WO2023103592A1

WO2023103592A1 - Device risk prediction method, electronic device and computer-readable storage medium

Info

Publication number: WO2023103592A1
Application number: PCT/CN2022/125757
Authority: WO
Inventors: 李君�
Original assignee: 中兴通讯股份有限公司
Priority date: 2021-12-06
Filing date: 2022-10-17
Publication date: 2023-06-15
Also published as: CN116307347A

Abstract

The embodiments of the present application relate to the field of information processing, in particular to a device risk prediction method, an electronic device and a computer-readable storage medium. The device risk prediction method comprises: respectively acquiring feature data of a plurality of devices; acquiring, according to the feature data, a gradient of a statistical algorithm model for predicting a device risk; encrypting the gradient of the statistical algorithm model and uploading same to a public server; receiving an updated gradient, which is issued by the public server, wherein the updated gradient is obtained by means of the public server performing aggregation according to gradients which are uploaded by N computing sub-nodes; updating the statistical algorithm model according to the updated gradient; and when the updated statistical algorithm model converges, outputting a result of device risk prediction.

Description

Device risk prediction method, electronic device and computer readable storage medium

related application

This application claims the priority of the Chinese patent application with application number 202111479950.3 filed on December 6, 2021.

technical field

The embodiments of the present application relate to the field of information processing, and in particular to a device risk prediction method, an electronic device, and a computer-readable storage medium.

Background technique

With the commercialization of the fifth generation mobile communication technology (5th Generation Mobile Communication Technology, 5G) gradually advancing, communication equipment overheating occurs more frequently. If the device has reported an overtemperature alarm and then handles it, it will seriously affect the normal operation of the device, for example, causing service interruption.

In order to prevent service interruption from affecting the service execution process, the device temperature can be detected for prevention. At present, the temperature data of a large number of devices is collected to detect the overheating of the device environment, but the algorithm used for detection requires a large amount of data. For a single sub-computing node (such as an office point) as the execution subject, The amount of data at a single site is small and single, which cannot meet the operating conditions required by the algorithm. Therefore, there is a problem of insufficient accuracy in the process of use.

Contents of the invention

The main purpose of the embodiments of the present application is to provide a device risk prediction method, an electronic device, and a computer-readable storage medium, so as to improve the accuracy of device risk prediction.

In order to achieve the above purpose, an embodiment of the present application provides a device risk prediction method, which is applied to a sub-computing node, including: respectively acquiring characteristic data of multiple devices; obtaining a statistical algorithm for predicting device risk according to the characteristic data Gradient of the model; encrypt and upload the gradient of the statistical algorithm model to the public server; receive the updated gradient issued by the public server, and the updated gradient is uploaded by the public server according to N sub-computing nodes The gradient is obtained by aggregation; the statistical algorithm model is updated according to the updated gradient; and the result of the equipment risk prediction is output when the updated statistical algorithm model converges.

In order to achieve the above purpose, the embodiment of the present application also provides a device risk prediction method, which is applied to the public server, including: receiving the gradient of the statistical algorithm model uploaded by the N sub-computing nodes; The attributes of the equipment all meet the preset public conditions, and the N is a positive integer greater than 1; the gradients of the statistical algorithm models encrypted and uploaded by the N sub-computing nodes are aggregated to obtain an updated gradient; The computing node sends the updated gradient.

To achieve the above purpose, an embodiment of the present application further provides an electronic device, including: at least one processor; and a memory connected to the at least one processor in communication; wherein, the memory stores information that can be used by the at least one processor Instructions executed by a processor, the instructions are executed by the at least one processor, so that the at least one processor can execute the above-mentioned equipment risk prediction method applied to sub-computing nodes, or can execute the above-mentioned method applied to public A device risk prediction approach for managers.

In order to achieve the above purpose, the embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the above-mentioned device risk prediction method applied to a sub-computing node is implemented, or The above-mentioned equipment risk prediction method applied to the public manager.

In the embodiment of the present application, it is predicted whether the equipment will be overheated, so as to prevent the equipment from being overheated due to the installation environment or other reasons, thereby affecting business execution and equipment life. In the case that the sub-computing nodes have few parameters for calculation, the statistical algorithm model is used to predict the overheating situation, and the encrypted data transmission is carried out with the public server, and the optimization data issued by the public server is updated and adjusted to increase participation. The amount of calculated data makes the prediction result of the current sub-computing node more accurate and improves the prediction efficiency of the sub-computing node.

Description of drawings

FIG. 1 is a flow chart of a method for predicting equipment risk according to an embodiment of the present application;

Fig. 2 is a schematic diagram of an equipment risk prediction method provided according to an embodiment of the present application;

Fig. 3 is a flow chart of a method for predicting equipment risk according to another embodiment of the present application;

Fig. 4 is a schematic diagram of an equipment risk prediction method provided according to another embodiment of the present application;

Fig. 5 is a schematic structural diagram of an electronic device provided according to an embodiment of the present application.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below with reference to the accompanying drawings. However, those of ordinary skill in the art can understand that in each embodiment of the application, many technical details are provided for readers to better understand the application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in this application can also be realized. The division of the following embodiments is for the convenience of description, and should not constitute any limitation to the specific implementation of the present application, and the embodiments can be combined and referred to each other on the premise of no contradiction.

With the gradual advancement of 5G commercial use, the overheating of communication equipment is also gradually increasing, and the processing process for different degrees of overheating of equipment is as follows: Minor—do power derating processing, after derating, the alarm disappears and try to power up, and the power recovers when the alarm disappears To the original value, if the alarm is still there, continue to derate processing. If the derating reaches the threshold (generally 3dB) and there is an alarm, report the alarm truthfully; if it is serious, turn off the power amplifier directly, and try to turn off some power amplifiers first. If the alarm disappears, try to turn on the power amplifier that is turned off. If the alarm is still there, continue to turn off the power amplifier. If there is still an alarm after the power amplifiers of all channels are turned off, report the alarm truthfully.

If the device has reported an overtemperature alarm and then handles it, it will seriously affect the normal operation of the device, for example, causing service interruption. There may be many reasons why the equipment is prone to overheating, such as model problems, high power due to heavy traffic, and closed installation environment. If the installation environment of the equipment can be detected, it can effectively predict the overheating of the equipment caused by poor heat dissipation in the installation environment, and reduce the impact and interruption of the business by measures such as equipment damage and power amplifier derating after an alarm occurs.

The current detection of device temperature includes monitoring and estimating the current temperature of the device, and there are few methods for predicting the device temperature. At present, there is an algorithm for detecting the overheating of the equipment environment by collecting temperature data of a large number of equipment, but the algorithm is based on the statistics of a large amount of data, and the data volume of a single site as the execution subject is small and single, which is accurate and applicable The problem of lack of sex and optimization efficiency.

An embodiment of the present application relates to a device risk prediction method, which is applied to sub-computing nodes. The specific process of the equipment risk prediction method in this embodiment can be shown in Figure 1, including:

Step 101, acquiring feature data of multiple devices respectively;

Step 102, obtaining the gradient of the statistical algorithm model for predicting equipment risk according to the feature data;

Step 103, uploading the gradient encryption of the statistical algorithm model to the public server;

Step 104, receiving the updated gradient issued by the public server, the updated gradient is obtained by the public server through aggregation according to the gradients uploaded by N sub-computing nodes;

Step 105, updating the statistical algorithm model according to the updated gradient;

Step 106, when the updated statistical algorithm model converges, output the result of equipment risk prediction.

The implementation details of the equipment risk prediction method of this embodiment are described in detail below, and the following content is only implementation details provided for easy understanding, and is not necessary for implementing this solution.

In step 101, feature data of multiple devices are acquired respectively. Wherein, the sub-computing node may be an office point, and the office point obtains characteristic data of devices within the current office point collected by the base station.

In one example, the attributes of the multiple devices meet the preset conditions; the preset conditions include: the models of the multiple devices are the same, the difference between the latitude coordinates of the multiple devices is within the first preset range, and the respective The temperature difference in the region is within the second preset range. That is, in a single sub-computing node or site, the attributes of the devices are similar, so that the sub-computing node can perform risk prediction on the device with the current attributes according to the acquired feature data.

In one example, the device includes a remote radio unit (Remote Radio Unit, RRU); in the case of the device being an RRU, the feature data of multiple devices includes each RRU: power, optical module temperature, baseband temperature, intermediate frequency temperature, Power supply temperature, board temperature, transceiver temperature, location, air temperature. In a specific implementation, the power is such as the average power of the power supply, and the temperature of the optical module is such as the average temperature of the optical/electrical module transceiver. Board temperature, such as the average temperature of the RRU board, transceiver temperature, such as the average RRU transmit (transport, TX)/receive (receive, RX) temperature or power amplifier temperature, location, such as latitude and longitude, and air temperature, such as the local air temperature where the device is located.

In addition, the device can be an electronic device whose position does not change in a large range, and is not limited to RRU, for example, it can also be an indoor baseband processing unit (Building Base band Unit, BBU). predict.

In step 102, the gradient of the statistical algorithm model used to predict equipment risk is obtained according to the feature data; that is, the acquired feature data is input into the statistical algorithm model for calculation and prediction, and the gradient in the calculation process is obtained, the gradient The mean value and deviation range in the set of characteristic data can be characterized, and the mean value and deviation range can be used to measure the operation results of the current statistical algorithm model. For example, if the ratio of the part of the current characteristic data within the deviation range of the mean value to the total number of the current characteristic data is less than the threshold value of the preset ratio, the calculation result of the current statistical algorithm model is not optimal and needs to be adjusted. That is, the gradient needs to be adjusted, that is, the model has not converged.

In an example, before obtaining the gradient of the statistical algorithm model for predicting equipment risk according to the feature data, it includes: performing data cleaning on the feature data of multiple devices. That is, when collecting characteristic data, it can be obtained through active detection or reporting of receiving equipment, and the format of characteristic data is not screened during the collection process, which can ensure the number of samples obtained.

In one example, data cleaning includes one of the following or any combination thereof: unifying the data format of feature data; deleting abnormal data in feature data; integrating feature data according to time granularity. Specifically, the data format of the characteristic data is unified, that is, since there are no restrictions on the data during the collection process, and there is no restriction on the way to collect the characteristic data, the received characteristic data may have inconsistent data formats. To facilitate calculation, firstly, the format of the acquired feature data is unified. Delete the abnormal data in the characteristic data. For computing nodes, there may be cases where the data format that can be processed is limited. If the obtained characteristic data is in a format that cannot be processed by the current computing node, it will be judged as abnormal data. You can Choose to delete it to avoid format problems affecting the calculation process. You can also choose to unify the format first, and delete it when unification cannot be achieved to ensure the amount of data in the calculation process; The preset range interval corresponding to the range interval and the minimum value can also be defined as abnormal data, which is deleted before processing. Integrate feature data according to time granularity. Since the reporting time granularity of various types of data is different, it is necessary to integrate all acquired feature data into one data stream, for example: add location data to each piece of collected temperature data, that is, Supplement the location data with coarser granularity of collection time to the temperature data with finer granularity of collection time. In a specific execution process, the granularity of feature data a is one piece of data every 15 minutes, and the granularity of feature data b is one piece of data every hour. After data integration, one data stream every 15 minutes contains feature data a and feature data b. , containing a is collected this time, and b is collected last time within one hour.

In step 103, the encrypted gradient of the statistical algorithm model is uploaded to the public server. That is to say, there is a preset encryption process in the sub-computing nodes. After the gradient obtained after the statistical algorithm model is executed, the obtained gradient is encrypted, and the encrypted gradient is uploaded to the public server.

In a specific implementation, after the gradient is obtained, the mean value and offset range are obtained according to the gradient, and the data of the characteristic parameters acquired by the current sub-computing node in the offset range of the mean value are judged to account for the characteristic parameters obtained by the current sub-computing node. The ratio of the total number of parameters, if the ratio is greater than the threshold of the preset ratio, it means that the current calculation result is optimal and meets the requirements, and the calculation result is output; if the ratio is not greater than the threshold of the preset ratio, it means the current calculation result It is not optimal and requires gradient adjustment, so perform the steps of uploading gradient encryption to the public server.

Among them, the process of uploading the gradient to the public server can refer to the federated learning algorithm, that is, refer to the federated learning algorithm to encrypt and transmit the gradient data between the sub-computing nodes and the public server to ensure the security of the gradient data. It can be understood that there is a decryption algorithm corresponding to the encryption algorithm in the current child node in the public server.

In step 104, the updated gradient issued by the public server is received, and the updated gradient is obtained by the public server through aggregation according to the gradients uploaded by the N sub-computing nodes. That is, the received updated gradient issued by the public server is obtained by integrating the gradients uploaded by the sub-computing nodes within the public server's own scope. Since there are multiple sub-computing nodes within the scope of the public server, and each sub-computing node also corresponds to multiple devices, for a public server, the received gradient is a value obtained through a large amount of sample data. Due to the abundant sample size, the aggregated value of the received gradient is closer to the optimal choice of the gradient of the statistical algorithm model of the current child node, and the original gradient value is updated with the aggregated gradient.

In addition, when the public server sends the updated gradient to the sub-computing node, it can also be encrypted, and the sub-computing node decrypts it after the sub-computing node receives it, further improving the security of data transmission.

In step 105, the statistical algorithm model is updated according to the updated gradient. That is, according to the updated gradient sent by the public server, the statistical algorithm model of the current computing node is updated, so that the output of the statistical algorithm model of the current sub-computing node is closer to the actual situation, wherein the statistical algorithm model such as statistical clustering The algorithm model constructs relatively close groups into a large group by continuously merging similar samples. For example, the gradient uploaded by each sub-computing node received in the public server is to obtain the mean value and offset range uploaded by each sub-computing node, aggregate each mean value and offset range, and obtain the updated mean value and offset range; update After the gradient is sent to the current sub-computing node, the current sub-computing node also gets the updated mean value and offset range from the public server; the statistical algorithm model of the current computing node is updated using the gradient delivered by the public server.

Among them, the gradient issued by the public server is used to update the statistical algorithm model of the current computing node, for example: according to the gradient issued by the public server, the parameters in the statistical algorithm model are updated, including updating various judgment criteria, such as the mean value and offset range , where the mean value and offset range include the relative mean value and offset range of temperature, latitude and longitude, power, etc. After the parameters are updated, it can be judged whether the current statistical algorithm model is converged. In addition, the criteria for judging abnormal data can also be updated.

In step 106, when the updated statistical algorithm model converges, the result of equipment risk prediction is output. After the statistical algorithm model of the current sub-computing node is updated according to the gradient issued by the public server, the calculation is performed again according to the characteristic data, and the gradient during this calculation process is obtained, and the current calculation is judged according to the mean value and offset range corresponding to the gradient. Whether the statistical algorithm model converges, and if it converges, the calculation result of the statistical algorithm model is the final equipment risk prediction result.

In an example, after updating the statistical algorithm model according to the updated gradient, it also includes: in the case that the updated statistical algorithm model does not converge, encrypting and uploading the updated statistical algorithm model according to the gradient obtained from the feature data To the public server; receive the updated gradient issued by the public server again, and update the statistical algorithm model again according to the updated gradient issued by the public server again until the updated statistical algorithm model converges, as shown in Figure 2. That is, if it is judged according to the average value and offset range obtained by the gradient in this calculation process, and the proportion of the acquired feature data within the offset range of the average value is still not greater than the threshold value of the preset ratio, then the calculation process in this calculation process The statistical algorithm model still does not converge, and it is necessary to continue to exchange information with the public server, recalculate the gradient according to the updated mean and offset range and other criteria, and upload it to the public server; until the updated statistical algorithm model converges, the output converges The results of the final equipment risk prediction.

Among them, the equipment risk prediction results of the statistical algorithm model include: normal, general overheating, high risk overheating, and severe overheating. Normal means that the device installation environment corresponding to the characteristic data obtained by the current sub-computing node is normal; general overheating means that there is an overheating risk. It is recommended to check the heat dissipation of the installation environment to determine the cause of the adverse effects of heat dissipation, and Renovate the installation environment when necessary to eliminate the impact of poor heat dissipation on equipment performance and hardware life; high risk of overheating indicates a high risk of overheating in the installation environment. It is recommended to renovate the environment when necessary to eliminate the impact of poor heat dissipation The impact of equipment performance, hardware life, etc.; serious overheating, indicating that the current installation environment of the equipment is very prone to overheating, and it is recommended to rectify it. The installation environment of the aforementioned equipment, for example, the installation environment of the RRU.

In order to facilitate the understanding of the foregoing implementation manner, a part of the implementation process is described below in combination with an execution scenario. It can be understood that it is only for illustration, and does not limit the relevant content of the embodiment.

In one example, a site is located in a high temperature zone, and RRU overheating alarms often occur, which is suspected to be caused by poor heat dissipation in the installation environment. It is necessary to check the installation environment of the RRUs, find out which RRUs are prone to overheating due to poor heat dissipation in the installation environment, and then rectify the installation environment. However, there are many types of RRUs in this site and the number of each type of RRU is very small, and the result obtained for each type of RRU is not accurate enough. Therefore, the following process is used for detection:

Firstly, select the equipment in the current site whose latitude coordinate difference is within the first preset range, the temperature difference of the region is within the second preset range, and the RRU model is the same, and obtain the characteristic data in the corresponding equipment; Secondly, the gradient calculated by the site according to the statistical algorithm model is encrypted and uploaded to the public server. The public server aggregates the data of multiple sites to calculate a new gradient and distributes it to each site. Among them, multiple sites under the public server The devices corresponding to the points have the same model and similar properties. Update the model according to the new gradient issued by the public server, and repeat until the model converges, and the output result when the model converges is the final prediction result.

Through the above process, whether the RRU is in an installation environment prone to overheating and the degree of overheating in the installation environment can be further checked whether the RRU that is seriously prone to overheating is installed with a beautification cover and the beautification cover has no cooling holes and is installed in a small, airtight enclosure. Space, etc., and rectify the problematic installation environment to avoid RRU operation interruption or equipment damage due to installation environment problems.

In another example, the business volume of a certain site is small, but the business needs to be expanded in summer. In order to avoid the overheating problem caused by the poor heat dissipation of the installation environment when the RRU power increases, it is hoped to predict the overheating degree of the RRU installation environment. Rectify in advance or determine which RRUs to expand. However, the daily operating power of this site is small, and the amount of data used for prediction is insufficient. Therefore, we choose to use the data of other sites with similar attributes to this site but with high operating power to perform model training together through the public server. Predict the RRUs that are prone to overheating at this site and make rectifications. For example, the current service volume of RRU1 in the site is small. By integrating the data of RRU1 at different operating powers in other sites, the temperature of RRU1 at different operating powers can be predicted to determine whether it is prone to overheating when the power increases. This method needs to collect historical data for a long time.

First, select a site with a larger capacity (for example, more cells) and a large number of RRUs of the same model as the RRU1, which corresponds to the same public server as the site where the current RRU1 is located to form a federated learning system. Create acquisition and calculation tasks at each site under the public server, and upload the encrypted gradient to the public server. The public server obtains a new gradient by integrating the data of each site. Send the new gradient to each site. Each site continuously updates the statistical clustering algorithm model according to the new gradient and uploads the updated new gradient until the model converges. When the model converges, the predicted result of the model is the final predicted result.

It can predict which RRUs in the current site are prone to overheating when the operating power is high, or the installation environment needs to be rectified. For example, the results show that RRU1 is prone to overheating. If the power continues to rise, an overtemperature alarm will be reported. You can choose to check and rectify the installation environment of RRU1 first, and expand business after rectification. Or the normal RRU increases the service volume in other installation environments.

In this embodiment, it is predicted whether the equipment will be overheated, so as to prevent the equipment from being overheated due to the installation environment or other reasons, thereby affecting service execution and equipment life. In the case that the sub-computing nodes have few parameters for calculation, the statistical algorithm model is used to predict the overheating situation, and the encrypted data transmission is carried out with the public server, and the optimization data issued by the public server is updated and adjusted to increase participation. The amount of calculated data makes the prediction result of the current sub-computing node more accurate and improves the prediction efficiency of the sub-computing node.

Another embodiment of the present application relates to a device risk prediction method, as shown in Figure 3, which is applied to a public server. The implementation details of the device risk prediction method in this embodiment will be described in detail below, and the following content is provided for convenience of understanding The implementation details are not necessary to implement this solution.

Step 201 , receiving gradients of statistical algorithm models uploaded by N sub-computing nodes; multiple devices acquired by N sub-computing nodes all meet preset common conditions, and N is a positive integer greater than 1. That is, the public server corresponds to N sub-computing nodes, and the attributes of all devices acquired by the N sub-computing nodes all meet preset public conditions.

In one example, the preset common condition is that the models of the multiple devices are the same, the difference between the latitude coordinates of the multiple devices is within the third preset range, and the temperature difference between the regions where the multiple devices are located is within the fourth preset range. within the set range. Wherein, the third preset range is not smaller than the first preset range, and the fourth preset range is not smaller than the second preset range. That is to say, the properties of multiple devices acquired in the same public server are similar, and the number of parameters in the sub-computing nodes can be expanded. The attribute range of multiple devices obtained by the sub-computing node in the public server can be consistent with the attribute range of the corresponding device in a sub-computing node, or can be appropriately expanded to expand the amount of data used by the sub-computing node for calculation .

Specifically, the number of devices required by the public server has a lower limit, and the number of devices acquired by the sub-computing nodes under the public server needs to be greater than the lower limit. For example, refer to the NA+NB+NC...NP>M formula, where M is the lower limit value, and NA to NP respectively represent the number of devices whose attributes selected by each sub-computing node meet the preset conditions.

In order to ensure that the number of devices whose attributes meet the preset public conditions is sufficient, sub-computing nodes with similar attributes of corresponding devices are selected as sub-computing nodes of the same system. Among them, taking the sub-computing node as the site and the equipment as the RRU as an example, that is, the attributes of the RRU in the site A are similar to the attributes of the RRU in the site B, and although the number of RRUs in the site C is large, the If the RRU attributes of RRUs are different, then A and B are more suitable as a computing system, corresponding to the same public server. It is necessary to find other sites that have similar RRU attributes to site C, and site C corresponds to the same public server. The relationship between Site A and Site B and the public server is shown in Figure 4. In addition, the sub-computing nodes corresponding to the public servers corresponding to Sites A and B are not limited to only Sites A and B.

In addition, the public server may be a northbound server.

In step 202, the gradients of the statistical algorithm models encrypted and uploaded by the N sub-computing nodes are aggregated to obtain updated gradients. That is, the N sub-computing nodes respectively collect the characteristic parameters of their corresponding devices, perform data cleaning, calculation, etc., and upload the calculated gradients to the public server. The public server aggregates the gradients uploaded by the sub-computing nodes obtained by the current public server, that is, performs gradient calculations based on all devices obtained by the sub-computing nodes under the current public server to increase accuracy.

Step 203 , sending the updated gradients to the N sub-computing nodes, and sending the updated gradients to the sub-computing nodes, so that the sub-computing nodes can update the statistical algorithm model and obtain prediction results. The prediction result generated when the statistical algorithm model converges is the final prediction result.

In this embodiment, a public server is provided to participate in the prediction process of the sub-computing nodes, obtain the gradients of each sub-computing node, and distribute them after integration to update the statistical algorithm model of the sub-computing nodes and expand the participation of each sub-computing node in the calculation The amount of data can improve the prediction accuracy of each sub-computing node.

Another embodiment of the present application relates to an electronic device, as shown in FIG. 5 , including: at least one processor 301; and a memory 302 communicatively connected to the at least one processor 301; wherein, the memory 302 stores Instructions that can be executed by the at least one processor 301, the instructions are executed by the at least one processor 301, so that the at least one processor 301 can execute the devices applied to sub-computing nodes in the above embodiments A risk prediction method, or a device risk prediction method applied to a public server in the various examples above.

Wherein, the memory and the processor are connected by a bus, and the bus may include any number of interconnected buses and bridges, and the bus connects one or more processors and various circuits of the memory together. The bus may also connect together various other circuits such as peripherals, voltage regulators, and power management circuits, all of which are well known in the art and therefore will not be further described herein. The bus interface provides an interface between the bus and the transceivers. A transceiver may be a single element or multiple elements, such as multiple receivers and transmitters, providing means for communicating with various other devices over a transmission medium. The data processed by the processor is transmitted on the wireless medium through the antenna, further, the antenna also receives the data and transmits the data to the processor.

The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. Instead, memory can be used to store data that the processor uses when performing operations.

Another embodiment of the present application relates to a computer-readable storage medium storing a computer program. The above method embodiments are implemented when the computer program is executed by the processor.

That is, those skilled in the art can understand that all or part of the steps in the method of the above-mentioned embodiments can be completed by instructing related hardware through a program, the program is stored in a storage medium, and includes several instructions to make a device ( It may be a single-chip microcomputer, a chip, etc.) or a processor (processor) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. .

Those of ordinary skill in the art can understand that the above-mentioned implementation modes are specific examples for realizing the present application, and in practical applications, various changes can be made to it in form and details without departing from the spirit and spirit of the present application. scope.

Claims

A device risk prediction method applied to sub-computing nodes, including:

Obtain feature data of multiple devices separately;

Obtaining the gradient of a statistical algorithm model for predicting equipment risk according to the characteristic data;

Upload the gradient encryption of the statistical algorithm model to the public server;

receiving the updated gradient issued by the public server, where the updated gradient is obtained by aggregation of the public server according to the gradients uploaded by N sub-computing nodes;

updating the statistical algorithm model according to the updated gradient;

When the updated statistical algorithm model converges, output the result of the equipment risk prediction.
The equipment risk prediction method according to claim 1, wherein, after updating the statistical algorithm model according to the updated gradient, further comprising:

In the case that the updated statistical algorithm model does not converge, uploading the gradient encryption obtained by the updated statistical algorithm model according to the characteristic data to the public server;

receiving the updated gradient sent again by the public server, and updating the statistical algorithm model again according to the updated gradient sent again by the public server until the updated statistical algorithm model converges.
The equipment risk prediction method according to claim 1, wherein, before obtaining the gradient of the statistical algorithm model for predicting equipment risk according to the characteristic data, comprising:

Data cleaning is performed on the feature data of the plurality of devices.
The equipment risk prediction method according to claim 3, wherein the data cleaning includes one of the following or any combination thereof:

Unify the data format of the feature data;

deleting abnormal data in the characteristic data;

The feature data is integrated according to time granularity.
The device risk prediction method according to claim 1, wherein the attributes of the multiple devices meet preset conditions; the preset conditions include: the models of the multiple devices are the same, and the latitude coordinates of the multiple devices The temperature difference is within the first preset range, and the temperature difference of the areas where the plurality of devices are located is within the second preset range.
The equipment risk prediction method according to any one of claims 1 to 5, wherein the equipment includes a remote radio unit (RRU);

In the case where the device is an RRU, the feature data of the plurality of devices includes each RRU: power, optical module temperature, baseband temperature, intermediate frequency temperature, power supply temperature, single board temperature, transceiver temperature, position, air temperature .
A device risk prediction method applied to public servers, including:

Receive the gradient of the statistical algorithm model uploaded by N sub-computing nodes; the attributes of multiple devices acquired by the N sub-computing nodes all meet the preset public conditions, and the N is a positive integer greater than 1;

Aggregating gradients of statistical algorithm models encrypted and uploaded by the N sub-computing nodes to obtain updated gradients;

Sending the updated gradients to the N sub-computing nodes.
The equipment risk prediction method according to claim 7, wherein the preset public conditions include:

The models of the multiple devices are the same, the difference between the latitude coordinates of the multiple devices is within a third preset range, and the temperature difference between the regions where the multiple devices are located is within a fourth preset range.
An electronic device comprising:

at least one processor; and,

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, the instructions are executed by the at least one processor, so that the at least one processor can perform the operation described in any one of claims 1 to 6 The equipment risk prediction method described above or the equipment risk prediction method described in any one of claims 7 to 8 can be implemented.
A computer-readable storage medium storing a computer program, wherein, when the computer program is executed by a processor, the device risk prediction method according to any one of claims 1 to 6 is implemented or the device risk prediction method according to any one of claims 7 to 8 is implemented. The equipment risk prediction method described in any one.