Deep learning-based object storage gateway access method

Abstract

The present invention relates to the technical field of computers. Specifically disclosed is a deep learning-based object storage gateway access method, comprising gateway node selection and gateway node capacity expansion and contraction. Specifically, the present invention comprises: collecting states and features of a gateway cluster and features of a data storage and, by means of a deep learning training model, selecting an appropriate gateway node to process a request sent by a user so as to balance the load of the user request to a gateway; and collecting traffic data, predicting a traffic flow trend by means of the model and, according to a prediction result and by means of an automation script, promptly expanding and contracting the capacity of gateway nodes, so as to avoid insufficient or excessive gateway node resources and effectively and reasonably utilize server resources.

Description

A deep learning-based object storage gateway access method

[0001] This application claims priority to Chinese patent application number CN202311710222.8, filed on December 13, 2023, entitled “A Method for Accessing Object Storage Gateways Based on Deep Learning,” the entire text of which is hereby incorporated by reference. Technical Field

[0002] The present invention relates to the field of computer technology, and in particular to an object storage gateway access method based on deep learning. Background Art

[0003] With the rapid development of the Industrial Internet, user data is exploding, and data storage services are becoming increasingly important. Object storage, a key cloud storage method, has gained popularity among many enterprises. As more and more enterprises embrace cloud computing, object storage faces increasing access pressure. Conventional master-slave models can no longer meet user access needs. Faced with this surge in access requests, optimizing load balancing within object storage gateways has become a pressing issue. Summary of the Invention

[0004] The purpose of the present invention is to provide an object storage gateway access method based on deep learning, including gateway node selection and gateway node expansion and contraction;

[0005] The gateway node selection includes: collecting the state and characteristics of the object storage gateway cluster and the characteristics of the objects to obtain an object dataset; training the object dataset using a deep learning model to obtain an object gateway selection model; and selecting a suitable gateway node using the object gateway selection model to process a request sent by a user;

[0006] The gateway node scaling includes: obtaining a feature information data set of user traffic in different time periods, training it using a recurrent neural network model, and obtaining a gateway traffic prediction model; using the gateway traffic prediction model to predict traffic trends, and realizing adaptive scaling of the gateway node of the object gateway according to the traffic trends.

[0007] Furthermore, the object features include the gateway node through which the object is accessed, the bucket where the object is located, whether it is segmented, and the cluster where the object is located;

[0008] The characteristics of the object storage gateway cluster include the number of rgw processes established on the gateway node, the physical resources of the gateway node, CPU, memory, network card bandwidth and the number of requests processed in the current time period.

[0009] Furthermore, by analyzing the log information of the object storage gateway, a characteristic information dataset of the user traffic in the different time periods is obtained.

[0010] Furthermore, the archived object gateway logs are analyzed to extract the number of user requests processed by the object storage gateway node in different time periods, the OP to which each request belongs, and the processing time of each request to obtain a characteristic information dataset of the user traffic.

[0011] Furthermore, before using the deep learning model for training, the object data set is converted into continuous data, and then the continuous data is normalized to obtain feature vectors, and the feature vectors are spliced ​​into a matrix;

[0012] And / or before using the recurrent neural network model for training, the feature information data set is converted into continuous data, the continuous data is normalized to obtain feature vectors, and the feature vectors are spliced ​​into a matrix.

[0013] Furthermore, the method of converting the data set into continuous data is one-hot encoding processing.

[0014] Furthermore, by collecting the status and characteristics of the cluster of the object storage gateway and the characteristics of the objects, the cluster of the object storage gateway is labeled to obtain a data set.

[0015] Furthermore, the label is that the gateway node is in a good state or the gateway node is in an alarm state.

[0016] Furthermore, the deep learning model includes a dilated convolutional network and a ResNeSt network.

[0017] Furthermore, the dataset is trained with a deep learning model to obtain an object gateway selection model, including:

[0018] The ResNeSt network is used as a basic feature extractor, and a dilated convolution module of the dilated convolutional network is introduced into subsequent layers of the ResNeSt network.

[0019] Furthermore, the recurrent neural network model is an LSTM model.

[0020] Furthermore, a fully connected neural network layer is added after the input layer of the LSTM model, and the initial weights are the same as the initial weights of the LSTM model.

[0021] Furthermore, according to the traffic trend, an automatic scaling script is called to implement adaptive scaling of the gateway node of the object gateway.

[0022] The beneficial effects of the present invention mainly include:

[0023] The present invention provides a solution for access selection of gateway nodes for object storage with the help of the dilated convolutional network and the ResNeSt network model, thereby reducing the load pressure on the gateway cluster. With the help of the optimized long short-term memory network model, the request traffic in the future time period is predicted, and the gateway node resources are adjusted dynamically in time to achieve effective and rational resource utilization. By monitoring and recording the status information of the gateway nodes, while ensuring the normal operation of the cluster, effective elements are provided for feature selection. Compared with traditional load balancing algorithms, by selecting a multi-dimensional feature training model, the rationality and real-time performance of gateway selection are increased. The problem of overfitting is avoided, and the object storage gateway cluster selection model and the traffic prediction model are kept dynamically updated.

[0024] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] FIG1 is a schematic diagram of the overall process of an object storage gateway access method based on deep learning of the present invention;

[0026] Figure 2 is a schematic diagram of a multi-scale feature enhancement module based on dilated convolution and ResNeSt network;

[0027] Figure 3 is a schematic diagram of the LSTM model structure;

[0028] Figure 4 is a schematic diagram of the LSTM model structure based on fully connected neural network optimization. DETAILED DESCRIPTION

[0029] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0031] It should be noted that the terms used herein are for describing particular embodiments only and are not intended to limit the exemplary embodiments according to the present invention.

[0032] The present invention provides an object storage gateway access method based on deep learning, including gateway node selection and gateway node expansion and contraction;

[0033] The gateway node selection includes: collecting the state and characteristics of the object storage gateway cluster and the characteristics of the objects to obtain an object dataset; training the object dataset using a deep learning model to obtain an object gateway selection model; and selecting a suitable gateway node using the object gateway selection model to process a request sent by a user;

[0034] The gateway node scaling includes: obtaining a feature information data set of user traffic in different time periods, training it using a recurrent neural network model, and obtaining a gateway traffic prediction model; using the gateway traffic prediction model to predict traffic trends, and realizing adaptive scaling of the gateway node of the object gateway according to the traffic trends.

[0035] It can be understood that this embodiment discloses the process of scaling up and down of gateway nodes, which includes two main steps: first, obtaining a characteristic information dataset of user traffic in different time periods; then, using a recurrent neural network model to train this dataset to obtain a gateway traffic prediction model. This prediction model can be used to predict the trend of traffic and realize adaptive scaling of the object gateway according to the trend of traffic.

[0036] The gateway node selection of this embodiment can make decisions intelligently based on the state of the object storage gateway cluster, object characteristics, and user requests, thereby improving the accuracy and efficiency of gateway node selection; user traffic is predicted through a recurrent neural network model, enabling adaptive scaling of gateway nodes. This means that the system can automatically adjust the number of gateway nodes based on real-time traffic conditions to cope with different load conditions, thereby improving the elasticity and stability of the system; through intelligent selection of gateway nodes, requests can be assigned to the most appropriate node for processing, thereby improving the performance and response speed of the system; through adaptive scaling, the number of gateway nodes can be dynamically adjusted based on real-time traffic conditions, avoiding resource waste and thus reducing costs; because the model training process can be adjusted according to actual conditions, this method has strong adaptability and can be customized for different scenarios and needs.

[0037] In some embodiments of the present application, the characteristics of the object include the gateway node through which the object is accessed, the bucket where the object is located, whether it is segmented, and the cluster where the object is located;

[0038] The characteristics of the object storage gateway cluster include the number of rgw processes established on the gateway node, the physical resources of the gateway node, CPU, memory, network card bandwidth and the number of requests processed in the current time period.

[0039] It is understandable that this embodiment combines the characteristics of objects and the cluster characteristics of object storage gateways to train deep learning models, which can enable the model to consider more comprehensive information, thereby improving the accuracy and flexibility of gateway node selection. Specific advantages include the following:

[0040] More comprehensive information consideration: including the object's access path, bucket, whether it is segmented, and the cluster in which the object is located, so that the model can take into account more contextual information and make more accurate gateway node selection decisions.

[0041] Dynamically respond to network status: By including features such as the number of RGW processes established on the gateway node, physical resources, CPU, memory, and network card bandwidth, the model can understand the actual status of the current gateway node, which helps to dynamically respond to changes in network status when selecting the gateway node.

[0042] Improved selection accuracy: Combining object and gateway node features allows the model to more comprehensively evaluate the availability and adaptability of each gateway node. This helps select the most appropriate gateway node to handle requests, improving selection accuracy.

[0043] In some embodiments of the present application, the characteristic information datasets of the user traffic in the different time periods are obtained by analyzing the log information of the object storage gateway.

[0044] It's understandable that by analyzing object storage gateway logs and obtaining datasets of user traffic characteristics over different time periods, the model can be made more realistic, capturing real-world user access behavior and traffic patterns, making model training more meaningful and valuable. By acquiring real-time user traffic datasets over different time periods, the system can flexibly respond to traffic bursts or peak periods, enabling adaptive scaling of gateway nodes and improving system resilience and stability.

[0045] In some embodiments of the present application, the archived object gateway log is analyzed to extract the number of user requests processed by the object storage gateway node in different time periods, the OP to which each request belongs, and the time for processing each request to obtain a characteristic information data set of the user traffic.

[0046] It can be understood that in this embodiment, by analyzing the processing time of each request, the performance of the gateway node in different time periods can be understood, which is very helpful for discovering performance bottlenecks and optimizing them; by understanding the number of requests and operation types in different time periods, the user's access patterns and behaviors can be analyzed, so as to better understand the usage of the website or application in different time periods; knowing the number of user requests processed by the object storage gateway node in each time period can help to make reasonable resource planning and adjustments to avoid insufficient resources or waste of resources; by understanding the characteristics of user traffic, more accurate capacity planning can be carried out to ensure that the gateway node can effectively process various requests and avoid performance degradation or service interruption due to insufficient capacity; based on the number of requests and processing time in different time periods, the model can be trained to predict traffic, thereby realizing adaptive expansion and contraction of the object storage gateway node and ensuring the stability and reliability of the system.

[0047] In some embodiments of the present application, before using a deep learning model for training, the object data set is converted into continuous data, the continuous data is normalized to obtain feature vectors, and the feature vectors are concatenated into a matrix;

[0048] And / or before using the recurrent neural network model for training, the feature information data set is converted into continuous data, the continuous data is normalized to obtain feature vectors, and the feature vectors are spliced ​​into a matrix.

[0049] Furthermore, the method of converting the data set into continuous data is one-hot encoding processing.

[0050] It is understandable that converting object data sets into continuous data can eliminate the discreteness in the data, making the data more suitable for the training of deep learning models, thereby improving the performance of the model; normalizing continuous data can make the distribution of data more stable, which is conducive to faster convergence of the model during training, and can also avoid problems such as gradient disappearance or gradient explosion; converting continuous data into feature vectors and splicing them into matrices can provide the model with more effective and structured input data, which helps the model learn and extract features more accurately.

[0051] In some embodiments of the present application, a data set is obtained by collecting the status and characteristics of the cluster of the object storage gateway and the characteristics of the objects, and labeling the cluster of the object storage gateway.

[0052] Furthermore, the label is that the gateway node is in a good state or the gateway node is in an alarm state.

[0053] It is understandable that based on the labeled dataset, the model can more accurately select the most appropriate gateway node to process user requests, thereby improving the efficiency and accuracy of gateway node selection.

[0054] In some embodiments of the present application, the deep learning model includes a dilated convolutional network and a ResNeSt network.

[0055] In some embodiments of the present application, the dataset is trained with a deep learning model to obtain an object gateway selection model, including:

[0056] The ResNeSt network is used as a basic feature extractor, and a dilated convolution module of the dilated convolutional network is introduced into subsequent layers of the ResNeSt network.

[0057] It is understandable that

[0058] As a basic feature extractor, the ResNeSt network has been proven to have powerful feature extraction capabilities in image processing tasks. By introducing dilated convolution modules in the subsequent layers of ResNeSt, the network's perception of the state and features of the object storage gateway can be further enhanced, thereby improving model performance. The dilated convolution modules in the dilated convolutional network can effectively expand the receptive field, allowing the network to obtain a wider range of contextual information while maintaining resolution. This is very helpful for understanding the overall state and feature distribution of the object storage gateway. The dilated convolution modules can help the network have stronger abstraction capabilities when processing high-level features, allowing the model to better understand and express the complex characteristics of the object storage gateway. By introducing dilated convolution modules in the subsequent layers of the ResNeSt network, the advantages of the ResNeSt network in feature extraction can be maintained while enhancing the ability to perceive specific details.

[0059] In some embodiments of the present application, the recurrent neural network model is an LSTM model.

[0060] It can be understood that using the LSTM model as a recurrent neural network model can effectively process data with time series information. It is particularly suitable for scenarios that process time series data, such as processing user traffic in different time periods in an object storage gateway.

[0061] In some embodiments of the present application, a fully connected neural network layer is added after the input layer of the LSTM model, and the initial weights are the same as the initial weights of the LSTM model.

[0062] It is understandable that combining LSTM and fully connected layers can give full play to the advantages of both. LSTM can process sequence data well, and the fully connected layer can introduce more nonlinear transformations on this basis, thereby improving the performance of the model.

[0063] In some embodiments of the present application, according to the traffic trend, an automatic scaling script is called to implement adaptive scaling of the gateway node of the object gateway.

[0064] It is understandable that calling the automatic scaling script can reduce the burden on operation and maintenance personnel. There is no need to manually monitor traffic and make adjustments, which reduces operation and maintenance costs and workload.

[0065] A preferred embodiment of the present application is:

[0066] A deep learning-based object storage gateway access method includes gateway node selection and scaling. Gateway node selection collects the status and characteristics of the gateway cluster and the characteristics of the data storage, trains them using a dilated convolutional network and a ResNeSt network, and uses the model to select an appropriate gateway node to handle requests. Gateway node scaling is achieved by analyzing gateway log information, obtaining traffic information from different time periods, and training it with an optimized LSTM model to obtain a gateway traffic prediction model. This prediction model is then used to adaptively scale the object gateway. The process of the object gateway access solution is shown in Figure 1.

[0067] Gateway node selection is achieved by collecting the status and characteristics of the gateway cluster and the characteristics of the data storage, training them through the dilated convolutional network and ResNeSt network, and using the model to select the appropriate gateway node to process the request. The detailed method flow is as follows:

[0068] 1. Obtaining a dataset: Collect feature information of the currently uploaded objects, including object features and cluster features. Object features include the gateway node through which the object is accessed, the bucket where the object is located, whether it is segmented, and the cluster where the object is located. Cluster features include the number of RGW processes established on the gateway node, the gateway node's physical resources, CPU, memory, network card bandwidth, and the number of requests processed in the current time period. Obtain a dataset.

[0069] 2. Obtain label data: Manually label the cluster status based on the log information of the gateway cluster. The labels are marked as two categories: gateway node status is good and gateway node status is alarm.

[0070] 3. Data preprocessing: First, perform one-hot encoding on discrete information features, normalize continuous data such as cluster capacity, and then concatenate the processed feature vectors into a matrix. For n cluster capacities, C0, C1…C n, the normalized data are:

[0071] 4. By introducing the dilated convolution strategy, the receptive field is expanded without losing the original resolution. This strategy can increase the network's receptive field without increasing the complexity of the algorithm parameters, thereby improving the network's ability to capture feature information. Given an input feature map, the dilated convolution is mathematically defined as Among them, x is the pixel position currently being processed, k is the kernel size, c is the dilation rate, w is the filter, and y is the output of the dilated convolution.

[0072] 5. Use the improved residual network ResNeSt based on ResNet as the backbone network, input the feature data into the ResNeSt model for learning and training, and obtain the gateway selection model. The ResNeSt model designs a split attention module, which splits the input into multiple distributed attention modules. For each cardinality, a channel-based attention mechanism is used to assign different weights to each channel. Then, the channel weight statistics are obtained through global average pooling, and the cardinality with channel weights is output. All outputs are spliced ​​together and recorded as V = Concat{V 1 ,V 2 ,...V k It enables cross-channel attention. Its unique structure solves a series of issues, such as gradient problems, that arise with deep networks. This allows for deeper network construction and strengthens cross-channel interactions. This paper adds a two-dimensional attention mechanism, channel and spatial, to the dilated convolution results to calculate the attention weights of the feature maps, implementing a multi-scale feature enhancement module. Figure 2 shows a multi-scale feature enhancement module based on dilated convolution and the ResNeSt network.

[0073] 6. Use the obtained model to select the appropriate gateway node to process the user's request.

[0074] Gateway node scaling is achieved by collecting user request information in the current time period, using an optimized long short-term memory network model to predict traffic trends in the future time period, dynamically adjusting object storage gateway nodes based on traffic prediction results, and using automated scripts to scale gateway nodes in a timely manner to prevent gateway node resource shortages or surpluses, thereby achieving effective and rational utilization of server resources. The detailed method flow is as follows:

[0075] 1. Obtaining a data set: Analyze the archived object gateway logs and extract characteristic information such as the number of user requests processed by the object storage gateway node in each time period, the OP to which each request belongs, and the processing time of each request to obtain a data set.

[0076] 2. Obtain label data: Manually label the cluster status based on the log information of the gateway cluster. The labels are marked as high business load pressure and low business load pressure.

[0077] 3. Data preprocessing: First, perform one-hot encoding on discrete information features, normalize continuous data such as cluster capacity, and then concatenate the processed feature vectors into a matrix. For n cluster capacities, C0, C1…C n , the normalized data are:

[0078] 4. The model is trained using a long short-term memory (LSTM) network. The LSTM is a feedback neural network that addresses the vanishing gradient problem of recurrent neural networks by improving their internal structure. LSTMs are highly adept at modeling time series and can effectively handle data over large timescales. Figure 3 shows the structure of an LSTM unit.

[0079] Suppose the input sequence is (x1,x2,...,x t ), the hidden layer state is (h1,h2,...,h t ), then at time t, f t =δ(W f ·[h t-1 ,x t ]+b f ); i t =δ(W i ·[h t-1 ,x t ]+b o ); O t =δ(W o ·[h t-1 ,x t ]+b o ); h t =O t tanh(C t ).

[0080] f t ,i t , O t They are respectively the forget gate unit, the input gate unit and the output gate unit, C t is the output layer of the hidden layer.

[0081] 5. Add a fully connected neural network layer after the input layer of the LSTM network, with the same initial weights as the LSTM network. This increases the depth of the entire network, extracting more comprehensive data features and avoiding issues such as a limited number of memory neurons and insufficient network layers that can affect prediction accuracy. The structure of the LSTM model optimized based on the fully connected neural network is shown in Figure 4.

[0082] 6. Use the obtained model to select the appropriate gateway node to process the user's request.

[0083] Example: Five resource pools with different Tianyi Cloud object storage architectures were selected to obtain status and configuration information for a total of 45 nodes in five object storage gateway clusters. Metadata for 500,000 storage objects was screened, and log information was analyzed and extracted to obtain 3 million records of object access. After converting the screened data into a feature matrix, the feature matrix was randomly selected as the training set for a ResNeSt network. Feature data obtained from log analysis was randomly selected and input into an optimized LSTM network as the training set. Ultimately, an object gateway cluster selection model and a gateway scaling traffic prediction model were developed, providing users with a deep learning-based object storage gateway access solution. Using this new object gateway access solution, the cluster selection model was dynamically updated, and status and log information for the object gateway cluster was simultaneously obtained. The results showed that the percentage of gateway cluster health alarms decreased significantly, and the latency in processing user requests during peak periods was also reduced.

[0084] In summary, it can be understood that the object storage gateway access method based on deep learning disclosed by the present invention provides a solution for the selection of gateway nodes for object storage with the help of void convolutional networks and ResNeSt network models, thereby reducing the load pressure on the gateway cluster. With the help of the optimized long short-term memory network model, the request traffic in the future time period is predicted, and the gateway node resources are adjusted dynamically in time to achieve effective and reasonable resource utilization. By monitoring and recording the status information of the gateway nodes, while ensuring the normal operation of the cluster, effective elements are provided for the selection of features. Compared with traditional load balancing algorithms, the rationality and real-time performance of gateway selection are increased by selecting a multi-dimensional feature training model. Maintaining the dynamic update of the cluster selection model makes full use of data features, continuously provides gateway cluster selection solutions and traffic predictions for new user requests, and has a certain improvement effect on the health status and load capacity of the gateway cluster.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

[0086] The system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiment can be combined into one module or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps and are not to be regarded as improper limitations of the present invention.

[0087] Those skilled in the art should be able to appreciate that, in conjunction with the modules and method steps of each example described in the embodiments disclosed herein, it is possible to implement them with electronic hardware, computer software, or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

Claims

1. A method for accessing an object storage gateway based on deep learning, characterized in that: Including gateway node selection and gateway node expansion and contraction; The gateway node selection includes: obtaining an object data set by collecting the state and characteristics of the cluster of the object storage gateway and the characteristics of the object; training the object data set using a deep learning model to obtain an object gateway selection model; and selecting a suitable gateway node using the object gateway selection model to process a request sent by a user; The gateway node expansion and contraction includes: obtaining a feature information data set of user traffic in different time periods, training using a recurrent neural network model, and obtaining a gateway traffic prediction model; using the gateway traffic prediction model to predict traffic trends, and realizing adaptive expansion and contraction of the gateway node of the object gateway according to the traffic trend.

2. The object storage gateway access method based on deep learning according to claim 1, characterized in that: The characteristics of the object include the gateway node through which the object is accessed, the bucket where the object is located, whether it is segmented, and the cluster where the object is located; The characteristics of the cluster of the object storage gateway include the number of rgw processes established on the gateway node, the physical resources of the gateway node, cpu, memory, network card bandwidth and the number of requests processed in the current time period.

3. The object storage gateway access method based on deep learning according to claim 2 is characterized in that: By analyzing the log information of the object storage gateway, a characteristic information data set of the user traffic in the different time periods is obtained.

4. The object storage gateway access method based on deep learning according to claim 3 is characterized in that: The archived object gateway logs are analyzed to extract the number of user requests processed by the object storage gateway node in different time periods, the OP to which each request belongs, and the processing time of each request, to obtain a characteristic information data set of the user traffic.

5. The object storage gateway access method based on deep learning according to claim 4 is characterized in that: Before using the deep learning model for training, the object data set is converted into continuous data, the continuous data is normalized to obtain feature vectors, and the feature vectors are spliced ​​into a matrix; And / or before using the recurrent neural network model for training, the feature information data set is converted into continuous data, and then the continuous data is normalized to obtain feature vectors, and the feature vectors are spliced ​​into a matrix.

6. The object storage gateway access method based on deep learning according to claim 5, characterized in that: The data set is obtained by collecting the status and characteristics of the cluster of the object storage gateway and the characteristics of the objects, and labeling the cluster of the object storage gateway.

7. The object storage gateway access method based on deep learning according to claim 6 is characterized in that: The deep learning model includes a dilated convolutional network and a ResNeSt network.

8. The object storage gateway access method based on deep learning according to claim 7, characterized in that: The data set is trained by a deep learning model to obtain an object gateway selection model, including: The ResNeSt network is used as a basic feature extractor, and a dilated convolution module of the dilated convolution network is introduced into subsequent layers of the ResNeSt network.

9. The object storage gateway access method based on deep learning according to claim 8, characterized in that: The recurrent neural network model is an LSTM model.

10. The object storage gateway access method based on deep learning according to claim 9, characterized in that: A fully connected neural network layer is added after the input layer of the LSTM model, and the initial weights are the same as the initial weights of the LSTM model.

Images