cache weight table and the target address of the new service cache; the connection converter will be informed of key information such as the access path of the old service device connection pool, the pre-defined connection conversion strategy, etc.

[0021] After the message is sent, the update controller immediately starts to monitor the update progress. For the work progress monitoring of the cache preheating processor, the update controller can adopt a polling or event-driven approach. If the polling method is adopted, the update controller will regularly send status query requests to the cache preheating processor at preset time intervals. After receiving the request, the cache preheating processor will return the current cache preheating progress information, such as the number of data batches processed, the proportion of data that has completed preheating, etc. If the event-driven method is adopted, the cache preheating processor will actively send an event notification containing progress information to the update controller after completing each key operation stage, such as completing the generation of the optimized cache weight table, completing the reading and writing of a batch of data, etc.

[0022] The update controller can also use a similar approach to monitor the progress of the connection converter. When processing the connection pool of legacy service devices, the connection converter records in real time information such as the number of connections processed, the number of successful and failed connections converted, and so on. The update controller monitors the real-time progress of the connection conversion process by polling this data or receiving progress event notifications proactively pushed by the connection converter. During monitoring, if the update controller detects anomalies in the cache warmup processor or the connection converter, such as extended periods of no progress updates or unusual fluctuations in progress, it immediately triggers an exception handling mechanism. It sends a pause command to the relevant components to prevent further escalation and records exception information, including the time of the exception, component name, and type of exception. The update controller also attempts to recover from the failure by reissuing commands and checking communication links. If recovery is unsuccessful, an alert is promptly sent to the system administrator for manual intervention.

[0023] After the cache warmup processor and connection converter complete their respective tasks and report completion to the update controller, the update controller performs a final status verification and data validation of the entire update process. It checks whether the required data has been correctly loaded into the new service's cache, whether the connection converter has successfully converted all normal connections and added them to the new service's connection pool, and whether abnormal connections have been properly handled. Only after confirming that all update tasks have been completed correctly does the update controller mark the entire IoT service update task as successful, concluding monitoring of the update process.

[0024] Specifically, in step S120, the cache preheating processor performs cache preheating on the new service according to the cache weight table generated in the old service. Accordingly, considering that when a new service is started, if data is not prepared in advance, a large delay may be faced when requesting data for the first time. Different data has different usage popularity and different importance to the service. The old service accumulates cache usage popularity data during operation, and the generated cache weight table can reflect the importance of various types of data. The cache preheating processor operates according to this table and can prioritize loading important and commonly used data for the new service to meet the demand for fast data access during the startup phase of the new service. When the Internet of Things service is updated, if data is not obtained until the new service is started, the service interruption time may be extended. By using the cache weight table to perform cache preheating in advance, the new service can directly obtain data from the cache after startup, quickly respond to requests, reduce the service unavailability time caused by data acquisition delays, and ensure the consistency of user experience during the service update process.

[0025] Figure 3 Flowchart of step S120 in the method for smooth update of IoT service according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 3 As shown, in step S120, the cache preheating processor performs cache preheating on the new service according to the cache weight table generated in the old service, including: S121, based on the attribute distribution of the cache weight table, fine-tuning the cache weight table to obtain an optimized cache weight table; S122, dividing the optimized cache weight table into multiple batches; S123, based on the priority order of the multiple batches, reading data from the cache of the old service and writing the data into the cache of the new service through the internal interface.

[0026] Specifically, step S121 fine-tunes the cache weight table based on the attribute distribution of the cache weight table to obtain an optimized cache weight table. It should be understood that the cache weight table generated for the old service is based on the old service's usage patterns and business requirements. The new service may differ from the old service in terms of functionality, architecture, or business logic. Directly using the old cache weight table may not accurately meet the needs of the new service. By analyzing the attribute distribution of the cache weight table and fine-tuning it, the weight table can be made more suitable for the actual usage scenarios of the new service, thereby improving the effectiveness of cached data. During service updates, the importance and frequency of data usage may change. For example, as the service evolves, some data that was previously less frequently used may become more critical in the new service. Fine-tuning the cache weight table can promptly reflect these data changes, ensuring that the new service prioritizes access to the most important data. However, traditional methods typically perform fine-tuning based on fixed models and algorithms, lacking flexibility and adaptability. In actual IoT services, business scenarios are complex and changing, and the distribution and characteristics of cache weight data will also change accordingly. Due to the lack of such adaptability, traditional methods may perform poorly in complex and changing business scenarios and fail to effectively optimize the cache weight table.

[0027] Based on this, in the process of fine-tuning the cache weight table based on the attribute distribution of the cache weight table to obtain an optimized cache weight table, the technical concept of this application is to first map the weight table into a feature map that characterizes the global distribution through convolutional coding, and then identify and strengthen the areas in the feature map that are strongly related to the business objectives of the new service based on the feature receptive field positioning technology (such as automatically enhancing the weight distribution of grid load forecast-related data when upgrading smart energy management services), and finally reconstruct the optimized weight table through feature decoding. Compared with the traditional fixed model, this solution can adapt to changes in business scenarios through a feature-driven dynamic modulation mechanism. For example, when a new service introduces a real-time decision module, it can automatically increase the priority of streaming data processing to avoid cache preheating deviations caused by weight table lags.

[0028] Figure 4 Flowchart of step S121 in the method for smooth update of IoT service according to an embodiment of the present application. Specifically, in the embodiment of the present application, Figure 4As shown, step S121, based on the attribute distribution of the cache weight table, fine-tunes the cache weight table to obtain an optimized cache weight table, including: S121-1, extracting the cache weight data spatial distribution features based on convolutional coding on the cache weight table to obtain a cache weight distribution feature map; S121-2, performing receptive field anchored fine-grained saliency modulation on the cache weight distribution feature map to obtain a cache weight distribution enhanced feature map; S121-3, performing feature decoding on the cache weight distribution enhanced feature map to obtain the optimized cache weight table.

[0029] Specifically, step S121-1 extracts the spatial distribution features of cache weight data from the cache weight table using convolutional coding to obtain a cache weight distribution feature map. It should be understood that optimizing the cache weight table requires fully considering the differences between the business logic, data access patterns, and real-time requirements of the new and old services. Traditional methods that directly use the old weight table or employ fixed algorithms for fine-tuning often fail to effectively capture the spatial correlation and dynamic changes between data, leading to a mismatch between cache pre-warming weight allocation and the actual needs of the new service. For example, in an intelligent transportation system, an old service may construct a weight table based on historical traffic statistics. However, if a new service introduces real-time accident warning functionality, it will need to increase the cache priority of data related to emergencies (such as the real-time status of sensors on specific road sections). In this case, traditional linear adjustment methods have difficulty identifying such data features with spatial locality and temporal correlation, resulting in insufficient pre-warming of critical data or redundant data occupying cache resources. Therefore, the present application extracts the spatial distribution features of cache weight data from the cache weight table using convolutional coding to obtain a cache weight distribution feature map. Through sliding calculations of multiple convolutional kernels in a convolutional neural network, spatial correlation features at different scales are extracted from the weight table. Shallow convolution captures the adjacent associations of data weights within a local area (such as the weight distribution of multiple electricity meter data associated with a substation in a smart grid), while deep convolution captures long-range dependencies across regions (such as the propagation of weights in hot spots in a city-level charging station network). During this process, the value of each node in the weight table is converted into a multi-channel feature vector corresponding to the location in the feature map. These feature vectors encode the spatial context of the node in the business scenario.

[0030] Specifically, step S121-2 performs receptive field-anchored fine-grained saliency modulation on the cache weight distribution feature map to obtain an enhanced cache weight distribution feature map. Furthermore, considering that dynamic optimization of the cache weight table requires precise identification of key data regions in new service scenarios, traditional methods often employ global uniform enhancement or fixed-rule region selection when adjusting the weight table, making it difficult to cope with sudden changes in local data weights caused by business logic iteration. For example, when a smart city traffic management system is upgraded to support real-time accident response, the dispersed road monitoring data in the original weight table needs to be significantly prioritized in specific high-accident areas (such as intersection sensor clusters). However, traditional coarse-grained enhancement methods may not be able to distinguish the weight differences between accident-related sensors and sensors on ordinary road sections, resulting in delayed loading of critical data during cache warm-up or excessive resource usage of redundant data. This limitation stems from the fact that static feature enhancement mechanisms cannot adequately capture the fine-grained features of data spatial distribution. In particular, when there are local differences in the business logic of new and old services, they cannot adaptively focus on high-value data regions. To this end, the present application performs receptive field-anchored fine-grained saliency modulation on the cache weight distribution feature map to obtain an enhanced cache weight distribution feature map. In this way, combined with dynamic receptive field adjustment, it is possible to accurately locate micro-features in the weight table that are strongly related to the new business (such as sudden changes in the data access pattern of a specific sensor cluster) and enhance them in a targeted manner in the feature space, rather than the "one-size-fits-all" global weight adjustment in traditional methods.

[0031] Specifically, in an embodiment of the present application, step S121-2 performs receptive field-anchored fine-grained significance modulation on the cache weight distribution feature map to obtain a cache weight distribution enhanced feature map, including: performing feature decoupling on the cache weight distribution feature map along the channel dimension to obtain a set of cache weight distribution feature pixel-level initial vectors; extracting the cache weight distribution feature pixel-level initial vector at the (i, j)th pixel position from the set of cache weight distribution feature pixel-level initial vectors as the cache weight distribution feature vector to be enhanced; based on the cache weight distribution feature vector to be enhanced, screening out a set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature from the set of cache weight distribution feature pixel-level initial vectors; based on the set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature, performing significance enhancement on the cache weight distribution feature vector to be enhanced to obtain an enhanced cache weight distribution feature pixel-level vector, wherein the enhanced cache weight distribution feature pixel-level vector is the channel feature vector at the (i, j)th pixel position of the cache weight distribution enhanced feature map.

[0032] Specifically, in an embodiment of the present application, feature decoupling is performed on the cache weight distribution feature map along the channel dimension to obtain a set of cache weight distribution feature pixel-level initial vectors. This process is expressed as follows: ;in, is the cache weight distribution feature map, is the set of real numbers, and They are The height and width of each feature matrix along the channel dimension, yes The number of channels, It is feature decoupling, It is each cache weight distribution feature pixel-level initialization vector in the set of cache weight distribution feature pixel-level initialization vectors.

[0033] It should be understandable that the traditional method is limited in fine-grained analysis due to the strong coupling of channel features. For example, in the weight table of the intelligent transportation system, different channels may encode data dimensions such as traffic flow, accident frequency, and signal light status. When a new service needs to enhance the accident response capability, it is necessary to accurately enhance the local area weights of the accident frequency-related channels. However, if the channel features are not decoupled, the correlation between accident frequency and channels such as traffic flow may mask the abnormal characteristics of a specific area (such as an area with frequent accidents but low traffic flow at a certain intersection), resulting in the inability to independently identify the significance of the accident channel in the area when adjusting the weight. Instead, it is disturbed by fluctuations in data from other channels, resulting in insufficient preheating of key data. This channel coupling problem essentially hinders the accurate capture of local differences in business scenarios. Based on this, the present application decouples the cache weight distribution feature map along the channel dimension to break the inherent correlation between channels. The decoupled pixel-level vector set can reveal the fine-grained business pattern that was originally masked by the coupled features, and obtain a set of pixel-level initial vectors of the cache weight distribution feature.

[0034] Specifically, in the embodiment of the present application, the cache weight distribution feature pixel-level initial vector at the (i, j)th pixel position is extracted from the set of cache weight distribution feature pixel-level initial vectors as the cache weight distribution feature vector to be enhanced. The process is expressed by the formula: ;in, yes The cache weight distribution feature pixel-level initial vector of the (i, j)th pixel position in , It will As the cache weight distribution feature vector to be enhanced.

[0035] Accordingly, the cache weight distribution feature pixel-level initial vector at the (i, j)th pixel position is extracted from the set of cache weight distribution feature pixel-level initial vectors as the cache weight distribution feature vector to be enhanced. This allows the construction of an analysis unit anchored by the target pixel. Through pixel-by-pixel processing, the unique business-sensitive features of each point can be accurately identified and enhanced when switching between new and old services. This avoids the misadjustment of weights caused by lumping together monitoring points with different mechanisms in traditional methods. This enables on-demand hierarchical loading of key data during the cache warm-up phase, significantly improving the decision-making accuracy and real-time response of IoT services after updates.

[0036] More specifically, in an embodiment of the present application, based on the cache weight distribution feature vector to be enhanced, a set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature is screened out from the set of pixel-level initial vectors of the cache weight distribution feature, including: The cache weight distribution feature vector to be enhanced is compressed to obtain a cache weight distribution feature distillation vector to be enhanced. The process is expressed as follows: ;in, It will As the cache weight distribution feature vector to be enhanced, To calculate the Euclidean norm of a vector, is the feature distillation vector of the cache weight distribution to be enhanced; Based on the feature distribution spatial structure characteristics of the cache weight distribution feature distillation vector to be enhanced, the size of the cache weight distribution feature receptive field of the cache weight distribution feature vector to be enhanced is determined. This process is expressed by the formula: ;in, To calculate the Euclidean norm of a vector, is the feature distillation vector of the cache weight distribution to be enhanced, is the logarithmic function value with base 2, yes The size of the receptive field of the cache weight distribution feature; Based on the size of the receptive field of the cache weight distribution feature, a set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature is filtered out from the set of pixel-level initial vectors of the cache weight distribution feature. This process is expressed as follows: ;in, It is a collection of pixel-level initial vectors within the local receptive field of the cache weight distribution feature. They are Middle , , , and The cache weight distribution features at each pixel position are pixel-level initial vectors within the local receptive field.

[0037] It should be understood that when processing feature vectors, if the original high-dimensional data is directly used for calculations, a large amount of non-critical information (such as sensor environmental noise and device status redundancy indicators) will interfere with the extraction of core business features. For example, in the new services of the smart logistics park, it is necessary to strengthen the temperature and humidity anomaly monitoring capabilities of cold chain transportation vehicles. However, the original feature vector is mixed with irrelevant dimensions such as vehicle position, speed, and fuel consumption. If it is directly used for weight adjustment without processing, the temperature mutation feature may be submerged in irrelevant data fluctuations, resulting in insufficient cache preheating priority in the cold chain abnormal area, affecting the efficiency of real-time alarms. Therefore, by compressing the information of the cache weight distribution feature vector to be enhanced, a distillation vector of the cache weight distribution feature to be enhanced is obtained. In other words, feature distillation is achieved through information compression, separating the core business features from the noise redundancy in the cache weight distribution feature vector to be enhanced. The lightweight, high-semantic feature representation established by feature distillation not only reduces the complexity of subsequent calculations, but more importantly, provides a high signal-to-noise ratio input for the adaptive determination of the receptive field size.

[0038] Accordingly, the rigid constraints of the traditional fixed-size receptive field become a bottleneck for accurately locating key business areas. For example, in the scenario of smart medical equipment monitoring, when the new service needs to strengthen the cache weight of the vital signs data in the ICU ward, the abnormal patterns of different vital signs indicators show differentiated spatial characteristics: the occasional premature beats of the electrocardiogram signal require a small receptive field to capture the instantaneous waveform mutation, while the continuous downward trend of blood oxygen saturation requires a large receptive field to correlate with historical baseline data. If a uniform size receptive field is used, the premature beat feature may be diluted by the surrounding stable signal, or the blood oxygen trend may fail to correlate with the evolution of the previous data, ultimately resulting in insufficient weight increase of key indicators during cache preheating, affecting the accuracy of real-time warning. Therefore, the present application determines the size of the cache weight distribution feature receptive field of the cache weight distribution feature vector to be enhanced based on the feature distribution spatial structure characteristics of the distilled vector of the cache weight distribution feature to be enhanced. This size-adaptive mechanism based on the nature of features enables cache weight enhancement to retain local key details while avoiding feature fragmentation or over-smoothing caused by improper perception range. Compared with the traditional fixed grid method, it significantly improves the spatial mapping accuracy of the weight table for the new service business logic, ensuring that the cache data distribution after the service update forms a fine-grained fit with the real-time business scenario.

[0039] Next, based on the size of the receptive field of the cache weight distribution feature, a set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature is filtered from the set of pixel-level initial vectors of the cache weight distribution feature. This semantically driven context aggregation avoids both the waste of resources caused by global processing (such as interference from sensor data in irrelevant rooms in smart home scenarios) and the omission of features caused by fixed-size receptive fields (such as the failure to cover camera data at accident-related intersections in Internet of Vehicles scenarios). Ultimately, this achieves precise enhancement of high-value data areas in the cache weight table, ensuring that the data sets loaded during the new service preheating phase are highly consistent with the upgraded business needs.

[0040] More specifically, in an embodiment of the present application, based on the set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature, the cache weight distribution feature vector to be enhanced is significantly enhanced to obtain an enhanced cache weight distribution feature pixel-level vector, including: performing adversarial semantic topology reconstruction on the set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature to obtain a first modulation weighting coefficient and a second modulation weighting coefficient; based on the first modulation weighting coefficient and the second modulation weighting coefficient, performing weighted fusion based on attention weights on the cache weight distribution feature vector to be enhanced and the set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature to obtain the enhanced cache weight distribution feature pixel-level vector. The processing process of this step is as follows: ;in, yes Middle The cache weight distribution feature of the pixel position is the pixel-level initial vector in the local receptive field, is the cache weight distribution feature score weight vector, is matrix multiplication, yes function, yes The corresponding cache weight distribution feature attention weight, and are the first modulation weighting coefficient and the second modulation weighting coefficient, respectively, yes The enhanced cache weight distribution feature pixel-level vector after enhancement is the channel feature vector of the (i, j)th pixel position of the cache weight distribution enhanced feature map.

[0041] Finally, based on the set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature, the cache weight distribution feature vector to be enhanced is saliency enhanced to obtain an enhanced cache weight distribution feature pixel-level vector. Here, the essence of saliency enhancement is to adjust the representation of the feature vector in the feature space so that the feature components related to saliency are amplified and the feature components related to non-saliency are suppressed or weakened. In other words, through weighted processing, irrelevant feature components (such as traffic flow data during normal time periods) are suppressed, and feature dimensions that are strongly related to business objectives (such as abnormal vehicle trajectory patterns during accident periods) are targeted and amplified, forming an enhanced pixel-level feature vector.

[0042] In particular, the first modulation weighting coefficient and the second modulation weighting coefficient The calculation process is as follows: First, the first modulation weighting coefficient and the second modulation weighting coefficient That is, for the cache weight distribution feature vector to be enhanced The corresponding set of pixel-level initial vectors in the local receptive field of the cache weight distribution feature In order to improve the gain optimization of significant correlation features and the attenuation control of non-significant correlation features, it is expected that the cache weight distribution feature has a set of pixel-level initial vectors within the local receptive field. It is able to have conformal representation, that is, it is expected that the geometric correspondence between its regional volume feature map and the interface contour feature encoding can be maintained.

[0043] Therefore, we first define the cache weight distribution area volume feature map vector as: .

[0044] The cache weight distribution interface topology feature encoding vector is defined as: .

[0045] Then, by adjusting the weighting coefficient and , so that the interface topology-volume tensor structure satisfies the canonical constraints of the exchange relation, that is, the cache weight distribution area volume feature map vector Interface topological feature encoding vector with cache weight distribution The Frobenius norm of the exchange residual vector between the two approaches the scaling factor A linear combination of the modulation weight coefficients: .

[0046] In this process, by reasonably setting the interface topology constraint criteria and utilizing the volume conformal exchange characteristics, the regularity specification standard is met, thereby achieving the pixel-level initial vector of the local receptive field of the cache weight distribution feature. Geometric conformal fusion representation significantly enhances the cache weight distribution feature vector to be enhanced The context-aware feature discrimination capability is improved while maintaining the geometric stability of the feature space under the approximate Euclidean metric framework of the fusion domain.

[0047] Because when performing weighted fusion, and The sum of is 1, and thus, the first modulation weighting coefficient and the second modulation weighting coefficient can be obtained through the above processing.

[0048] Specifically, step S121-3, feature decoding is performed on the cache weight distribution enhanced feature map to obtain the optimized cache weight table. Specifically, in an embodiment of the present application, feature decoding is performed on the cache weight distribution enhanced feature map to obtain the optimized cache weight table, including: using a decoder to feature decode the cache weight distribution enhanced feature map to obtain the optimized cache weight table. It should be understood that although the cache weight distribution enhanced feature map highlights the features that are strongly related to the new service business objectives through convolutional coding and fine-grained saliency modulation, it is still an abstract feature expression form and cannot be directly applied to the cache preheating process. Cache preheating of IoT services requires specific cache weight data to guide the data loading strategy, so the feature map must be converted into an operational cache weight table, and the decoder is the key tool to achieve this conversion. By decoding the cache weight distribution enhanced feature map through the decoder, the key features highlighted in the feature map, the adjusted weight relationship and other information can be converted into specific cache weight data, and then the optimized cache weight table can be constructed. This weight table accurately matches the functions, architecture, and business logic of the new service, allowing the cache preheating processor to prioritize loading important and commonly used data required for the startup phase of the new service based on this table, thereby improving the response speed and operating efficiency of the new service.

[0049] Specifically, a decoder is used to perform feature decoding on the cache weight distribution enhancement feature map to obtain the optimized cache weight table. The process is as follows: Once the cached weight distribution enhanced feature map is generated, the decoder begins its work. The decoder is typically a specially trained neural network module, whose structure and parameters are designed and optimized based on task requirements and data characteristics. Before decoding begins, the decoder needs to load pre-trained model parameters. These parameters contain key information for effectively decoding the feature map and guide the decoder to accurately convert the feature map into an optimized cache weight table.

[0050] The decoder takes as input a cached weight distribution enhanced feature map. This feature map is obtained through convolutional coding and fine-grained saliency modulation. It contains feature information that is strongly related to the business objectives of the new service, as well as the adjusted weight relationship, but it is still an abstract form of expression. The first layer of the decoder performs preliminary processing on the input feature map. Through specific convolution kernels and activation functions, the channel and spatial dimensions of the feature map are adjusted to extract more representative features. For example, a transposed convolution operation may be used to restore the size of the feature map to a dimension similar to the original cached weight table. At the same time, the features are reorganized and integrated to enhance the correlation between features.

[0051] As processing progresses, the decoder gradually transforms abstract features into specific weight data. In the intermediate layers, features are further fused and mapped through a series of fully connected layers and nonlinear transformations. These layers associate features in the feature map with actual cached weight data based on patterns learned during training. For example, for regions that are significantly enhanced in the feature map, the decoder maps them to higher weight values ​​to indicate their importance in the new service; for relatively unimportant regions, the decoder generates lower weight values.

[0052] The final decoder layer outputs the same structure as the optimized cache weight table. This output may be a two-dimensional matrix, where each element corresponds to the weight of a different data point in the cache. However, the weights obtained at this point may require some post-processing. For example, the weights may be normalized to ensure that the sum of all weights is 1. This facilitates the subsequent allocation of resources based on the weight ratio during the cache warm-up process. Furthermore, the weights may be thresholded, setting some excessively small weights to 0 to reduce unnecessary waste of computational resources.

[0053] After the post-processing operation is completed, the result is an optimized cache weight table. This weight table accurately matches the functionality, architecture, and business logic of the new service and provides accurate guidance for the cache warm-up processor.

[0054] In summary, step S121 is explained. It first maps the weight table into a feature map representing the global distribution through convolutional coding. Then, based on feature receptive field positioning technology, it identifies and strengthens areas in the feature map that are strongly correlated with the new service's business objectives (for example, when upgrading smart energy management services, the weight distribution of grid load forecasting-related data is automatically enhanced). Finally, through feature decoding, it reconstructs the optimized weight table. Compared to traditional fixed models, this solution, through a feature-driven dynamic modulation mechanism, can adapt to changes in business scenarios. For example, when a new service introduces a real-time decision module, it can automatically increase the priority of streaming data processing to avoid cache warm-up deviations caused by weight table lags.

[0055] Specifically, step S122 divides the optimized cache weight table into multiple batches. Accordingly, considering that the optimized cache weight table may contain a large amount of data, if all the data that need to be preheated are loaded into the cache of the new service at one time, it will cause huge pressure on system resources (such as memory, network bandwidth, etc.), which may cause system performance degradation or even failure. The present application can disperse the pressure of data loading by dividing the optimized cache weight table into multiple batches, so that the system can smoothly perform cache preheating operations. That is, batch processing allows the cache preheating processor to read data from the cache of the old service and write it to the cache of the new service in sequence according to the priority order of the batches. In this way, the transmission and processing of large amounts of data can be broken down into multiple smaller tasks, each task processing a part of the data, thereby improving the overall processing efficiency.

[0056] Specifically, before dividing batches, the system first evaluates its own resources, including memory capacity, network bandwidth, and CPU processing power. It also analyzes and optimizes the data in the cache weight table, categorizing and labeling the data based on its frequency of use and its importance to the launch and operation of the new service. For example, data that will be frequently accessed when a new service is launched is marked as high priority, while data that will only be used in specific business scenarios or later is marked as low priority.

[0057] When dividing batches, they will be allocated based on the total amount of data, resource carrying capacity and data priority. If the system memory is limited, the amount of data in each batch will be controlled to avoid memory overflow caused by loading too much data at one time. For high-priority data, priority is given to the front batches, and the amount of data in the batch is relatively small to ensure that these critical data can be quickly loaded into the new service cache. Low-priority data is arranged in the back batches, and the amount of data in the batch can be appropriately increased. Assuming that there are 100 data in the optimized cache weight table, it is divided into 5 batches based on resource evaluation and data priority. Among them, the first two batches each contain 15 high-priority data, the middle two batches each contain 20 medium-priority data, and the last batch contains 30 low-priority data.

[0058] After the division is completed, a unique identifier will be generated for each batch, and a corresponding index will be established to facilitate quick positioning in subsequent data reading and writing operations. At the same time, the batch division information will be recorded in the system log, including the start and end data identifiers, priority, estimated processing time, etc. of each batch, to facilitate system monitoring and management. During the execution process, the cache preheating processor will process them in batch order. When processing each batch, the corresponding data is first read from the old service cache, and written to the new service cache through the internal interface. After completing a batch of data processing, a completion notification is sent to the update controller. The update controller monitors the progress of the entire cache preheating accordingly to ensure that data loading is carried out in an orderly manner, avoid excessive pressure on system resources, and improve the efficiency and stability of new service startup and operation.

[0059] Specifically, in step S123, based on the priority order of the multiple batches, data is read from the cache of the old service and written into the cache of the new service through an internal interface. It should be understood that different data has different importance for the normal operation of the new service. By setting a priority order for each batch, it can be ensured that data that is critical to the new service can be loaded into the cache first. In this way, the new service can process key business as soon as possible after startup, thereby improving the availability and response speed of the service. In particular, loading data in order of priority can make the cache warm-up process more efficient. For example, loading frequently accessed data first can satisfy the requests of most users at the initial startup of the new service, reducing user waiting time.

[0060] After determining the batch priority order for the optimized cache weight table, the cache warmup processor begins data read and write operations based on this order. Before initiating data transfer, the cache warmup processor first establishes stable connections with both the old and new service caches to ensure smooth data transfer. Simultaneously, it initializes its internal data processing modules to prepare for data reading, transfer, and writing.

[0061] For the first high-priority batch, the cache warmup processor locates the data storage location for that batch in the old service cache based on the batch index information. Assuming the old service cache uses a distributed storage architecture, the cache warmup processor accurately locates the corresponding data block based on the node address and data identifier in the index. When reading data, it uses efficient data reading algorithms, such as batch reading, to reduce read times and improve read efficiency. After reading the data, it transfers it to the new service cache via an internal interface. This internal interface typically uses a high-speed communication protocol to ensure fast and stable data transmission. Before writing the data to the new service cache, the cache warmup processor performs necessary format conversion and validation to ensure data integrity and correctness. If the new service cache has specific data format requirements, the cache warmup processor will adapt the format accordingly. If any data errors or missing data are detected during the validation process, the cache warmup processor will attempt to reread the data or take appropriate action based on the error handling strategy.

[0062] After completing the first batch of data writing, the cache preheating processor will send feedback information on the completion of batch processing to the update controller, including details such as the amount of data written and whether there are any errors. After receiving the feedback, the update controller will record the processing results of the batch and monitor the progress of the entire cache preheating. Then, the cache preheating processor will process the next batch of data according to the same process. As the batch progresses, the requirements for the timeliness of data processing for batches with gradually lower priorities are relatively reduced, but the accurate writing of data must still be guaranteed.

[0063] During the entire data transmission process, resource usage, such as network bandwidth and memory usage, will be monitored in real time. If resources are found to be tight, the data transmission rate may be adjusted or the data transmission of some low-priority batches may be suspended to prioritize the processing of high-priority data and the stable operation of the system. When all batches of data are successfully read from the old service cache and written to the new service cache, the cache preheating processor will send a signal to the update controller that all batches have been processed. After the update controller confirms it, it indicates that the data transmission task of the cache preheating phase has been successfully completed. The new service can use this preheated data to quickly start and run efficiently, providing users with a stable and smooth service experience.

[0064] Specifically, in step S130, the connection converter accesses the device connection pool in the old service according to a predefined connection conversion policy. Specifically, the predefined connection conversion policy includes connection conversion priority, whether to initiate a reconnection for failed device connections, and whether to ensure the final consistency of connection conversions. Accordingly, due to the complex device connection landscape in the IoT environment, different devices vary in terms of their importance to the service, data volume, and usage frequency. Connection conversion priorities can differentiate between different device connections, ensuring that critical device connections are prioritized for conversion, thus ensuring the normal operation of core services. Prioritizing device connections responsible for critical data collection can avoid business interruptions. During the connection conversion process, various issues can cause conversion failures. If left unaddressed, some device connections may be lost. Defining whether to initiate a reconnection for failed device connections allows for remedial measures to be taken, improving the success rate of connection conversions and ensuring that as many devices as possible can successfully access the new service. Furthermore, data consistency is crucial for IoT services. Ensuring the final consistency of connection conversions ensures data integrity and accuracy during the transition between the old and new services. In financial IoT scenarios, ensuring the consistency of transaction data during the connection conversion process is crucial to avoiding transaction errors and data corruption.

[0065] After the connection converter is started, it first reads and parses the predefined connection conversion policy. This policy is typically stored in a system configuration file or database, and the connection converter accesses this information through the corresponding interface. Connection conversion priority is determined based on factors such as the device's importance to the business, data transmission volume, and frequency of use. For example, in industrial automation scenarios, devices responsible for core production process data collection have higher connection priority; in smart home scenarios, devices related to security monitoring have higher connection priority than ordinary household appliances. Based on these preset rules, the connection converter prioritizes the connections in the legacy service device connection pool and builds a priority queue.

[0066] Next, the connection converter begins reading connections from the legacy service device's connection pool. It processes each connection in order of priority. When reading a connection, it establishes a stable communication connection with the legacy service device's connection pool to ensure the accuracy and integrity of the data read. For each connection read, the connection converter performs subsequent operations based on the connection conversion strategy.

[0067] During the connection conversion process, the connection converter performs a health check on each connection. It sends specific detection messages or performs a handshake protocol to determine whether the connection is functioning properly. If the connection is functioning properly, the connection converter converts it to a connection format compatible with the new service. This may involve modifying connection parameters or adapting the protocol. For example, if the new service uses a new communication protocol, the connection converter will convert the protocol of the old connection to the new one. Once the conversion is complete, the connection converter adds the connection to the new service's device connection pool.

[0068] However, if the connection check detects an anomaly, the connection converter will decide whether to initiate a reconnection based on a pre-defined policy. If the policy allows reconnection, the connection converter will attempt to reestablish the connection with the device. It will first close the current abnormal connection and then re-initiate the connection request based on the device's connection information. During the reconnection process, the connection converter will make multiple attempts and set a reasonable retry interval. If a successful connection is still not successful after multiple retries, the connection converter will record the abnormal connection information, including device identification, abnormality type, number of retries, etc., and handle it according to the preset exception handling mechanism, such as reporting the abnormal connection information to the system administrator for subsequent investigation and repair.

[0069] Throughout the entire connection conversion process, ensuring eventual consistency is crucial. The connection converter employs various techniques to achieve this. For example, it uses a transaction management mechanism to process the connection conversion operation as an atomic transaction. When converting multiple connections, either all connections are successfully converted and added to the new service connection pool, or, if an error occurs, all completed conversion operations are rolled back to ensure consistency between the old and new service connection states. Furthermore, the connection converter verifies and backs up data during the conversion process. Before the connection conversion, the relevant data in the old service connection pool is backed up; after the conversion is complete, the data in the new service connection pool is verified to ensure that no data was lost or corrupted during the conversion process. If any data inconsistency is discovered, the connection converter promptly initiates a data repair process, comparing the backup data with the current data to identify and repair any discrepancies, ensuring eventual consistency during the connection conversion.

[0070] When the connection converter has processed all connections in the old service device's connection pool, it sends a connection conversion completion message to the update controller, along with a detailed report of the connection conversion, including the number of successfully converted connections, the number of failed connections, and the reasons for the failures. Based on this information, the update controller determines whether the connection conversion was successful and proceeds with subsequent operations.

[0071] Specifically, in step S140, the connection converter performs a health check and connection lifecycle management on each connection in the device connection pool. If the connection is normal, the corresponding connection is converted to a connection available for the new service and added to the new service's device connection pool. If the connection is abnormal, the corresponding connection is handled. It should be understood that some connections in the old service's device connection pool may have potential risks, such as unstable connections due to network fluctuations or abnormal connections caused by device failures. Directly converting these connections to the new service would introduce problems into the new service, affecting its performance and stability. Therefore, performing a health check before conversion can screen out healthy connections and ensure high-quality connections for the new service. Connection lifecycle management covers the creation, use, and release of connections. During the update process, standardized connection lifecycle management ensures that each connection is properly handled. Timely conversion of normal connections maintains the connection status between the device and the service, avoids service interruptions caused by improper connection handling, and ensures the continuity of IoT services during the update. Handling abnormal connections when anomalies occur can prevent them from negatively impacting the new service. Unhandled anomalies such as connection leaks and timeouts can consume system resources and even cause system crashes. Timely exception handling ensures stable system operation and enhances the reliability of the entire IoT service system. Therefore, converting normal connections and adding them to the new service connection pool is a key step in achieving a smooth transition between old and new services. This allows devices to quickly access new services during service updates, minimizing connection interruptions and making the service update process virtually imperceptible to users, ensuring smooth IoT service operation.

[0072] Upon startup, the connection converter first retrieves information about all connections from the legacy service device's connection pool. It establishes a stable communication link with the legacy service device's connection pool to accurately read detailed data about each connection, such as the connection identifier, device address, port number, communication protocol, and current connection status. After obtaining this information, the connection converter performs a health check on each connection. This health check utilizes a combination of detection methods. First, the connection converter sends specific test messages, such as heartbeat packets, to the connection. Heartbeat packets are lightweight messages used to quickly verify connection activity. Receiving a heartbeat response from a connection within a specified timeframe indicates that the connection is functioning to a certain extent. Second, the connection converter also performs a simple data transmission test. It sends some pre-set test data and checks whether the receiving end can correctly receive and parse the data. These two methods provide a comprehensive assessment of the connection's health.

[0073] During health checks, the connection converter also manages the connection lifecycle. The connection lifecycle covers multiple stages, including creation, use, maintenance, and release. During this process, the connection converter records the lifecycle status of each connection. Connections undergoing health checks are marked as "Checking." If the health check passes, the connection status is updated to "Convertible." If the check fails, the connection status is marked as "Abnormal."

[0074] When the connection is judged to be normal, the connection converter will convert it into a connection that can be used by the new service. Since the new and old services may have differences in communication protocols, data formats, etc., the connection converter needs to perform a series of adaptation operations. For example, if the new service adopts a new encryption protocol, the connection converter will upgrade the protocol of the old connection and convert it into a connection format that meets the requirements of the new service. After the conversion is completed, the connection converter will add the connection to the device connection pool of the new service. Before joining, it will check the status of the new service connection pool again to ensure that the connection pool has sufficient resources to accommodate the new connection and will not cause the connection pool to be overloaded. At the same time, the connection converter will update the relevant metadata of the connection, such as the identifier of the service to which it belongs, so that the new service can correctly identify and manage the connection.

[0075] If the connection is judged to be abnormal during the health check, the connection converter will immediately handle the exception. The exception handling method depends on the type of exception and the pre-set strategy. For some exceptions caused by temporary network failures, the connection converter will try to reconnect. It will first disconnect the current abnormal connection, and then re-initiate the connection request based on the device's connection configuration information. During the reconnection process, a reasonable number of retries and retry intervals will be set to avoid excessive retries and waste of resources. If the connection cannot be restored after multiple retries, or the exception is caused by unrecoverable reasons such as device hardware failure, the connection converter will mark the connection as "unavailable" and record detailed exception information, including the time the exception occurred, the type of exception, and related identification of the connection. These records will be stored in the system's log file for subsequent troubleshooting and analysis. At the same time, the connection converter will send an alert notification to the system administrator to inform him that there is an abnormal connection that needs to be handled.

[0076] Throughout the entire processing process, the connection converter continuously monitors changes in the connection status. For connections that have been added to the new service connection pool, it regularly performs health checks to ensure that the connection continues to operate stably in the new service environment. For connections in an abnormal state, the connection converter decides whether to try to restore the connection again or completely release related resources based on the system administrator's processing results or preset rules. When all connections have undergone health checks, connection conversion, or exception handling, the connection converter sends a connection processing completion report to the update controller. The report includes information such as the number of successfully converted connections, the number of abnormal connections, and the handling status, so that the update controller can accurately understand the results of the connection conversion and promote the smooth update process of the IoT service.

[0077] In summary, a method for smooth update of IoT services based on an embodiment of the present application is explained. Specifically, the update controller starts the update process, sends a start message to the cache preheating processor and the connection converter, and monitors the progress. The cache preheating processor implements cache preheating for the new service based on the cache weight table of the old service to improve data access efficiency. At the same time, the connection converter processes the old service device connection pool according to a predefined strategy, and performs health checks and lifecycle management. Normal connections are converted and added to the connection pool of the new service, and abnormal connections are processed accordingly. This process ensures the effective loading of data and the smooth transition of connections during service updates, reduces data access delays and connection interruptions caused by service upgrades, enhances the stability and availability of the system, and achieves efficient and seamless service migration.

[0078] Figure 5 FIG. 1 is a block diagram of a system for smoothly updating an Internet of Things service according to an embodiment of the present application. Figure 5 As shown, the system 100 for smooth update of IoT services according to an embodiment of the present application includes: an update module 110, which is used for the update controller to send a start update message to the cache preheating processor and the connection converter and start monitoring the update progress; a preheating module 120, which is used for the cache preheating processor to preheat the cache for the new service according to the cache weight table generated in the old service; a reading module 130, which is used to read the device connection pool in the old service through the connection converter according to a predefined connection conversion strategy; a checking module 140, which is used to perform health checks and connection lifecycle management on each connection in the device connection pool through the connection converter. If the connection is normal, the corresponding connection is converted into a connection available for the new service and added to the device connection pool of the new service; if the connection is abnormal, the corresponding connection is handled abnormally.

[0079] Here, those skilled in the art will appreciate that the specific operations of each step in the system for smooth update of the Internet of Things service have been described above with reference to Figures 1 to 4The method for smooth update of IoT services has been described in detail, and therefore, its repeated description will be omitted.

[0080] In summary, it is intended that the above detailed description be considered illustrative rather than restrictive, and it should be understood that the above embodiments are intended to be merely illustrative of the present invention and not to limit the scope of protection of the present invention. After reading the contents of the present invention, a skilled person may make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A method for smoothly updating an Internet of Things service, characterized in that: include: The update controller sends a start update message to the cache warmup processor and the connection switch and starts monitoring the update progress; The cache warming processor warms up the cache for the new service based on the cache weight table generated in the old service; The connection converter reads the device connection pool in the old service according to the predefined connection conversion strategy; The connection converter performs health checks and connection lifecycle management on each connection in the device connection pool. If the connection is normal, the corresponding connection is converted into a connection available for the new service and added to the device connection pool of the new service; if the connection is abnormal, the corresponding connection is handled abnormally.

2. The method for smoothly updating an Internet of Things service according to claim 1, characterized in that: The predefined connection conversion strategy includes connection conversion priority, whether to initiate reconnection of a device connection that fails in conversion, and whether to ensure the final consistency of the connection conversion.

3. The method for smoothly updating Internet of Things services according to claim 2, characterized in that: The cache warming processor warms up the cache for the new service based on the cache weight table generated in the old service, including: Based on the attribute distribution of the cache weight table, fine-tuning the cache weight table to obtain an optimized cache weight table; Dividing the optimized cache weight table into multiple batches; Based on the priority order of the multiple batches, data is read from the cache of the old service and written into the cache of the new service through an internal interface.

4. The method for smoothly updating Internet of Things services according to claim 3, characterized in that: Based on the attribute distribution of the cache weight table, fine-tuning the cache weight table to obtain an optimized cache weight table includes: Performing convolutional coding-based cache weight data spatial distribution feature extraction on the cache weight table to obtain a cache weight distribution feature map; Performing receptive field anchored fine-grained saliency modulation on the cache weight distribution feature map to obtain a cache weight distribution enhanced feature map; Feature decoding is performed on the cache weight distribution enhancement feature map to obtain the optimized cache weight table.

5. The method for smoothly updating Internet of Things services according to claim 4, characterized in that: Performing receptive field anchored fine-grained saliency modulation on the cache weight distribution feature map to obtain a cache weight distribution enhanced feature map, including: Performing feature decoupling on the cache weight distribution feature map along the channel dimension to obtain a set of cache weight distribution feature pixel-level initial vectors; Extracting the cache weight distribution feature pixel-level initial vector at the (i, j)th pixel position from the set of cache weight distribution feature pixel-level initial vectors as the cache weight distribution feature vector to be enhanced; Based on the cache weight distribution feature vector to be enhanced, filtering out a set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature from the set of pixel-level initial vectors of the cache weight distribution feature; Based on a set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature, the cache weight distribution feature vector to be enhanced is significantly enhanced to obtain an enhanced cache weight distribution feature pixel-level vector, wherein the enhanced cache weight distribution feature pixel-level vector is the channel feature vector of the (i, j)th pixel position of the cache weight distribution enhanced feature map.

6. The method for smoothly updating Internet of Things services according to claim 5, characterized in that: Based on the cache weight distribution feature vector to be enhanced, a set of pixel-level initial vectors within a local receptive field of the cache weight distribution feature is screened out from the set of pixel-level initial vectors of the cache weight distribution feature, including: Performing information compression on the cache weight distribution feature vector to be enhanced to obtain a cache weight distribution feature distillation vector to be enhanced; Determining the size of the cache weight distribution feature receptive field of the cache weight distribution feature vector to be enhanced based on the feature distribution spatial structure characteristics of the cache weight distribution feature distillation vector to be enhanced; Based on the size of the receptive field of the cache weight distribution feature, a set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature is filtered out from the set of pixel-level initial vectors of the cache weight distribution feature.

7. The method for smoothly updating Internet of Things services according to claim 6, characterized in that: Based on a set of pixel-level initial vectors within a local receptive field of the cache weight distribution feature, the cache weight distribution feature vector to be enhanced is significantly enhanced to obtain an enhanced cache weight distribution feature pixel-level vector, including: Performing adversarial semantic topology reconstruction on a set of pixel-level initial vectors within a local receptive field of the cache weight distribution feature to obtain a first modulation weighting coefficient and a second modulation weighting coefficient; Based on the first modulation weighting coefficient and the second modulation weighting coefficient, the cache weight distribution feature vector to be enhanced and the set of pixel-level initial vectors within the local receptive field of the cache weight distribution feature are weighted fused based on attention weights to obtain the enhanced cache weight distribution feature pixel-level vector.

8. The method for smoothly updating Internet of Things services according to claim 7, characterized in that: Feature decoding is performed on the cache weight distribution enhancement feature map to obtain the optimized cache weight table, including: using a decoder to feature decode the cache weight distribution enhancement feature map to obtain the optimized cache weight table.

9. A system for smooth update of Internet of Things services, characterized in that: include: An update module is used for the update controller to send a start update message to the cache warm-up processor and the connection converter and start monitoring the update progress; The preheating module is used for the cache preheating processor to preheat the cache to the new service according to the cache weight table generated in the old service; A reading module, configured to read a device connection pool in an old service through a connection converter according to a predefined connection conversion strategy; The inspection module is used to perform health checks and connection lifecycle management on each connection in the device connection pool through the connection converter. If the connection is normal, the corresponding connection is converted into a connection available for the new service and added to the device connection pool of the new service; if the connection is abnormal, the corresponding connection is handled abnormally.

Images