An information management method, system, terminal and storage medium based on a cloud platform

Abstract

The present application relates to an information management method, system, terminal and storage medium based on a cloud platform, and relates to the field of communication technologies. The method includes: in response to receiving a configuration request, generating a configuration model according to the configuration request, and obtaining model configuration requirements corresponding to the configuration model; pruning and quantifying the configuration model according to the model configuration requirements to obtain at least two quantized configuration models, wherein different quantized configuration models consume different amounts of computing resources; obtaining configuration information of edge nodes; grouping the edge nodes according to the quantized configuration models and the configuration information to obtain edge node groups and target quantized configuration models corresponding to the edge node groups one by one; and transmitting the target quantized configuration models to the edge node groups. The present application has the effects of reducing the influence of the hardware conditions of edge nodes and improving the configurability of the model.

Description

Technical Field

[0001] This application relates to the field of communication technologies, and in particular, to an information management method, system, terminal, and storage medium based on a cloud platform. Background Art

[0002] A cloud platform is a service platform based on cloud computing technology, which provides computing resources, storage resources, network resources, and various application services through the Internet. The information management configuration of the cloud platform has become increasingly important.

[0003] In related technologies, when configuring user models with the same function for edge nodes, the data center will integrate the existing model architectures to obtain a unified user model. After that, the data center will send the user model to each edge node to achieve the deployment of the user model.

[0004] Regarding the above related technologies, the inventors believe that due to different hardware conditions of edge nodes, when configuring models, the unified model will make it difficult for edge nodes with poor hardware conditions to fully utilize the model. Summary of the Invention

[0005] In order to reduce the influence of the hardware conditions of edge nodes and improve the configurability of models, this application provides an information management method, system, terminal, and storage medium based on a cloud platform.

[0006] In a first aspect, this application provides an information management method based on a cloud platform, adopting the following technical solution:

[0007] An information management method based on a cloud platform includes:

[0008] In response to receiving a configuration request, generate a configuration model according to the configuration request, and obtain the model configuration requirements corresponding to the configuration model;

[0009] Prune and quantize the configuration model according to the model configuration requirements to obtain at least two quantized configuration models, where different quantized configuration models consume different amounts of computing;

[0010] Obtain the configuration information of the edge node;

[0011] Group the edge nodes according to the quantized configuration model and the configuration information to obtain edge node groups and target quantized configuration models corresponding to the edge node groups one by one;

[0012] Transmit the target quantized configuration model to the edge node group.

[0013] By adopting the above technical solution, pruning and quantifying the configuration model according to the model configuration requirements to obtain at least two quantized configuration models. And for edge nodes with different configuration information, allocate target quantized configuration models. So that the target quantized configuration model can adapt to edge nodes with different configurations, reduce the influence of the hardware conditions of edge nodes, and improve the configurability of the model.

[0014] Optionally, extract device performance information from the configuration information;

[0015] Obtain the network transmission performance of the edge node according to the device performance information;

[0016] Predict the transmission duration of the target quantized configuration model according to the target quantized configuration model and the network transmission performance;

[0017] When the transmission duration is greater than the preset duration, add a network networking for the edge node.

[0018] By adopting the above technical solution, predict the transmission duration of the target quantized configuration model according to the target quantized configuration model and the network transmission performance. And when the transmission duration is greater than the preset duration, add a network networking for the edge node to improve the network transmission performance between the data center and the edge node and shorten the time for the data center to deploy the configuration model to the edge node.

[0019] Optionally, split the target quantized configuration model into several functional modules;

[0020] Statistical data capacity of the functional modules;

[0021] Classify the several functional modules into a first functional module and a second functional module according to the data capacity and the capacity threshold, the data capacity of the first functional module is greater than the capacity threshold, and the data capacity of the second functional module is less than the capacity threshold;

[0022] Transmit the first functional module to the edge node group through the network networking;

[0023] Transmit the second functional module to the edge node group through the host networking.

[0024] By adopting the above technical solution, transmit the first functional module and the second functional module to the edge node through different networking methods, make full use of the transmission characteristics of different networkings, and improve the transmission efficiency of the first functional module and the second functional module.

[0025] Optionally, elect a group representative node from the edge node group;

[0026] Statistically group edge nodes with the same target quantization configuration model to obtain a combined group representative node;

[0027] Split the functional module into several sub-modules, where the functional module includes the first functional module and the second functional module;

[0028] Statistically count the number of management nodes corresponding to the group representative nodes;

[0029] Send the sub-modules to the group representative nodes according to the number of management nodes, where the number of sub-modules sent is negatively correlated with the number of management nodes, and the group representative nodes are used to obtain the functional module by sharing the sub-modules.

[0030] By adopting the above technical solution, sending sub-modules to the group representative nodes according to the number of management nodes enables the group representative nodes managing more nodes to obtain fewer sub-modules, reducing the losses caused by information security incidents occurring in the group representative nodes. Moreover, the sub-modules obtained by different group representative nodes are different, and the impact brought by model leakage can also be reduced by adopting the form of distributed deployment.

[0031] Optionally, in response to detecting an abnormal condition in the transmission network of the first group representative node, according to the configuration information of the first group representative node, search for a matching candidate group representative node in the edge node group corresponding to the first group representative node;

[0032] Calculate the transmission distance between the candidate group representative node and the data center;

[0033] Take the candidate group representative node corresponding to the minimum value in the transmission distances as the second group representative node;

[0034] Stop sending the sub-modules to the first group representative node and send the sub-modules to the second group representative node.

[0035] By adopting the above technical solution, when an abnormal condition occurs in the transmission network of the first group representative node, use the second group representative node to replace the first group representative node to ensure that the process of deploying the target quantization configuration model is not interrupted, improving the data transmission efficiency and the stability of the transmission process.

[0036] Optionally, in response to receiving a node abnormal signal, statistically count the target group representative nodes that send the node abnormal signal, where the node abnormal signal is sent when the model obtained by combining the target group representative nodes is inconsistent with the target quantization configuration model;

[0037] In the case where the number of the node abnormal signals is greater than a preset number threshold, determine the abnormal group representative nodes according to the target group representative nodes;

[0038] Remove the abnormal grouped representative node from the abnormal edge node group where it is located;

[0039] Send a warning signal to the abnormal grouped representative node.

[0040] By adopting the above technical solution, after generating a node abnormal signal, the abnormal grouped representative node will be judged according to the abnormal signal sent by the target grouped representative node, and the abnormal grouped representative node will be removed to ensure the safety of the remaining edge nodes, and an action model for automatically discovering edge node vulnerabilities will be realized.

[0041] Optionally, obtain the access record of the abnormal grouped representative node;

[0042] Extract the abnormal visitor and the abnormal behavior record of the abnormal visitor from the access record;

[0043] Send the abnormal visitor and the abnormal behavior record to the edge node;

[0044] Receive the retrieval result returned by the edge node;

[0045] Obtain candidate abnormal edge nodes according to the retrieval result.

[0046] By adopting the above technical solution, through the abnormal visitors of the abnormal grouped representative node and the abnormal behavior records of the abnormal visitors, check whether there are similar problems in other edge nodes, so as to find candidate abnormal edge nodes that may have problems for relevant personnel to further verify.

[0047] In a second aspect, the present application provides an information management system based on a cloud platform, adopting the following technical solution:

[0048] An information management system based on a cloud platform, comprising:

[0049] An acquisition module, configured to acquire a configuration request, configuration information, a preset duration, a data capacity, a capacity threshold, a grouped representative node, the number of management nodes, a transmission distance, a node abnormal signal, a preset quantity threshold, a warning signal, an access record, and a retrieval result;

[0050] A memory, configured to store a program of the information management method based on the cloud platform;

[0051] A processor, the program in the memory can be loaded and executed by the processor and implement the information management method based on the cloud platform.

[0052] By adopting the above technical solution, pruning and quantifying the configuration model according to the model configuration requirements to obtain at least two quantized configuration models. And for edge nodes with different configuration information, allocate target quantized configuration models. So that the target quantized configuration model can adapt to edge nodes with different configurations, reduce the influence of the hardware conditions of the edge nodes, and improve the configurability of the model.

[0053] In a third aspect, the present application provides an intelligent terminal, adopting the following technical solution:

[0054] An intelligent terminal includes a memory and a processor, and a computer program capable of being loaded and executed by the processor for any one of the above-mentioned methods is stored on the memory.

[0055] In a fourth aspect, the present application provides a computer storage medium that can store corresponding programs, and has the characteristics of being convenient to reduce the influence of the hardware conditions of edge nodes and improve the configurability of the model. The technical solution is as follows:

[0056] A computer-readable storage medium stores a computer program capable of being loaded and executed by the processor for any one of the above-mentioned information management methods based on the cloud platform.

[0057] In summary, the present application includes at least one of the following beneficial technical effects:

[0058] 1. Prune and quantify the configuration model according to the model configuration requirements to obtain at least two quantized configuration models. And for edge nodes with different configuration information, allocate target quantized configuration models. So that the target quantized configuration model can adapt to edge nodes with different configurations, reduce the influence of the hardware conditions of the edge nodes, and improve the configurability of the model;

[0059] 2. Predict the transmission duration of the target quantized configuration model according to the target quantized configuration model and the network transmission performance. And when the transmission duration is greater than the preset duration, add a network connection for the edge node to improve the network transmission performance between the data center and the edge node, and shorten the time for the data center to deploy the configuration model to the edge node;

[0060] 3. When an abnormal situation occurs in the transmission network of the first group representative node, use the second group representative node to replace the first group representative node to ensure that the process of deploying the target quantized configuration model is not interrupted, and improve the transmission efficiency of data and the stability of the transmission process. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 is a schematic flowchart of an information management method based on a cloud platform provided by an embodiment of the present application.

[0062] Figure 2It is a schematic flowchart of a networking configuration method for an edge node provided by an embodiment of the present application.

[0063] Figure 3 It is a schematic flowchart of a transmission method for a configuration model provided by an embodiment of the present application.

[0064] Figure 4 It is a schematic flowchart of a data transmission method based on a cloud platform provided by an embodiment of the present application.

[0065] Figure 5 It is a schematic flowchart of an exception management method based on a cloud platform provided by an embodiment of the present application.

[0066] Figure 6 It is a schematic flowchart of a first exception warning method based on a cloud platform provided by an embodiment of the present application.

[0067] Figure 7 It is a schematic flowchart of a second exception warning method based on a cloud platform provided by an embodiment of the present application.

[0068] Figure 8 It is a schematic structural diagram of an information management system based on a cloud platform provided by an embodiment of the present application. Detailed implementation manners

[0069] To make the objectives, technical solutions and advantages of the present application clearer and more understandable, the following further elaborates on the present application in conjunction with the appended Figure 1 to the appended Figure 8 and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0070] An embodiment of the present application discloses an information management method based on a cloud platform. This method is executed by a data center. Referring to Figure 1 , this method includes:

[0071] Step S101: In response to receiving a configuration request, generate a configuration model according to the configuration request, and obtain the model configuration requirements corresponding to the configuration model.

[0072] The cloud platform includes a data center and edge nodes. The data center is used to uniformly manage multiple edge nodes, and the data center can provide services to the edge nodes. Further, the data center can provide services to the edge nodes in the form of a deployment model. Among them, in the case where a model has been deployed on the edge node, a model upgrade service can also be provided to the edge node.

[0073] The configuration request is an operation to request the data center to deploy a model to the edge node. Optionally, the configuration request includes at least one of the model type, the deployment time limit of the model, the model number, and the model version.

[0074] The configuration model can be any type of model, and the present application does not make specific limitations thereto.

[0075] The model configuration requirements refer to the hardware requirements for the computer device when deploying the configuration model on the computer device. Optionally, the model configuration requirements include at least one of GPU (Graphics Processing Unit), TPU (Tensor Processing Unit), memory, and database storage capacity.

[0076] Step S102: Prune and quantize the configuration model according to the model configuration requirements to obtain at least two quantized configuration models, where different quantized configuration models consume different amounts of computation.

[0077] Pruning is used to remove unimportant parameters or neuron structures in the configuration model. On the premise of maintaining the performance of the configuration model, the complexity of the configuration model is reduced.

[0078] Quantization is used to reduce the numerical precision of the model parameters of the configuration model, thereby reducing the storage and computation overhead.

[0079] Exemplarily, generate quantization model configuration requirements according to the model configuration requirements, where the quantization model configuration requirements are lower than the model configuration requirements. Prune and quantize the configuration model according to the quantization model configuration requirements to obtain at least two quantized configuration models. For example, reduce the memory in the model configuration requirements to obtain the quantization model configuration requirements.

[0080] Further, evaluate the parameters (which can also be neuron structures) in the configuration model according to a preset standard to obtain the importance scores of the parameters, and the importance scores are positively correlated with the importance degrees of the parameters. Classify the parameters into necessary parameters and non-necessary parameters according to the importance scores. Remove the non-necessary parameters in ascending order to obtain a candidate configuration model. Compare whether the model configuration requirements of the candidate configuration model are less than the quantization model configuration requirements. If so, use the candidate configuration model as the quantized configuration model. If not, continue to execute the step of removing non-necessary parameters.

[0081] Step S103: Obtain the configuration information of the edge node.

[0082] The content included in the configuration information of the edge node is the same as the model configuration requirements, but the numerical values of the configuration information of the edge node and the model configuration requirements are different.

[0083] Optionally, the data center sends a configuration request to the edge node. After receiving the configuration request, the edge node sends the configuration information to the data center.

[0084] Step S104: Group the edge nodes according to the quantization configuration model and the configuration information, obtain the edge node groups and the target quantization configuration models that correspond one by one to the edge node groups.

[0085] In some embodiments, count the first model configuration requirements corresponding to the quantization configuration model. Generate a first model configuration requirement interval according to the first model configuration requirements, and the model configuration requirements within the first model configuration requirement interval are not greater than the first model configuration requirements. Taking the first model configuration requirement interval as the classification criterion, divide the edge nodes located within the same first model configuration requirement interval into the same group, obtain the edge node groups and the target quantization configuration models that correspond one by one to the edge node groups.

[0086] Furthermore, to improve the fault tolerance rate, a quantity upper limit can be set for the number of edge nodes in the edge node groups. When the number of edge nodes divided into the quantization configuration model reaches the edge node quantity upper limit, another group of edge node groups can be set. At this time, the same quantization configuration model will correspond to at least two groups of edge node groups.

[0087] Step S105: Transmit the target quantization configuration model to the edge node groups.

[0088] Optionally, the data center can transmit the target quantization configuration model dispersedly or as a whole.

[0089] By adopting the above technical solution, prune and quantize the configuration model according to the model configuration requirements, and obtain at least two quantization configuration models. And for the edge nodes with different configuration information, allocate the target quantization configuration models. So that the target quantization configuration models can adapt to edge nodes with different configurations, reduce the influence of the hardware conditions of the edge nodes, and improve the configurability of the model.

[0090] In the following embodiments, when deploying the target quantization configuration model, it is necessary to consider that the network transmission performances of each edge node are different. To ensure that each edge node can complete the deployment of the model within a certain time, it is necessary to optimize the network status of the edge nodes according to the actual situation. Therefore, the embodiments of the present application disclose a networking configuration method for edge nodes. Refer to Figure 2 , this method includes:

[0091] Step S201: Extract device performance information from the configuration information.

[0092] The device performance information refers to the device parameters on the computer device that affect network communication. Exemplarily, the device performance information includes at least one of CPU, memory, network card, network interface performance, and database read and write speed.

[0093] Step S202: Obtain the network transmission performance of the edge nodes according to the device performance information.

[0094] The network transmission performance is used to represent the network data transmission speed between the data center and the edge nodes.

[0095] In some embodiments, according to the CPU, memory, and database read and write speeds, the network latency of the edge node is calculated. According to the network interface performance, the network bandwidth between the edge node and the data center is calculated. The network latency and network bandwidth are scored to obtain a latency score and a bandwidth score. The latency score and the bandwidth score are weighted and calculated to obtain the network transmission performance.

[0096] Step S203: Predict the transmission duration of the target quantization configuration model according to the target quantization configuration model and the network transmission performance.

[0097] Exemplarily, calculate the transmission volume of the target quantization configuration model. According to the transmission volume and the network transmission performance, the transmission duration is calculated.

[0098] Step S204: When the transmission duration is greater than the preset duration, add a network networking for the edge node.

[0099] Network networking refers to a virtual network implemented at the network device (such as a switch, router, etc.) level, and a logical network is constructed through physical network devices. Optionally, the network networking configures a tunneling protocol on the network device to enable the encapsulation and transmission of data packets between network devices.

[0100] In some other embodiments, when the transmission duration is less than the preset duration, it indicates that the network transmission condition of the edge node is good and there is no need to add network networking.

[0101] By adopting the above technical solution, according to the target quantization configuration model and the network transmission performance, the transmission duration of the target quantization configuration model is predicted. And when the transmission duration is greater than the preset duration, a network networking is added for the edge node to improve the network transmission performance between the data center and the edge node and shorten the time for the data center to deploy the configuration model to the edge node.

[0102] In the following embodiments, when transmitting the configuration model, to improve the transmission efficiency, different modules can be transmitted for different transmission methods to improve the transmission efficiency. Therefore, an embodiment of the present application discloses a method for transmitting a configuration model. Referring to Figure 3 , the method includes:

[0103] Step S301: Split the target quantization configuration model into several functional modules.

[0104] Exemplarily, obtain the functional layers of the target quantization configuration model, where the functional layers include at least one of a fully connected layer, a convolutional layer, a recurrent layer, and an output layer. According to the functions of each functional layer, split the functional layers with the same function into the same functional module.

[0105] Step S302: Statistically calculate the data capacity of the functional module.

[0106] The data capacity is used to represent the occupied space of the model's module in the computer device.

[0107] Exemplarily, statistically calculate the number of parameters and data types in the functional module. According to the number of parameters and data types, calculate the data capacity of the functional module. Among them, different data types correspond to different basic data capacities. For example, one parameter of data type A corresponds to a data capacity, then b data of data type A will consume a * b data capacity.

[0108] Step S303: Classify several functional modules into a first functional module and a second functional module according to the data capacity and the capacity threshold. The data capacity of the first functional module is greater than the capacity threshold, and the data capacity of the second functional module is less than the capacity threshold.

[0109] The capacity threshold is a preset empirical value, and relevant personnel can adjust the specific value of the capacity threshold according to actual needs.

[0110] Exemplarily, if the data capacity of the sub-module is greater than the capacity threshold, classify the sub-module as the first functional module. If the data capacity of the sub-module is less than the capacity threshold, classify the sub-module as the second functional module.

[0111] Step S304: Through network networking, transmit the first functional module to the edge node in groups.

[0112] It should be noted that there is no order of precedence between Step S304 and Step S305. Step S304 can be executed first, Step S305 can be executed first, or Step S304 and Step S305 can be executed simultaneously.

[0113] Step S305: Through host networking, transmit the second functional module to the edge node in groups.

[0114] Host networking refers to a virtual network implemented at the host level (such as virtual machines or containers), and a logical network is created between hosts through software. Exemplarily, run a virtual network proxy on the edge node and encapsulate and transmit data packets through tunneling technology.

[0115] Among them, network networking has better network processing efficiency compared to host networking and is suitable for transmitting large-volume data.

[0116] By adopting the above technical solutions, the first functional module and the second functional module are transmitted to the edge nodes through different networking methods, making full use of the transmission characteristics of different networkings to improve the transmission efficiency of the first functional module and the second functional module.

[0117] In the following embodiments, when transmitting data from the data center to the edge nodes, it is necessary to consider how to reduce the adverse effects caused by data leakage. An embodiment of the present application discloses a data transmission method based on a cloud platform. Refer to Figure 4 , the method includes:

[0118] Step S401: Select a group representative node from the edge node groups.

[0119] The group representative node can be any edge node in the edge node group.

[0120] Exemplarily, select the edge node with the best configuration information in the edge node group as the group representative node. Or, the group representative node is elected by voting among the edge nodes in the edge node group. Or. Select the edge node with the best network transmission performance in the edge node group as the group representative node.

[0121] Step S402: Count the edge node groups with the same target quantization configuration model to obtain a group representative node combination.

[0122] The group representative node combination is composed of group representative nodes, and the group representative nodes in the same group representative node combination correspond to the same target quantization configuration model.

[0123] Exemplarily, count the edge node groups with the same target quantization configuration model, and select the group representative nodes from these edge node groups to form a group representative node combination.

[0124] Step S403: Split the functional module into several sub-modules, where the functional module includes a first functional module and a second functional module.

[0125] Further, this step includes two cases: splitting the first functional module into several sub-modules and splitting the second functional module into several sub-modules.

[0126] Exemplarily, taking the functional module as a fully connected layer as an example, according to the output dimension of the fully connected layer, split the fully connected layer into several sub-layers to obtain several sub-modules. For example, if the output dimension of the fully connected layer is m, the fully connected layer can be split into two sub-layers, and the output dimension of each sub-layer is m / 2. Or, if the weight matrix of the fully connected layer is an n×m matrix, it is decomposed into the product of several small matrices through matrix decomposition technology to achieve the splitting of the fully connected layer.

[0127] Further, the difference between the data capacities of the split sub-modules is not less than a preset data capacity, so as to ensure that the data capacities of each sub-module are close, which is beneficial to subsequent allocation and sending of sub-modules.

[0128] Step S404: Count the number of management nodes corresponding to the grouped representative nodes.

[0129] The number of management nodes refers to the total number of edge nodes in the edge node group where the grouped representative node is located.

[0130] Exemplarily, the data center sends work query information to the edge nodes in the edge node group. The work query information is used to query whether the edge nodes are in a working state. The data center counts the work status information returned by the edge nodes to obtain the number of management nodes.

[0131] Exemplarily, the data center sends node number query information to the grouped representative nodes. After receiving the data query information, the grouped representative nodes send work query information to the edge nodes in the same edge node group. The grouped representative nodes count the work status information returned by the edge nodes to obtain the number of management nodes. The grouped representative nodes send the number of management nodes to the data center.

[0132] Step S405: Send sub-modules to the grouped representative nodes according to the number of management nodes. Among them, the sending quantity of the sub-modules is negatively correlated with the number of management nodes. The grouped representative nodes are used to combine the shared sub-modules to obtain functional modules.

[0133] Optionally, sort the grouped representative nodes in ascending order according to the number of management nodes to obtain the sorted grouped representative nodes. Allocate the sending quantities of the sub-modules of the grouped representative nodes according to the sorted grouped representative nodes. On the one hand, when the total number of grouped representative nodes is greater than the number of sub-modules, denote the number of sub-modules as a and the total number of grouped representative nodes as b, where b is greater than a. Randomly select a - 1 sub-modules from the sub-modules. Allocate the a - 1 sub-modules to the 1st to the a - 1th grouped representative nodes in the sorted grouped representative nodes, so that there is a one-to-one correspondence between each sub-module and the grouped representative node. Allocate the remaining sub-modules after taking out the a - 1 sub-modules to the a - th to the b - th grouped representative nodes. On the other hand, when the total number of grouped representative nodes is less than or equal to the number of sub-modules, calculate the division of the number of sub-modules by the total number of grouped representative nodes to obtain the quotient and the remainder. Denote the quotient as c and the remainder as d. If the remainder is 0, randomly select b groups of sub-modules from the sub-modules, with each group of sub-modules including c - 1 sub-modules. Evenly allocate the b groups of sub-modules to the grouped representative nodes. Randomly select p sub-modules from the remaining b sub-modules and allocate them to the 1st grouped representative node, where p = [b / 2] + 1, and [b / 2] represents the floor operation on b / 2. Allocate the remaining b - p sub-modules to the 2nd to the b - th grouped representative nodes. If the remainder is not 0, randomly select b groups of sub-modules from the modules, with each group of sub-modules including c sub-modules. Randomly select q sub-modules from the remaining d sub-modules and allocate them to the 1st grouped representative node, where q = [d / 2] + 1. Allocate the remaining d - q sub-modules to the 2nd to the b - th grouped representative nodes.

[0134] Further, after the transmission of the sub-modules is completed, the grouped representative nodes will share the sub-modules they hold with other grouped representative nodes. At this time, the grouped representative nodes can combine the sub-modules to obtain all the complete functional modules, and combine the functional modules into a complete target quantization configuration model. Then, the grouped representative nodes deploy the target quantization configuration model to other edge nodes in the edge node group.

[0135] Further, before transmitting the sub-modules, a check code can also be added to the sub-modules. The grouped representative nodes can judge whether the target quantization configuration model is complete through the check code.

[0136] By adopting the above technical solutions, sub-modules are sent to the grouped representative nodes according to the number of management nodes, so that the grouped representative nodes managing more nodes obtain fewer sub-modules, reducing the losses caused by information security incidents of the grouped representative nodes. Moreover, the sub-modules obtained by different grouped representative nodes are different, and the distributed deployment form can also reduce the impact of model leakage.

[0137] In the following embodiments, when a problem occurs in the transmission to the group representative node, other edge nodes can be selected from the edge node group to replace the group representative node, so as to improve the transmission stability and anti-interference ability. Therefore, the embodiments of the present application disclose an exception management method based on a cloud platform. Referring to Figure 5 , the method includes:

[0138] Step S501: In response to detecting an abnormal condition in the transmission network of the first group representative node, according to the configuration information of the first group representative node, search for a matching candidate group representative node in the edge node group corresponding to the first group representative node.

[0139] The abnormal condition includes at least one of communication interruption, network delay and jitter, data synchronization failure, data tampering or loss, and network attack.

[0140] Exemplarily, according to the configuration information of the first group representative node and the configuration information of other edge nodes in the edge node group, a configuration difference is obtained. The edge nodes with a configuration difference less than a preset difference threshold are taken as candidate group representative nodes.

[0141] Step S502: Calculate the transmission distance between the candidate group representative node and the data center.

[0142] The transmission distance is determined according to the information transmission method between the candidate group representative node and the data center. For example, when the candidate group representative node and the data center communicate through an optical fiber, the transmission distance refers to the physical distance between the candidate group representative node and the data center. When the candidate group representative node and the data center communicate through a wireless network, the transmission distance refers to the network logical distance between the candidate group representative node and the data center.

[0143] Step S503: Take the candidate group representative node corresponding to the minimum value in the transmission distance as the second group representative node.

[0144] In some embodiments, if there are at least two candidate group representative nodes corresponding to the minimum value in the transmission distance, the second group representative node is randomly selected.

[0145] Step S504: Stop sending sub-modules to the first group representative node and send sub-modules to the second group representative node.

[0146] In some embodiments, if the abnormal condition is any one of communication interruption, network delay and jitter, or data synchronization failure, then after stopping sending sub-modules to the first group representative node, the first group representative node sends the sub-modules that the data center has transmitted to the first group representative node to the second group representative node.

[0147] In some embodiments, the data center may send a complete sub-module to the second group representative.

[0148] By adopting the above technical solution, when an abnormal condition occurs in the transmission network of the first group representative node, the second group representative node is used to replace the first group representative node, ensuring that the process of deploying the target quantization configuration model is not interrupted, and improving the transmission efficiency of data and the stability of the transmission process.

[0149] In the following embodiments, the edge node can also detect whether there is a security problem with the group representative node by judging the integrity of the target quantization configuration model. Therefore, an embodiment of the present application discloses an abnormal warning method based on a cloud platform. Refer to Figure 6 , the method includes:

[0150] Step S601: In response to receiving a node abnormal signal, count the target group representative nodes that send the node abnormal signal, where the node abnormal signal is sent when the model combined by the target group representative nodes is inconsistent with the target quantization configuration model.

[0151] Exemplarily, before transmitting the sub-module, the data center will add a check code to the sub-module. The group representative node can judge whether the target quantization configuration model is complete through the check code. Further, the edge node can verify the check code of each sub-module to judge whether the sub-module is available.

[0152] The node abnormal signal includes the number of the abnormal sub-model.

[0153] Step S602: When the number of node abnormal signals is greater than a preset number threshold, determine the abnormal group representative node according to the target group representative node.

[0154] The preset number threshold is a preset empirical value, and relevant personnel can adjust the value of the preset number threshold according to actual needs.

[0155] Exemplarily, determine the abnormal sub-model number according to the node abnormal signal. Determine the group representative node corresponding to the abnormal sub-model number to obtain the abnormal group representative node.

[0156] In some other embodiments, during the process of transmitting data, accidental events such as data loss or garbled codes may occur. Therefore, when the number of node abnormal signals is less than the preset number threshold, the above accidental events are likely to occur, so the step of determining the abnormal group representative node according to the target group representative node is not performed.

[0157] Step S603: Remove the abnormal group representative node from the abnormal edge node group where it is located.

[0158] Exemplarily, disconnect the communication connection between the data center and the abnormal grouping representative node. Alternatively, the data center stops receiving data or information from the abnormal grouping representative.

[0159] Step S604: Send a warning signal to the abnormal grouping representative node.

[0160] The warning signal is used to notify the abnormal grouping representative node of the existence of a security risk.

[0161] In some other embodiments, determine the user terminal corresponding to the abnormal grouping representative node. Send a warning signal to the user terminal.

[0162] By adopting the above technical solution, after generating a node abnormal signal, the abnormal grouping representative node is determined according to the abnormal signal sent by the target grouping representative node, and the abnormal grouping representative node is removed to ensure the safety of the remaining edge nodes, and an action model for automatically discovering edge node vulnerabilities is realized.

[0163] In the following embodiments, security risks existing in other edge nodes can also be found through the abnormal grouping representative node. Therefore, the embodiments of the present application disclose a second abnormal warning method based on a cloud platform. Refer to Figure 7 , the method includes:

[0164] Step S701: Obtain the access record of the abnormal grouping representative node.

[0165] The access record includes a login record, a file access record, a network access record, and a service access record. Among them, the login record is used to record the detailed information of the user or system logging in to the node, and the login record includes at least one of the login time, login IP address, login user, login method, and login success or failure status. The file access record is used to record the read and write operations on the files on the node. The file access record includes at least one of the file path, operating user, operation time, and operation type. The network access record is used to record the network communication between the abnormal grouping representative node and other devices, and the network access record includes at least one of the source IP, target IP, communication port, communication protocol, transmission time, and connection status. The service access record is used to record the access behavior of the services running on the abnormal grouping representative node, and the service access record includes at least one of the access time, request path, client IP, response status code, and response time.

[0166] Step S702: Extract the abnormal visitors and the abnormal behavior records of the abnormal visitors from the access record.

[0167] Exemplarily, set abnormal behaviors. Extract abnormal visitors that match the abnormal behaviors from the access records, and obtain the abnormal behavior records corresponding to the abnormal visitors. Among them, the abnormal behaviors may include at least one of high-frequency access, access during unconventional periods, frequent login failures, access to unconventional paths, and abnormal IP addresses.

[0168] Step S703: Send the abnormal visitors and the abnormal behavior records to the edge node.

[0169] Furthermore, after the edge node receives the abnormal visitors and the abnormal behavior records, the edge node retrieves them in its own access records to obtain a retrieval result. If the abnormal visitors or the abnormal behavior records are retrieved, the retrieval result indicates a risk. If neither the abnormal visitors nor the abnormal behavior records are retrieved, the retrieval result indicates no risk.

[0170] Step S704: Receive the retrieval result returned by the edge node.

[0171] Step S705: Obtain candidate abnormal edge nodes according to the retrieval result.

[0172] Exemplarily, extract the risk retrieval results that indicate no risk, and determine the edge nodes corresponding to the risk retrieval results to obtain candidate abnormal edge nodes.

[0173] By adopting the above technical solution, through the abnormal visitors represented by the abnormal groups and the abnormal behavior records of the abnormal visitors, check whether there are similar problems in other edge nodes, so as to find candidate abnormal edge nodes that may have problems for relevant personnel to further verify.

[0174] Based on the same inventive concept, an embodiment of the present application provides an information management system based on a cloud platform. Please refer to Figure 8 , the system includes:

[0175] An acquisition module 801, configured to acquire a configuration request, configuration information, a preset duration, a data capacity, a capacity threshold, a grouped representative node, the number of management nodes, a transmission distance, a node abnormal signal, a preset quantity threshold, a warning signal, an access record, and a retrieval result;

[0176] A memory 802, configured to store a program of the information management method based on the cloud platform;

[0177] A processor 803, and the program in the memory can be loaded and executed by the processor to implement the information management method based on the cloud platform.

[0178] By adopting the above technical solution, pruning and quantifying the configuration model according to the model configuration requirements to obtain at least two quantized configuration models. And for edge nodes with different configuration information, assign the target quantized configuration model. So that the target quantized configuration model can adapt to edge nodes with different configurations, reduce the influence of the hardware conditions of the edge nodes, and improve the configurability of the model.

[0179] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the above division of each functional module is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0180] The embodiments of the present application provide a computer-readable storage medium storing a computer program that can be loaded and executed by a processor to perform an information management method based on a cloud platform.

[0181] Computer storage media include, for example: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0182] Based on the same inventive concept, the embodiments of the present application provide an intelligent terminal, including a memory and a processor, and a computer program that can be loaded and executed by the processor to perform an information management method based on a cloud platform is stored on the memory.

[0183] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the above division of each functional module is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0184] The above are all the preferred embodiments of the present application. Without limiting the protection scope of the present application accordingly, any feature disclosed in this specification (including the abstract and drawings), unless specifically described, can be replaced by other equivalent or similar-purpose alternative features. That is, unless specifically described, each feature is only an example of a series of equivalent or similar features.

Claims

1. A cloud platform-based information management method, characterized in that: The cloud platform includes a data center and an edge node. The method is performed by the data center. The method includes: In response to receiving a configuration request, generating a configuration model according to the configuration request, and obtaining a model configuration requirement corresponding to the configuration model; Pruning and quantizing the configuration model according to the model configuration requirement to obtain at least two quantized configuration models, wherein different quantized configuration models consume different amounts of computation; Get the configuration information of the edge node; The edge nodes are grouped according to the quantitative configuration model and the configuration information to obtain edge node groups and target quantitative configuration models corresponding to the edge node groups one by one; Transmitting the target quantization configuration model to the edge nodes in groups; Before transmitting the target quantization configuration model to the edge nodes in groups, the method further includes: extracting device performance information from the configuration information; Acquire the network transmission performance of the edge node according to the device performance information; Predicting the transmission duration of the target quantization configuration model according to the target quantization configuration model and the network transmission performance; When the transmission duration is longer than a preset duration, adding a network group for the edge node; The transmitting the target quantization configuration model to the edge nodes in groups includes: Splitting the target quantization configuration model into several functional modules; Counting the data capacity of the functional modules; According to the data capacity and the capacity threshold, classify the plurality of functional modules to obtain a first functional module and a second functional module, wherein the data capacity of the first functional module is greater than the capacity threshold, and the data capacity of the second functional module is less than the capacity threshold; Transmitting the first functional module to the edge node in groups through the network; The second functional module is transmitted to the edge node in groups through the host networking.

2. The cloud platform-based information management method according to claim 1, characterized in that: The method further comprises: Selecting a group representative node from the edge node group; Count the edge nodes with the same target quantitative configuration model and group them to obtain the group representative node combination; Splitting a functional module into a plurality of sub-modules, wherein the functional module includes the first functional module and the second functional module; Counting the number of management nodes corresponding to the group representative nodes; The submodules are sent to the group representative nodes according to the number of management nodes, wherein the number of submodules sent is negatively correlated with the number of management nodes, and the group representative nodes are used to obtain the functional modules by sharing the submodule combination.

3. The cloud platform-based information management method according to claim 2, characterized in that: The method further comprises: In response to monitoring that an abnormal condition occurs in the transmission network of the first group representative node, searching for a matching candidate group representative node in the edge node group corresponding to the first group representative node according to the configuration information of the first group representative node; Calculating the transmission distance between the candidate group representative node and the data center; Taking the candidate group representative node corresponding to the minimum value of the transmission distance as the second group representative node; Stop sending the submodule to the first group representative node, and send the submodule to the second group representative node.

4. The cloud platform-based information management method according to claim 2, characterized in that: The method further comprises: In response to receiving a node abnormality signal, counting the target group representative nodes that send the node abnormality signal, wherein the node abnormality signal is sent when a model obtained by combining the target group representative nodes is inconsistent with the target quantization configuration model; When the number of abnormal signals of the node is greater than a preset number threshold, determining an abnormal group representative node according to the target group representative node; Removing the abnormal group representative node from the abnormal edge node group in which it is located; A warning signal is sent to the abnormal group representative node.

5. The cloud platform-based information management method according to claim 4, characterized in that: The method further comprises: Obtaining access records of representative nodes of the abnormal group; Extracting abnormal visitors and abnormal behavior records of the abnormal visitors from the access records; Sending the abnormal visitor and the abnormal behavior record to the edge node; Receiving the search result returned by the edge node; A candidate abnormal edge node is obtained according to the search result.

6. An information management system based on a cloud platform, characterized in that: For executing the cloud platform-based information management method according to any one of claims 1 to 5, the system comprises: An acquisition module is used to obtain configuration requests, configuration information, preset duration, data capacity, capacity threshold, group representative nodes, number of management nodes, transmission distance, node abnormality signals, preset number threshold, warning signals, access records and retrieval results; A memory, used to store a program of the cloud platform-based information management method; The program in the memory can be loaded and executed by the processor to implement the cloud platform-based information management method.

7. An intelligent terminal, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 5.

Images